Skip to contentSkip to conversation

qemu-img convert: Fix abort with unaligned image size

PATCH In review Kevin Wolf 2 participants 7 messages Jul 16, 2020 – Jul 17, 2020 Block layer

lore
Share & tools
lore

Kevin Wolf

author
#0 lore raw

v2 cover qemu-img convert: Fix abort with unaligned image size

Patch 1 fixes the assertion failure by failing gracefully when opening an image whose size isn't aligned to the required request alignment.

Patch 2 relaxes the restrictions for NFS, which actually supports byte alignment, but incorrectly gets a 4k request alignment in the file-posix block driver.

v2:
- Don't fail opening unaligned images, but requesting WRITE permission without RESIZE. This keeps qcow2 images with unaligned metadata at EOF working. [Max]

Kevin Wolf (2):
block: Require aligned image size to avoid assertion failure
file-posix: Allow byte-aligned O_DIRECT with NFS

 block.c            | 16 ++++++++++++++++
 block/file-posix.c | 26 +++++++++++++++++++++++++-
 2 files changed, 41 insertions(+), 1 deletion(-)
Signature · 1 line
2.25.4
author
#1 lore raw

PATCH v2 1/2 block: Require aligned image size to avoid assertion failure

Unaligned requests will automatically be aligned to bl.request_alignment and we can't extend write requests to access space beyond the end of the image without resizing the image, so if we have the WRITE permission, but not the RESIZE one, it's required that the image size is aligned.

Failing to meet this requirement could cause assertion failures like this if RESIZE permissions weren't requested:

qemu-img: block/io.c:1910: bdrv_co_write_req_prepare: Assertion `end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE' failed.

This was e.g. triggered by qemu-img converting to a target image with 4k request alignment when the image was only aligned to 512 bytes, but not to 4k.

Turn this into a graceful error in bdrv_check_perm() so that WRITE without RESIZE can only be taken if the image size is aligned. If a user holds both permissions and drops only RESIZE, the function will return an error, but bdrv_child_try_set_perm() will ignore the failure silently if permissions are only requested to be relaxed and just keep both permissions while returning success.

---
 block.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
block.c +16 -0
index 35a372df57..6371928edb 100644
@@ -2025,6 +2025,22 @@ static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
         return -EPERM;
     }
 
+    /*
+     * Unaligned requests will automatically be aligned to bl.request_alignment
+     * and without RESIZE we can't extend requests to write to space beyond the
+     * end of the image, so it's required that the image size is aligned.
+     */
+    if ((cumulative_perms & BLK_PERM_WRITE) &&
+        !(cumulative_perms & BLK_PERM_RESIZE))
+    {
+        if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
+            error_setg(errp, "Cannot get 'write' permission without 'resize': "
+                             "Image size is not a multiple of request "
+                             "alignment");
+            return -EPERM;
+        }
+    }
+
     /* Check this node */
     if (!drv) {
         return 0;
Signature · 1 line
2.25.4
Reviewed-by on v2 1/2
#2 lore raw

Re: [PATCH for-5.1 v2 1/2] block: Require aligned image size to avoid assertion failure

Quoted 31 lines · Kevin Wolf

Unaligned requests will automatically be aligned to bl.request_alignment and we can't extend write requests to access space beyond the end of the image without resizing the image, so if we have the WRITE permission, but not the RESIZE one, it's required that the image size is aligned.

Failing to meet this requirement could cause assertion failures like this if RESIZE permissions weren't requested:

qemu-img: block/io.c:1910: bdrv_co_write_req_prepare: Assertion `end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE' failed.

This was e.g. triggered by qemu-img converting to a target image with 4k request alignment when the image was only aligned to 512 bytes, but not to 4k.

Turn this into a graceful error in bdrv_check_perm() so that WRITE without RESIZE can only be taken if the image size is aligned. If a user holds both permissions and drops only RESIZE, the function will return an error, but bdrv_child_try_set_perm() will ignore the failure silently if permissions are only requested to be relaxed and just keep both permissions while returning success.

---
 block.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
block.c +0 -0
index 35a372df57..6371928edb 100644

block.c static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q, · line 2033

Quoted 9 lines
block.c +6 -0
@@ -2025,6 +2025,22 @@ static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
         return -EPERM;
     }
 
+    /*
+     * Unaligned requests will automatically be aligned to bl.request_alignment
+     * and without RESIZE we can't extend requests to write to space beyond the
+     * end of the image, so it's required that the image size is aligned.
+     */
+    if ((cumulative_perms & BLK_PERM_WRITE) &&

What about WRITE_UNCHANGED? I think this would only matter with nodes that can have backing files (i.e., qcow2 in practice) because WRITE_UNCHANGED is only used by COR and block jobs doing something with a backing chain, so it shouldn’t matter in practice, but, well.

So, either way:

Quoted 14 lines
block.c +10 -0
+        !(cumulative_perms & BLK_PERM_RESIZE))
+    {
+        if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
+            error_setg(errp, "Cannot get 'write' permission without 'resize': "
+                             "Image size is not a multiple of request "
+                             "alignment");
+            return -EPERM;
+        }
+    }
+
     /* Check this node */
     if (!drv) {
         return 0;
author
#3 lore raw

Re: [PATCH for-5.1 v2 1/2] block: Require aligned image size to avoid assertion failure

Am 17.07.2020 um 13:02 hat Max Reitz geschrieben:

Quoted 35 lines · Kevin Wolf
Quoted 34 lines

Unaligned requests will automatically be aligned to bl.request_alignment and we can't extend write requests to access space beyond the end of the image without resizing the image, so if we have the WRITE permission, but not the RESIZE one, it's required that the image size is aligned.

Failing to meet this requirement could cause assertion failures like this if RESIZE permissions weren't requested:

qemu-img: block/io.c:1910: bdrv_co_write_req_prepare: Assertion `end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE' failed.

This was e.g. triggered by qemu-img converting to a target image with 4k request alignment when the image was only aligned to 512 bytes, but not to 4k.

Turn this into a graceful error in bdrv_check_perm() so that WRITE without RESIZE can only be taken if the image size is aligned. If a user holds both permissions and drops only RESIZE, the function will return an error, but bdrv_child_try_set_perm() will ignore the failure silently if permissions are only requested to be relaxed and just keep both permissions while returning success.

---
 block.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
block.c +0 -0
index 35a372df57..6371928edb 100644
@@ -2025,6 +2025,22 @@ static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
         return -EPERM;
     }
 
Quoted 11 lines

block.c static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q, · line 2033

Quoted 6 lines
block.c +6 -0
+    /*
+     * Unaligned requests will automatically be aligned to bl.request_alignment
+     * and without RESIZE we can't extend requests to write to space beyond the
+     * end of the image, so it's required that the image size is aligned.
+     */
+    if ((cumulative_perms & BLK_PERM_WRITE) &&

What about WRITE_UNCHANGED? I think this would only matter with nodes that can have backing files (i.e., qcow2 in practice) because WRITE_UNCHANGED is only used by COR and block jobs doing something with a backing chain, so it shouldn’t matter in practice, but, well.

So basically just replacing the line with this?

    if ((cumulative_perms & (BLK_PERM_WRITE | BDRV_PERM_WRITE_UNCHANGED)) &&

I can do that while applying if it is what you mean.

Quoted 3 lines

So, either way:

Thanks!

Kevin

Quoted 16 lines
Quoted 14 lines
block.c +10 -0
+        !(cumulative_perms & BLK_PERM_RESIZE))
+    {
+        if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
+            error_setg(errp, "Cannot get 'write' permission without 'resize': "
+                             "Image size is not a multiple of request "
+                             "alignment");
+            return -EPERM;
+        }
+    }
+
     /* Check this node */
     if (!drv) {
         return 0;
#4 lore raw

Re: [PATCH for-5.1 v2 1/2] block: Require aligned image size to avoid assertion failure

Quoted 48 lines · Kevin Wolf

Am 17.07.2020 um 13:02 hat Max Reitz geschrieben:

Quoted 35 lines · Kevin Wolf
Quoted 34 lines

Unaligned requests will automatically be aligned to bl.request_alignment and we can't extend write requests to access space beyond the end of the image without resizing the image, so if we have the WRITE permission, but not the RESIZE one, it's required that the image size is aligned.

Failing to meet this requirement could cause assertion failures like this if RESIZE permissions weren't requested:

qemu-img: block/io.c:1910: bdrv_co_write_req_prepare: Assertion `end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE' failed.

This was e.g. triggered by qemu-img converting to a target image with 4k request alignment when the image was only aligned to 512 bytes, but not to 4k.

Turn this into a graceful error in bdrv_check_perm() so that WRITE without RESIZE can only be taken if the image size is aligned. If a user holds both permissions and drops only RESIZE, the function will return an error, but bdrv_child_try_set_perm() will ignore the failure silently if permissions are only requested to be relaxed and just keep both permissions while returning success.

---
 block.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
block.c +0 -0
index 35a372df57..6371928edb 100644
@@ -2025,6 +2025,22 @@ static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
         return -EPERM;
     }
 
Quoted 11 lines

block.c static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q, · line 2033

Quoted 6 lines
block.c +6 -0
+    /*
+     * Unaligned requests will automatically be aligned to bl.request_alignment
+     * and without RESIZE we can't extend requests to write to space beyond the
+     * end of the image, so it's required that the image size is aligned.
+     */
+    if ((cumulative_perms & BLK_PERM_WRITE) &&

What about WRITE_UNCHANGED? I think this would only matter with nodes that can have backing files (i.e., qcow2 in practice) because WRITE_UNCHANGED is only used by COR and block jobs doing something with a backing chain, so it shouldn’t matter in practice, but, well.

Quoted 5 lines

So basically just replacing the line with this?

    if ((cumulative_perms & (BLK_PERM_WRITE | BDRV_PERM_WRITE_UNCHANGED)) &&

I can do that while applying if it is what you mean.

Yes. :)

Quoted 27 lines
Quoted 3 lines

So, either way:

Thanks!

Kevin

Quoted 16 lines
Quoted 14 lines
block.c +10 -0
+        !(cumulative_perms & BLK_PERM_RESIZE))
+    {
+        if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
+            error_setg(errp, "Cannot get 'write' permission without 'resize': "
+                             "Image size is not a multiple of request "
+                             "alignment");
+            return -EPERM;
+        }
+    }
+
     /* Check this node */
     if (!drv) {
         return 0;
author
#5 lore raw

PATCH v2 2/2 file-posix: Allow byte-aligned O_DIRECT with NFS

In reply to Kevin Wolf

Since commit a6b257a08e3 ('file-posix: Handle undetectable alignment'), we assume that if we open a file with O_DIRECT and alignment probing returns 1, we just couldn't find out the real alignment requirement because some filesystems make the requirement only for allocated blocks. In this case, a safe default of 4k is used.

This is too strict for NFS, which does actually allow byte-aligned requests even with O_DIRECT. Because we can't distinguish both cases with generic code, let's just look at the file system magic and disable s->needs_alignment for NFS. This way, O_DIRECT can still be used on NFS for images that are not aligned to 4k.

---
 block/file-posix.c | 26 +++++++++++++++++++++++++-
 1 file changed, 25 insertions(+), 1 deletion(-)
block/file-posix.c +25 -1
index 8067e238cb..ae8190edab 100644
@@ -62,10 +62,12 @@
 #include <sys/ioctl.h>
 #include <sys/param.h>
 #include <sys/syscall.h>
+#include <sys/vfs.h>
 #include <linux/cdrom.h>
 #include <linux/fd.h>
 #include <linux/fs.h>
 #include <linux/hdreg.h>
+#include <linux/magic.h>
 #include <scsi/sg.h>
 #ifdef __s390__
 #include <asm/dasd.h>
@@ -300,6 +302,28 @@ static int probe_physical_blocksize(int fd, unsigned int *blk_size)
 #endif
 }
 
+/*
+ * Returns true if no alignment restrictions are necessary even for files
+ * opened with O_DIRECT.
+ *
+ * raw_probe_alignment() probes the required alignment and assume that 1 means
+ * the probing failed, so it falls back to a safe default of 4k. This can be
+ * avoided if we know that byte alignment is okay for the file.
+ */
+static bool dio_byte_aligned(int fd)
+{
+#ifdef __linux__
+    struct statfs buf;
+    int ret;
+
+    ret = fstatfs(fd, &buf);
+    if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) {
+        return true;
+    }
+#endif
+    return false;
+}
+
 /* Check if read is allowed with given memory buffer and length.
  *
  * This function is used to check O_DIRECT memory buffer and request alignment.
@@ -629,7 +653,7 @@ static int raw_open_common(BlockDriverState *bs, QDict *options,
 
     s->has_discard = true;
     s->has_write_zeroes = true;
-    if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
+    if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) {
         s->needs_alignment = true;
     }
 
Signature · 1 line
2.25.4
Reviewed-by on v2 2/2
#6 lore raw

Re: [PATCH for-5.1 v2 2/2] file-posix: Allow byte-aligned O_DIRECT with NFS

Quoted 12 lines · Kevin Wolf

Since commit a6b257a08e3 ('file-posix: Handle undetectable alignment'), we assume that if we open a file with O_DIRECT and alignment probing returns 1, we just couldn't find out the real alignment requirement because some filesystems make the requirement only for allocated blocks. In this case, a safe default of 4k is used.

This is too strict for NFS, which does actually allow byte-aligned requests even with O_DIRECT. Because we can't distinguish both cases with generic code, let's just look at the file system magic and disable s->needs_alignment for NFS. This way, O_DIRECT can still be used on NFS for images that are not aligned to 4k.

Quoted 5 lines
---
 block/file-posix.c | 26 +++++++++++++++++++++++++-
 1 file changed, 25 insertions(+), 1 deletion(-)