Block patches
Hi Anthony,
this bunch of patches contains some fixes that I think qualify for stable, especially the first two. Please cherry-pick them into stable-0.12 when you have pulled:
468f932 block: fix sector comparism in multiwrite_req_compare 49eb7b7 virtio-blk: fix barrier support
5fdf2a6 block: fix aio_flush segfaults for read-only protocols (e.g. curl)
Kevin
The following changes since commit 5a4bb580cdb10b066f9fd67658b31cac4a4ea5e5:
Richard Henderson (1):
target-sparc: Simplify ICC generation.
are available in the git repository at:
git://repo.or.cz/qemu/kevin.git for-anthony
Avi Kivity (1):
block: fix aio_flush segfaults for read-only protocols (e.g. curl)
Christoph Hellwig (2):
virtio-blk: fix barrier support
block: fix sector comparism in multiwrite_req_compare
Kevin Wolf (1):
vvfat: More build fixes with DEBUG
Nicholas Bellinger (2):
block: Make find_image_format() return 'raw' BlockDriver for SG_IO devices
block: Add SG_IO device check in refresh_total_sectors()
Riccardo Magliocchetti (1):
vvfat: Fix compilation with DEBUG defined
Stefan Hajnoczi (1):
virtio-blk: Avoid zeroing every request structure
aio.c | 4 +++- block.c | 23 ++++++++++++++++++++++- block/vvfat.c | 10 +++++++--- hw/virtio-blk.c | 19 ++++++++++++++++--- 4 files changed, 48 insertions(+), 8 deletions(-)
PATCH 1/8 block: fix aio_flush segfaults for read-only protocols (e.g. curl)
From: Avi Kivity <[email protected]>
Not all block format drivers expose an io_flush method (reasonable for read-only protocols), so calling io_flush there will immediately segfault.
Fix by checking for the method's existence before calling it.
--- aio.c | 4 +++- 1 files changed, 3 insertions(+), 1 deletions(-)
aio.c +3 -1
@@ -113,7 +113,9 @@ void qemu_aio_flush(void)qemu_aio_wait (); QLIST_FOREACH(node, &aio_handlers, node) { - ret |= node->io_flush (node->opaque); +if (node->io_flush) { + ret |= node->io_flush (node->opaque); + } } }while (qemu_bh_poll () || ret >0 ); }
Signature · 1 line
1.6.6.1
PATCH 2/8 virtio-blk: Avoid zeroing every request structure
In reply to Kevin WolfFrom: Stefan Hajnoczi <[email protected]>
The VirtIOBlockRequest structure is about 40 KB in size. This patch avoids zeroing every request by only initializing fields that are read. The other fields are either written to or may not be used at all.
Oprofile shows about 10% of CPU samples in memset called by virtio_blk_alloc_request(). The workload is
dd if=/dev/vda of=/dev/null iflag=direct bs=8k running concurrently 4 times. This patch makes memset disappear to the bottom of the profile.
--- hw/virtio-blk.c | 4 +++- 1 files changed, 3 insertions(+), 1 deletions(-)
hw/virtio-blk.c +3 -1
@@ -105,8 +105,10 @@ static void virtio_blk_flush_complete(void *opaque, int ret)static VirtIOBlockReq *virtio_blk_alloc_request (VirtIOBlock *s) { -VirtIOBlockReq *req =qemu_mallocz (sizeof (*req)); +VirtIOBlockReq *req =qemu_malloc (sizeof (*req)); req->dev = s; + req->qiov.size =0 ; + req->next =NULL ;return req; }
Signature · 1 line
1.6.6.1
PATCH 3/8 virtio-blk: fix barrier support
In reply to Kevin WolfFrom: Christoph Hellwig <[email protected]>
Before issuing the barrier to the block driver we need to flush our oustanding queue of write requests, as the flush is supposed to be issued after them.
--- hw/virtio-blk.c | 15 +++++++++++++-- 1 files changed, 13 insertions(+), 2 deletions(-)
hw/virtio-blk.c +13 -2
@@ -240,10 +240,20 @@ static void do_multiwrite(BlockDriverState *bs, BlockRequest *blkreq, } } -static void virtio_blk_handle_flush (VirtIOBlockReq *req) +static void virtio_blk_handle_flush (BlockRequest *blkreq,int *num_writes, +VirtIOBlockReq *req,BlockDriverState **old_bs) {BlockDriverAIOCB *acb; +/* +* Make sure all outstanding writes are posted to the backing device. +*/ +if (*old_bs !=NULL ) { +do_multiwrite (*old_bs, blkreq, *num_writes); + } + *num_writes =0 ; + *old_bs = req->dev->bs; + acb =bdrv_aio_flush (req->dev->bs, virtio_blk_flush_complete, req);if (!acb) {virtio_blk_req_complete (req,VIRTIO_BLK_S_IOERR );
@@ -316,7 +326,8 @@ static void virtio_blk_handle_request(VirtIOBlockReq *req, req->in = (void *)req->elem.in_sg[req->elem.in_num -1 ].iov_base;if (req->out->type &VIRTIO_BLK_T_FLUSH ) { -virtio_blk_handle_flush (req); +virtio_blk_handle_flush (mrb->blkreq, &mrb->num_writes, + req, &mrb->old_bs); }else if (req->out->type &VIRTIO_BLK_T_SCSI_CMD ) {virtio_blk_handle_scsi (req); }else if (req->out->type &VIRTIO_BLK_T_OUT ) {
Signature · 1 line
1.6.6.1
PATCH 4/8 block: fix sector comparism in multiwrite_req_compare
In reply to Kevin WolfFrom: Christoph Hellwig <[email protected]>
The difference between the start sectors of two requests can be larger than the size of the "int" type, which can lead to a not correctly sorted multiwrite array and thus spurious I/O errors and filesystem corruption due to incorrect request merges.
So instead of doing the cute sector arithmetics trick spell out the exact comparisms.
Spotted by Kevin Wolf based on a testcase from Michael Tokarev.
--- block.c | 14 +++++++++++++- 1 files changed, 13 insertions(+), 1 deletions(-)
block.c +13 -1
@@ -1929,7 +1929,19 @@ static void multiwrite_cb(void *opaque, int ret)static int multiwrite_req_compare (const void *a,const void *b) { -return (((BlockRequest *) a)->sector - ((BlockRequest *) b)->sector); +const BlockRequest *req1 = a, *req2 = b; + +/* +* Note that we can't simply subtract req2->sector from req1->sector +* here as that could overflow the return value. +*/ +if (req1->sector > req2->sector) { +return 1 ; + }else if (req1->sector < req2->sector) { +return -1 ; + }else { +return 0 ; + } }/*
Signature · 1 line
1.6.6.1
PATCH 5/8 block: Make find_image_format() return 'raw' BlockDriver for SG_IO devices
In reply to Kevin WolfFrom: Nicholas Bellinger <[email protected]>
This patch adds a special BlockDriverState->sg check in block.c:find_image_format() after bdrv_file_open() -> block/raw-posix.c:hdev_open() has been called to determine if we are dealing with a Linux host scsi-generic device.
The patch then returns the BlockDriver * from bdrv_find_format("raw"), skipping the subsequent bdrv_read() and rest of find_image_format().
--- block.c | 5 +++++ 1 files changed, 5 insertions(+), 0 deletions(-)
block.c +5 -0
@@ -329,6 +329,11 @@ static BlockDriver *find_image_format(const char *filename) ret =bdrv_file_open (&bs, filename,0 );if (ret <0 )return NULL ; + +/* Return the raw BlockDriver * to scsi-generic devices */ +if (bs->sg) +return bdrv_find_format ("raw" ); + ret =bdrv_pread (bs,0 , buf,sizeof (buf));bdrv_delete (bs);if (ret <0 ) {
Signature · 1 line
1.6.6.1
PATCH 6/8 block: Add SG_IO device check in refresh_total_sectors()
In reply to Kevin WolfFrom: Nicholas Bellinger <[email protected]>
This patch adds a special case check for scsi-generic devices in refresh_total_sectors() to skip the subsequent BlockDriver->bdrv_getlength() that will be returning -ESPIPE from block/raw-posic.c:raw_getlength() for BlockDriverState->sg=1 devices.
--- block.c | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-)
block.c +4 -0
@@ -361,6 +361,10 @@ static int refresh_total_sectors(BlockDriverState *bs, int64_t hint) {BlockDriver *drv = bs->drv; +/* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ +if (bs->sg) +return 0 ; +/* query actual device if possible, otherwise just trust the hint */ if (drv->bdrv_getlength) {int64_t length = drv->bdrv_getlength (bs);
Signature · 1 line
1.6.6.1
PATCH 7/8 vvfat: Fix compilation with DEBUG defined
In reply to Kevin WolfFrom: Riccardo Magliocchetti <[email protected]>
gcc does not like passing a NULL where an int value is expected:
block/vvfat.c: In function ‘checkpoint’:block/vvfat.c:2868: error: passing argument 2 of ‘remove_mapping’ makes
integer from pointer without a cast
--- block/vvfat.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
block/vvfat.c +1 -1
@@ -2865,7 +2865,7 @@ static void checkpoint(void) {return ;/* avoid compiler warnings: */ hexdump (NULL ,100 ); -remove_mapping (vvv,NULL ); +remove_mapping (vvv,0 );print_mapping (NULL );print_direntry (NULL ); }
Signature · 1 line
1.6.6.1
PATCH 8/8 vvfat: More build fixes with DEBUG
In reply to Kevin WolfCasting a pointer to an int doesn't work on 64 bit platforms. Use the %p printf conversion specifier instead.
--- block/vvfat.c | 8 ++++++-- 1 files changed, 6 insertions(+), 2 deletions(-)
block/vvfat.c +6 -2
@@ -1244,7 +1244,7 @@ static void print_direntry(const direntry_t* direntry)int j =0 ;char buffer[1024 ]; -fprintf (stderr,"direntry 0x%x: " , (int )direntry); +fprintf (stderr,"direntry %p: " , direntry);if (!direntry)return ;if (is_long_name (direntry)) {
@@ -1273,7 +1273,11 @@ static void print_direntry(const direntry_t* direntry)static void print_mapping (const mapping_t * mapping) { -fprintf (stderr,"mapping (0x%x): begin, end = %d, %d, dir_index = %d, first_mapping_index = %d, name = %s, mode = 0x%x, " , (int )mapping, mapping->begin, mapping->end, mapping->dir_index, mapping->first_mapping_index, mapping->path, mapping->mode); +fprintf (stderr,"mapping (%p): begin, end = %d, %d, dir_index = %d, " +"first_mapping_index = %d, name = %s, mode = 0x%x, " , + mapping, mapping->begin, mapping->end, mapping->dir_index, + mapping->first_mapping_index, mapping->path, mapping->mode); +if (mapping->mode &MODE_DIRECTORY )fprintf (stderr,"parent_mapping_index = %d, first_dir_index = %d\n" , mapping->info.dir.parent_mapping_index, mapping->info.dir.first_dir_index);else
Signature · 1 line
1.6.6.1