block: Fix resize (extending) of short overlays
v7 cover block: Fix resize (extending) of short overlays
v7:
- Allocate smaller zero buffer [Vladimir]
- Added missing error_setg_errno() [Max]
- Code cleanup in the iotest, enabled mapping for 'metadata' [Vladimir]
- Don't assign to errp twice [Eric]
v6:
- qcow2: Don't round up end offset [Eric]
- qcow2: Use different error messages for different error paths
- Request BDRV_REQ_ZERO_WRITE only if the backing file actually covers (part of) the new area (fix regression from v3) [Max]
- New patch: Forward ZERO_WRITE flag for full preallocation [Max]
v5:
- Fixed file-win32 [Patchew]
- Fixed zeroing in qcow2 for unaligned requests + tests [Vladimir]
- Made raw-format code more consistent [Eric]
- Leave output for existing iotests cases unchanged [Vladimir]
v4:
- Rewrote the series to move the actual zeroing to the block drivers so that error paths can work correctly and potentially long-running fallback to writing explicit zeroes is avoided.
- Fixed output filtering order in the test case [Max]
v3:
- Don't allow blocking the monitor for a zero write in block_resize (even though we can already blockfor other reasons there). This is mainly responsible for the increased complexity compared to v2. Personally, I think this is not an improvement over v2, but if this is what it takes to fix a corruption issue in 4.2... [Max]
- Don't use huge image files in the test case [Vladimir]
v2:
- Switched order of bs->total_sectors update and zero write [Vladimir]
- Fixed coding style [Vladimir]
- Changed the commit message to contain what was in the cover letter
- Test all preallocation modes
- Test allocation status with qemu-io 'map' [Vladimir]
Kevin Wolf (10):
block: Add flags to BlockDriver.bdrv_co_truncate()
block: Add flags to bdrv(_co)_truncate()
block-backend: Add flags to blk_truncate()
qcow2: Support BDRV_REQ_ZERO_WRITE for truncate
raw-format: Support BDRV_REQ_ZERO_WRITE for truncate
file-posix: Support BDRV_REQ_ZERO_WRITE for truncate
block: truncate: Don't make backing file data visible
iotests: Filter testfiles out in filter_img_info()
iotests: Test committing to short backing file
qcow2: Forward ZERO_WRITE flag for full preallocation
include/block/block.h | 5 +- include/block/block_int.h | 10 +- include/sysemu/block-backend.h | 2 +- block.c | 3 +- block/block-backend.c | 4 +- block/commit.c | 4 +- block/crypto.c | 7 +- block/file-posix.c | 6 +- block/file-win32.c | 2 +- block/gluster.c | 1 + block/io.c | 43 +++++- block/iscsi.c | 2 +- block/mirror.c | 2 +- block/nfs.c | 3 +- block/parallels.c | 6 +- block/qcow.c | 4 +- block/qcow2-cluster.c | 2 +- block/qcow2-refcount.c | 2 +- block/qcow2.c | 73 +++++++-- block/qed.c | 3 +- block/raw-format.c | 6 +- block/rbd.c | 1 + block/sheepdog.c | 4 +- block/ssh.c | 2 +- block/vdi.c | 2 +- block/vhdx-log.c | 2 +- block/vhdx.c | 6 +- block/vmdk.c | 8 +- block/vpc.c | 2 +- blockdev.c | 2 +- qemu-img.c | 2 +- qemu-io-cmds.c | 2 +- tests/test-block-iothread.c | 9 +- tests/qemu-iotests/iotests.py | 5 +- tests/qemu-iotests/274 | 155 +++++++++++++++++++ tests/qemu-iotests/274.out | 268 +++++++++++++++++++++++++++++++++ tests/qemu-iotests/group | 1 + 37 files changed, 597 insertions(+), 64 deletions(-) create mode 100755 tests/qemu-iotests/274 create mode 100644 tests/qemu-iotests/274.out
Signature · 1 line
2.25.3
PATCH v7 5/10 raw-format: Support BDRV_REQ_ZERO_WRITE for truncate
The raw format driver can simply forward the flag and let its bs->file child take care of actually providing the zeros.
--- block/raw-format.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
block/raw-format.c +3 -1
@@ -387,7 +387,7 @@ static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, s->size = offset; offset += s->offset; -return bdrv_co_truncate (bs->file, offset, exact, prealloc,0 , errp); +return bdrv_co_truncate (bs->file, offset, exact, prealloc, flags, errp); }static void raw_eject (BlockDriverState *bs,bool eject_flag)
@@ -445,6 +445,8 @@ static int raw_open(BlockDriverState *bs, QDict *options, int flags, bs->supported_zero_flags =BDRV_REQ_WRITE_UNCHANGED | ((BDRV_REQ_FUA |BDRV_REQ_MAY_UNMAP |BDRV_REQ_NO_FALLBACK ) & bs->file->bs->supported_zero_flags); + bs->supported_truncate_flags = bs->file->bs->supported_truncate_flags & +BDRV_REQ_ZERO_WRITE ;if (bs->probed && !bdrv_is_read_only (bs)) {bdrv_refresh_filename (bs->file->bs);
Signature · 1 line
2.25.3
PATCH v7 3/10 block-backend: Add flags to blk_truncate()
In reply to Kevin WolfNow that node level interface bdrv_truncate() supports passing request flags to the block driver, expose this on the BlockBackend level, too.
--- include/sysemu/block-backend.h | 2 +- block.c | 3 ++- block/block-backend.c | 4 ++-- block/commit.c | 4 ++-- block/crypto.c | 2 +- block/mirror.c | 2 +- block/qcow2.c | 4 ++-- block/qed.c | 2 +- block/vdi.c | 2 +- block/vhdx.c | 4 ++-- block/vmdk.c | 6 +++--- block/vpc.c | 2 +- blockdev.c | 2 +- qemu-img.c | 2 +- qemu-io-cmds.c | 2 +- 15 files changed, 22 insertions(+), 21 deletions(-)
include/sysemu/block-backend.h +1 -1
@@ -237,7 +237,7 @@ int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset,int blk_pwrite_compressed (BlockBackend *blk,int64_t offset,const void *buf,int bytes);int blk_truncate (BlockBackend *blk,int64_t offset,bool exact, -PreallocMode prealloc,Error **errp); +PreallocMode prealloc,BdrvRequestFlags flags,Error **errp);int blk_pdiscard (BlockBackend *blk,int64_t offset,int bytes);int blk_save_vmstate (BlockBackend *blk,const uint8_t *buf,int64_t pos,int size);
block.c +2 -1
@@ -548,7 +548,8 @@ static int64_t create_file_fallback_truncate(BlockBackend *blk,int64_t size;int ret; - ret =blk_truncate (blk, minimum_size, false,PREALLOC_MODE_OFF , &local_err); + ret =blk_truncate (blk, minimum_size, false,PREALLOC_MODE_OFF ,0 , + &local_err);if (ret <0 && ret != -ENOTSUP ) {error_propagate (errp, local_err);return ret;
block/block-backend.c +2 -2
@@ -2137,14 +2137,14 @@ int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf, }int blk_truncate (BlockBackend *blk,int64_t offset,bool exact, -PreallocMode prealloc,Error **errp) +PreallocMode prealloc,BdrvRequestFlags flags,Error **errp) {if (!blk_is_available (blk)) {error_setg (errp,"No medium inserted" );return -ENOMEDIUM ; } -return bdrv_truncate (blk->root, offset, exact, prealloc,0 , errp); +return bdrv_truncate (blk->root, offset, exact, prealloc, flags, errp); } int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
block/commit.c +2 -2
@@ -133,7 +133,7 @@ static int coroutine_fn commit_run(Job *job, Error **errp) }if (base_len < len) { - ret =blk_truncate (s->base, len, false,PREALLOC_MODE_OFF ,NULL ); + ret =blk_truncate (s->base, len, false,PREALLOC_MODE_OFF ,0 ,NULL );if (ret) { goto out; }
@@ -458,7 +458,7 @@ int bdrv_commit(BlockDriverState *bs)* grow the backing file image if possible. If not possible, * we must return an error */ if (length > backing_length) { - ret =blk_truncate (backing, length, false,PREALLOC_MODE_OFF , + ret =blk_truncate (backing, length, false,PREALLOC_MODE_OFF ,0 , &local_err);if (ret <0 ) {error_report_err (local_err);
block/crypto.c +1 -1
@@ -115,7 +115,7 @@ static ssize_t block_crypto_init_func(QCryptoBlock *block,* which will be used by the crypto header */ return blk_truncate (data->blk, data->size + headerlen, false, - data->prealloc, errp); + data->prealloc,0 , errp); }
block/mirror.c +1 -1
@@ -900,7 +900,7 @@ static int coroutine_fn mirror_run(Job *job, Error **errp)if (s->bdev_length > base_length) { ret =blk_truncate (s->target, s->bdev_length, false, -PREALLOC_MODE_OFF ,NULL ); +PREALLOC_MODE_OFF ,0 ,NULL );if (ret <0 ) { goto immediate_exit; }
block/qcow2.c +2 -2
@@ -3511,7 +3511,7 @@ qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)/* Okay, now that we have a valid image, let's give it the right size */ ret =blk_truncate (blk, qcow2_opts->size, false, qcow2_opts->preallocation, - errp); +0 , errp);if (ret <0 ) {error_prepend (errp,"Could not resize image: " ); goto out;
@@ -5374,7 +5374,7 @@ static int qcow2_amend_options(BlockDriverState *bs, QemuOpts *opts,* Amending image options should ensure that the image has * exactly the given new values, so pass exact=true here. */ - ret =blk_truncate (blk, new_size, true,PREALLOC_MODE_OFF , errp); + ret =blk_truncate (blk, new_size, true,PREALLOC_MODE_OFF ,0 , errp);blk_unref (blk);if (ret <0 ) {return ret;
block/qed.c +1 -1
@@ -677,7 +677,7 @@ static int coroutine_fn bdrv_qed_co_create(BlockdevCreateOptions *opts,* The QED format associates file length with allocation status, * so a new file (which is empty) must have a length of 0. */ - ret =blk_truncate (blk,0 , true,PREALLOC_MODE_OFF , errp); + ret =blk_truncate (blk,0 , true,PREALLOC_MODE_OFF ,0 , errp);if (ret <0 ) { goto out; }
block/vdi.c +1 -1
@@ -875,7 +875,7 @@ static int coroutine_fn vdi_co_do_create(BlockdevCreateOptions *create_options,if (image_type ==VDI_TYPE_STATIC ) { ret =blk_truncate (blk, offset + blocks * block_size, false, -PREALLOC_MODE_OFF , errp); +PREALLOC_MODE_OFF ,0 , errp);if (ret <0 ) {error_prepend (errp,"Failed to statically allocate file" ); goto exit;
block/vhdx.c +2 -2
@@ -1703,13 +1703,13 @@ static int vhdx_create_bat(BlockBackend *blk, BDRVVHDXState *s,/* All zeroes, so we can just extend the file - the end of the BAT * is the furthest thing we have written yet */ ret =blk_truncate (blk, data_file_offset, false,PREALLOC_MODE_OFF , - errp); +0 , errp);if (ret <0 ) { goto exit; } }else if (type ==VHDX_TYPE_FIXED ) { ret =blk_truncate (blk, data_file_offset + image_size, false, -PREALLOC_MODE_OFF , errp); +PREALLOC_MODE_OFF ,0 , errp);if (ret <0 ) { goto exit; }
block/vmdk.c +3 -3
@@ -2118,7 +2118,7 @@ static int vmdk_init_extent(BlockBackend *blk,int gd_buf_size;if (flat) { - ret =blk_truncate (blk, filesize, false,PREALLOC_MODE_OFF , errp); + ret =blk_truncate (blk, filesize, false,PREALLOC_MODE_OFF ,0 , errp); goto exit; } magic =cpu_to_be32 (VMDK4_MAGIC );
@@ -2182,7 +2182,7 @@ static int vmdk_init_extent(BlockBackend *blk, } ret =blk_truncate (blk,le64_to_cpu (header.grain_offset) <<9 , false, -PREALLOC_MODE_OFF , errp); +PREALLOC_MODE_OFF ,0 , errp);if (ret <0 ) { goto exit; }
@@ -2523,7 +2523,7 @@ static int coroutine_fn vmdk_co_do_create(int64_t size,/* bdrv_pwrite write padding zeros to align to sector, we don't need that * for description file */ if (desc_offset ==0 ) { - ret =blk_truncate (blk, desc_len, false,PREALLOC_MODE_OFF , errp); + ret =blk_truncate (blk, desc_len, false,PREALLOC_MODE_OFF ,0 , errp);if (ret <0 ) { goto exit; }
block/vpc.c +1 -1
@@ -898,7 +898,7 @@ static int create_fixed_disk(BlockBackend *blk, uint8_t *buf,/* Add footer to total size */ total_size +=HEADER_SIZE ; - ret =blk_truncate (blk, total_size, false,PREALLOC_MODE_OFF , errp); + ret =blk_truncate (blk, total_size, false,PREALLOC_MODE_OFF ,0 , errp);if (ret <0 ) {return ret; }
blockdev.c +1 -1
@@ -2741,7 +2741,7 @@ void qmp_block_resize(bool has_device, const char *device, }bdrv_drained_begin (bs); - ret =blk_truncate (blk, size, false,PREALLOC_MODE_OFF , errp); + ret =blk_truncate (blk, size, false,PREALLOC_MODE_OFF ,0 , errp);bdrv_drained_end (bs); out:
qemu-img.c +1 -1
@@ -3866,7 +3866,7 @@ static int img_resize(int argc, char **argv)* resizing, so pass @exact=true. It is of no use to report * success when the image has not actually been resized. */ - ret =blk_truncate (blk, total_size, true, prealloc, &err); + ret =blk_truncate (blk, total_size, true, prealloc,0 , &err);if (!ret) {qprintf (quiet,"Image resized.\n" ); }else {
qemu-io-cmds.c +1 -1
@@ -1715,7 +1715,7 @@ static int truncate_f(BlockBackend *blk, int argc, char **argv)* exact=true. It is better to err on the "emit more errors" side * than to be overly permissive. */ - ret =blk_truncate (blk, offset, true,PREALLOC_MODE_OFF , &local_err); + ret =blk_truncate (blk, offset, true,PREALLOC_MODE_OFF ,0 , &local_err);if (ret <0 ) {error_report_err (local_err);return ret;
Signature · 1 line
2.25.3
PATCH v7 4/10 qcow2: Support BDRV_REQ_ZERO_WRITE for truncate
In reply to Kevin WolfIf BDRV_REQ_ZERO_WRITE is set and we're extending the image, calling qcow2_cluster_zeroize() with flags=0 does the right thing: It doesn't undo any previous preallocation, but just adds the zero flag to all relevant L2 entries. If an external data file is in use, a write_zeroes request to the data file is made instead.
--- block/qcow2-cluster.c | 2 +- block/qcow2.c | 34 ++++++++++++++++++++++++++++++++++ 2 files changed, 35 insertions(+), 1 deletion(-)
block/qcow2-cluster.c +1 -1
@@ -1795,7 +1795,7 @@ int qcow2_cluster_zeroize(BlockDriverState *bs, uint64_t offset,/* Caller must pass aligned values, except at image end */ assert (QEMU_IS_ALIGNED (offset, s->cluster_size));assert (QEMU_IS_ALIGNED (end_offset, s->cluster_size) || - end_offset == bs->total_sectors <<BDRV_SECTOR_BITS ); + end_offset >= bs->total_sectors <<BDRV_SECTOR_BITS );/* The zero flag is only supported by version 3 and newer */ if (s->qcow_version <3 ) {
block/qcow2.c +34 -0
@@ -1726,6 +1726,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, bs->supported_zero_flags = header.version >=3 ?BDRV_REQ_MAY_UNMAP |BDRV_REQ_NO_FALLBACK :0 ; + bs->supported_truncate_flags =BDRV_REQ_ZERO_WRITE ;/* Repair image if dirty */ if (!(flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) && !bs->read_only &&
@@ -4214,6 +4215,39 @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,g_assert_not_reached (); } +if ((flags &BDRV_REQ_ZERO_WRITE ) && offset > old_length) { +uint64_t zero_start =QEMU_ALIGN_UP (old_length, s->cluster_size); + +/* +* Use zero clusters as much as we can. qcow2_cluster_zeroize() +* requires a cluster-aligned start. The end may be unaligned if it is +* at the end of the image (which it is here). +*/ + ret =qcow2_cluster_zeroize (bs, zero_start, offset - zero_start,0 ); +if (ret <0 ) { +error_setg_errno (errp, -ret,"Failed to zero out new clusters" ); + goto fail; + } + +/* Write explicit zeros for the unaligned head */ +if (zero_start > old_length) { +uint64_t len = zero_start - old_length; +uint8_t *buf =qemu_blockalign0 (bs, len); +QEMUIOVector qiov; +qemu_iovec_init_buf (&qiov, buf, len); + +qemu_co_mutex_unlock (&s->lock); + ret =qcow2_co_pwritev_part (bs, old_length, len, &qiov,0 ,0 ); +qemu_co_mutex_lock (&s->lock); + +qemu_vfree (buf); +if (ret <0 ) { +error_setg_errno (errp, -ret,"Failed to zero out the new area" ); + goto fail; + } + } + } +if (prealloc !=PREALLOC_MODE_OFF ) {/* Flush metadata before actually changing the image size */ ret =qcow2_write_caches (bs);
Signature · 1 line
2.25.3
PATCH v7 1/10 block: Add flags to BlockDriver.bdrv_co_truncate()
In reply to Kevin WolfThis adds a new BdrvRequestFlags parameter to the .bdrv_co_truncate() driver callbacks, and a supported_truncate_flags field in BlockDriverState that allows drivers to advertise support for request flags in the context of truncate.
For now, we always pass 0 and no drivers declare support for any flag.
--- include/block/block_int.h | 10 +++++++++- block/crypto.c | 3 ++- block/file-posix.c | 2 +- block/file-win32.c | 2 +- block/gluster.c | 1 + block/io.c | 8 +++++++- block/iscsi.c | 2 +- block/nfs.c | 3 ++- block/qcow2.c | 2 +- block/qed.c | 1 + block/raw-format.c | 2 +- block/rbd.c | 1 + block/sheepdog.c | 4 ++-- block/ssh.c | 2 +- tests/test-block-iothread.c | 3 ++- 15 files changed, 33 insertions(+), 13 deletions(-)
include/block/block_int.h +9 -1
@@ -355,7 +355,7 @@ struct BlockDriver {*/ int coroutine_fn (*bdrv_co_truncate)(BlockDriverState *bs, int64_t offset,bool exact,PreallocMode prealloc, -Error **errp); +BdrvRequestFlags flags,Error **errp);int64_t (*bdrv_getlength)(BlockDriverState *bs);bool has_variable_length;
@@ -847,6 +847,14 @@ struct BlockDriverState {/* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA, * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED) */ unsigned int supported_zero_flags; +/* +* Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE). +* +* If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure +* that any added space reads as all zeros. If this can't be guaranteed, +* the operation must fail. +*/ +unsigned int supported_truncate_flags;/* the following member gives a name to every node on the bs graph. */ char node_name[32 ];
block/crypto.c +2 -1
@@ -299,7 +299,8 @@ static int block_crypto_co_create_generic(BlockDriverState *bs, static int coroutine_fnblock_crypto_co_truncate (BlockDriverState *bs,int64_t offset,bool exact, -PreallocMode prealloc,Error **errp) +PreallocMode prealloc,BdrvRequestFlags flags, +Error **errp) {BlockCrypto *crypto = bs->opaque; uint64_t payload_offset =
block/file-posix.c +1 -1
@@ -2080,7 +2080,7 @@ raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,bool exact,PreallocMode prealloc, -Error **errp) +BdrvRequestFlags flags,Error **errp) {BDRVRawState *s = bs->opaque;struct stat st;
block/file-win32.c +1 -1
@@ -469,7 +469,7 @@ static void raw_close(BlockDriverState *bs) static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,bool exact,PreallocMode prealloc, -Error **errp) +BdrvRequestFlags flags,Error **errp) {BDRVRawState *s = bs->opaque;LONG low, high;
block/gluster.c +1 -0
@@ -1228,6 +1228,7 @@ static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,int64_t offset, bool exact, PreallocMode prealloc, + BdrvRequestFlags flags, Error **errp) { BDRVGlusterState *s = bs->opaque;
block/io.c +7 -1
@@ -3344,6 +3344,7 @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact,BlockDriverState *bs = child->bs;BlockDriver *drv = bs->drv;BdrvTrackedRequest req; +BdrvRequestFlags flags =0 ;int64_t old_size, new_bytes;int ret;
@@ -3394,7 +3395,12 @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, }if (drv->bdrv_co_truncate) { - ret = drv->bdrv_co_truncate (bs, offset, exact, prealloc, errp); +if (flags & ~bs->supported_truncate_flags) { +error_setg (errp,"Block driver does not support requested flags" ); + ret = -ENOTSUP ; + goto out; + } + ret = drv->bdrv_co_truncate (bs, offset, exact, prealloc, flags, errp); }else if (bs->file && drv->is_filter) { ret =bdrv_co_truncate (bs->file, offset, exact, prealloc, errp); }else {
block/iscsi.c +1 -1
@@ -2124,7 +2124,7 @@ static void iscsi_reopen_commit(BDRVReopenState *reopen_state) static int coroutine_fn iscsi_co_truncate(BlockDriverState *bs, int64_t offset,bool exact,PreallocMode prealloc, -Error **errp) +BdrvRequestFlags flags,Error **errp) {IscsiLun *iscsilun = bs->opaque;int64_t cur_length;
block/nfs.c +2 -1
@@ -755,7 +755,8 @@ static int64_t nfs_get_allocated_file_size(BlockDriverState *bs) static int coroutine_fnnfs_file_co_truncate (BlockDriverState *bs,int64_t offset,bool exact, -PreallocMode prealloc,Error **errp) +PreallocMode prealloc,BdrvRequestFlags flags, +Error **errp) {NFSClient *client = bs->opaque;int ret;
block/qcow2.c +1 -1
@@ -3964,7 +3964,7 @@ fail: static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,bool exact,PreallocMode prealloc, -Error **errp) +BdrvRequestFlags flags,Error **errp) {BDRVQcow2State *s = bs->opaque;uint64_t old_length;
block/qed.c +1 -0
@@ -1467,6 +1467,7 @@ static int coroutine_fn bdrv_qed_co_truncate(BlockDriverState *bs,int64_t offset, bool exact, PreallocMode prealloc, + BdrvRequestFlags flags, Error **errp) { BDRVQEDState *s = bs->opaque;
block/raw-format.c +1 -1
@@ -371,7 +371,7 @@ static void raw_refresh_limits(BlockDriverState *bs, Error **errp) static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,bool exact,PreallocMode prealloc, -Error **errp) +BdrvRequestFlags flags,Error **errp) {BDRVRawState *s = bs->opaque;
block/rbd.c +1 -0
@@ -1108,6 +1108,7 @@ static int coroutine_fn qemu_rbd_co_truncate(BlockDriverState *bs,int64_t offset, bool exact, PreallocMode prealloc, + BdrvRequestFlags flags, Error **errp) { int r;
block/sheepdog.c +2 -2
@@ -2288,7 +2288,7 @@ static int64_t sd_getlength(BlockDriverState *bs) static int coroutine_fn sd_co_truncate(BlockDriverState *bs, int64_t offset,bool exact,PreallocMode prealloc, -Error **errp) +BdrvRequestFlags flags,Error **errp) {BDRVSheepdogState *s = bs->opaque;int ret, fd;
@@ -2604,7 +2604,7 @@ static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,assert (!flags);if (offset > s->inode.vdi_size) { - ret =sd_co_truncate (bs, offset, false,PREALLOC_MODE_OFF ,NULL ); + ret =sd_co_truncate (bs, offset, false,PREALLOC_MODE_OFF ,0 ,NULL );if (ret <0 ) {return ret; }
block/ssh.c +1 -1
@@ -1298,7 +1298,7 @@ static int64_t ssh_getlength(BlockDriverState *bs) static int coroutine_fn ssh_co_truncate(BlockDriverState *bs, int64_t offset,bool exact,PreallocMode prealloc, -Error **errp) +BdrvRequestFlags flags,Error **errp) {BDRVSSHState *s = bs->opaque;
tests/test-block-iothread.c +2 -1
@@ -46,7 +46,8 @@ static int coroutine_fn bdrv_test_co_pdiscard(BlockDriverState *bs, static int coroutine_fnbdrv_test_co_truncate (BlockDriverState *bs,int64_t offset,bool exact, -PreallocMode prealloc,Error **errp) +PreallocMode prealloc,BdrvRequestFlags flags, +Error **errp) {return 0 ; }
Signature · 1 line
2.25.3
PATCH v7 2/10 block: Add flags to bdrv(_co)_truncate()
In reply to Kevin WolfNow that block drivers can support flags for .bdrv_co_truncate, expose the parameter in the node level interfaces bdrv_co_truncate() and bdrv_truncate().
--- include/block/block.h | 5 +++-- block/block-backend.c | 2 +- block/crypto.c | 2 +- block/io.c | 12 +++++++----- block/parallels.c | 6 +++--- block/qcow.c | 4 ++-- block/qcow2-refcount.c | 2 +- block/qcow2.c | 15 +++++++++------ block/raw-format.c | 2 +- block/vhdx-log.c | 2 +- block/vhdx.c | 2 +- block/vmdk.c | 2 +- tests/test-block-iothread.c | 6 +++--- 13 files changed, 34 insertions(+), 28 deletions(-)
include/block/block.h +3 -2
@@ -339,9 +339,10 @@ BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,void bdrv_refresh_filename (BlockDriverState *bs); int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, -PreallocMode prealloc,Error **errp); +PreallocMode prealloc,BdrvRequestFlags flags, +Error **errp);int bdrv_truncate (BdrvChild *child,int64_t offset,bool exact, -PreallocMode prealloc,Error **errp); +PreallocMode prealloc,BdrvRequestFlags flags,Error **errp);int64_t bdrv_nb_sectors (BlockDriverState *bs);int64_t bdrv_getlength (BlockDriverState *bs);
block/block-backend.c +1 -1
@@ -2144,7 +2144,7 @@ int blk_truncate(BlockBackend *blk, int64_t offset, bool exact,return -ENOMEDIUM ; } -return bdrv_truncate (blk->root, offset, exact, prealloc, errp); +return bdrv_truncate (blk->root, offset, exact, prealloc,0 , errp); } int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf,
block/crypto.c +1 -1
@@ -313,7 +313,7 @@ block_crypto_co_truncate(BlockDriverState *bs, int64_t offset, bool exact, offset += payload_offset; -return bdrv_co_truncate (bs->file, offset, exact, prealloc, errp); +return bdrv_co_truncate (bs->file, offset, exact, prealloc,0 , errp); }static void block_crypto_close (BlockDriverState *bs)
block/io.c +7 -5
@@ -3339,12 +3339,12 @@ static void bdrv_parent_cb_resize(BlockDriverState *bs)* 'offset' bytes in length. */ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, -PreallocMode prealloc,Error **errp) +PreallocMode prealloc,BdrvRequestFlags flags, +Error **errp) {BlockDriverState *bs = child->bs;BlockDriver *drv = bs->drv;BdrvTrackedRequest req; -BdrvRequestFlags flags =0 ;int64_t old_size, new_bytes;int ret;
@@ -3402,7 +3402,7 @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, } ret = drv->bdrv_co_truncate (bs, offset, exact, prealloc, flags, errp); }else if (bs->file && drv->is_filter) { - ret =bdrv_co_truncate (bs->file, offset, exact, prealloc, errp); + ret =bdrv_co_truncate (bs->file, offset, exact, prealloc, flags, errp); }else {error_setg (errp,"Image format driver does not support resize" ); ret = -ENOTSUP ;
@@ -3435,6 +3435,7 @@ typedef struct TruncateCo {int64_t offset;bool exact;PreallocMode prealloc; +BdrvRequestFlags flags;Error **errp;int ret; } TruncateCo;
@@ -3443,12 +3444,12 @@ static void coroutine_fn bdrv_truncate_co_entry(void *opaque) {TruncateCo *tco = opaque; tco->ret =bdrv_co_truncate (tco->child, tco->offset, tco->exact, - tco->prealloc, tco->errp); + tco->prealloc, tco->flags, tco->errp);aio_wait_kick (); }int bdrv_truncate (BdrvChild *child,int64_t offset,bool exact, -PreallocMode prealloc,Error **errp) +PreallocMode prealloc,BdrvRequestFlags flags,Error **errp) {Coroutine *co;TruncateCo tco = {
@@ -3456,6 +3457,7 @@ int bdrv_truncate(BdrvChild *child, int64_t offset, bool exact, .offset = offset, .exact = exact, .prealloc = prealloc, + .flags = flags, .errp = errp, .ret = NOT_DONE, };
block/parallels.c +3 -3
@@ -203,7 +203,7 @@ static int64_t allocate_clusters(BlockDriverState *bs, int64_t sector_num, }else { ret =bdrv_truncate (bs->file, (s->data_end + space) <<BDRV_SECTOR_BITS , - false,PREALLOC_MODE_OFF ,NULL ); + false,PREALLOC_MODE_OFF ,0 ,NULL ); }if (ret <0 ) {return ret;
@@ -493,7 +493,7 @@ static int coroutine_fn parallels_co_check(BlockDriverState *bs,* That means we have to pass exact=true. */ ret =bdrv_truncate (bs->file, res->image_end_offset, true, -PREALLOC_MODE_OFF , &local_err); +PREALLOC_MODE_OFF ,0 , &local_err);if (ret <0 ) {error_report_err (local_err); res->check_errors++;
@@ -889,7 +889,7 @@ static void parallels_close(BlockDriverState *bs)/* errors are ignored, so we might as well pass exact=true */ bdrv_truncate (bs->file, s->data_end <<BDRV_SECTOR_BITS , true, -PREALLOC_MODE_OFF ,NULL ); +PREALLOC_MODE_OFF ,0 ,NULL ); }g_free (s->bat_dirty_bmap);
block/qcow.c +2 -2
@@ -480,7 +480,7 @@ static int get_cluster_offset(BlockDriverState *bs,return -E2BIG ; } ret =bdrv_truncate (bs->file, cluster_offset + s->cluster_size, - false,PREALLOC_MODE_OFF ,NULL ); + false,PREALLOC_MODE_OFF ,0 ,NULL );if (ret <0 ) {return ret; }
@@ -1035,7 +1035,7 @@ static int qcow_make_empty(BlockDriverState *bs) l1_length) < 0)return -1 ; ret =bdrv_truncate (bs->file, s->l1_table_offset + l1_length, false, -PREALLOC_MODE_OFF ,NULL ); +PREALLOC_MODE_OFF ,0 ,NULL );if (ret <0 )return ret;
block/qcow2-refcount.c +1 -1
@@ -2018,7 +2018,7 @@ static int check_refblocks(BlockDriverState *bs, BdrvCheckResult *res, } ret =bdrv_truncate (bs->file, offset + s->cluster_size, false, -PREALLOC_MODE_OFF , &local_err); +PREALLOC_MODE_OFF ,0 , &local_err);if (ret <0 ) {error_report_err (local_err); goto resize_fail;
block/qcow2.c +9 -6
@@ -3095,7 +3095,7 @@ static int coroutine_fn preallocate_co(BlockDriverState *bs, uint64_t offset, mode =PREALLOC_MODE_OFF ; } ret =bdrv_co_truncate (s->data_file, host_offset + cur_bytes, false, - mode, errp); + mode,0 , errp);if (ret <0 ) {return ret; }
@@ -4061,7 +4061,7 @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,* always fulfilled, so there is no need to pass it on.) */ bdrv_co_truncate (bs->file, (last_cluster +1 ) * s->cluster_size, - false,PREALLOC_MODE_OFF , &local_err); + false,PREALLOC_MODE_OFF ,0 , &local_err);if (local_err) {warn_reportf_err (local_err,"Failed to truncate the tail of the image: " );
@@ -4083,7 +4083,8 @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,* file should be resized to the exact target size, too, * so we pass @exact here. */ - ret =bdrv_co_truncate (s->data_file, offset, exact, prealloc, errp); + ret =bdrv_co_truncate (s->data_file, offset, exact, prealloc,0 , + errp);if (ret <0 ) { goto fail; }
@@ -4169,7 +4170,8 @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, new_file_size = allocation_start + nb_new_data_clusters * s->cluster_size;/* Image file grows, so @exact does not matter */ - ret =bdrv_co_truncate (bs->file, new_file_size, false, prealloc, errp); + ret =bdrv_co_truncate (bs->file, new_file_size, false, prealloc,0 , + errp);if (ret <0 ) {error_prepend (errp,"Failed to resize underlying file: " ); qcow2_free_clusters(bs, allocation_start,
@@ -4348,7 +4350,8 @@ qcow2_co_pwritev_compressed_part(BlockDriverState *bs,if (len <0 ) {return len; } -return bdrv_co_truncate (bs->file, len, false,PREALLOC_MODE_OFF ,NULL ); +return bdrv_co_truncate (bs->file, len, false,PREALLOC_MODE_OFF ,0 , +NULL ); }if (offset_into_cluster (s, offset)) {
@@ -4563,7 +4566,7 @@ static int make_completely_empty(BlockDriverState *bs) } ret =bdrv_truncate (bs->file, (3 + l1_clusters) * s->cluster_size, false, -PREALLOC_MODE_OFF , &local_err); +PREALLOC_MODE_OFF ,0 , &local_err);if (ret <0 ) {error_report_err (local_err); goto fail;
block/raw-format.c +1 -1
@@ -387,7 +387,7 @@ static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, s->size = offset; offset += s->offset; -return bdrv_co_truncate (bs->file, offset, exact, prealloc, errp); +return bdrv_co_truncate (bs->file, offset, exact, prealloc,0 , errp); }static void raw_eject (BlockDriverState *bs,bool eject_flag)
block/vhdx-log.c +1 -1
@@ -558,7 +558,7 @@ static int vhdx_log_flush(BlockDriverState *bs, BDRVVHDXState *s, goto exit; } ret =bdrv_truncate (bs->file, new_file_size, false, -PREALLOC_MODE_OFF ,NULL ); +PREALLOC_MODE_OFF ,0 ,NULL );if (ret <0 ) { goto exit; }
block/vhdx.c +1 -1
@@ -1264,7 +1264,7 @@ static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s, }return bdrv_truncate (bs->file, *new_offset + s->block_size, false, -PREALLOC_MODE_OFF ,NULL ); +PREALLOC_MODE_OFF ,0 ,NULL ); }/*
block/vmdk.c +1 -1
@@ -2077,7 +2077,7 @@ vmdk_co_pwritev_compressed(BlockDriverState *bs, uint64_t offset, } length =QEMU_ALIGN_UP (length,BDRV_SECTOR_SIZE ); ret =bdrv_truncate (s->extents[i].file, length, false, -PREALLOC_MODE_OFF ,NULL ); +PREALLOC_MODE_OFF ,0 ,NULL );if (ret <0 ) {return ret; }
tests/test-block-iothread.c +3 -3
@@ -186,18 +186,18 @@ static void test_sync_op_truncate(BdrvChild *c)int ret;/* Normal success path */ - ret =bdrv_truncate (c,65536 , false,PREALLOC_MODE_OFF ,NULL ); + ret =bdrv_truncate (c,65536 , false,PREALLOC_MODE_OFF ,0 ,NULL ); g_assert_cmpint(ret, ==, 0);/* Early error: Negative offset */ - ret =bdrv_truncate (c,-2 , false,PREALLOC_MODE_OFF ,NULL ); + ret =bdrv_truncate (c,-2 , false,PREALLOC_MODE_OFF ,0 ,NULL ); g_assert_cmpint(ret, ==, -EINVAL);/* Error: Read-only image */ c->bs->read_only = true; c->bs->open_flags &= ~BDRV_O_RDWR ; - ret =bdrv_truncate (c,65536 , false,PREALLOC_MODE_OFF ,NULL ); + ret =bdrv_truncate (c,65536 , false,PREALLOC_MODE_OFF ,0 ,NULL ); g_assert_cmpint(ret, ==, -EACCES); c->bs->read_only = false;
Signature · 1 line
2.25.3
PATCH v7 7/10 block: truncate: Don't make backing file data visible
In reply to Kevin WolfWhen extending the size of an image that has a backing file larger than its old size, make sure that the backing file data doesn't become visible in the guest, but the added area is properly zeroed out.
Consider the following scenario where the overlay is shorter than its backing file:
base.qcow2: AAAAAAAA
overlay.qcow2: BBBB
When resizing (extending) overlay.qcow2, the new blocks should not stay unallocated and make the additional As from base.qcow2 visible like before this patch, but zeros should be read.
A similar case happens with the various variants of a commit job when an intermediate file is short (- for unallocated):
base.qcow2: A-A-AAAA
mid.qcow2: BB-B
top.qcow2: C--C--C-After commit top.qcow2 to mid.qcow2, the following happens:
mid.qcow2: CB-C00C0 (correct result)
mid.qcow2: CB-C--C- (before this fix)
Without the fix, blocks that previously read as zeros on top.qcow2 suddenly turn into A.
--- block/io.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+)
block/io.c +25 -0
@@ -3394,6 +3394,31 @@ int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, goto out; } +/* +* If the image has a backing file that is large enough that it would +* provide data for the new area, we cannot leave it unallocated because +* then the backing file content would become visible. Instead, zero-fill +* the new area. +* +* Note that if the image has a backing file, but was opened without the +* backing file, taking care of keeping things consistent with that backing +* file is the user's responsibility. +*/ +if (new_bytes && bs->backing) { +int64_t backing_len; + + backing_len =bdrv_getlength (backing_bs (bs)); +if (backing_len <0 ) { + ret = backing_len; +error_setg_errno (errp, -ret,"Could not get backing file size" ); + goto out; + } + +if (backing_len > old_size) { + flags |=BDRV_REQ_ZERO_WRITE ; + } + } +if (drv->bdrv_co_truncate) {if (flags & ~bs->supported_truncate_flags) {error_setg (errp,"Block driver does not support requested flags" );
Signature · 1 line
2.25.3
PATCH v7 6/10 file-posix: Support BDRV_REQ_ZERO_WRITE for truncate
In reply to Kevin WolfFor regular files, we always get BDRV_REQ_ZERO_WRITE behaviour from the OS, so we can advertise the flag and just ignore it.
--- block/file-posix.c | 4 ++++ 1 file changed, 4 insertions(+)
block/file-posix.c +4 -0
@@ -702,6 +702,10 @@ static int raw_open_common(BlockDriverState *bs, QDict *options,#endif bs->supported_zero_flags =BDRV_REQ_MAY_UNMAP |BDRV_REQ_NO_FALLBACK ; +if (S_ISREG (st.st_mode)) { +/* When extending regular files, we get zeros from the OS */ + bs->supported_truncate_flags =BDRV_REQ_ZERO_WRITE ; + } ret =0 ; fail:if (filename && (bdrv_flags &BDRV_O_TEMPORARY )) {
Signature · 1 line
2.25.3
PATCH v7 8/10 iotests: Filter testfiles out in filter_img_info()
In reply to Kevin WolfWe want to keep TEST_IMG for the full path of the main test image, but filter_testfiles() must be called for other test images before replacing other things like the image format because the test directory path could contain the format as a substring.
Insert a filter_testfiles() call between both.
--- tests/qemu-iotests/iotests.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
tests/qemu-iotests/iotests.py +3 -2
@@ -338,8 +338,9 @@ def filter_img_info(output, filename): for line in output.split('\n'): if 'disk size' in line or 'actual-size' in line: continue - line = line.replace(filename, 'TEST_IMG') \ - .replace(imgfmt, 'IMGFMT') + line = line.replace(filename, 'TEST_IMG') + line = filter_testfiles(line) + line = line.replace(imgfmt, 'IMGFMT') line = re.sub('iters: [0-9]+', 'iters: XXX', line) line = re.sub('uuid: [-a-f0-9]+', 'uuid: XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX', line) line = re.sub('cid: [0-9]+', 'cid: XXXXXXXXXX', line)
Signature · 1 line
2.25.3
PATCH v7 9/10 iotests: Test committing to short backing file
In reply to Kevin Wolf--- tests/qemu-iotests/274 | 155 +++++++++++++++++++++ tests/qemu-iotests/274.out | 268 +++++++++++++++++++++++++++++++++++++ tests/qemu-iotests/group | 1 + 3 files changed, 424 insertions(+) create mode 100755 tests/qemu-iotests/274 create mode 100644 tests/qemu-iotests/274.out
tests/qemu-iotests/274 new file +155 -0
@@ -0,0 +1,155 @@ +#!/usr/bin/env python3 +# +# Copyright (C) 2019 Red Hat, Inc. +# +# This program is free software; you can redistribute it and/or modify +# it under the terms of the GNU General Public License as published by +# the Free Software Foundation; either version 2 of the License, or +# (at your option) any later version. +# +# This program is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +# GNU General Public License for more details. +# +# You should have received a copy of the GNU General Public License +# along with this program. If not, see <http://www.gnu.org/licenses/>. +# +# Creator/Owner: Kevin Wolf <[email protected]> +# +# Some tests for short backing files and short overlays + +import iotests + +iotests.verify_image_format(supported_fmts=['qcow2']) +iotests.verify_platform(['linux']) + +size_short = 1 * 1024 * 1024 +size_long = 2 * 1024 * 1024 +size_diff = size_long - size_short + +def create_chain() -> None: + iotests.qemu_img_log('create', '-f', iotests.imgfmt, base, + str(size_long)) + iotests.qemu_img_log('create', '-f', iotests.imgfmt, '-b', base, mid, + str(size_short)) + iotests.qemu_img_log('create', '-f', iotests.imgfmt, '-b', mid, top, + str(size_long)) + + iotests.qemu_io_log('-c', 'write -P 1 0 %d' % size_long, base) + +def create_vm() -> iotests.VM: + vm = iotests.VM() + vm.add_blockdev('file,filename=%s,node-name=base-file' % base) + vm.add_blockdev('%s,file=base-file,node-name=base' % iotests.imgfmt) + vm.add_blockdev('file,filename=%s,node-name=mid-file' % mid) + vm.add_blockdev('%s,file=mid-file,node-name=mid,backing=base' + % iotests.imgfmt) + vm.add_drive(top, 'backing=mid,node-name=top') + return vm + +with iotests.FilePath('base') as base, \ + iotests.FilePath('mid') as mid, \ + iotests.FilePath('top') as top: + + iotests.log('== Commit tests ==') + + create_chain() + + iotests.log('=== Check visible data ===') + + iotests.qemu_io_log('-c', 'read -P 1 0 %d' % size_short, top) + iotests.qemu_io_log('-c', 'read -P 0 %d %d' % (size_short, size_diff), top) + + iotests.log('=== Checking allocation status ===') + + iotests.qemu_io_log('-c', 'alloc 0 %d' % size_short, + '-c', 'alloc %d %d' % (size_short, size_diff), + base) + + iotests.qemu_io_log('-c', 'alloc 0 %d' % size_short, + '-c', 'alloc %d %d' % (size_short, size_diff), + mid) + + iotests.qemu_io_log('-c', 'alloc 0 %d' % size_short, + '-c', 'alloc %d %d' % (size_short, size_diff), + top) + + iotests.log('=== Checking map ===') + + iotests.qemu_img_log('map', '--output=json', base) + iotests.qemu_img_log('map', '--output=human', base) + iotests.qemu_img_log('map', '--output=json', mid) + iotests.qemu_img_log('map', '--output=human', mid) + iotests.qemu_img_log('map', '--output=json', top) + iotests.qemu_img_log('map', '--output=human', top) + + iotests.log('=== Testing qemu-img commit (top -> mid) ===') + + iotests.qemu_img_log('commit', top) + iotests.img_info_log(mid) + iotests.qemu_io_log('-c', 'read -P 1 0 %d' % size_short, mid) + iotests.qemu_io_log('-c', 'read -P 0 %d %d' % (size_short, size_diff), mid) + + iotests.log('=== Testing HMP commit (top -> mid) ===') + + create_chain() + with create_vm() as vm: + vm.launch() + vm.qmp_log('human-monitor-command', command_line='commit drive0') + + iotests.img_info_log(mid) + iotests.qemu_io_log('-c', 'read -P 1 0 %d' % size_short, mid) + iotests.qemu_io_log('-c', 'read -P 0 %d %d' % (size_short, size_diff), mid) + + iotests.log('=== Testing QMP active commit (top -> mid) ===') + + create_chain() + with create_vm() as vm: + vm.launch() + vm.qmp_log('block-commit', device='top', base_node='mid', + job_id='job0', auto_dismiss=False) + vm.run_job('job0', wait=5) + + iotests.img_info_log(mid) + iotests.qemu_io_log('-c', 'read -P 1 0 %d' % size_short, mid) + iotests.qemu_io_log('-c', 'read -P 0 %d %d' % (size_short, size_diff), mid) + + + iotests.log('== Resize tests ==') + + # Use different sizes for different allocation modes: + # + # We want to have at least one test where 32 bit truncation in the size of + # the overlapping area becomes visible. This is covered by the + # prealloc='off' case (1G to 6G is an overlap of 5G). + # + # However, we can only do this for modes that don't preallocate data + # because otherwise we might run out of space on the test host. + # + # We also want to test some unaligned combinations. + for (prealloc, base_size, top_size_old, top_size_new, off) in [ + ('off', '6G', '1G', '8G', '5G'), + ('metadata', '32G', '30G', '33G', '31G'), + ('falloc', '10M', '5M', '15M', '9M'), + ('full', '16M', '8M', '12M', '11M'), + ('off', '384k', '253k', '512k', '253k'), + ('off', '400k', '256k', '512k', '336k'), + ('off', '512k', '256k', '500k', '436k')]: + + iotests.log('=== preallocation=%s ===' % prealloc) + iotests.qemu_img_log('create', '-f', iotests.imgfmt, base, base_size) + iotests.qemu_img_log('create', '-f', iotests.imgfmt, '-b', base, top, + top_size_old) + iotests.qemu_io_log('-c', 'write -P 1 %s 64k' % off, base) + + # After this, 0 to base_size should be allocated/zeroed. + # + # In theory, leaving base_size to top_size_new unallocated would be + # correct, but in practice, if we zero out anything, we zero out + # everything up to top_size_new. + iotests.qemu_img_log('resize', '-f', iotests.imgfmt, + '--preallocation', prealloc, top, top_size_new) + iotests.qemu_io_log('-c', 'read -P 0 %s 64k' % off, top) + iotests.qemu_io_log('-c', 'map', top) + iotests.qemu_img_log('map', '--output=json', top)
tests/qemu-iotests/274.out new file +268 -0
268 lines not on this page · raw message
tests/qemu-iotests/group +1 -0
@@ -286,6 +286,7 @@ 270 rw backing quick 272 rw 273 backing quick +274 rw backing 277 rw quick 279 rw backing quick 280 rw migration quick
Signature · 1 line
2.25.3
Re: [PATCH v7 04/10] qcow2: Support BDRV_REQ_ZERO_WRITE for truncate
In reply to Kevin WolfQuoted 11 lines · Kevin Wolf
If
BDRV_REQ_ZERO_WRITEis set and we're extending the image, callingqcow2_cluster_zeroize()with flags=0 does the right thing: It doesn't undo any previous preallocation, but just adds the zero flag to all relevant L2 entries. If an external data file is in use, awrite_zeroesrequest to the data file is made instead.Signed-off-by Kevin Wolf <[email protected]>--- block/qcow2-cluster.c | 2 +- block/qcow2.c | 34 ++++++++++++++++++++++++++++++++++ 2 files changed, 35 insertions(+), 1 deletion(-)
[...]
Quoted 4 lines
block/qcow2.c +0
-0
[...]
Quoted 1 line
block/qcow2.c +0
-0@@ -4214,6 +4215,39 @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,
[...]
block/qcow2.c static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset, · line 4235
Quoted 4 lines
block/qcow2.c +4
-0+/* Write explicit zeros for the unaligned head */ +if (zero_start > old_length) { +uint64_t len = zero_start - old_length; +uint8_t *buf =qemu_blockalign0 (bs, len);
I wonder whether I should raise the question of why this should be block-aligned when we make no effort to align the offset its written to (and we know it isn’t aligned to qcow2 clusters at least).
I probably should not.
Quoted 13 lines
block/qcow2.c +13
-0+QEMUIOVector qiov; +qemu_iovec_init_buf (&qiov, buf, len); + +qemu_co_mutex_unlock (&s->lock); + ret =qcow2_co_pwritev_part (bs, old_length, len, &qiov,0 ,0 ); +qemu_co_mutex_lock (&s->lock); + +qemu_vfree (buf); +if (ret <0 ) { +error_setg_errno (errp, -ret,"Failed to zero out the new area" ); + goto fail; + } + }
Re: [PATCH v7 07/10] block: truncate: Don't make backing file data visible
In reply to Kevin WolfQuoted 26 lines · Kevin Wolf
When extending the size of an image that has a backing file larger than its old size, make sure that the backing file data doesn't become visible in the guest, but the added area is properly zeroed out.
Consider the following scenario where the overlay is shorter than its backing file:
base.qcow2: AAAAAAAA
overlay.qcow2: BBBBWhen resizing (extending) overlay.qcow2, the new blocks should not stay unallocated and make the additional As from base.qcow2 visible like before this patch, but zeros should be read.
A similar case happens with the various variants of a commit job when an intermediate file is short (- for unallocated):
base.qcow2: A-A-AAAA mid.qcow2: BB-B top.qcow2: C--C--C-After commit top.qcow2 to mid.qcow2, the following happens:
mid.qcow2: CB-C00C0 (correct result)
mid.qcow2: CB-C--C- (before this fix)
Quoted 8 lines
Without the fix, blocks that previously read as zeros on top.qcow2 suddenly turn into A.
Signed-off-by Kevin Wolf <[email protected]>Reviewed-by Vladimir Sementsov-Ogievskiy <[email protected]>--- block/io.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+)
Re: [PATCH v7 09/10] iotests: Test committing to short backing file
In reply to Kevin WolfQuoted 8 lines · Kevin Wolf
Signed-off-by Kevin Wolf <[email protected]>--- tests/qemu-iotests/274 | 155 +++++++++++++++++++++ tests/qemu-iotests/274.out | 268 +++++++++++++++++++++++++++++++++++++ tests/qemu-iotests/group | 1 + 3 files changed, 424 insertions(+) create mode 100755 tests/qemu-iotests/274 create mode 100644 tests/qemu-iotests/274.out
Quoted 5 lines · Kevin Wolf
v7:
- Allocate smaller zero buffer [Vladimir]
- Added missingerror_setg_errno()[Max]
- Code cleanup in the iotest, enabled mapping for 'metadata' [Vladimir]
- Don't assign to errp twice [Eric]
I would have liked to see that change in patch 10, but the mail seems to have disappeared somewhere. :?
Max
PATCH v7 10/10 qcow2: Forward ZERO_WRITE flag for full preallocation
In reply to Kevin WolfThe BDRV_REQ_ZERO_WRITE is currently implemented in a way that first the image is possibly preallocated and then the zero flag is added to all clusters. This means that a copy-on-write operation may be needed when writing to these clusters, despite having used preallocation, negating one of the major benefits of preallocation.
Instead, try to forward the BDRV_REQ_ZERO_WRITE to the protocol driver, and if the protocol driver can ensure that the new area reads as zeros, we can skip setting the zero flag in the qcow2 layer.
Unfortunately, the same approach doesn't work for metadata preallocation, so we'll still set the zero flag there.
--- block/qcow2.c | 22 +++++++++++++++++++--- tests/qemu-iotests/274.out | 4 ++-- 2 files changed, 21 insertions(+), 5 deletions(-)
block/qcow2.c +19 -3
@@ -4170,9 +4170,25 @@ static int coroutine_fn qcow2_co_truncate(BlockDriverState *bs, int64_t offset,/* Allocate the data area */ new_file_size = allocation_start + nb_new_data_clusters * s->cluster_size; -/* Image file grows, so @exact does not matter */ - ret =bdrv_co_truncate (bs->file, new_file_size, false, prealloc,0 , - errp); +/* +* Image file grows, so @exact does not matter. +* +* If we need to zero out the new area, try first whether the protocol +* driver can already take care of this. +*/ +if (flags &BDRV_REQ_ZERO_WRITE ) { + ret =bdrv_co_truncate (bs->file, new_file_size, false, prealloc, +BDRV_REQ_ZERO_WRITE ,NULL ); +if (ret >=0 ) { + flags &= ~BDRV_REQ_ZERO_WRITE ; + } + }else { + ret =-1 ; + } +if (ret <0 ) { + ret =bdrv_co_truncate (bs->file, new_file_size, false, prealloc,0 , + errp); + }if (ret <0 ) {error_prepend (errp,"Failed to resize underlying file: " ); qcow2_free_clusters(bs, allocation_start,
tests/qemu-iotests/274.out +2 -2
@@ -187,7 +187,7 @@ read 65536/65536 bytes at offset 9437184 10 MiB (0xa00000) bytes allocated at offset 5 MiB (0x500000) [{ "start": 0, "length": 5242880, "depth": 1, "zero": true, "data": false}, -{ "start": 5242880, "length": 10485760, "depth": 0, "zero": true, "data": false, "offset": 327680}] +{ "start": 5242880, "length": 10485760, "depth": 0, "zero": false, "data": true, "offset": 327680}] === preallocation=full === Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=16777216 cluster_size=65536 lazy_refcounts=off refcount_bits=16
@@ -206,7 +206,7 @@ read 65536/65536 bytes at offset 11534336 4 MiB (0x400000) bytes allocated at offset 8 MiB (0x800000) [{ "start": 0, "length": 8388608, "depth": 1, "zero": true, "data": false}, -{ "start": 8388608, "length": 4194304, "depth": 0, "zero": true, "data": false, "offset": 327680}] +{ "start": 8388608, "length": 4194304, "depth": 0, "zero": false, "data": true, "offset": 327680}] === preallocation=off === Formatting 'TEST_DIR/PID-base', fmt=qcow2 size=393216 cluster_size=65536 lazy_refcounts=off refcount_bits=16
Signature · 1 line
2.25.3
Re: [PATCH v7 10/10] qcow2: Forward ZERO_WRITE flag for full preallocation
Quoted 13 lines · Kevin Wolf
The
BDRV_REQ_ZERO_WRITEis currently implemented in a way that first the image is possibly preallocated and then the zero flag is added to all clusters. This means that a copy-on-write operation may be needed when writing to these clusters, despite having used preallocation, negating one of the major benefits of preallocation.Instead, try to forward the
BDRV_REQ_ZERO_WRITEto the protocol driver, and if the protocol driver can ensure that the new area reads as zeros, we can skip setting the zero flag in the qcow2 layer.Unfortunately, the same approach doesn't work for metadata preallocation, so we'll still set the zero flag there.
Quoted 6 lines
Signed-off-by Kevin Wolf <[email protected]>Reviewed-by Max Reitz <[email protected]>--- block/qcow2.c | 22 +++++++++++++++++++--- tests/qemu-iotests/274.out | 4 ++-- 2 files changed, 21 insertions(+), 5 deletions(-)
Thanks for the resend! :)
Max
Re: [PATCH v7 04/10] qcow2: Support BDRV_REQ_ZERO_WRITE for truncate
In reply to Kevin WolfQuoted 6 lines · Kevin Wolf
If
BDRV_REQ_ZERO_WRITEis set and we're extending the image, callingqcow2_cluster_zeroize()with flags=0 does the right thing: It doesn't undo any previous preallocation, but just adds the zero flag to all relevant L2 entries. If an external data file is in use, awrite_zeroesrequest to the data file is made instead.
Quoted 6 lines
Signed-off-by Kevin Wolf <[email protected]>--- block/qcow2-cluster.c | 2 +- block/qcow2.c | 34 ++++++++++++++++++++++++++++++++++ 2 files changed, 35 insertions(+), 1 deletion(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v7 09/10] iotests: Test committing to short backing file
In reply to Kevin WolfQuoted 1 line · Kevin Wolf
Signed-off-by Kevin Wolf <[email protected]>
Quoted 1 line
---
[..]
Review
Quoted 2 lines
(quoted diff) +2
-0+ + # After this, 0 to base_size should be allocated/zeroed.
Actually, top_size_old to base_size, yes? (sorry, nitpicking, missed on previous review).
Quoted 9 lines
tests/qemu-iotests/274 new file +9
-0+ # + # In theory, leaving base_size to top_size_new unallocated would be + # correct, but in practice, if we zero out anything, we zero out + # everything up to top_size_new. + iotests.qemu_img_log('resize', '-f', iotests.imgfmt, + '--preallocation', prealloc, top, top_size_new) + iotests.qemu_io_log('-c', 'read -P 0 %s 64k' % off, top) + iotests.qemu_io_log('-c', 'map', top) + iotests.qemu_img_log('map', '--output=json', top)
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v7 10/10] qcow2: Forward ZERO_WRITE flag for full preallocation
In reply to Kevin WolfQuoted 10 lines · Kevin Wolf
The
BDRV_REQ_ZERO_WRITEis currently implemented in a way that first the image is possibly preallocated and then the zero flag is added to all clusters. This means that a copy-on-write operation may be needed when writing to these clusters, despite having used preallocation, negating one of the major benefits of preallocation.Instead, try to forward the
BDRV_REQ_ZERO_WRITEto the protocol driver, and if the protocol driver can ensure that the new area reads as zeros, we can skip setting the zero flag in the qcow2 layer.
Quoted 5 lines
Unfortunately, the same approach doesn't work for metadata preallocation, so we'll still set the zero flag there.
Signed-off-by Kevin Wolf<[email protected]>Reviewed-by Max Reitz<[email protected]>
Signature · 2 lines
Best regards, Vladimir
Am 24.04.2020 um 14:54 hat Kevin Wolf geschrieben:
Quoted 5 lines
v7:
- Allocate smaller zero buffer [Vladimir]
- Added missingerror_setg_errno()[Max]
- Code cleanup in the iotest, enabled mapping for 'metadata' [Vladimir]
- Don't assign to errp twice [Eric]
Thanks for the review, applied to the block branch.
Kevin
Re: [PATCH v7 04/10] qcow2: Support BDRV_REQ_ZERO_WRITE for truncate
In reply to Kevin WolfQuoted 15 lines · Kevin Wolf
If
BDRV_REQ_ZERO_WRITEis set and we're extending the image, callingqcow2_cluster_zeroize()with flags=0 does the right thing: It doesn't undo any previous preallocation, but just adds the zero flag to all relevant L2 entries. If an external data file is in use, awrite_zeroesrequest to the data file is made instead.Signed-off-by Kevin Wolf <[email protected]>--- block/qcow2-cluster.c | 2 +- block/qcow2.c | 34 ++++++++++++++++++++++++++++++++++ 2 files changed, 35 insertions(+), 1 deletion(-)+++
b/block/qcow2.c
block/qcow2.c static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, · line 1729
Quoted 4 lines
block/qcow2.c +1
-0@@ -1726,6 +1726,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, bs->supported_zero_flags = header.version >=3 ?BDRV_REQ_MAY_UNMAP |BDRV_REQ_NO_FALLBACK :0 ; + bs->supported_truncate_flags =BDRV_REQ_ZERO_WRITE ;
Is this really what we want for encrypted files, or would it be better as:
if (bs->encrypted) {
bs->supported_truncate_flags = 0;
} else {
bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
}At the qcow2 level, we can guarantee a read of 0 even for an encrypted image, but is that really what we want? Is setting the qcow2 zero flag on the cluster done at the decrypted level (at which point we may be leaking information about guest contents via anyone that can read the qcow2 metadata) or at the encrypted level (at which point it's useless information, because knowing the underlying file reads as zero still decrypts into garbage)?
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v7 04/10] qcow2: Support BDRV_REQ_ZERO_WRITE for truncate
Am 28.04.2020 um 18:28 hat Eric Blake geschrieben:
Quoted 28 lines · Kevin Wolf
Quoted 15 lines
If
BDRV_REQ_ZERO_WRITEis set and we're extending the image, callingqcow2_cluster_zeroize()with flags=0 does the right thing: It doesn't undo any previous preallocation, but just adds the zero flag to all relevant L2 entries. If an external data file is in use, awrite_zeroesrequest to the data file is made instead.Signed-off-by Kevin Wolf <[email protected]>--- block/qcow2-cluster.c | 2 +- block/qcow2.c | 34 ++++++++++++++++++++++++++++++++++ 2 files changed, 35 insertions(+), 1 deletion(-)+++
b/block/qcow2.cblock/qcow2.c static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, · line 1728
Quoted 3 lines
block/qcow2.c +1
-0@@ -1726,6 +1726,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, bs->supported_zero_flags = header.version >=3 ?BDRV_REQ_MAY_UNMAP |BDRV_REQ_NO_FALLBACK :0 ; + bs->supported_truncate_flags =BDRV_REQ_ZERO_WRITE ;Is this really what we want for encrypted files, or would it be better as:
if (bs->encrypted) { bs->supported_truncate_flags = 0; } else { bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; }
Quoted 7 lines
At the qcow2 level, we can guarantee a read of 0 even for an encrypted image, but is that really what we want? Is setting the qcow2 zero flag on the cluster done at the decrypted level (at which point we may be leaking information about guest contents via anyone that can read the qcow2 metadata) or at the encrypted level (at which point it's useless information, because knowing the underlying file reads as zero still decrypts into garbage)?
The zero flag means that the guest reads zeros, even with encrypted files. I'm not sure if it's worse than exposing the information which clusters are allocated and which are unallocated, which we have always been doing and which is hard to avoid without encrypting all the metadata, too. But it does reveal some information.
If we think that exposing zero flags is worse than exposing the allocation status, I would still not use your solution above. In that case, the full fix would be returning -ENOTSUP from .bdrv_co_pwrite_zeroes() to cover all other callers, too.
If we think that allocation status and zero flags are of comparable importance, then we need to fix either both or nothing. Hiding all of this information probably means encrypting at least the L2 tables and potentially all of the metadata apart from the header. This would obviously require an incompatible feature flag (and some effort to implement it).
Kevin
Re: [PATCH v7 04/10] qcow2: Support BDRV_REQ_ZERO_WRITE for truncate
Quoted 43 lines · Kevin Wolf
Am 28.04.2020 um 18:28 hat Eric Blake geschrieben:
Quoted 28 lines · Kevin Wolf
Quoted 15 lines
If
BDRV_REQ_ZERO_WRITEis set and we're extending the image, callingqcow2_cluster_zeroize()with flags=0 does the right thing: It doesn't undo any previous preallocation, but just adds the zero flag to all relevant L2 entries. If an external data file is in use, awrite_zeroesrequest to the data file is made instead.Signed-off-by Kevin Wolf <[email protected]>--- block/qcow2-cluster.c | 2 +- block/qcow2.c | 34 ++++++++++++++++++++++++++++++++++ 2 files changed, 35 insertions(+), 1 deletion(-)+++
b/block/qcow2.cblock/qcow2.c static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, · line 1728
Quoted 3 lines
block/qcow2.c +1
-0@@ -1726,6 +1726,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, bs->supported_zero_flags = header.version >=3 ?BDRV_REQ_MAY_UNMAP |BDRV_REQ_NO_FALLBACK :0 ; + bs->supported_truncate_flags =BDRV_REQ_ZERO_WRITE ;Is this really what we want for encrypted files, or would it be better as:
if (bs->encrypted) { bs->supported_truncate_flags = 0; } else { bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; }Quoted 7 lines
At the qcow2 level, we can guarantee a read of 0 even for an encrypted image, but is that really what we want? Is setting the qcow2 zero flag on the cluster done at the decrypted level (at which point we may be leaking information about guest contents via anyone that can read the qcow2 metadata) or at the encrypted level (at which point it's useless information, because knowing the underlying file reads as zero still decrypts into garbage)?
The zero flag means that the guest reads zeros, even with encrypted files. I'm not sure if it's worse than exposing the information which clusters are allocated and which are unallocated, which we have always been doing and which is hard to avoid without encrypting all the metadata, too. But it does reveal some information.
Quoted 4 lines
If we think that exposing zero flags is worse than exposing the allocation status, I would still not use your solution above. In that case, the full fix would be returning
-ENOTSUPfrom .bdrv_co_pwrite_zeroes() to cover all other callers, too.
Indeed, it also makes me wonder if we should support truncate(BDRV_REQ_ZERO_WRITE|BDRV_REQ_NO_FALLBACK), to differentiate whether a truncation request is aiming more to be fast (NO_FALLBACK set, fail immediately with -ENOTSUP on encryption) or complete (NO_FALLBACK clear, go ahead and write guest-visible zeroes, which populates the format layer). In other words, maybe we want a knob that the user can set on encrypted volumes on whether to allow zero flags in the qcow2 image.
Quoted 7 lines
If we think that allocation status and zero flags are of comparable importance, then we need to fix either both or nothing. Hiding all of this information probably means encrypting at least the L2 tables and potentially all of the metadata apart from the header. This would obviously require an incompatible feature flag (and some effort to implement it).
Indeed, my question is broad enough that it does not hold up _this_ series, so much as providing food for thought on what else we may need to add for encrypted qcow2 images as a future series, to make it easier to adjust the slider between the extremes of performance vs. minimal data leaks when using encryption.
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org