2013/11/12 Fam Zheng <f...@redhat.com>
> On 2013年11月07日 16:08, Chunyan Liu wrote:
>
>> Set NOCOW flag to newly created images to solve performance issues on
>> btrfs.
>>
>> Btrfs has terrible performance when hosting VM images, even more when the
>> guest
>> in those VM are also using btrfs as file system. One way to mitigate this
>> bad
>> performance is to turn off COW attributes on VM files (since having copy
>> on
>> write for this kind of data is not useful).
>>
>> Signed-off-by: Chunyan Liu <cy...@suse.com>
>> ---
>> qemu-img.c | 15 +++++++++++++++
>> 1 files changed, 15 insertions(+), 0 deletions(-)
>>
>> diff --git a/qemu-img.c b/qemu-img.c
>> index bf3fb4f..d43e8f1 100644
>> --- a/qemu-img.c
>> +++ b/qemu-img.c
>> @@ -34,11 +34,17 @@
>> #include <getopt.h>
>> #include <stdio.h>
>> #include <stdarg.h>
>> +#include <linux/fs.h>
>> +#include <sys/ioctl.h>
>>
>
> This should be in #ifdef to not break build on Windows.
>
>
> #ifdef _WIN32
>> #include <windows.h>
>> #endif
>>
>> +#ifndef FS_NOCOW_FL
>> +#define FS_NOCOW_FL 0x00800000 /* Do not cow file */
>> +#endif
>> +
>> typedef struct img_cmd_t {
>> const char *name;
>> int (*handler)(int argc, char **argv);
>> @@ -340,6 +346,7 @@ static int img_create(int argc, char **argv)
>> char *options = NULL;
>> Error *local_err = NULL;
>> bool quiet = false;
>> + int fd, attr;
>>
>> for(;;) {
>> c = getopt(argc, argv, "F:b:f:he6o:q");
>> @@ -417,6 +424,14 @@ static int img_create(int argc, char **argv)
>> return 1;
>> }
>>
>> + /* set NOCOW by default to solve performance issue on btrfs */
>> + fd = qemu_open(filename, O_RDONLY|O_NONBLOCK);
>> + if (fd >= 0) {
>> + attr = FS_NOCOW_FL;
>> + ioctl(fd, FS_IOC_SETFLAGS, &attr);
>> + qemu_close(fd);
>> + }
>> +
>> return 0;
>> }
>>
>>
>> "man chattr" says:
> ... For btrfs, the 'C' flag should be set on new or empty files. If it is
> set on a file which already has data blocks, it is undefined when the
> blocks assigned to the file will be fully stable...
>
> But you are setting the attr after image creation, so what's the
> difference here? just want to make sure this does what's expected.
>
I expected such question. Following the man page, I should add changes in
each related block driver's drv_create, that seems too much changes. Just
to make the change as little as possible, I asked in #btrfs irc about the
reason why must empty and if it affects in our case. Maybe I misunderstand
the reply, but now I check btrfs/ioctl.c, I should say I made a mistake
here. If the file size is not 0, the COW flag won't be removed. Sorry!
Still need to change in block driver's drv_create function. I'll revise.
Thanks for correction.
> Thanks,
> Fam
>
>
