Message ID | 1311680948-7648-10-git-send-email-kwolf@redhat.com |
---|---|
State | New |
Headers | show |
2011/7/26 Kevin Wolf <kwolf@redhat.com>: > In order to be able to transparently replace bdrv_read calls by bdrv_co_read, > reading beyond EOF must produce zeros instead of short reads for AIO, too. > > Signed-off-by: Kevin Wolf <kwolf@redhat.com> > --- > posix-aio-compat.c | 19 +++++++++++++++++++ > 1 files changed, 19 insertions(+), 0 deletions(-) > > diff --git a/posix-aio-compat.c b/posix-aio-compat.c > index 788d113..8dc00cb 100644 > --- a/posix-aio-compat.c > +++ b/posix-aio-compat.c > @@ -198,6 +198,12 @@ static ssize_t handle_aiocb_rw_vector(struct qemu_paiocb *aiocb) > return len; > } > > +/* > + * Read/writes the data to/from a given linear buffer. > + * > + * Returns the number of bytes handles or -errno in case of an error. Short > + * reads are only returned if the end of the file is reached. > + */ > static ssize_t handle_aiocb_rw_linear(struct qemu_paiocb *aiocb, char *buf) > { > ssize_t offset = 0; > @@ -334,6 +340,19 @@ static void *aio_thread(void *unused) > > switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { > case QEMU_AIO_READ: > + ret = handle_aiocb_rw(aiocb); > + if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->common.bs->growable) { > + /* A short read means that we have reached EOF. Pad the buffer > + * with zeros for bytes after EOF. */ > + QEMUIOVector qiov; > + > + qemu_iovec_init_external(&qiov, aiocb->aio_iov, > + aiocb->aio_niov); > + qemu_iovec_memset_skip(&qiov, 0, aiocb->aio_nbytes - ret, ret); > + > + ret = aiocb->aio_nbytes; > + } > + break; > case QEMU_AIO_WRITE: > ret = handle_aiocb_rw(aiocb); > break; > -- > 1.7.6 > Still not tested but I think to know what does it solve :) I think Linux AIO require same attention. Frediano
Am 26.07.2011 15:55, schrieb Frediano Ziglio: > 2011/7/26 Kevin Wolf <kwolf@redhat.com>: >> In order to be able to transparently replace bdrv_read calls by bdrv_co_read, >> reading beyond EOF must produce zeros instead of short reads for AIO, too. >> >> Signed-off-by: Kevin Wolf <kwolf@redhat.com> >> --- >> posix-aio-compat.c | 19 +++++++++++++++++++ >> 1 files changed, 19 insertions(+), 0 deletions(-) >> >> diff --git a/posix-aio-compat.c b/posix-aio-compat.c >> index 788d113..8dc00cb 100644 >> --- a/posix-aio-compat.c >> +++ b/posix-aio-compat.c >> @@ -198,6 +198,12 @@ static ssize_t handle_aiocb_rw_vector(struct qemu_paiocb *aiocb) >> return len; >> } >> >> +/* >> + * Read/writes the data to/from a given linear buffer. >> + * >> + * Returns the number of bytes handles or -errno in case of an error. Short >> + * reads are only returned if the end of the file is reached. >> + */ >> static ssize_t handle_aiocb_rw_linear(struct qemu_paiocb *aiocb, char *buf) >> { >> ssize_t offset = 0; >> @@ -334,6 +340,19 @@ static void *aio_thread(void *unused) >> >> switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { >> case QEMU_AIO_READ: >> + ret = handle_aiocb_rw(aiocb); >> + if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->common.bs->growable) { >> + /* A short read means that we have reached EOF. Pad the buffer >> + * with zeros for bytes after EOF. */ >> + QEMUIOVector qiov; >> + >> + qemu_iovec_init_external(&qiov, aiocb->aio_iov, >> + aiocb->aio_niov); >> + qemu_iovec_memset_skip(&qiov, 0, aiocb->aio_nbytes - ret, ret); >> + >> + ret = aiocb->aio_nbytes; >> + } >> + break; >> case QEMU_AIO_WRITE: >> ret = handle_aiocb_rw(aiocb); >> break; >> -- >> 1.7.6 >> > > Still not tested but I think to know what does it solve :) > > I think Linux AIO require same attention. In theory yes, but it's not as easy and for some reason I couldn't reproduce it with Linux AIO (maybe the problematic requests are misaligned so that it falls back to posix-aio-compat.c), so I decided to ignore it for now. Patches are welcome. ;-) Kevin
diff --git a/posix-aio-compat.c b/posix-aio-compat.c index 788d113..8dc00cb 100644 --- a/posix-aio-compat.c +++ b/posix-aio-compat.c @@ -198,6 +198,12 @@ static ssize_t handle_aiocb_rw_vector(struct qemu_paiocb *aiocb) return len; } +/* + * Read/writes the data to/from a given linear buffer. + * + * Returns the number of bytes handles or -errno in case of an error. Short + * reads are only returned if the end of the file is reached. + */ static ssize_t handle_aiocb_rw_linear(struct qemu_paiocb *aiocb, char *buf) { ssize_t offset = 0; @@ -334,6 +340,19 @@ static void *aio_thread(void *unused) switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { case QEMU_AIO_READ: + ret = handle_aiocb_rw(aiocb); + if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->common.bs->growable) { + /* A short read means that we have reached EOF. Pad the buffer + * with zeros for bytes after EOF. */ + QEMUIOVector qiov; + + qemu_iovec_init_external(&qiov, aiocb->aio_iov, + aiocb->aio_niov); + qemu_iovec_memset_skip(&qiov, 0, aiocb->aio_nbytes - ret, ret); + + ret = aiocb->aio_nbytes; + } + break; case QEMU_AIO_WRITE: ret = handle_aiocb_rw(aiocb); break;
In order to be able to transparently replace bdrv_read calls by bdrv_co_read, reading beyond EOF must produce zeros instead of short reads for AIO, too. Signed-off-by: Kevin Wolf <kwolf@redhat.com> --- posix-aio-compat.c | 19 +++++++++++++++++++ 1 files changed, 19 insertions(+), 0 deletions(-)