O_DIRECT is under control of the caller (though the --bypass-cache option). I am running in a constrained memory environment and when I don't use O_DIRECT the dump stalls at 5% load av hits 41 and the system becomes unresponsive.
Simon
From: Michal Prívozník <mprivozn@xxxxxxxxxx>
Sent: 20 August 2021 16:09 To: Simon Rowe <simon.rowe@xxxxxxxxxxx>; libvir-list@xxxxxxxxxx <libvir-list@xxxxxxxxxx> Subject: Re: [PATCH 1/2] iohelper: skip lseek() and ftruncate() on block devices On 8/20/21 10:39 AM, Simon Rowe wrote:
> Signed-off-by: Simon Rowe <simon.rowe@xxxxxxxxxxx> > --- > src/util/iohelper.c | 10 ++++++++-- > 1 file changed, 8 insertions(+), 2 deletions(-) > > diff --git a/src/util/iohelper.c b/src/util/iohelper.c > index b8810d16d3..e6eb178fde 100644 > --- a/src/util/iohelper.c > +++ b/src/util/iohelper.c > @@ -28,6 +28,8 @@ > #include <unistd.h> > #include <fcntl.h> > #include <stdlib.h> > +#include <sys/types.h> > +#include <sys/stat.h> > > #include "virthread.h" > #include "virfile.h" > @@ -56,6 +58,8 @@ runIO(const char *path, int fd, int oflags) > unsigned long long total = 0; > bool direct = O_DIRECT && ((oflags & O_DIRECT) != 0); > off_t end = 0; > + struct stat sb; > + bool isBlockDev = false; > > #if WITH_POSIX_MEMALIGN > if (posix_memalign(&base, alignMask + 1, buflen)) > @@ -86,9 +90,11 @@ runIO(const char *path, int fd, int oflags) > fdinname = "stdin"; > fdout = fd; > fdoutname = path; > + if (fstat(fd, &sb) == 0) > + isBlockDev = S_ISBLK(sb.st_mode); > /* To make the implementation simpler, we give up on any > * attempt to use O_DIRECT in a non-trivial manner. */ > - if (direct && (end = lseek(fd, 0, SEEK_END)) != 0) { > + if (!isBlockDev && direct && (end = lseek(fd, 0, SEEK_END)) != 0) { > virReportSystemError(end < 0 ? errno : EINVAL, "%s", > _("O_DIRECT write needs empty seekable file")); > goto cleanup; > @@ -140,7 +146,7 @@ runIO(const char *path, int fd, int oflags) > goto cleanup; > } > > - if (ftruncate(fd, total) < 0) { > + if (!isBlockDev && ftruncate(fd, total) < 0) { > virReportSystemError(errno, _("Unable to truncate %s"), fdoutname); > goto cleanup; > } > IIUC, O_DIRECT is no good for block devices. I wonder whether the caller (doCoreDump()) should learn whether the path is a block device and don't set O_DIRECT flag instead of changing iohelper. Michal |