Users may give "-mem-path" a read only file and expect the file to be mapped read-write privately. Allow this but give a warning since other users may surprise when the ram file is readonly and qemu suddenly aborts elsewhere.
Suggested-by: David Hildenbrand <da...@redhat.com> Signed-off-by: Thiner Logoer <logoerthin...@163.com> --- See the previous version at: https://lore.kernel.org/qemu-devel/96a462ec-6f9d-fd83-f697-73e132432...@redhat.com/T/ verified, this patch works for my setup, both functionality and the warning are expected behavior. Also another problem when I look at the file_ram_open When readonly is true and the path is a directory, the open will succeed but any later operations will fail since it is a directory fd. This may require additional commits which is out of my scope. Merely record the question here. softmmu/physmem.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+) diff --git a/softmmu/physmem.c b/softmmu/physmem.c index 3df73542e1..e8279d69d4 100644 --- a/softmmu/physmem.c +++ b/softmmu/physmem.c @@ -1296,6 +1296,7 @@ static int file_ram_open(const char *path, char *sanitized_name; char *c; int fd = -1; + bool first_trial = true; *created = false; for (;;) { @@ -1332,6 +1333,18 @@ static int file_ram_open(const char *path, break; } g_free(filename); + } else if (first_trial && !readonly && errno == EACCES) { + /* @path may be a read only file */ + fd = open(path, O_RDONLY); + if (fd >= 0) { + /* + * Sometimes this behavior is not desired. Fire a warning but + * continue. + */ + warn_report("backing store %s is opened readonly because the" + "file is not writable", path); + break; + } } if (errno != EEXIST && errno != EINTR) { error_setg_errno(errp, errno, @@ -1343,6 +1356,7 @@ static int file_ram_open(const char *path, * Try again on EINTR and EEXIST. The latter happens when * something else creates the file between our two open(). */ + first_trial = false; } return fd; -- 2.40.1
