Re: [PATCH v5 09/11] 9p: darwin: Implement compatibility for mknodat

[Christian Schoenebeck <qemu_oss@crudebyte.com>]

On Dienstag, 8. Februar 2022 16:57:55 CET Will Cohen wrote:
> My inclination is to go with the __builtin_available(macOS 10.12, *) path,
> if acceptable, since it partially mirrors the API_AVAILABLE macro idea. I

OTOH that's duplication of the ">= macOS 10.12" info, plus __builtin_available
is direct use of a clang-only extension, whereas API_AVAILABLE() works (or
more precisely: doesn't error out at least) with other compilers like GCC as
well. GCC is sometimes used for cross-compilation.

Moreover, I would also add an error message in this case, e.g.:

    if (!pthread_fchdir_np) {
        error_report_once("pthread_fchdir_np() is not available on this macOS version");
        return -ENOTSUPP;	
    }

I should elaborate why I think this is needed: you are already doing a Meson
check for the existence of pthread_fchdir_np(), but the system where QEMU is
compiled and the systems where the compiled binary will be running, might be
different ones (i.e. different macOS versions).

Best regards,
Christian Schoenebeck

> guess it's perhaps a tradeoff between predicting the future unknown
> availability of functions versus just ensuring a minimum macOS version and
> hoping for the best. With any luck, the distinction between the two
> approaches will be moot, if we try to assume that a future macOS version
> that removes this also provides mknodat.
> 
> On Tue, Feb 8, 2022 at 10:03 AM Christian Schoenebeck <
> 
> qemu_oss@crudebyte.com> wrote:
> > On Dienstag, 8. Februar 2022 14:36:42 CET Will Cohen wrote:
> > > On Mon, Feb 7, 2022 at 5:56 PM Christian Schoenebeck
> > > <qemu_oss@crudebyte.com>
> > > 
> > > wrote:
> > > > On Montag, 7. Februar 2022 23:40:22 CET Will Cohen wrote:
> > > > > From: Keno Fischer <keno@juliacomputing.com>
> > > > > 
> > > > > Darwin does not support mknodat. However, to avoid race conditions
> > > > > with later setting the permissions, we must avoid using mknod on
> > > > > the full path instead. We could try to fchdir, but that would cause
> > > > > problems if multiple threads try to call mknodat at the same time.
> > > > > However, luckily there is a solution: Darwin includes a function
> > > > > that sets the cwd for the current thread only.
> > > > > This should suffice to use mknod safely.
> > > > > 
> > > > > This function (pthread_fchdir_np) is protected by a check in
> > > > > meson in a patch later in tihs series.
> > > > > 
> > > > > Signed-off-by: Keno Fischer <keno@juliacomputing.com>
> > > > > Signed-off-by: Michael Roitzsch <reactorcontrol@icloud.com>
> > > > > [Will Cohen: - Adjust coding style
> > > > > 
> > > > >              - Replace clang references with gcc
> > > > >              - Note radar filed with Apple for missing syscall
> > > > >              - Replace direct syscall with pthread_fchdir_np and
> > > > >              
> > > > >                adjust patch notes accordingly
> > > > >              
> > > > >              - Move qemu_mknodat from 9p-util to osdep and os-posix]
> > > > > 
> > > > > Signed-off-by: Will Cohen <wwcohen@gmail.com>
> > > > > ---
> > > > 
> > > > Like already mentioned by me moments ago on previous v4 (just echoing)
> > 
> > ...
> > 
> > > > >  hw/9pfs/9p-local.c   |  4 ++--
> > > > >  include/qemu/osdep.h | 10 ++++++++++
> > > > >  os-posix.c           | 34 ++++++++++++++++++++++++++++++++++
> > > > >  3 files changed, 46 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/hw/9pfs/9p-local.c b/hw/9pfs/9p-local.c
> > > > > index a0d08e5216..d42ce6d8b8 100644
> > > > > --- a/hw/9pfs/9p-local.c
> > > > > +++ b/hw/9pfs/9p-local.c
> > > > > @@ -682,7 +682,7 @@ static int local_mknod(FsContext *fs_ctx,
> > 
> > V9fsPath
> > 
> > > > > *dir_path,
> > > > > 
> > > > >      if (fs_ctx->export_flags & V9FS_SM_MAPPED ||
> > > > >      
> > > > >          fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
> > > > > 
> > > > > -        err = mknodat(dirfd, name, fs_ctx->fmode | S_IFREG, 0);
> > > > > +        err = qemu_mknodat(dirfd, name, fs_ctx->fmode | S_IFREG,
> > > > > 0);
> > > > > 
> > > > >          if (err == -1) {
> > > > >          
> > > > >              goto out;
> > > > >          
> > > > >          }
> > > > > 
> > > > > @@ -697,7 +697,7 @@ static int local_mknod(FsContext *fs_ctx,
> > 
> > V9fsPath
> > 
> > > > > *dir_path, }
> > > > > 
> > > > >      } else if (fs_ctx->export_flags & V9FS_SM_PASSTHROUGH ||
> > > > >      
> > > > >                 fs_ctx->export_flags & V9FS_SM_NONE) {
> > > > > 
> > > > > -        err = mknodat(dirfd, name, credp->fc_mode, credp->fc_rdev);
> > > > > +        err = qemu_mknodat(dirfd, name, credp->fc_mode,
> > > > > credp->fc_rdev);
> > > > > 
> > > > >          if (err == -1) {
> > > > >          
> > > > >              goto out;
> > > > >          
> > > > >          }
> > > > > 
> > > > > diff --git a/include/qemu/osdep.h b/include/qemu/osdep.h
> > > > > index d1660d67fa..f3a8367ece 100644
> > > > > --- a/include/qemu/osdep.h
> > > > > +++ b/include/qemu/osdep.h
> > > > > @@ -810,3 +810,13 @@ static inline int
> > > > > platform_does_not_support_system(const char *command) #endif
> > > > > 
> > > > >  #endif
> > > > > 
> > > > > +
> > > > > +/*
> > > > > + * As long as mknodat is not available on macOS, this workaround
> > > > > + * using pthread_fchdir_np is needed. qemu_mknodat is defined in
> > > > > + * os-posix.c
> > > > > + */
> > > > > +#ifdef CONFIG_DARWIN
> > > > > +int pthread_fchdir_np(int fd);
> > > > > +#endif
> > > > 
> > > > I would make that:
> > > > 
> > > > #ifdef CONFIG_DARWIN
> > > > int pthread_fchdir_np(int fd) API_AVAILABLE(macosx(10.12));
> > > > #endif
> > > > 
> > > > here and ...
> > > > 
> > > > > +int qemu_mknodat(int dirfd, const char *filename, mode_t mode,
> > > > > dev_t
> > > > 
> > > > dev);
> > > > 
> > > > > diff --git a/os-posix.c b/os-posix.c
> > > > > index ae6c9f2a5e..95c1607065 100644
> > > > > --- a/os-posix.c
> > > > > +++ b/os-posix.c
> > > > > @@ -24,6 +24,7 @@
> > > > > 
> > > > >   */
> > > > >  
> > > > >  #include "qemu/osdep.h"
> > > > > 
> > > > > +#include <os/availability.h>
> > > > > 
> > > > >  #include <sys/wait.h>
> > > > >  #include <pwd.h>
> > > > >  #include <grp.h>
> > > > > 
> > > > > @@ -332,3 +333,36 @@ int os_mlock(void)
> > > > > 
> > > > >      return -ENOSYS;
> > > > >  
> > > > >  #endif
> > > > >  }
> > > > > 
> > > > > +
> > > > > +/*
> > > > > + * As long as mknodat is not available on macOS, this workaround
> > > > > + * using pthread_fchdir_np is needed.
> > > > > + *
> > > > > + * Radar filed with Apple for implementing mknodat:
> > > > > + * rdar://FB9862426 (https://openradar.appspot.com/FB9862426)
> > > > > + */
> > > > > +#ifdef CONFIG_DARWIN
> > > > > +
> > > > > +int pthread_fchdir_np(int fd) API_AVAILABLE(macosx(10.12));
> > > > 
> > > > ... drop the duplicate declaration of pthread_fchdir_np() here.
> > > 
> > > Trying this out, it reminds me that this use of API_AVAILABLE in
> > 
> > os-posix.c
> > 
> > > relies on the added #include <os/availability.h>.
> > > 
> > > Leaving the include out leads to:
> > > .../include/qemu/osdep.h:820:31: error: expected function body after
> > > function declarator
> > > int pthread_fchdir_np(int fd) API_AVAILABLE(macosx(10.12));
> > > 
> > >                               ^
> > > 
> > > 1 error generated.
> > > ninja: build stopped: subcommand failed.
> > > make[1]: *** [run-ninja] Error 1
> > > make: *** [all] Error 2
> > > 
> > > The admonition against modifying osdep.h's includes too much led me to
> > > steer away from putting it all in there. If there's no issue with adding
> > > +#include <os/availability.h> to osdep.h, then this change makes sense
> > > to
> > > me.
> > 
> > If you embed that include into ifdefs, sure!
> > 
> > #ifdef CONFIG_DARWIN
> > /* defines API_AVAILABLE(...) */
> > #include <os/availability.h>
> > #endif
> > 
> > One more thing though ...
> > 
> > > > > +
> > > > > +int qemu_mknodat(int dirfd, const char *filename, mode_t mode,
> > > > > dev_t
> > > > 
> > > > dev)
> > > > 
> > > > > +{
> > > > > +    int preserved_errno, err;
> > 
> > pthread_fchdir_np() is weakly linked. So I guess here should be a check
> > 
> > like:
> >         if (!pthread_fchdir_np) {
> >         
> >                 return -ENOTSUPP;
> >         
> >         }
> > 
> > Before trying to call pthread_fchdir_np() below. As already discussed with
> > the
> > Chromium [1] example, some do that a bit differently by using
> > 
> > __builtin_available():
> >         if (__builtin_available(macOS 10.12, *)) {
> >         
> >                 return -ENOTSUPP;
> >         
> >         }
> > 
> > Which makes me wonder why they are not doing a simple NULL check?
> > 
> > [1]
> > https://chromium.googlesource.com/chromium/src/+/lkgr/base/process/launch_
> > mac.cc#110> 
> > > > > +    if (pthread_fchdir_np(dirfd) < 0) {
> > > > > +        return -1;
> > > > > +    }
> > > > > +    err = mknod(filename, mode, dev);
> > > > > +    preserved_errno = errno;
> > > > > +    /* Stop using the thread-local cwd */
> > > > > +    pthread_fchdir_np(-1);
> > > > > +    if (err < 0) {
> > > > > +        errno = preserved_errno;
> > > > > +    }
> > > > > +    return err;
> > > > > +}
> > > > > +#else
> > > > > +int qemu_mknodat(int dirfd, const char *filename, mode_t mode,
> > > > > dev_t
> > > > 
> > > > dev)
> > > > 
> > > > > +{
> > > > > +    return mknodat(dirfd, filename, mode, dev);
> > > > > +}
> > > > > +#endif