Skip to contentSkip to conversation

9p: Add support for darwin

PATCH Superseded Will Cohen 2 participants 20 messages Feb 20, 2022 – Feb 25, 2022 Arm/Amlogic Meson SoC support

v8 of 9 Superseded Latest is v9 →

lore
Share & tools
lore

Will Cohen

author
#0 lore raw

v8 cover 9p: Add support for darwin

This is a followup to https://lists.gnu.org/archive/html/qemu-devel/2022-02/msg04298.html,
adding 9p server support for Darwin.

Since v7, no functional changes have been made to this patch set, but Patch 9/11 (9p: darwin: Implement compatibility for mknodat) was rebased to apply cleanly on top of the most recent changes to 9pfs, which affected the code changes osdep.h directly above patch 9’s additions.

WIth these changes, v8 correctly applies and functions on the latest mainline qemu.

Keno Fischer (10):
9p: linux: Fix a couple Linux assumptions 9p: Rename 9p-util -> 9p-util-linux
9p: darwin: Handle struct stat(fs) differences 9p: darwin: Handle struct dirent differences 9p: darwin: Ignore O_{NOATIME, DIRECT}
9p: darwin: Move XATTR_SIZE_MAX->P9_XATTR_SIZE_MAX 9p: darwin: *xattr_nofollow implementations 9p: darwin: Compatibility for f/l*xattr 9p: darwin: Implement compatibility for mknodat 9p: darwin: meson: Allow VirtFS on Darwin

Will Cohen (1):
9p: darwin: Adjust assumption on virtio-9p-test

 fsdev/file-op-9p.h                     |  9 +++-
 fsdev/meson.build                      |  1 +
 hw/9pfs/9p-local.c                     | 27 ++++++++---
 hw/9pfs/9p-proxy.c                     | 38 +++++++++++++--
 hw/9pfs/9p-synth.c                     |  6 +++
 hw/9pfs/9p-util-darwin.c               | 64 ++++++++++++++++++++++++++
 hw/9pfs/{9p-util.c => 9p-util-linux.c} |  2 +-
 hw/9pfs/9p-util.h                      | 35 ++++++++++++++
 hw/9pfs/9p.c                           | 42 ++++++++++++++---
 hw/9pfs/9p.h                           | 18 ++++++++
 hw/9pfs/codir.c                        |  4 +-
 hw/9pfs/meson.build                    |  3 +-
 include/qemu/osdep.h                   | 12 +++++
 include/qemu/xattr.h                   |  4 +-
 meson.build                            | 15 ++++--
 os-posix.c                             | 35 ++++++++++++++
 tests/qtest/virtio-9p-test.c           |  2 +-
 17 files changed, 292 insertions(+), 25 deletions(-)
 create mode 100644 hw/9pfs/9p-util-darwin.c
 rename hw/9pfs/{9p-util.c => 9p-util-linux.c} (97%)
Signature · 1 line
2.35.1
author
#7 lore raw

PATCH v8 5/11 9p: darwin: Ignore O_{NOATIME, DIRECT}

From: Keno Fischer <[email protected]>

Darwin doesn't have either of these flags. Darwin does have F_NOCACHE, which is similar to O_DIRECT, but has different enough semantics that other projects don't generally map them automatically. In any case, we don't support O_DIRECT on Linux at the moment either.

[Michael Roitzsch: - Rebase for NixOS]
[Will Cohen: - Adjust coding style]
---
 hw/9pfs/9p-util.h |  2 ++
 hw/9pfs/9p.c      | 13 ++++++++++++-
 2 files changed, 14 insertions(+), 1 deletion(-)
hw/9pfs/9p-util.h +2 -0
index d41f37f085..0e445b5d52 100644
@@ -41,6 +41,7 @@ again:
     fd = openat(dirfd, name, flags | O_NOFOLLOW | O_NOCTTY | O_NONBLOCK,
                 mode);
     if (fd == -1) {
+#ifndef CONFIG_DARWIN
         if (errno == EPERM && (flags & O_NOATIME)) {
             /*
              * The client passed O_NOATIME but we lack permissions to honor it.
@@ -53,6 +54,7 @@ again:
             flags &= ~O_NOATIME;
             goto again;
         }
+#endif
         return -1;
     }
 
hw/9pfs/9p.c +12 -1
index caf3b240fe..14e84c3bcf 100644
@@ -138,11 +138,20 @@ static int dotl_to_open_flags(int flags)
         { P9_DOTL_NONBLOCK, O_NONBLOCK } ,
         { P9_DOTL_DSYNC, O_DSYNC },
         { P9_DOTL_FASYNC, FASYNC },
+#ifndef CONFIG_DARWIN
+        { P9_DOTL_NOATIME, O_NOATIME },
+        /*
+         *  On Darwin, we could map to F_NOCACHE, which is
+         *  similar, but doesn't quite have the same
+         *  semantics. However, we don't support O_DIRECT
+         *  even on linux at the moment, so we just ignore
+         *  it here.
+         */
         { P9_DOTL_DIRECT, O_DIRECT },
+#endif
         { P9_DOTL_LARGEFILE, O_LARGEFILE },
         { P9_DOTL_DIRECTORY, O_DIRECTORY },
         { P9_DOTL_NOFOLLOW, O_NOFOLLOW },
-        { P9_DOTL_NOATIME, O_NOATIME },
         { P9_DOTL_SYNC, O_SYNC },
     };
 
@@ -171,10 +180,12 @@ static int get_dotl_openflags(V9fsState *s, int oflags)
      */
     flags = dotl_to_open_flags(oflags);
     flags &= ~(O_NOCTTY | O_ASYNC | O_CREAT);
+#ifndef CONFIG_DARWIN
     /*
      * Ignore direct disk access hint until the server supports it.
      */
     flags &= ~O_DIRECT;
+#endif
     return flags;
 }
 
Signature · 1 line
2.35.1
author
#2 lore raw

PATCH v8 2/11 9p: Rename 9p-util -> 9p-util-linux

In reply to Will Cohen

From: Keno Fischer <[email protected]>

The current file only has the Linux versions of these functions. Rename the file accordingly and update the Makefile to only build it on Linux. A Darwin version of these will follow later in the series.

---
 hw/9pfs/{9p-util.c => 9p-util-linux.c} | 2 +-
 hw/9pfs/meson.build                    | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
 rename hw/9pfs/{9p-util.c => 9p-util-linux.c} (97%)
hw/9pfs/9p-util-linux.c renamed from hw/9pfs/9p-util.c +1 -1
similarity index 97%
rename from hw/9pfs/9p-util.c
rename to hw/9pfs/9p-util-linux.c
index 3221d9b498..398614a5d0 100644
@@ -1,5 +1,5 @@
 /*
- * 9p utilities
+ * 9p utilities (Linux Implementation)
  *
  * Copyright IBM, Corp. 2017
  *
hw/9pfs/meson.build +1 -1
index 99be5d9119..1b28e70040 100644
@@ -4,7 +4,6 @@ fs_ss.add(files(
   '9p-posix-acl.c',
   '9p-proxy.c',
   '9p-synth.c',
-  '9p-util.c',
   '9p-xattr-user.c',
   '9p-xattr.c',
   '9p.c',
@@ -14,6 +13,7 @@ fs_ss.add(files(
   'coth.c',
   'coxattr.c',
 ))
+fs_ss.add(when: 'CONFIG_LINUX', if_true: files('9p-util-linux.c'))
 fs_ss.add(when: 'CONFIG_XEN', if_true: files('xen-9p-backend.c'))
 softmmu_ss.add_all(when: 'CONFIG_FSDEV_9P', if_true: fs_ss)
 
Signature · 1 line
2.35.1
author
#3 lore raw

PATCH v8 3/11 9p: darwin: Handle struct stat(fs) differences

In reply to Will Cohen

From: Keno Fischer <[email protected]>

[Will Cohen: - Note lack of f_namelen and f_frsize on Darwin
- Ensure that tv_sec and tv_nsec are both initialized for Darwin and non-Darwin]

---
 hw/9pfs/9p-proxy.c | 22 ++++++++++++++++++++--
 hw/9pfs/9p-synth.c |  2 ++
 hw/9pfs/9p.c       | 16 ++++++++++++++--
 3 files changed, 36 insertions(+), 4 deletions(-)
hw/9pfs/9p-proxy.c +20 -2
index 09bd9f1464..b1664080d8 100644
@@ -123,10 +123,16 @@ static void prstatfs_to_statfs(struct statfs *stfs, ProxyStatFS *prstfs)
     stfs->f_bavail = prstfs->f_bavail;
     stfs->f_files = prstfs->f_files;
     stfs->f_ffree = prstfs->f_ffree;
+#ifdef CONFIG_DARWIN
+    /* f_namelen and f_frsize do not exist on Darwin */
+    stfs->f_fsid.val[0] = prstfs->f_fsid[0] & 0xFFFFFFFFU;
+    stfs->f_fsid.val[1] = prstfs->f_fsid[1] >> 32 & 0xFFFFFFFFU;
+#else
     stfs->f_fsid.__val[0] = prstfs->f_fsid[0] & 0xFFFFFFFFU;
     stfs->f_fsid.__val[1] = prstfs->f_fsid[1] >> 32 & 0xFFFFFFFFU;
     stfs->f_namelen = prstfs->f_namelen;
     stfs->f_frsize = prstfs->f_frsize;
+#endif
 }
 
 /* Converts proxy_stat structure to VFS stat structure */
@@ -143,12 +149,24 @@ static void prstat_to_stat(struct stat *stbuf, ProxyStat *prstat)
    stbuf->st_size = prstat->st_size;
    stbuf->st_blksize = prstat->st_blksize;
    stbuf->st_blocks = prstat->st_blocks;
+   stbuf->st_atime = prstat->st_atim_sec;
+   stbuf->st_mtime = prstat->st_mtim_sec;
+   stbuf->st_ctime = prstat->st_ctim_sec;
+#ifdef CONFIG_DARWIN
+   stbuf->st_atimespec.tv_sec = prstat->st_atim_sec;
+   stbuf->st_mtimespec.tv_sec = prstat->st_mtim_sec;
+   stbuf->st_ctimespec.tv_sec = prstat->st_ctim_sec;
+   stbuf->st_atimespec.tv_nsec = prstat->st_atim_nsec;
+   stbuf->st_mtimespec.tv_nsec = prstat->st_mtim_nsec;
+   stbuf->st_ctimespec.tv_nsec = prstat->st_ctim_nsec;
+#else
    stbuf->st_atim.tv_sec = prstat->st_atim_sec;
+   stbuf->st_mtim.tv_sec = prstat->st_mtim_sec;
+   stbuf->st_ctim.tv_sec = prstat->st_ctim_sec;
    stbuf->st_atim.tv_nsec = prstat->st_atim_nsec;
-   stbuf->st_mtime = prstat->st_mtim_sec;
    stbuf->st_mtim.tv_nsec = prstat->st_mtim_nsec;
-   stbuf->st_ctime = prstat->st_ctim_sec;
    stbuf->st_ctim.tv_nsec = prstat->st_ctim_nsec;
+#endif
 }
 
 /*
hw/9pfs/9p-synth.c +2 -0
index 7a7cd5c5ba..bf9b0c5ddd 100644
@@ -439,7 +439,9 @@ static int synth_statfs(FsContext *s, V9fsPath *fs_path,
     stbuf->f_bsize = 512;
     stbuf->f_blocks = 0;
     stbuf->f_files = synth_node_count;
+#ifndef CONFIG_DARWIN
     stbuf->f_namelen = NAME_MAX;
+#endif
     return 0;
 }
 
hw/9pfs/9p.c +14 -2
index 9c63e14b28..1563d7b7c6 100644
@@ -1313,11 +1313,17 @@ static int stat_to_v9stat_dotl(V9fsPDU *pdu, const struct stat *stbuf,
     v9lstat->st_blksize = stat_to_iounit(pdu, stbuf);
     v9lstat->st_blocks = stbuf->st_blocks;
     v9lstat->st_atime_sec = stbuf->st_atime;
-    v9lstat->st_atime_nsec = stbuf->st_atim.tv_nsec;
     v9lstat->st_mtime_sec = stbuf->st_mtime;
-    v9lstat->st_mtime_nsec = stbuf->st_mtim.tv_nsec;
     v9lstat->st_ctime_sec = stbuf->st_ctime;
+#ifdef CONFIG_DARWIN
+    v9lstat->st_atime_nsec = stbuf->st_atimespec.tv_nsec;
+    v9lstat->st_mtime_nsec = stbuf->st_mtimespec.tv_nsec;
+    v9lstat->st_ctime_nsec = stbuf->st_ctimespec.tv_nsec;
+#else
+    v9lstat->st_atime_nsec = stbuf->st_atim.tv_nsec;
+    v9lstat->st_mtime_nsec = stbuf->st_mtim.tv_nsec;
     v9lstat->st_ctime_nsec = stbuf->st_ctim.tv_nsec;
+#endif
     /* Currently we only support BASIC fields in stat */
     v9lstat->st_result_mask = P9_STATS_BASIC;
 
@@ -3519,9 +3525,15 @@ static int v9fs_fill_statfs(V9fsState *s, V9fsPDU *pdu, struct statfs *stbuf)
     f_bavail = stbuf->f_bavail / bsize_factor;
     f_files  = stbuf->f_files;
     f_ffree  = stbuf->f_ffree;
+#ifdef CONFIG_DARWIN
+    fsid_val = (unsigned int)stbuf->f_fsid.val[0] |
+               (unsigned long long)stbuf->f_fsid.val[1] << 32;
+    f_namelen = NAME_MAX;
+#else
     fsid_val = (unsigned int) stbuf->f_fsid.__val[0] |
                (unsigned long long)stbuf->f_fsid.__val[1] << 32;
     f_namelen = stbuf->f_namelen;
+#endif
 
     return pdu_marshal(pdu, offset, "ddqqqqqqd",
                        f_type, f_bsize, f_blocks, f_bfree,
Signature · 1 line
2.35.1
author
#1 lore raw

PATCH v8 1/11 9p: linux: Fix a couple Linux assumptions

In reply to Will Cohen

From: Keno Fischer <[email protected]>

- Guard Linux only headers.
- Add qemu/statfs.h header to abstract over the which headers are needed for struct statfs
- Define ENOATTR only if not only defined (it's defined in system headers on Darwin).

[Michael Roitzsch: - Rebase for NixOS]

While it might at first appear that fsdev/virtfs-proxy-header.c would need similar adjustment for darwin as file-op-9p here, a later patch in this series disables virtfs-proxy-helper for non-Linux. Allowing virtfs-proxy-helper on darwin could potentially be an additional optimization later.

[Will Cohen: - Fix headers for Alpine
- Integrate statfs.h back into file-op-9p.h
- Remove superfluous header guards from file-opt-9p
- Add note about virtfs-proxy-helper being disabled on non-Linux for this patch series]

---
 fsdev/file-op-9p.h   | 9 ++++++++-
 hw/9pfs/9p-local.c   | 2 ++
 hw/9pfs/9p.c         | 4 ++++
 include/qemu/xattr.h | 4 +++-
 4 files changed, 17 insertions(+), 2 deletions(-)
fsdev/file-op-9p.h +8 -1
index 8fd89f0447..4997677460 100644
@@ -16,10 +16,17 @@
 
 #include <dirent.h>
 #include <utime.h>
-#include <sys/vfs.h>
 #include "qemu-fsdev-throttle.h"
 #include "p9array.h"
 
+#ifdef CONFIG_LINUX
+# include <sys/vfs.h>
+#endif
+#ifdef CONFIG_DARWIN
+# include <sys/param.h>
+# include <sys/mount.h>
+#endif
+
 #define SM_LOCAL_MODE_BITS    0600
 #define SM_LOCAL_DIR_MODE_BITS    0700
 
hw/9pfs/9p-local.c +2 -0
index 210d9e7705..1a5e3eed73 100644
@@ -32,10 +32,12 @@
 #include "qemu/error-report.h"
 #include "qemu/option.h"
 #include <libgen.h>
+#ifdef CONFIG_LINUX
 #include <linux/fs.h>
 #ifdef CONFIG_LINUX_MAGIC_H
 #include <linux/magic.h>
 #endif
+#endif
 #include <sys/ioctl.h>
 
 #ifndef XFS_SUPER_MAGIC
hw/9pfs/9p.c +4 -0
index 15b3f4d385..9c63e14b28 100644
@@ -32,7 +32,11 @@
 #include "migration/blocker.h"
 #include "qemu/xxhash.h"
 #include <math.h>
+#ifdef CONFIG_LINUX
 #include <linux/limits.h>
+#else
+#include <limits.h>
+#endif
 
 int open_fd_hw;
 int total_open_fd;
include/qemu/xattr.h +3 -1
index a83fe8e749..f1d0f7be74 100644
@@ -22,7 +22,9 @@
 #ifdef CONFIG_LIBATTR
 #  include <attr/xattr.h>
 #else
-#  define ENOATTR ENODATA
+#  if !defined(ENOATTR)
+#    define ENOATTR ENODATA
+#  endif
 #  include <sys/xattr.h>
 #endif
 
Signature · 1 line
2.35.1
author
#4 lore raw

PATCH v8 4/11 9p: darwin: Handle struct dirent differences

In reply to Will Cohen

From: Keno Fischer <[email protected]>

On darwin d_seekoff exists, but is optional and does not seem to be commonly used by file systems. Use telldir instead to obtain the seek offset and inject it into d_seekoff, and create a qemu_dirent_off helper to call it appropriately when appropriate.

[Michael Roitzsch: - Rebase for NixOS]

[Will Cohen: - Adjust to pass testing
- Ensure that d_seekoff is filled using telldir on darwin, and create qemu_dirent_off helper to decide which to access]

[Fabian Franz: - Add telldir error handling for darwin]

[Will Cohen: - Ensure that telldir error handling uses
signed int
- Cleanup of telldir error handling
- Remove superfluous error handling for qemu_dirent_off
- Adjust formatting
- Use qemu_dirent_off in codir.c]

---
 hw/9pfs/9p-local.c |  9 +++++++++
 hw/9pfs/9p-proxy.c | 16 +++++++++++++++-
 hw/9pfs/9p-synth.c |  4 ++++
 hw/9pfs/9p-util.h  | 16 ++++++++++++++++
 hw/9pfs/9p.c       |  7 +++++--
 hw/9pfs/codir.c    |  4 +++-
 6 files changed, 52 insertions(+), 4 deletions(-)
hw/9pfs/9p-local.c +9 -0
index 1a5e3eed73..f3272f0b43 100644
@@ -562,6 +562,15 @@ again:
     if (!entry) {
         return NULL;
     }
+#ifdef CONFIG_DARWIN
+    int off;
+    off = telldir(fs->dir.stream);
+    /* If telldir fails, fail the entire readdir call */
+    if (off < 0) {
+        return NULL;
+    }
+    entry->d_seekoff = off;
+#endif
 
     if (ctx->export_flags & V9FS_SM_MAPPED) {
         entry->d_type = DT_UNKNOWN;
hw/9pfs/9p-proxy.c +15 -1
index b1664080d8..8b4b5cf7dc 100644
@@ -706,7 +706,21 @@ static off_t proxy_telldir(FsContext *ctx, V9fsFidOpenState *fs)
 
 static struct dirent *proxy_readdir(FsContext *ctx, V9fsFidOpenState *fs)
 {
-    return readdir(fs->dir.stream);
+    struct dirent *entry;
+    entry = readdir(fs->dir.stream);
+#ifdef CONFIG_DARWIN
+    if (!entry) {
+        return NULL;
+    }
+    int td;
+    td = telldir(fs->dir.stream);
+    /* If telldir fails, fail the entire readdir call */
+    if (td < 0) {
+        return NULL;
+    }
+    entry->d_seekoff = td;
+#endif
+    return entry;
 }
 
 static void proxy_seekdir(FsContext *ctx, V9fsFidOpenState *fs, off_t off)
hw/9pfs/9p-synth.c +4 -0
index bf9b0c5ddd..b3080e415b 100644
@@ -234,7 +234,11 @@ static void synth_direntry(V9fsSynthNode *node,
              offsetof(struct dirent, d_name) + sz);
     memcpy(entry->d_name, node->name, sz);
     entry->d_ino = node->attr->inode;
+#ifdef CONFIG_DARWIN
+    entry->d_seekoff = off + 1;
+#else
     entry->d_off = off + 1;
+#endif
 }
 
 static struct dirent *synth_get_dentry(V9fsSynthNode *dir,
hw/9pfs/9p-util.h +16 -0
index 546f46dc7d..d41f37f085 100644
@@ -79,3 +79,19 @@ ssize_t fremovexattrat_nofollow(int dirfd, const char *filename,
                                 const char *name);
 
 #endif
+
+
+/**
+ * Darwin has d_seekoff, which appears to function similarly to d_off.
+ * However, it does not appear to be supported on all file systems,
+ * so ensure it is manually injected earlier and call here when
+ * needed.
+ */
+inline off_t qemu_dirent_off(struct dirent *dent)
+{
+#ifdef CONFIG_DARWIN
+    return dent->d_seekoff;
+#else
+    return dent->d_off;
+#endif
+}
hw/9pfs/9p.c +5 -2
index 1563d7b7c6..caf3b240fe 100644
@@ -27,6 +27,7 @@
 #include "virtio-9p.h"
 #include "fsdev/qemu-fsdev.h"
 #include "9p-xattr.h"
+#include "9p-util.h"
 #include "coth.h"
 #include "trace.h"
 #include "migration/blocker.h"
@@ -2281,7 +2282,7 @@ static int coroutine_fn v9fs_do_readdir_with_stat(V9fsPDU *pdu,
         count += len;
         v9fs_stat_free(&v9stat);
         v9fs_path_free(&path);
-        saved_dir_pos = dent->d_off;
+        saved_dir_pos = qemu_dirent_off(dent);
     }
 
     v9fs_readdir_unlock(&fidp->fs.dir);
@@ -2420,6 +2421,7 @@ static int coroutine_fn v9fs_do_readdir(V9fsPDU *pdu, V9fsFidState *fidp,
     V9fsString name;
     int len, err = 0;
     int32_t count = 0;
+    off_t off;
     struct dirent *dent;
     struct stat *st;
     struct V9fsDirEnt *entries = NULL;
@@ -2480,12 +2482,13 @@ static int coroutine_fn v9fs_do_readdir(V9fsPDU *pdu, V9fsFidState *fidp,
             qid.version = 0;
         }
 
+        off = qemu_dirent_off(dent);
         v9fs_string_init(&name);
         v9fs_string_sprintf(&name, "%s", dent->d_name);
 
         /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
         len = pdu_marshal(pdu, 11 + count, "Qqbs",
-                          &qid, dent->d_off,
+                          &qid, off,
                           dent->d_type, &name);
 
         v9fs_string_free(&name);
hw/9pfs/codir.c +3 -1
index c0873bde16..f96d8ac4e6 100644
@@ -22,6 +22,8 @@
 #include "qemu/coroutine.h"
 #include "qemu/main-loop.h"
 #include "coth.h"
+#include "9p-xattr.h"
+#include "9p-util.h"
 
 /*
  * Intended to be called from bottom-half (e.g. background I/O thread)
@@ -166,7 +168,7 @@ static int do_readdir_many(V9fsPDU *pdu, V9fsFidState *fidp,
         }
 
         size += len;
-        saved_dir_pos = dent->d_off;
+        saved_dir_pos = qemu_dirent_off(dent);
     }
 
     /* restore (last) saved position */
Signature · 1 line
2.35.1
author
#8 lore raw

PATCH v8 6/11 9p: darwin: Move XATTR_SIZE_MAX->P9_XATTR_SIZE_MAX

In reply to Will Cohen

From: Keno Fischer <[email protected]>

Because XATTR_SIZE_MAX is not defined on Darwin, create a cross-platform P9_XATTR_SIZE_MAX instead.

[Will Cohen: - Adjust coding style
- Lower XATTR_SIZE_MAX to 64k
- Add explanatory context related to XATTR_SIZE_MAX]
[Fabian Franz: - Move XATTR_SIZE_MAX reference from 9p.c to

                 P9_XATTR_SIZE_MAX in 9p.h]

[Will Cohen: - For P9_XATTR_MAX, ensure that Linux uses
XATTR_SIZE_MAX, Darwin uses 64k, and error out for undefined hosts]

---
 hw/9pfs/9p.c |  2 +-
 hw/9pfs/9p.h | 18 ++++++++++++++++++
 2 files changed, 19 insertions(+), 1 deletion(-)
hw/9pfs/9p.c +1 -1
index 14e84c3bcf..7405352c37 100644
@@ -3949,7 +3949,7 @@ static void coroutine_fn v9fs_xattrcreate(void *opaque)
         rflags |= XATTR_REPLACE;
     }
 
-    if (size > XATTR_SIZE_MAX) {
+    if (size > P9_XATTR_SIZE_MAX) {
         err = -E2BIG;
         goto out_nofid;
     }
hw/9pfs/9p.h +18 -0
index 1567b67841..94b273b3d0 100644
@@ -479,4 +479,22 @@ struct V9fsTransport {
     void        (*push_and_notify)(V9fsPDU *pdu);
 };
 
+#if defined(XATTR_SIZE_MAX)
+/* Linux */
+#define P9_XATTR_SIZE_MAX XATTR_SIZE_MAX
+#elif defined(CONFIG_DARWIN)
+/*
+ * Darwin doesn't seem to define a maximum xattr size in its user
+ * space header, so manually configure it across platforms as 64k.
+ *
+ * Having no limit at all can lead to QEMU crashing during large g_malloc()
+ * calls. Because QEMU does not currently support macOS guests, the below
+ * preliminary solution only works due to its being a reflection of the limit of
+ * Linux guests.
+ */
+#define P9_XATTR_SIZE_MAX 65536
+#else
+#error Missing definition for P9_XATTR_SIZE_MAX for this host system
+#endif
+
 #endif
Signature · 1 line
2.35.1
author
#11 lore raw

PATCH v8 9/11 9p: darwin: Implement compatibility for mknodat

In reply to Will Cohen

From: Keno Fischer <[email protected]>

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 this series.

[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
- Move pthread_fchdir_np declaration only to osdep
- Declare pthread_fchdir_np with
- __attribute__((weak_import)) to allow checking for its presence before usage
- Move declarations above cplusplus guard
- Add CONFIG_PTHREAD_FCHDIR_NP to meson and check for presence in osdep.h and os-posix.c
- Rebase to apply cleanly on top of the 2022-02-10 changes to 9pfs]

---
 hw/9pfs/9p-local.c   |  4 ++--
 include/qemu/osdep.h | 12 ++++++++++++
 meson.build          |  1 +
 os-posix.c           | 35 +++++++++++++++++++++++++++++++++++
 4 files changed, 50 insertions(+), 2 deletions(-)
hw/9pfs/9p-local.c +2 -2
index a0d08e5216..d42ce6d8b8 100644
@@ -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;
         }
include/qemu/osdep.h +12 -0
index ce12f64853..c0f442d791 100644
@@ -818,6 +818,18 @@ static inline int platform_does_not_support_system(const char *command)
  */
 struct dirent *qemu_dirent_dup(struct dirent *dent);
 
+/*
+ * 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. pthread_fchdir_np is weakly linked here as a guard
+ * in case it disappears in future macOS versions, because it is
+ * is a private API.
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+int pthread_fchdir_np(int fd) __attribute__((weak_import));
+#endif
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t dev);
+
 #ifdef __cplusplus
 }
 #endif
meson.build +1 -0
index ae5f7eec6e..6fdc0281ad 100644
@@ -1557,6 +1557,7 @@ config_host_data.set('CONFIG_POSIX_FALLOCATE', cc.has_function('posix_fallocate'
 config_host_data.set('CONFIG_POSIX_MEMALIGN', cc.has_function('posix_memalign'))
 config_host_data.set('CONFIG_PPOLL', cc.has_function('ppoll'))
 config_host_data.set('CONFIG_PREADV', cc.has_function('preadv', prefix: '#include <sys/uio.h>'))
+config_host_data.set('CONFIG_PTHREAD_FCHDIR_NP', cc.has_function('pthread_fchdir_np'))
 config_host_data.set('CONFIG_SEM_TIMEDWAIT', cc.has_function('sem_timedwait', dependencies: threads))
 config_host_data.set('CONFIG_SENDFILE', cc.has_function('sendfile'))
 config_host_data.set('CONFIG_SETNS', cc.has_function('setns') and cc.has_function('unshare'))
os-posix.c +35 -0
index ae6c9f2a5e..ccc3d1e9d3 100644
@@ -332,3 +332,38 @@ 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)
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t dev)
+{
+    int preserved_errno, err;
+    if (!pthread_fchdir_np) {
+        error_report_once("pthread_fchdir_np() is not available on this version of macOS");
+        return -ENOTSUP;
+    }
+    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
Signature · 1 line
2.35.1
author
#10 lore raw

PATCH v8 8/11 9p: darwin: Compatibility for f/l*xattr

In reply to Will Cohen

From: Keno Fischer <[email protected]>

On darwin fgetxattr takes two extra optional arguments, and the l* variants are not defined (in favor of an extra flag to the regular variants.

[Michael Roitzsch: - Rebase for NixOS]
---
 hw/9pfs/9p-local.c | 12 ++++++++----
 hw/9pfs/9p-util.h  | 17 +++++++++++++++++
 2 files changed, 25 insertions(+), 4 deletions(-)
hw/9pfs/9p-local.c +8 -4
index f3272f0b43..a0d08e5216 100644
@@ -790,16 +790,20 @@ static int local_fstat(FsContext *fs_ctx, int fid_type,
         mode_t tmp_mode;
         dev_t tmp_dev;
 
-        if (fgetxattr(fd, "user.virtfs.uid", &tmp_uid, sizeof(uid_t)) > 0) {
+        if (qemu_fgetxattr(fd, "user.virtfs.uid",
+                           &tmp_uid, sizeof(uid_t)) > 0) {
             stbuf->st_uid = le32_to_cpu(tmp_uid);
         }
-        if (fgetxattr(fd, "user.virtfs.gid", &tmp_gid, sizeof(gid_t)) > 0) {
+        if (qemu_fgetxattr(fd, "user.virtfs.gid",
+                           &tmp_gid, sizeof(gid_t)) > 0) {
             stbuf->st_gid = le32_to_cpu(tmp_gid);
         }
-        if (fgetxattr(fd, "user.virtfs.mode", &tmp_mode, sizeof(mode_t)) > 0) {
+        if (qemu_fgetxattr(fd, "user.virtfs.mode",
+                           &tmp_mode, sizeof(mode_t)) > 0) {
             stbuf->st_mode = le32_to_cpu(tmp_mode);
         }
-        if (fgetxattr(fd, "user.virtfs.rdev", &tmp_dev, sizeof(dev_t)) > 0) {
+        if (qemu_fgetxattr(fd, "user.virtfs.rdev",
+                           &tmp_dev, sizeof(dev_t)) > 0) {
             stbuf->st_rdev = le64_to_cpu(tmp_dev);
         }
     } else if (fs_ctx->export_flags & V9FS_SM_MAPPED_FILE) {
hw/9pfs/9p-util.h +17 -0
index 0e445b5d52..82399702b9 100644
@@ -19,6 +19,23 @@
 #define O_PATH_9P_UTIL 0
 #endif
 
+#ifdef CONFIG_DARWIN
+#define qemu_fgetxattr(...) fgetxattr(__VA_ARGS__, 0, 0)
+#define qemu_lgetxattr(...) getxattr(__VA_ARGS__, 0, XATTR_NOFOLLOW)
+#define qemu_llistxattr(...) listxattr(__VA_ARGS__, XATTR_NOFOLLOW)
+#define qemu_lremovexattr(...) removexattr(__VA_ARGS__, XATTR_NOFOLLOW)
+static inline int qemu_lsetxattr(const char *path, const char *name,
+                                 const void *value, size_t size, int flags) {
+    return setxattr(path, name, value, size, 0, flags | XATTR_NOFOLLOW);
+}
+#else
+#define qemu_fgetxattr fgetxattr
+#define qemu_lgetxattr lgetxattr
+#define qemu_llistxattr llistxattr
+#define qemu_lremovexattr lremovexattr
+#define qemu_lsetxattr lsetxattr
+#endif
+
 static inline void close_preserve_errno(int fd)
 {
     int serrno = errno;
Signature · 1 line
2.35.1
author
#9 lore raw

PATCH v8 7/11 9p: darwin: *xattr_nofollow implementations

In reply to Will Cohen

From: Keno Fischer <[email protected]>

This implements the darwin equivalent of the functions that were moved to 9p-util(-linux) earlier in this series in the new 9p-util-darwin file.

[Michael Roitzsch: - Rebase for NixOS]
---
 hw/9pfs/9p-util-darwin.c | 64 ++++++++++++++++++++++++++++++++++++++++
 hw/9pfs/meson.build      |  1 +
 2 files changed, 65 insertions(+)
 create mode 100644 hw/9pfs/9p-util-darwin.c
hw/9pfs/9p-util-darwin.c new file +64 -0
new file mode 100644
index 0000000000..cdb4c9e24c
@@ -0,0 +1,64 @@
+/*
+ * 9p utilities (Darwin Implementation)
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/xattr.h"
+#include "9p-util.h"
+
+ssize_t fgetxattrat_nofollow(int dirfd, const char *filename, const char *name,
+                             void *value, size_t size)
+{
+    int ret;
+    int fd = openat_file(dirfd, filename,
+                         O_RDONLY | O_PATH_9P_UTIL | O_NOFOLLOW, 0);
+    if (fd == -1) {
+        return -1;
+    }
+    ret = fgetxattr(fd, name, value, size, 0, 0);
+    close_preserve_errno(fd);
+    return ret;
+}
+
+ssize_t flistxattrat_nofollow(int dirfd, const char *filename,
+                              char *list, size_t size)
+{
+    int ret;
+    int fd = openat_file(dirfd, filename,
+                         O_RDONLY | O_PATH_9P_UTIL | O_NOFOLLOW, 0);
+    if (fd == -1) {
+        return -1;
+    }
+    ret = flistxattr(fd, list, size, 0);
+    close_preserve_errno(fd);
+    return ret;
+}
+
+ssize_t fremovexattrat_nofollow(int dirfd, const char *filename,
+                                const char *name)
+{
+    int ret;
+    int fd = openat_file(dirfd, filename, O_PATH_9P_UTIL | O_NOFOLLOW, 0);
+    if (fd == -1) {
+        return -1;
+    }
+    ret = fremovexattr(fd, name, 0);
+    close_preserve_errno(fd);
+    return ret;
+}
+
+int fsetxattrat_nofollow(int dirfd, const char *filename, const char *name,
+                         void *value, size_t size, int flags)
+{
+    int ret;
+    int fd = openat_file(dirfd, filename, O_PATH_9P_UTIL | O_NOFOLLOW, 0);
+    if (fd == -1) {
+        return -1;
+    }
+    ret = fsetxattr(fd, name, value, size, 0, flags);
+    close_preserve_errno(fd);
+    return ret;
+}
hw/9pfs/meson.build +1 -0
index 1b28e70040..12443b6ad5 100644
@@ -14,6 +14,7 @@ fs_ss.add(files(
   'coxattr.c',
 ))
 fs_ss.add(when: 'CONFIG_LINUX', if_true: files('9p-util-linux.c'))
+fs_ss.add(when: 'CONFIG_DARWIN', if_true: files('9p-util-darwin.c'))
 fs_ss.add(when: 'CONFIG_XEN', if_true: files('xen-9p-backend.c'))
 softmmu_ss.add_all(when: 'CONFIG_FSDEV_9P', if_true: fs_ss)
 
Signature · 1 line
2.35.1
author
#18 lore raw

PATCH v8 10/11 9p: darwin: Adjust assumption on virtio-9p-test

In reply to Will Cohen

The previous test depended on the assumption that P9_DOTL_AT_REMOVEDIR and AT_REMOVEDIR have the same value.

While this is true on Linux, it is not true everywhere, and leads to an incorrect test failure on unlink_at, noticed when adding 9p to darwin:

Received response 7 (RLERROR) instead of 77 (RUNLINKAT) Rlerror has errno 22 (Invalid argument)
**

ERROR:../tests/qtest/virtio-9p-test.c:305:v9fs_req_recv: assertion failed (hdr.id == id): (7 == 77) Bail out!

ERROR:../tests/qtest/virtio-9p-test.c:305:v9fs_req_recv: assertion failed (hdr.id == id): (7 == 77)

[Will Cohen: - Add explanation of patch and description
of pre-patch test failure]

[Will Cohen: - Move this patch before 9p: darwin: meson
patch to avoid qtest breakage during bisecting]

---
 tests/qtest/virtio-9p-test.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/qtest/virtio-9p-test.c +1 -1
index 502e5ad0c7..01ca076afe 100644
@@ -1253,7 +1253,7 @@ static void fs_unlinkat_dir(void *obj, void *data, QGuestAllocator *t_alloc)
     /* ... and is actually a directory */
     g_assert((st.st_mode & S_IFMT) == S_IFDIR);
 
-    do_unlinkat(v9p, "/", "02", AT_REMOVEDIR);
+    do_unlinkat(v9p, "/", "02", P9_DOTL_AT_REMOVEDIR);
     /* directory should be gone now */
     g_assert(stat(new_dir, &st) != 0);
 }
Signature · 1 line
2.35.1
author
#19 lore raw

PATCH v8 11/11 9p: darwin: meson: Allow VirtFS on Darwin

In reply to Will Cohen

From: Keno Fischer <[email protected]>

To allow VirtFS on darwin, we need to check that pthread_fchdir_np is available, which has only been available since macOS 10.12.

Additionally, virtfs_proxy_helper is disabled on Darwin. This patch series does not currently provide an implementation of the proxy-helper, but this functionality could be implemented later on.

[Michael Roitzsch: - Rebase for NixOS]
[Will Cohen: - Rebase to master]

[Will Cohen: - Add check for pthread_fchdir_np to virtfs
- Add comments to patch commit
- Note that virtfs_proxy_helper does not work on macOS
- Fully adjust meson virtfs error note to specify macOS]

---
 fsdev/meson.build |  1 +
 meson.build       | 14 ++++++++++----
 2 files changed, 11 insertions(+), 4 deletions(-)
fsdev/meson.build +1 -0
index adf57cc43e..b632b66348 100644
@@ -7,6 +7,7 @@ fsdev_ss.add(when: ['CONFIG_FSDEV_9P'], if_true: files(
   'qemu-fsdev.c',
 ), if_false: files('qemu-fsdev-dummy.c'))
 softmmu_ss.add_all(when: 'CONFIG_LINUX', if_true: fsdev_ss)
+softmmu_ss.add_all(when: 'CONFIG_DARWIN', if_true: fsdev_ss)
 
 if have_virtfs_proxy_helper
   executable('virtfs-proxy-helper',
meson.build +10 -4
index 6fdc0281ad..b6f5c57487 100644
@@ -1416,17 +1416,23 @@ if not get_option('dbus_display').disabled()
   endif
 endif
 
-have_virtfs = (targetos == 'linux' and
+if targetos == 'darwin' and cc.has_function('pthread_fchdir_np')
+  have_virtfs = have_system
+else
+  have_virtfs = (targetos == 'linux' and
     have_system and
     libattr.found() and
     libcap_ng.found())
+endif
 
-have_virtfs_proxy_helper = have_virtfs and have_tools
+have_virtfs_proxy_helper = targetos != 'darwin' and have_virtfs and have_tools
 
 if get_option('virtfs').enabled()
   if not have_virtfs
-    if targetos != 'linux'
-      error('virtio-9p (virtfs) requires Linux')
+    if targetos != 'linux' and targetos != 'darwin'
+      error('virtio-9p (virtfs) requires Linux or macOS')
+    elif targetos == 'darwin' and not cc.has_function('pthread_fchdir_np')
+      error('virtio-9p (virtfs) on macOS requires the presence of pthread_fchdir_np')
     elif not libcap_ng.found() or not libattr.found()
       error('virtio-9p (virtfs) requires libcap-ng-devel and libattr-devel')
     elif not have_system
Signature · 1 line
2.35.1
#5 lore raw

Re: [PATCH v8 04/11] 9p: darwin: Handle struct dirent differences

In reply to Will Cohen
Quoted 23 lines · On Sonntag, 20. Februar 2022 17:50:49 CET Will Cohen wrote:

From: Keno Fischer <[email protected]>

On darwin d_seekoff exists, but is optional and does not seem to be commonly used by file systems. Use telldir instead to obtain the seek offset and inject it into d_seekoff, and create a qemu_dirent_off helper to call it appropriately when appropriate.

[Michael Roitzsch: - Rebase for NixOS]

[Will Cohen: - Adjust to pass testing
- Ensure that d_seekoff is filled using telldir on darwin, and create qemu_dirent_off helper to decide which to access]

[Fabian Franz: - Add telldir error handling for darwin]

[Will Cohen: - Ensure that telldir error handling uses
signed int
- Cleanup of telldir error handling
- Remove superfluous error handling for qemu_dirent_off
- Adjust formatting
- Use qemu_dirent_off in codir.c]

Quoted 2 lines

---

This patch does not compile on Linux ...

Quoted 77 lines
 hw/9pfs/9p-local.c |  9 +++++++++
 hw/9pfs/9p-proxy.c | 16 +++++++++++++++-
 hw/9pfs/9p-synth.c |  4 ++++
 hw/9pfs/9p-util.h  | 16 ++++++++++++++++
 hw/9pfs/9p.c       |  7 +++++--
 hw/9pfs/codir.c    |  4 +++-
 6 files changed, 52 insertions(+), 4 deletions(-)
hw/9pfs/9p-local.c +9 -0
index 1a5e3eed73..f3272f0b43 100644
@@ -562,6 +562,15 @@ again:
     if (!entry) {
         return NULL;
     }
+#ifdef CONFIG_DARWIN
+    int off;
+    off = telldir(fs->dir.stream);
+    /* If telldir fails, fail the entire readdir call */
+    if (off < 0) {
+        return NULL;
+    }
+    entry->d_seekoff = off;
+#endif
 
     if (ctx->export_flags & V9FS_SM_MAPPED) {
         entry->d_type = DT_UNKNOWN;
hw/9pfs/9p-proxy.c +15 -1
index b1664080d8..8b4b5cf7dc 100644
@@ -706,7 +706,21 @@ static off_t proxy_telldir(FsContext *ctx,
 V9fsFidOpenState *fs)
 
 static struct dirent *proxy_readdir(FsContext *ctx, V9fsFidOpenState *fs)
 {
-    return readdir(fs->dir.stream);
+    struct dirent *entry;
+    entry = readdir(fs->dir.stream);
+#ifdef CONFIG_DARWIN
+    if (!entry) {
+        return NULL;
+    }
+    int td;
+    td = telldir(fs->dir.stream);
+    /* If telldir fails, fail the entire readdir call */
+    if (td < 0) {
+        return NULL;
+    }
+    entry->d_seekoff = td;
+#endif
+    return entry;
 }
 

static void proxy_seekdir(FsContext *ctx, V9fsFidOpenState *fs, off_t off)

hw/9pfs/9p-synth.c +4 -0
index bf9b0c5ddd..b3080e415b 100644
@@ -234,7 +234,11 @@ static void synth_direntry(V9fsSynthNode *node,
              offsetof(struct dirent, d_name) + sz);
     memcpy(entry->d_name, node->name, sz);
     entry->d_ino = node->attr->inode;
+#ifdef CONFIG_DARWIN
+    entry->d_seekoff = off + 1;
+#else
     entry->d_off = off + 1;
+#endif
 }
 
 static struct dirent *synth_get_dentry(V9fsSynthNode *dir,
hw/9pfs/9p-util.h +0 -0
index 546f46dc7d..d41f37f085 100644

hw/9pfs/9p-util.h ssize_t fremovexattrat_nofollow(int dirfd, const char · line 81

Quoted 3 lines
hw/9pfs/9p-util.h +0 -0
@@ -79,3 +79,19 @@ ssize_t fremovexattrat_nofollow(int dirfd, const char
 *filename, const char *name);
 
 #endif

... ^- this is the end of file #endif, so qemu_dirent_off() should be above that #endif, and ...

hw/9pfs/9p-util.h ssize_t fremovexattrat_nofollow(int dirfd, const char *filename, · line 90

Quoted 9 lines
hw/9pfs/9p-util.h +9 -0
+
+
+/**
+ * Darwin has d_seekoff, which appears to function similarly to d_off.
+ * However, it does not appear to be supported on all file systems,
+ * so ensure it is manually injected earlier and call here when
+ * needed.
+ */
+inline off_t qemu_dirent_off(struct dirent *dent)

... this function declaration misses the 'static' keyword, which is mandatory to prevent a linker error.

Best regards,
Christian Schoenebeck

Quoted 73 lines
hw/9pfs/9p-util.h +7 -0
+{
+#ifdef CONFIG_DARWIN
+    return dent->d_seekoff;
+#else
+    return dent->d_off;
+#endif
+}
hw/9pfs/9p.c +5 -2
index 1563d7b7c6..caf3b240fe 100644
@@ -27,6 +27,7 @@
 #include "virtio-9p.h"
 #include "fsdev/qemu-fsdev.h"
 #include "9p-xattr.h"
+#include "9p-util.h"
 #include "coth.h"
 #include "trace.h"
 #include "migration/blocker.h"
@@ -2281,7 +2282,7 @@ static int coroutine_fn
 v9fs_do_readdir_with_stat(V9fsPDU *pdu, count += len;
         v9fs_stat_free(&v9stat);
         v9fs_path_free(&path);
-        saved_dir_pos = dent->d_off;
+        saved_dir_pos = qemu_dirent_off(dent);
     }
 
     v9fs_readdir_unlock(&fidp->fs.dir);
@@ -2420,6 +2421,7 @@ static int coroutine_fn v9fs_do_readdir(V9fsPDU *pdu,
 V9fsFidState *fidp, V9fsString name;
     int len, err = 0;
     int32_t count = 0;
+    off_t off;
     struct dirent *dent;
     struct stat *st;
     struct V9fsDirEnt *entries = NULL;
@@ -2480,12 +2482,13 @@ static int coroutine_fn v9fs_do_readdir(V9fsPDU
 *pdu, V9fsFidState *fidp, qid.version = 0;
         }
 
+        off = qemu_dirent_off(dent);
         v9fs_string_init(&name);
         v9fs_string_sprintf(&name, "%s", dent->d_name);
 
         /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
         len = pdu_marshal(pdu, 11 + count, "Qqbs",
-                          &qid, dent->d_off,
+                          &qid, off,
                           dent->d_type, &name);
 
         v9fs_string_free(&name);
hw/9pfs/codir.c +3 -1
index c0873bde16..f96d8ac4e6 100644
@@ -22,6 +22,8 @@
 #include "qemu/coroutine.h"
 #include "qemu/main-loop.h"
 #include "coth.h"
+#include "9p-xattr.h"
+#include "9p-util.h"
 
 /*
  * Intended to be called from bottom-half (e.g. background I/O thread)
@@ -166,7 +168,7 @@ static int do_readdir_many(V9fsPDU *pdu, V9fsFidState
 *fidp, }
 
         size += len;
-        saved_dir_pos = dent->d_off;
+        saved_dir_pos = qemu_dirent_off(dent);
     }
 
     /* restore (last) saved position */
author
#6 lore raw

Re: [PATCH v8 04/11] 9p: darwin: Handle struct dirent differences

Apologies — I tested on Mac but should have done Linux too. Will revise.

Quoted 210 lines · Christian Schoenebeck
Quoted 23 lines · On Sonntag, 20. Februar 2022 17:50:49 CET Will Cohen wrote:

From: Keno Fischer <[email protected]>

On darwin d_seekoff exists, but is optional and does not seem to be commonly used by file systems. Use telldir instead to obtain the seek offset and inject it into d_seekoff, and create a qemu_dirent_off helper to call it appropriately when appropriate.

[Michael Roitzsch: - Rebase for NixOS]

[Will Cohen: - Adjust to pass testing
- Ensure that d_seekoff is filled using telldir on darwin, and create qemu_dirent_off helper to decide which to access]

[Fabian Franz: - Add telldir error handling for darwin]

[Will Cohen: - Ensure that telldir error handling uses
signed int
- Cleanup of telldir error handling
- Remove superfluous error handling for qemu_dirent_off
- Adjust formatting
- Use qemu_dirent_off in codir.c]

Quoted 2 lines

---

This patch does not compile on Linux ...

Quoted 33 lines
 hw/9pfs/9p-local.c |  9 +++++++++
 hw/9pfs/9p-proxy.c | 16 +++++++++++++++-
 hw/9pfs/9p-synth.c |  4 ++++
 hw/9pfs/9p-util.h  | 16 ++++++++++++++++
 hw/9pfs/9p.c       |  7 +++++--
 hw/9pfs/codir.c    |  4 +++-
 6 files changed, 52 insertions(+), 4 deletions(-)
hw/9pfs/9p-local.c +9 -0
index 1a5e3eed73..f3272f0b43 100644
@@ -562,6 +562,15 @@ again:
     if (!entry) {
         return NULL;
     }
+#ifdef CONFIG_DARWIN
+    int off;
+    off = telldir(fs->dir.stream);
+    /* If telldir fails, fail the entire readdir call */
+    if (off < 0) {
+        return NULL;
+    }
+    entry->d_seekoff = off;
+#endif
 
     if (ctx->export_flags & V9FS_SM_MAPPED) {
         entry->d_type = DT_UNKNOWN;
hw/9pfs/9p-proxy.c +0 -0
index b1664080d8..8b4b5cf7dc 100644

hw/9pfs/9p-proxy.c static off_t proxy_telldir(FsContext *ctx, · line 708

Quoted 3 lines
hw/9pfs/9p-proxy.c +0 -0
@@ -706,7 +706,21 @@ static off_t proxy_telldir(FsContext *ctx,
 V9fsFidOpenState *fs)
 
 static struct dirent *proxy_readdir(FsContext *ctx, V9fsFidOpenState

*fs)

Quoted 19 lines
hw/9pfs/9p-proxy.c +15 -1
 {
-    return readdir(fs->dir.stream);
+    struct dirent *entry;
+    entry = readdir(fs->dir.stream);
+#ifdef CONFIG_DARWIN
+    if (!entry) {
+        return NULL;
+    }
+    int td;
+    td = telldir(fs->dir.stream);
+    /* If telldir fails, fail the entire readdir call */
+    if (td < 0) {
+        return NULL;
+    }
+    entry->d_seekoff = td;
+#endif
+    return entry;
 }
 
Quoted 1 line

static void proxy_seekdir(FsContext *ctx, V9fsFidOpenState *fs, off_t

off)

Quoted 21 lines
hw/9pfs/9p-synth.c +4 -0
index bf9b0c5ddd..b3080e415b 100644
@@ -234,7 +234,11 @@ static void synth_direntry(V9fsSynthNode *node,
              offsetof(struct dirent, d_name) + sz);
     memcpy(entry->d_name, node->name, sz);
     entry->d_ino = node->attr->inode;
+#ifdef CONFIG_DARWIN
+    entry->d_seekoff = off + 1;
+#else
     entry->d_off = off + 1;
+#endif
 }
 
 static struct dirent *synth_get_dentry(V9fsSynthNode *dir,
hw/9pfs/9p-util.h +0 -0
index 546f46dc7d..d41f37f085 100644

hw/9pfs/9p-util.h ssize_t fremovexattrat_nofollow(int dirfd, const char · line 81

Quoted 3 lines
hw/9pfs/9p-util.h +0 -0
@@ -79,3 +79,19 @@ ssize_t fremovexattrat_nofollow(int dirfd, const char
 *filename, const char *name);
 
 #endif

... ^- this is the end of file #endif, so qemu_dirent_off() should be above
that #endif, and ...

hw/9pfs/9p-util.h ssize_t fremovexattrat_nofollow(int dirfd, const char *filename, · line 90

Quoted 9 lines
hw/9pfs/9p-util.h +9 -0
+
+
+/**
+ * Darwin has d_seekoff, which appears to function similarly to d_off.
+ * However, it does not appear to be supported on all file systems,
+ * so ensure it is manually injected earlier and call here when
+ * needed.
+ */
+inline off_t qemu_dirent_off(struct dirent *dent)

... this function declaration misses the 'static' keyword, which is mandatory
to prevent a linker error.

Best regards,
Christian Schoenebeck

hw/9pfs/9p.c static int coroutine_fn v9fs_do_readdir(V9fsPDU

Quoted 29 lines
hw/9pfs/9p-util.h +7 -0
+{
+#ifdef CONFIG_DARWIN
+    return dent->d_seekoff;
+#else
+    return dent->d_off;
+#endif
+}
hw/9pfs/9p.c +2 -1
index 1563d7b7c6..caf3b240fe 100644
@@ -27,6 +27,7 @@
 #include "virtio-9p.h"
 #include "fsdev/qemu-fsdev.h"
 #include "9p-xattr.h"
+#include "9p-util.h"
 #include "coth.h"
 #include "trace.h"
 #include "migration/blocker.h"
@@ -2281,7 +2282,7 @@ static int coroutine_fn
 v9fs_do_readdir_with_stat(V9fsPDU *pdu, count += len;
         v9fs_stat_free(&v9stat);
         v9fs_path_free(&path);
-        saved_dir_pos = dent->d_off;
+        saved_dir_pos = qemu_dirent_off(dent);
     }
 
     v9fs_readdir_unlock(&fidp->fs.dir);
@@ -2420,6 +2421,7 @@ static int coroutine_fn v9fs_do_readdir(V9fsPDU

*pdu,

Quoted 44 lines
hw/9pfs/9p.c +3 -1
 V9fsFidState *fidp, V9fsString name;
     int len, err = 0;
     int32_t count = 0;
+    off_t off;
     struct dirent *dent;
     struct stat *st;
     struct V9fsDirEnt *entries = NULL;
@@ -2480,12 +2482,13 @@ static int coroutine_fn v9fs_do_readdir(V9fsPDU
 *pdu, V9fsFidState *fidp, qid.version = 0;
         }
 
+        off = qemu_dirent_off(dent);
         v9fs_string_init(&name);
         v9fs_string_sprintf(&name, "%s", dent->d_name);
 
         /* 11 = 7 + 4 (7 = start offset, 4 = space for storing count) */
         len = pdu_marshal(pdu, 11 + count, "Qqbs",
-                          &qid, dent->d_off,
+                          &qid, off,
                           dent->d_type, &name);
 
         v9fs_string_free(&name);
hw/9pfs/codir.c +3 -1
index c0873bde16..f96d8ac4e6 100644
@@ -22,6 +22,8 @@
 #include "qemu/coroutine.h"
 #include "qemu/main-loop.h"
 #include "coth.h"
+#include "9p-xattr.h"
+#include "9p-util.h"
 
 /*
  * Intended to be called from bottom-half (e.g. background I/O thread)
@@ -166,7 +168,7 @@ static int do_readdir_many(V9fsPDU *pdu, V9fsFidState
 *fidp, }
 
         size += len;
-        saved_dir_pos = dent->d_off;
+        saved_dir_pos = qemu_dirent_off(dent);
     }
 
     /* restore (last) saved position */
#12 lore raw

Re: [PATCH v8 09/11] 9p: darwin: Implement compatibility for mknodat

In reply to Will Cohen
Quoted 123 lines · On Sonntag, 20. Februar 2022 17:50:54 CET Will Cohen wrote:

From: Keno Fischer <[email protected]>

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 this series.

[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
- Move pthread_fchdir_np declaration only to osdep
- Declare pthread_fchdir_np with
- __attribute__((weak_import)) to allow checking for its presence before usage
- Move declarations above cplusplus guard
- Add CONFIG_PTHREAD_FCHDIR_NP to meson and check for presence in osdep.h and os-posix.c
- Rebase to apply cleanly on top of the 2022-02-10 changes to 9pfs]

---
 hw/9pfs/9p-local.c   |  4 ++--
 include/qemu/osdep.h | 12 ++++++++++++
 meson.build          |  1 +
 os-posix.c           | 35 +++++++++++++++++++++++++++++++++++
 4 files changed, 50 insertions(+), 2 deletions(-)
hw/9pfs/9p-local.c +1 -1
index a0d08e5216..d42ce6d8b8 100644
@@ -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;
         }
hw/9pfs/9p-local.c +1 -1
@@ -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;
         }
include/qemu/osdep.h +0 -0
index ce12f64853..c0f442d791 100644
@@ -818,6 +818,18 @@ static inline int

platform_does_not_support_system(const char *command) */
struct dirent *qemu_dirent_dup(struct dirent *dent);

include/qemu/osdep.h +12 -0
+/*
+ * 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. pthread_fchdir_np is weakly linked here as a guard
+ * in case it disappears in future macOS versions, because it is
+ * is a private API.
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+int pthread_fchdir_np(int fd) __attribute__((weak_import));
+#endif
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t dev);
+
 #ifdef __cplusplus
 }
 #endif
meson.build +0 -0
index ae5f7eec6e..6fdc0281ad 100644
@@ -1557,6 +1557,7 @@ config_host_data.set('CONFIG_POSIX_FALLOCATE',

cc.has_function('posix_fallocate' config_host_data.set('CONFIG_POSIX_MEMALIGN', cc.has_function('posix_memalign')) config_host_data.set('CONFIG_PPOLL', cc.has_function('ppoll'))
config_host_data.set('CONFIG_PREADV', cc.has_function('preadv', prefix:
'#include <sys/uio.h>')) +config_host_data.set('CONFIG_PTHREAD_FCHDIR_NP', cc.has_function('pthread_fchdir_np')) config_host_data.set('CONFIG_SEM_TIMEDWAIT', cc.has_function('sem_timedwait', dependencies: threads)) config_host_data.set('CONFIG_SENDFILE', cc.has_function('sendfile')) config_host_data.set('CONFIG_SETNS', cc.has_function('setns') and cc.has_function('unshare')) diff --git a/os-posix.c b/os-posix.c index ae6c9f2a5e..ccc3d1e9d3 100644

os-posix.c +15 -0
@@ -332,3 +332,38 @@ 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)
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t dev)
+{
+    int preserved_errno, err;
+    if (!pthread_fchdir_np) {
+        error_report_once("pthread_fchdir_np() is not available on this
Quoted 1 line

version of macOS");

Code style error:

ERROR: line over 90 characters
#226: FILE: os-posix.c:349:
+ error_report_once("pthread_fchdir_np() is not available on this
version of macOS");

BTW, does anybody know what happened to our Patchew bot that used to automatically report such errors to the list?

Best regards,
Christian Schoenebeck

Quoted 20 lines
os-posix.c +20 -0
+        return -ENOTSUP;
+    }
+    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
#13 lore raw

Re: [PATCH v8 09/11] 9p: darwin: Implement compatibility for mknodat

In reply to Will Cohen
Quoted 74 lines · On Sonntag, 20. Februar 2022 17:50:54 CET Will Cohen wrote:

From: Keno Fischer <[email protected]>

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 this series.

[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
- Move pthread_fchdir_np declaration only to osdep
- Declare pthread_fchdir_np with
- __attribute__((weak_import)) to allow checking for its presence before usage
- Move declarations above cplusplus guard
- Add CONFIG_PTHREAD_FCHDIR_NP to meson and check for presence in osdep.h and os-posix.c
- Rebase to apply cleanly on top of the 2022-02-10 changes to 9pfs]

---
 hw/9pfs/9p-local.c   |  4 ++--
 include/qemu/osdep.h | 12 ++++++++++++
 meson.build          |  1 +
 os-posix.c           | 35 +++++++++++++++++++++++++++++++++++
 4 files changed, 50 insertions(+), 2 deletions(-)
hw/9pfs/9p-local.c +1 -1
index a0d08e5216..d42ce6d8b8 100644
@@ -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;
         }
hw/9pfs/9p-local.c +1 -1
@@ -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;
         }
include/qemu/osdep.h +0 -0
index ce12f64853..c0f442d791 100644
@@ -818,6 +818,18 @@ static inline int

platform_does_not_support_system(const char *command) */
struct dirent *qemu_dirent_dup(struct dirent *dent);

include/qemu/osdep.h +5 -0
+/*
+ * 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. pthread_fchdir_np is weakly linked here as a guard
+ * in case it disappears in future macOS versions, because it is

include/qemu/osdep.h static inline int platform_does_not_support_system(const char *command) · line 835

Quoted 10 lines
include/qemu/osdep.h +7 -0
+ * is a private API.
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+int pthread_fchdir_np(int fd) __attribute__((weak_import));
+#endif
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t dev);
+
 #ifdef __cplusplus
 }
 #endif

Peter just informed me that osdep.h is not the right place to add something like this:
https://lore.kernel.org/qemu-devel/CAFEAcA_=HAUNomKD2wurSVaAHa5mrk22A1oHKLWUDjk7v6Khmg@mail.gmail.com/

So this should be moved into a different header file.

Quoted 59 lines
meson.build +0 -0
index ae5f7eec6e..6fdc0281ad 100644
@@ -1557,6 +1557,7 @@ config_host_data.set('CONFIG_POSIX_FALLOCATE',

cc.has_function('posix_fallocate' config_host_data.set('CONFIG_POSIX_MEMALIGN', cc.has_function('posix_memalign')) config_host_data.set('CONFIG_PPOLL', cc.has_function('ppoll'))
config_host_data.set('CONFIG_PREADV', cc.has_function('preadv', prefix:
'#include <sys/uio.h>')) +config_host_data.set('CONFIG_PTHREAD_FCHDIR_NP', cc.has_function('pthread_fchdir_np')) config_host_data.set('CONFIG_SEM_TIMEDWAIT', cc.has_function('sem_timedwait', dependencies: threads)) config_host_data.set('CONFIG_SENDFILE', cc.has_function('sendfile')) config_host_data.set('CONFIG_SETNS', cc.has_function('setns') and cc.has_function('unshare')) diff --git a/os-posix.c b/os-posix.c index ae6c9f2a5e..ccc3d1e9d3 100644

os-posix.c +15 -0
@@ -332,3 +332,38 @@ 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)
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t dev)
+{
+    int preserved_errno, err;
+    if (!pthread_fchdir_np) {
+        error_report_once("pthread_fchdir_np() is not available on this

version of macOS"); + return -ENOTSUP;

os-posix.c +19 -0
+    }
+    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
author
#14 lore raw

Re: [PATCH v8 09/11] 9p: darwin: Implement compatibility for mknodat

Quoted 87 lines · Christian Schoenebeck
Quoted 70 lines · On Sonntag, 20. Februar 2022 17:50:54 CET Will Cohen wrote:

From: Keno Fischer <[email protected]>

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 this series.

[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
- Move pthread_fchdir_np declaration only to osdep
- Declare pthread_fchdir_np with
- __attribute__((weak_import)) to allow checking for its presence before usage
- Move declarations above cplusplus guard
- Add CONFIG_PTHREAD_FCHDIR_NP to meson and check for presence in osdep.h and os-posix.c
- Rebase to apply cleanly on top of the 2022-02-10 changes to 9pfs]

---
 hw/9pfs/9p-local.c   |  4 ++--
 include/qemu/osdep.h | 12 ++++++++++++
 meson.build          |  1 +
 os-posix.c           | 35 +++++++++++++++++++++++++++++++++++
 4 files changed, 50 insertions(+), 2 deletions(-)
hw/9pfs/9p-local.c +1 -1
index a0d08e5216..d42ce6d8b8 100644
@@ -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;
         }
hw/9pfs/9p-local.c +1 -1
@@ -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;
         }
include/qemu/osdep.h +0 -0
index ce12f64853..c0f442d791 100644
@@ -818,6 +818,18 @@ static inline int

platform_does_not_support_system(const char *command) */
struct dirent *qemu_dirent_dup(struct dirent *dent);

include/qemu/osdep.h +1 -0
+/*

include/qemu/osdep.h static inline int platform_does_not_support_system(const char *command)

Quoted 10 lines
include/qemu/osdep.h +10 -0
+ * 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. pthread_fchdir_np is weakly linked here as a guard
+ * in case it disappears in future macOS versions, because it is
+ * is a private API.
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+int pthread_fchdir_np(int fd) __attribute__((weak_import));
+#endif
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t

dev);

Quoted 4 lines

+
#ifdef __cplusplus
}
#endif

Quoted 7 lines

Peter just informed me that osdep.h is not the right place to add something like this:

https://lore.kernel.org/qemu-devel/CAFEAcA_=HAUNomKD2wurSVaAHa5mrk22A1oHKLWUDjk7v6Khmg@mail.gmail.com/

So this should be moved into a different header file.

Does that logic apply here too? In some ways I'd say that mknodat falls into the category of "things without which code would work on most platforms but fail to compile or misbehave on a minority of host OSes." If osdep.h is the wrong place, should it go back to 9p-util, where it lived before it moved to osdep?

Quoted 65 lines
Quoted 11 lines
meson.build +0 -0
index ae5f7eec6e..6fdc0281ad 100644
@@ -1557,6 +1557,7 @@ config_host_data.set('CONFIG_POSIX_FALLOCATE',

cc.has_function('posix_fallocate' config_host_data.set('CONFIG_POSIX_MEMALIGN', cc.has_function('posix_memalign')) config_host_data.set('CONFIG_PPOLL', cc.has_function('ppoll'))
config_host_data.set('CONFIG_PREADV', cc.has_function('preadv', prefix:
'#include <sys/uio.h>'))

+config_host_data.set('CONFIG_PTHREAD_FCHDIR_NP',

Quoted 14 lines

cc.has_function('pthread_fchdir_np')) config_host_data.set('CONFIG_SEM_TIMEDWAIT', cc.has_function('sem_timedwait', dependencies: threads)) config_host_data.set('CONFIG_SENDFILE', cc.has_function('sendfile')) config_host_data.set('CONFIG_SETNS', cc.has_function('setns') and cc.has_function('unshare')) diff --git a/os-posix.c b/os-posix.c index ae6c9f2a5e..ccc3d1e9d3 100644

os-posix.c +1 -0
@@ -332,3 +332,38 @@ int os_mlock(void)
     return -ENOSYS;
 #endif
 }
+

os-posix.c int os_mlock(void) · line 345

Quoted 10 lines
os-posix.c +10 -0
+/*
+ * 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)
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t

dev)

Quoted 10 lines
os-posix.c +4 -0
+{
+    int preserved_errno, err;
+    if (!pthread_fchdir_np) {
+        error_report_once("pthread_fchdir_np() is not available on this

version of macOS"); + return -ENOTSUP;

os-posix.c +5 -0
+    }
+    if (pthread_fchdir_np(dirfd) < 0) {
+        return -1;
+    }
+    err = mknod(filename, mode, dev);

os-posix.c int os_mlock(void)

Quoted 10 lines
os-posix.c +10 -0
+    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)

Quoted 4 lines
os-posix.c +4 -0
+{
+    return mknodat(dirfd, filename, mode, dev);
+}
+#endif
#15 lore raw

Re: [PATCH v8 09/11] 9p: darwin: Implement compatibility for mknodat

Quoted 125 lines · On Freitag, 25. Februar 2022 15:00:40 CET Will Cohen wrote:

On Tue, Feb 22, 2022 at 9:27 AM Christian Schoenebeck <

Quoted 122 lines · [email protected]> wrote:
Quoted 95 lines · On Sonntag, 20. Februar 2022 17:50:54 CET Will Cohen wrote:

From: Keno Fischer <[email protected]>

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 this series.

[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
- Move pthread_fchdir_np declaration only to osdep
- Declare pthread_fchdir_np with
- __attribute__((weak_import)) to allow checking for

its presence before usage

- Move declarations above cplusplus guard
- Add CONFIG_PTHREAD_FCHDIR_NP to meson and check for

presence in osdep.h and os-posix.c

- Rebase to apply cleanly on top of the 2022-02-10

changes to 9pfs]

---

 hw/9pfs/9p-local.c   |  4 ++--
 include/qemu/osdep.h | 12 ++++++++++++
 meson.build          |  1 +
 os-posix.c           | 35 +++++++++++++++++++++++++++++++++++
 4 files changed, 50 insertions(+), 2 deletions(-)
hw/9pfs/9p-local.c +1 -1
index a0d08e5216..d42ce6d8b8 100644
@@ -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;
         
         }
hw/9pfs/9p-local.c +1 -1
@@ -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;
         
         }
include/qemu/osdep.h +0 -0
index ce12f64853..c0f442d791 100644
@@ -818,6 +818,18 @@ static inline int

platform_does_not_support_system(const char *command) */

struct dirent *qemu_dirent_dup(struct dirent *dent);

include/qemu/osdep.h +1 -0
+/*

include/qemu/osdep.h static inline int platform_does_not_support_system(const char *command)

Quoted 10 lines
include/qemu/osdep.h +10 -0
+ * 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. pthread_fchdir_np is weakly linked here as a guard
+ * in case it disappears in future macOS versions, because it is
+ * is a private API.
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+int pthread_fchdir_np(int fd) __attribute__((weak_import));
+#endif
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t

dev);

Quoted 5 lines

+

#ifdef __cplusplus
}
#endif

Peter just informed me that osdep.h is not the right place to add something
like this:

https://lore.kernel.org/qemu-devel/CAFEAcA_=HAUNomKD2wurSVaAHa5mrk22A1oHKL
[email protected]/

Quoted 7 lines
Quoted 1 line

So this should be moved into a different header file.

Does that logic apply here too? In some ways I'd say that mknodat falls into the category of "things without which code would work on most platforms but fail to compile or misbehave on a minority of host OSes." If osdep.h is the wrong place, should it go back to 9p-util, where it lived before it moved to osdep?

The other function Peter wanted out from osdep.h would fall into that category as well. I understand it like "only put things into osdep.h as very last resort". CCing Peter on this. If he doesn't give you green light, or on doubt, then yes please move it from osdep -> 9p-util.

Will, are you planning to send a v9 of this series soon? Next week I need to add changes that would collide with these here. So if you intend to send it soon, then I just wait so that we can finally queue these macOS patches here for QEMU 7.0, before I would go along with my changes.

Best regards,
Christian Schoenebeck

Quoted 72 lines
Quoted 72 lines
Quoted 10 lines
meson.build +0 -0
index ae5f7eec6e..6fdc0281ad 100644
@@ -1557,6 +1557,7 @@ config_host_data.set('CONFIG_POSIX_FALLOCATE',

cc.has_function('posix_fallocate' config_host_data.set('CONFIG_POSIX_MEMALIGN', cc.has_function('posix_memalign')) config_host_data.set('CONFIG_PPOLL', cc.has_function('ppoll'))

Quoted 2 lines

config_host_data.set('CONFIG_PREADV', cc.has_function('preadv', prefix:
'#include <sys/uio.h>'))

+config_host_data.set('CONFIG_PTHREAD_FCHDIR_NP',

Quoted 17 lines

cc.has_function('pthread_fchdir_np')) config_host_data.set('CONFIG_SEM_TIMEDWAIT', cc.has_function('sem_timedwait', dependencies: threads)) config_host_data.set('CONFIG_SENDFILE', cc.has_function('sendfile')) config_host_data.set('CONFIG_SETNS', cc.has_function('setns') and cc.has_function('unshare')) diff --git a/os-posix.c b/os-posix.c index ae6c9f2a5e..ccc3d1e9d3 100644

os-posix.c +1 -0
@@ -332,3 +332,38 @@ int os_mlock(void)
 
     return -ENOSYS;
 
 #endif
 }
 
+

os-posix.c int os_mlock(void)

Quoted 10 lines
os-posix.c +10 -0
+/*
+ * 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)
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t

dev)

Quoted 10 lines
os-posix.c +4 -0
+{
+    int preserved_errno, err;
+    if (!pthread_fchdir_np) {
+        error_report_once("pthread_fchdir_np() is not available on this

version of macOS"); + return -ENOTSUP;

os-posix.c +5 -0
+    }
+    if (pthread_fchdir_np(dirfd) < 0) {
+        return -1;
+    }
+    err = mknod(filename, mode, dev);

os-posix.c int os_mlock(void)

Quoted 10 lines
os-posix.c +10 -0
+    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)

Quoted 4 lines
os-posix.c +4 -0
+{
+    return mknodat(dirfd, filename, mode, dev);
+}
+#endif
author
#16 lore raw

Re: [PATCH v8 09/11] 9p: darwin: Implement compatibility for mknodat

Quoted 153 lines · Christian Schoenebeck
Quoted 52 lines · On Freitag, 25. Februar 2022 15:00:40 CET Will Cohen wrote:

On Tue, Feb 22, 2022 at 9:27 AM Christian Schoenebeck <

Quoted 49 lines · [email protected]> wrote:
Quoted 48 lines · On Sonntag, 20. Februar 2022 17:50:54 CET Will Cohen wrote:

From: Keno Fischer <[email protected]>

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 this series.

[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
- Move pthread_fchdir_np declaration only to osdep
- Declare pthread_fchdir_np with
- __attribute__((weak_import)) to allow checking for

its presence before usage

- Move declarations above cplusplus guard
- Add CONFIG_PTHREAD_FCHDIR_NP to meson and check for

presence in osdep.h and os-posix.c

- Rebase to apply cleanly on top of the 2022-02-10

changes to 9pfs]

---

 hw/9pfs/9p-local.c   |  4 ++--
 include/qemu/osdep.h | 12 ++++++++++++
 meson.build          |  1 +
 os-posix.c           | 35 +++++++++++++++++++++++++++++++++++
 4 files changed, 50 insertions(+), 2 deletions(-)
Quoted 5 lines
Quoted 5 lines

hw/9pfs/9p-local.c static int local_mknod(FsContext *fs_ctx,

Quoted 5 lines
hw/9pfs/9p-local.c +0 -0
index a0d08e5216..d42ce6d8b8 100644
@@ -682,7 +682,7 @@ static int local_mknod(FsContext *fs_ctx,

V9fsPath

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
hw/9pfs/9p-local.c +1 -1
 *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);
Quoted 7 lines
Quoted 7 lines

hw/9pfs/9p-local.c static int local_mknod(FsContext *fs_ctx,

Quoted 7 lines
         if (err == -1) {

             goto out;

         }
hw/9pfs/9p-local.c +0 -0
@@ -697,7 +697,7 @@ static int local_mknod(FsContext *fs_ctx,

V9fsPath

Quoted 45 lines
Quoted 45 lines
Quoted 26 lines
hw/9pfs/9p-local.c +1 -1
 *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;

         }
include/qemu/osdep.h +0 -0
index ce12f64853..c0f442d791 100644
@@ -818,6 +818,18 @@ static inline int

platform_does_not_support_system(const char *command) */

struct dirent *qemu_dirent_dup(struct dirent *dent);

include/qemu/osdep.h +1 -0
+/*

include/qemu/osdep.h static inline int platform_does_not_support_system(const char *command)

Quoted 10 lines
include/qemu/osdep.h +10 -0
+ * 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. pthread_fchdir_np is weakly linked here as a guard
+ * in case it disappears in future macOS versions, because it is
+ * is a private API.
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+int pthread_fchdir_np(int fd) __attribute__((weak_import));
+#endif
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t

dev);

Quoted 5 lines

+

#ifdef __cplusplus
}
#endif

Quoted 5 lines
Quoted 5 lines

Peter just informed me that osdep.h is not the right place to add something
like this:

https://lore.kernel.org/qemu-devel/CAFEAcA_=HAUNomKD2wurSVaAHa5mrk22A1oHKL

Quoted 7 lines
Quoted 3 lines

[email protected]/

So this should be moved into a different header file.

Does that logic apply here too? In some ways I'd say that mknodat falls into the category of "things without which code would work on most platforms but fail to compile or misbehave on a minority of host OSes."

If

Quoted 2 lines

osdep.h is the wrong place, should it go back to 9p-util, where it lived before it moved to osdep?

The other function Peter wanted out from osdep.h would fall into that category
as well. I understand it like "only put things into osdep.h as very last resort". CCing Peter on this. If he doesn't give you green light, or on doubt,
then yes please move it from osdep -> 9p-util.

Will, are you planning to send a v9 of this series soon? Next week I need to
add changes that would collide with these here. So if you intend to send it
soon, then I just wait so that we can finally queue these macOS patches here
for QEMU 7.0, before I would go along with my changes.

Quoted 3 lines

Best regards,
Christian Schoenebeck

Prepping a v9 now based on all the other changes + the formatting issues. I'll send this weekend if that's okay!

Quoted 79 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
meson.build +0 -0
index ae5f7eec6e..6fdc0281ad 100644
@@ -1557,6 +1557,7 @@ config_host_data.set('CONFIG_POSIX_FALLOCATE',

cc.has_function('posix_fallocate'
config_host_data.set('CONFIG_POSIX_MEMALIGN',
cc.has_function('posix_memalign'))

config_host_data.set('CONFIG_PPOLL',

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

cc.has_function('ppoll'))

config_host_data.set('CONFIG_PREADV', cc.has_function('preadv',

prefix:

Quoted 34 lines
Quoted 34 lines
Quoted 1 line

'#include <sys/uio.h>'))

+config_host_data.set('CONFIG_PTHREAD_FCHDIR_NP',

Quoted 17 lines

cc.has_function('pthread_fchdir_np')) config_host_data.set('CONFIG_SEM_TIMEDWAIT', cc.has_function('sem_timedwait', dependencies: threads)) config_host_data.set('CONFIG_SENDFILE', cc.has_function('sendfile')) config_host_data.set('CONFIG_SETNS', cc.has_function('setns') and cc.has_function('unshare')) diff --git a/os-posix.c b/os-posix.c index ae6c9f2a5e..ccc3d1e9d3 100644

os-posix.c +1 -0
@@ -332,3 +332,38 @@ int os_mlock(void)
 
     return -ENOSYS;
 
 #endif
 }
 
+

os-posix.c int os_mlock(void)

Quoted 10 lines
os-posix.c +10 -0
+/*
+ * 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)
+ */
+#if defined CONFIG_DARWIN && defined CONFIG_PTHREAD_FCHDIR_NP
+
+int qemu_mknodat(int dirfd, const char *filename, mode_t mode, dev_t

dev)

Quoted 4 lines
Quoted 4 lines

os-posix.c int os_mlock(void)

Quoted 4 lines
os-posix.c +4 -0
+{
+    int preserved_errno, err;
+    if (!pthread_fchdir_np) {
+        error_report_once("pthread_fchdir_np() is not available on

this

Quoted 23 lines
Quoted 23 lines
Quoted 6 lines

version of macOS"); + return -ENOTSUP;

os-posix.c +5 -0
+    }
+    if (pthread_fchdir_np(dirfd) < 0) {
+        return -1;
+    }
+    err = mknod(filename, mode, dev);

os-posix.c int os_mlock(void)

Quoted 10 lines
os-posix.c +10 -0
+    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)

Quoted 4 lines
os-posix.c +4 -0
+{
+    return mknodat(dirfd, filename, mode, dev);
+}
+#endif
#17 lore raw

Re: [PATCH v8 09/11] 9p: darwin: Implement compatibility for mknodat

Quoted 5 lines · On Freitag, 25. Februar 2022 17:38:51 CET Will Cohen wrote:

On Fri, Feb 25, 2022 at 11:31 AM Christian Schoenebeck <

Quoted 2 lines · [email protected]> wrote:
Quoted 1 line · On Freitag, 25. Februar 2022 15:00:40 CET Will Cohen wrote:

On Tue, Feb 22, 2022 at 9:27 AM Christian Schoenebeck

[...]

Quoted 18 lines
Quoted 18 lines
Quoted 2 lines

osdep.h is the wrong place, should it go back to 9p-util, where it lived before it moved to osdep?

The other function Peter wanted out from osdep.h would fall into that category
as well. I understand it like "only put things into osdep.h as very last resort". CCing Peter on this. If he doesn't give you green light, or on doubt,
then yes please move it from osdep -> 9p-util.

Will, are you planning to send a v9 of this series soon? Next week I need to
add changes that would collide with these here. So if you intend to send it
soon, then I just wait so that we can finally queue these macOS patches here
for QEMU 7.0, before I would go along with my changes.

Quoted 5 lines
Quoted 2 lines

Best regards,
Christian Schoenebeck

Prepping a v9 now based on all the other changes + the formatting issues. I'll send this weekend if that's okay!

Thumbs up! :)

Best regards,
Christian Schoenebeck