changeset 51649:ba205f944cb4 stable

mmap: add a `is_mmap_safe` method to vfs

This will be useful to safeguard mmap usage to void SIGBUS when repositories
lives on a NFS drive.

--- a/mercurial/util.py	Mon Jun 24 13:14:05 2024 +0200
+++ b/mercurial/util.py	Wed Jul 03 12:22:48 2024 +0200
@@ -441,6 +441,13 @@
 
 
 def mmapread(fp, size=None):
+    """Read a file content using mmap
+
+    The responsability of checking the file system is mmap safe is the
+    responsability of the caller.
+
+    In some case, a normal string might be returned.
+    """
     if size == 0:
         # size of 0 to mmap.mmap() means "all data"
         # rather than "zero bytes", so special case that.

--- a/mercurial/vfs.py	Mon Jun 24 13:14:05 2024 +0200
+++ b/mercurial/vfs.py	Wed Jul 03 12:22:48 2024 +0200
@@ -189,6 +189,35 @@
     def lstat(self, path: Optional[bytes] = None):
         return os.lstat(self.join(path))
 
+    def is_mmap_safe(self, path: Optional[bytes] = None) -> bool:
+        """return True if it is safe to read a file content as mmap
+
+        This focus on the file system aspect of such safety, the application
+        logic around that file is not taken into account, so caller need to
+        make sure the file won't be truncated in a way that will create SIGBUS
+        on access.
+
+
+        The initial motivation for this logic is that if mmap is used on NFS
+        and somebody deletes the mapped file (e.g. by renaming on top of it),
+        then you get SIGBUS, which can be pretty disruptive: we get core dump
+        reports, and the process terminates without writing to the blackbox.
+
+        Instead in this situation we prefer to read the file normally.
+        The risk of ESTALE in the middle of the read remains, but it's
+        smaller because we read sooner and the error should be reported
+        just as any other error.
+
+        Note that python standard library does not offer the necessary function
+        to detect the file stem bits. So this detection rely on compiled bits
+        and is not available in pure python.
+        """
+        # XXX Since we already assume a vfs to address a consistent file system
+        # in other location, we could determine the fstype once for the root
+        # and cache that value.
+        fstype = util.getfstype(self.join(path))
+        return fstype is not None and fstype != b'nfs'
+
     def listdir(self, path: Optional[bytes] = None):
         return os.listdir(self.join(path))