RAM-backed small-file cache with auto-flush; mount without PUT-Range; cache_threshold option; robust unmount

DAV-requirements.md

@@ -25,9 +25,8 @@ Renaming files, creating directories, removing files/directories
 ## For writing files.
 
 Writing files is a delicate operation, we should take care to do it
-correctly. Right now, the driver checks if we're talking to an Apache
-or SabreDAV implementation because they are the only ones that implement
-partial put.
+correctly. The driver checks if we're talking to an Apache or SabreDAV
+implementation because they are the only ones that implement partial updates.
 
 - If-Match: * / If-None-Match: * support (RFC2616).  
   If-Match: * is used with the PUT method to prevent files being written
@@ -36,14 +35,20 @@ partial put.
   if it already exists.  
   Though this is very basic, there are servers that do not implement this,
   or not correctly.
-- Partial PUT support.  
+- Partial PUT support (preferred).  
   This means writing just a part of a file, updating it in-place, instead
-  of replacing an existing file. webdavFS detects what webserver it is
-  talking  to. If it's Apache it uses PUT + Content-Range, if it's
+  of replacing an existing file. webdavfs detects what webserver it is
+  talking to. If it's Apache it uses PUT + Content-Range, if it's
   SabreDAV it uses PATCH + X-Update-Range. For more info, see:  
   https://blog.sphere.chronosempire.org.uk/2012/11/21/webdav-and-the-http-patch-nightmare  
   http://sabre.io/dav/http-patch/  
 
+- Fallback: full PUT on close.  
+  If partial PUT is not available, webdavfs supports read–write by using a
+  per‑open in‑RAM buffer. The full file is uploaded with a normal PUT on
+  fsync/close or after a short inactivity period. Servers only need to support
+  standard `PUT`, `GET`, and conditional `If-Match`/`If-None-Match`.
+
 ## Partial PUT support as a standard
 
 There seems to be some movement in this space. RFC9110 mentions

README.md

@@ -2,15 +2,13 @@
 
 ## A FUSE filesystem for WEBDAV shares.
 
-Most filesystem drivers for Webdav shares act somewhat like a mirror;
-if a file is read it's first downloaded then cached in its entirety
-on a local drive, then read from there. Writing files is similar or
-even worse- a partial update to a file might involve downloading it first,
-modifying it, then uploading it again. In many cases that is not optimal.
+Most filesystem drivers for WebDAV shares mirror files locally and operate on
+on-disk caches. Partial updates often involve downloading the full file,
+modifying it, then uploading it again.
 
-This filesystem driver behaves like a network filesystem. It doesn't
-cache anything locally, it just sends out partial reads/writes over the
-network.
+webdavfs behaves like a network filesystem and supports efficient partial I/O
+when the server allows it. It also has a robust RAM-backed mode for servers
+without partial write support.
 
 For that to work, you need partial write support- and unfortunately,
 there is no standard for that. See
@@ -22,13 +20,11 @@ used by e.g. NextCloud) for partial writes. So we detect if it's Apache or
 SabreDav we're talking to and then use their specific methods to partially
 update files.
 
-If no support for partial writes is detected, mount.webdavfs will
-print a warning and mount the filesystem read-only. In that case you can
-also use the `rwdirops` mount option, this will make metadata writable
-(i.e. you can use rm / mv / mkdir / rmdir) but you still won't be able
-to write to files.
-
-But if you only need to read files it's still way faster than davfs2 :)
+If no support for partial writes is detected, webdavfs still allows full
+read–write operation by buffering file content in RAM per open handle, and
+flushing the entire file with a single PUT on close (or fsync). There is also
+an inactivity auto‑flush (default 10s) to push changes even if the application
+keeps the file open.
 
 ## What is working
 
@@ -38,6 +34,10 @@ Basic filesystem operations.
 - directories: mkdir rmdir readdir
 - query filesystem size (df / vfsstat)
 
+Small files (by default ≤64 MiB) use a per‑open in‑RAM cache for reads/writes,
+and upload on close/fsync/auto‑flush. Large files use ranged I/O when the
+server supports partial updates.
+
 ## What is not yet working
 
 - locking
@@ -47,7 +47,9 @@ Basic filesystem operations.
 - change permissions (all files are 644, all dirs are 755)
 - change user/group
 - devices / fifos / chardev / blockdev etc
-- truncate(2) / ftruncate(2) for lengths between 1 .. currentfilesize - 1
+- truncate(2) / ftruncate(2) shrinking a file while using ranged I/O.
+  Shrinking is supported for handles using the RAM cache (small files or
+  when the server lacks partial write support).
 
 This is basically because these are mostly just missing properties
 from webdav.
@@ -96,6 +98,7 @@ Using it is simple as:
 ```
 # mount -t webdavfs -ousername=you,password=pass https://webdav.where.ever/subdir /mnt
 ```
+On exit (Ctrl+C or SIGTERM), webdavfs unmounts the mountpoint automatically.
 
 ## Command line options
 
@@ -133,6 +136,15 @@ Using it is simple as:
 | maxidleconns          | Maximum number of idle connections (default 8)
 | sabredav_partialupdate | Use the sabredav partialupdate protocol even when
 |                        | the remote server doesn't advertise support (DANGEROUS)
+| cache_threshold        | Size in bytes above which files use non‑cached ranged
+|                        | I/O (if available). Default 67108864 (64 MiB).
+| cache_threshold_mb     | Same as cache_threshold but specified in MiB. |
+
+Notes:
+- If the server does not support partial writes, all files use the in‑RAM
+  per‑open cache regardless of size.
+- The in‑RAM cache auto‑flushes after 10 seconds of inactivity per open handle,
+  or on fsync/close.
 
 If the webdavfs program is called via `mount -t webdavfs` or as `mount.webdav`,
 it will fork, re-exec and run in the background. In that case it will remove
@@ -168,4 +180,3 @@ experience and better performance, here are a few ideas:
 - DELETE Depth 0 for collections (no delete if non-empty)
 - return updated PROPSTAT information after operations
   like PUT / DELETE / MKCOL / MOVE
-

cache.go

@@ -9,7 +9,6 @@ func (nd *Node) statInfoFresh() bool {
 	return nd.LastStat.Add(statCacheTime).After(now)
 }
 
-func (nd* Node) statInfoTouch() {
+func (nd *Node) statInfoTouch() {
 	nd.LastStat = time.Now()
 }
-

daemon.go

@@ -1,4 +1,3 @@
-
 package main
 
 import (
@@ -83,8 +82,8 @@ func Daemonize() error {
 
 	// now re-exec ourselves.
 	attrs := os.ProcAttr{
-		Files: []*os.File{ devnull, wout, werr },
-		Sys: &syscall.SysProcAttr{ Setsid: true },
+		Files: []*os.File{devnull, wout, werr},
+		Sys:   &syscall.SysProcAttr{Setsid: true},
 	}
 	os.Setenv(isDaemonEnv, "YES")
 	proc, err := os.StartProcess(binary, os.Args, &attrs)
@@ -126,7 +125,7 @@ func Daemonize() error {
 }
 
 // Returns true when this process is a daemonized child.
-func IsDaemon() (bool) {
+func IsDaemon() bool {
 	return os.Getenv(isDaemonEnv) != ""
 }
 
@@ -141,4 +140,3 @@ func Detach() error {
 	fh.Close()
 	return nil
 }
-

debug.go

@@ -1,4 +1,3 @@
-
 package main
 
 import (
@@ -8,7 +7,8 @@ import (
 )
 
 var dbgChan = make(chan string, 8)
-func init () {
+
+func init() {
 	go func() {
 		for {
 			line := <-dbgChan
@@ -28,4 +28,3 @@ func dbgJson(obj interface{}) string {
 	}
 	return fmt.Sprintf("%+v", obj)
 }
-