curlhttpfs :: FUSE HTTP Filesystem

Example:

$ make curlhttpfs
gcc -Wall -I/usr/include/fuse/ -D_FILE_OFFSET_BITS=64 curlhttpfs.c -o curlhttpfs -lcurl -lpthread -lfuse

$ mkdir remote
$ ./curlhttpfs -o base_url=http://tthsum.devs.nu,ro remote
$ mount | tail -n1
curlhttpfs on .../curlhttpfs/remote type fuse.curlhttpfs (ro,nosuid,nodev,user=walter)

$ cat remote/index.html | file -
/dev/stdin: HTML document, ASCII text

$ wget -qO- http://tthsum.devs.nu/pkg/tthsum-1.2.0-win32-bin.zip | md5sum
a8c756a3ae93732d93380fb383754dce  -
$ md5sum remote/pkg/tthsum-1.2.0-win32-bin.zip
a8c756a3ae93732d93380fb383754dce  remote/pkg/tthsum-1.2.0-win32-bin.zip

Bugs/Limitations:

• Due to the way the kernel traverses the filesystem (using getattr on individual path elements) we have no clean way to discern files from directories. Right now we use the period ('.') as heuristic. If there is a period in the path-element, we treat it as a file; otherwise it is a directory.

• Many webservers do not handle HTTP range requests. (Especially for dynamically generated content.) Right now support for those servers is broken.

• Getting large files with small block sizes, yields a tremendous amount of work for both the server and the client. Compare these:

  $ dd if=remote/pkg/tthsum-1.2.0-win32-bin.zip of=tmp.log
  313+1 records in
  313+1 records out
  160636 bytes (161 kB) copied, 1.01863 s, 158 kB/s
  
  $ dd if=remote/pkg/tthsum-1.2.0-win32-bin.zip of=tmp.log bs=1M
  0+1 records in
  0+1 records out
  160636 bytes (161 kB) copied, 0.0257208 s, 6.2 MB/s
  

• The filesystem should be mounted 'ro' automatically.

• Vim(1) will create an seemingly endless loop trying to write swapfiles with different names on the RO-filesystem. Don't try to write any files there.

• There are a few unchecked buffers here and there, they should be fixed: strcpy, strcat, sprintf. Buffer length of URL should be increased. Structure names and properties could be refactored for better naming and clarity.

• Logging should be optional (mount option).

• Should add curl options to specify timeouts and think about how to handle those.