Subversion Repositories HelenOS

Rev

Go to most recent revision | Show changed files | Details | Compare with Previous | Blame | RSS feed

Filtering Options

Rev Age Author Path Log message Diff
4585 5464 d 17 h jermar /trunk/uspace/ Introduce VFS_IN_FSTAT and VFS_OUT_STAT.
Provide libc fstat() and devfs_stat().
This functionality replaces VFS_IN_NODE
and VFS_IN/OUT_DEVICE. FAT and TMPFS
still do not implement this and VFS_IN_STAT
and stat() need implementation as well.
 
4584 5464 d 22 h jermar /trunk/uspace/ Separate the VFS input and output protocols.  
4566 5467 d 15 h jermar /trunk/uspace/srv/vfs/ Do not use the "pending" fibril mechanism in VFS  
4555 5475 d 14 h jermar /trunk/uspace/srv/vfs/ Finish converting VFS to fibril synchronization.  
4551 5475 d 15 h jermar /trunk/uspace/srv/vfs/ Handle each VFS request using a dedicated connection.  
4539 5477 d 17 h jermar /trunk/uspace/srv/vfs/ Handle pending mounts using only one fibril.  
4520 5482 d 17 h jermar /trunk/uspace/srv/vfs/ Synchronize access to pending_req.  
4518 5483 d 14 h jermar /trunk/uspace/srv/vfs/ Make VFS use the new synchronization for fibrils. Now there should be no (or
only secondary) fibril serialization. Code reorganized not to hold the phone
lock during async_wait_for() in most cases. Tested on ia32. On amd64, VFS
crashes, but I think it is an unrelated problem.
 
4509 5484 d 22 h decky /trunk/uspace/ rename libadt -> adt  
4463 5489 d 17 h decky /trunk/uspace/srv/vfs/ VFS rewrite:
- add new VFS methods
- VFS_NODE for getting VFS node for a file descriptor
- VFS_OPEN_NODE for opening an i-node directly
- VFS_DEVICE for getting the underlying device of a file (if any)
- VFS_SYNC for syncing buffers
- L_OPEN flag is set to underlying filesystem in VFS_LOOKUP when opening a file to support initialization (if any)
- VFS_CLOSE is now propagated to underlying filesystem to support cleanup (if any)
 
4445 5491 d 19 h decky /trunk/uspace/srv/vfs/ revert commit 4443 and implement a much finer solution as suggested by Jakub Jermar in commit 4444  
4443 5494 d 18 h decky /trunk/uspace/srv/vfs/ serialization of VFS_OPEN implementation to avoid lockup by concurrent open() operations from multiple tasks  
4437 5496 d 2 h jermar /trunk/uspace/srv/vfs/ Fix function definition.  
4431 5500 d 20 h jermar /trunk/uspace/srv/vfs/ Hold the namespace rwlock during the entire operation of vfs_mount_internal().
Get rid of rootfs futex, which is not needed anymore.
Do not instantiate VFS node for the mounted root if the mount failed.
 
4427 5502 d 16 h jermar /trunk/uspace/srv/vfs/ Fix rename() on a non-root file system.  
4409 5504 d 14 h jermar /trunk/uspace/ Rudimentary support for non-root mounts.  
4368 5515 d 3 h jermar /trunk/uspace/ Lookup a node's parent lexically rather than using the broken L_PARENT flag of
VFS_LOOKUP.
 
4366 5515 d 15 h jermar /trunk/uspace/srv/vfs/ Allow rename("a", "aa");  
4305 5535 d 21 h jermar /trunk/uspace/ Support mount options.
Let tmpfs decide whether to start empty or restored a TMPFS dump based on the mount option.
 
4302 5535 d 22 h jermar /trunk/uspace/ Make IPC_M_PING part of the VFS_MOUNT protocol again.  
4301 5535 d 23 h jermar /trunk/uspace/srv/vfs/ Return the actual error code instead of EREFUSED.  
4279 5539 d 17 h svoboda /trunk/uspace/ Nuke strncmp().  
4264 5544 d 15 h svoboda /trunk/uspace/ Nuke strcpy() and strcmp().  
4159 5564 d 15 h jermar /trunk/uspace/srv/vfs/ Just for pro forma, initialize the the link before use.  
4010 5582 d 18 h decky /trunk/uspace/srv/vfs/ support for pending (blocking) mounts (waiting for the presence of the filesystem implementation)
the mount point and filesystem type arguments of VFS_MOUNT were swapped, the IPC_M_PING was eliminated
small cleanups
 
3653 5675 d 20 h jermar /trunk/uspace/srv/vfs/ Hold the namespace_rwlock during readdir().  
3488 5726 d 17 h svoboda /trunk/uspace/ Fix assorted warnings.  
3352 5774 d 16 h jermar /trunk/uspace/srv/ Add reference to the mounted FS root.

This makes it possible to keep the FS root logically unlinked (i.e. lnkcnt == 0)
because it prevents VFS from attempting to destroy it during closedir().
 
3255 5788 d 2 h jermar /trunk/uspace/ Fix the VFS protocol so that the client can determine that VFS_MOUNT failed due
to a request to mount an unregistered file system.
 
3215 5820 d 23 h jermar /trunk/uspace/srv/vfs/ Instead of aborting or crashing, return EBADF when an invalid file descriptor
is passed to vfs_close().
 
3109 5851 d 21 h jermar /trunk/uspace/srv/ Split the 'mount another filesystem here' and 'you are being mounted and the
device is this' mount semantics. Add VFS_MOUNTED VFS operation that corresponds
to the latter and reserve VFS_MOUNT only for the former. Because of this
change, the VFS server does not maintain the mr_node VFS node for the name space
root anymore and the VFS_LOOKUP operation is now not meant to be used on
unmounted file system, not even for looking up the root node of unmounted file
systems. In the light of these changes, TMPFS is now initialized from
tmpfs_mounted() function.
 
3079 5854 d 21 h decky /trunk/uspace/srv/vfs/ disable extensive debugging output  
2958 5870 d 14 h jermar /trunk/uspace/srv/ Make VFS_MOUNT call even when mounting the root file system.  
2957 5870 d 16 h jermar /trunk/uspace/srv/vfs/ Cram VFS_MOUNT into one message. This creates a hidden demand for the sixth IPC
argument as there might be a desire to pass some options/flags in the future.
 
2788 5922 d 4 h jermar /trunk/uspace/ Return EBUSY if trying to mount something over an already mounted root FS.
(Contributed by Tim Post)
 
2770 5938 d 16 h jermar /trunk/uspace/ Introduce fs_handle_t, dev_handle_t and fs_index_t.  
2766 5940 d 16 h jermar /trunk/uspace/srv/vfs/ Grab the nodes_futex when manipulating node link counts in rename() and
_unlink().

Fix VFS futex definitions and declarations by replacing atomic_t by futex_t.
 
2763 5940 d 19 h jermar /trunk/uspace/ Support for rename().  
2752 5947 d 21 h jermar /trunk/uspace/srv/vfs/ Make VFS canonify path names on lookup.  
2749 5950 d 14 h jermar /trunk/uspace/srv/vfs/ contents_rwlock should not be acquired as a futex but rather as a rwlock.