This patch employs one solution to resolve two independent but related
issues. Both issues are the result of one fundamental aspect of the
way VM's memory mapping works: VM uses its cache to map in blocks for
memory-mapped file regions, and for blocks already in the VM cache, VM
does not go to the file system before mapping them in. To preserve
consistency between the FS and VM caches, VM relies on being informed
about all updates to file contents through the block cache. The two
issues are both the result of VM not being properly informed about
such updates:
1. Once a file system provides libminixfs with an inode association
(inode number + inode offset) for a disk block, this association
is not broken until a new inode association is provided for it.
If a block is freed and reallocated as a metadata (non-inode)
block, its old association is maintained, and may be supplied to
VM's secondary cache. Due to reuse of inodes, it is possible
that the same inode association becomes valid for an actual file
block again. In that case, when that new file is memory-mapped,
under certain circumstances, VM may end up using the metadata
block to satisfy a page fault on the file, due to the stale inode
association. The result is a corrupted memory mapping, with the
application seeing data other than the current file contents
mapped in at the file block.
2. When a hole is created in a file, the underlying block is freed
from the device, but VM is not informed of this update, and thus,
if VM's cache contains the block with its previous inode
association, this block will remain there. As a result, if an
application subsequently memory-maps the file, VM will map in the
old block at the position of the hole, rather than an all-zeroes
block. Thus, again, the result is a corrupted memory mapping.
This patch resolves both issues by making the file system inform the
minixfs library about blocks being freed, so that libminixfs can
break the inode association for that block, both in its own cache and
in the VM cache. Since libminixfs does not know whether VM has the
block in its cache or not, it makes a call to VM for each block being
freed. Thus, this change introduces more calls to VM, but it solves
the correctness issues at hand; optimizations may be introduced
later. On the upside, all freed blocks are now marked as clean,
which should result in fewer blocks being written back to the device,
and the blocks are removed from the caches entirely, which should
result in slightly better cache usage.
This patch is necessary but not sufficient to resolve the situation
with respect to memory mapping of file holes in general. Therefore,
this patch extends test 74 with a (rather particular but effective)
test for the first issue, but not yet with a test for the second one.
This fixes#90.
Change-Id: Iad8b134d2f88a884f15d3fc303e463280749c467
There are currently no devices out there that require this change.
The change is merely needed to support subsequent changes.
Change-Id: I64214c5f46ff4a2260815d15c15e4a17709b9036
This barrier ensures that all fields of an asynchronously sent
message are properly initialized before the message is marked as
valid.
Change-Id: I7b9590c11c4e040c8f992f1dd2581e54201bf214
Previously, there was a tiny chance that tickdelay(3) would return
early or that it would fail to reinstate a previous alarm.
- sys_setalarm(2) now returns TMR_NEVER instead of 0 for the time
left if no previous alarm was set;
- sys_setalarm(2) now also returns the current time, to allow the
caller to determine whether it got an alarm notification for the
alarm it set or for a previous alarm that has just gone off;
- tickdelay(3) now makes use of these facilities.
Change-Id: Id4f8fe19a61ca8574f43131964e6f0317f613f49
Extended by David van Moolenbroek to continue using static buffers
for short inode names, so as to prevent important file system
services such as procfs from running out of memory at runtime.
Change-Id: I6f841741ee9944fc87dbdb78b5cdaa2abee9da76
As part of its built-in mmap emulation support for "none" file system
services, libfsdriver clears the VM cache upon exit. However, for
trivial file systems which do not even support reading from files, the
the VM cache need to be cleared either. With this patch, the VM cache
is cleared only when modified, so that such trivial file systems need
not be given CLEARCACHE permission.
Change-Id: I518c092443455302b9b9728f10a3f894d2c8036b
While putnode requests should always succeed, very simple file system
services may not care about reference counts and thus about putnode
requests at all. For this reason, we now default to an OK response if
no fdr_putnode implementation is given.
Change-Id: I01f6421abf4546a1f69d8c21900a92d6acc45745
The stat.st_ino field must always be filled with the inode number
given as part of the fdr_stat request anyway, so libfsdriver can
simply fill in the number and allow the file system not to bother.
Change-Id: Ia7a849d0b23dfc83010df0d48fa26e4225427694
This change requires a small patch to libc, in order to avoid that
libminc has to pull in a large chunk of libc just for mktime(3).
Change-Id: I48e598b3716eff626cac461f78a41e32334e6b28
Bad logic introduced as part of the fsdriver changes could cause
getdents to terminate early in these libraries.
Issue reported by r0ller.
Change-Id: If450d5ea85e830584878d8a4ec0f00519355a353
. bitcode fixes
. switch to compiler-rt instead of netbsd libc functions
or libgcc for support functions for both x86 and arm
. minor build fixes
. allow build with llvm without crossbuilding llvm itself
. can now build minix/arm using llvm and eabi - without C++
support for now (hence crossbuilding llvm itself is turned off
for minix/arm)
Change-Id: If5c44ef766f5b4fc4394d4586ecc289927a0d6eb
The new implementation of this library provides abstractions for
network drivers, and should be used for all network drivers from now
on. It provides the following functionality:
- a function call table abstraction, hiding the details of the
datalink protocol with simple parameters;
- a state machine for sending and receiving packets, freeing the
actual driver from keeping track of pending requests;
- an abstraction for copying data from and to the network driver,
freeing the actual driver from dealing with I/O vectors while at
the same time providing a copy implementation which is more
efficient than most current driver implementations;
- a generalized implementation of zero-copy port-based I/O;
- a clearer set of policies and defaults.
While the concept is very similar to lib{block,char,fs,input}driver,
one main difference is that libnetdriver now also takes care of SEF
initialization, mainly so that aspects such as recovery policies and
live-update aspects can be changed for all network drivers in a
single place. As always, for the case that the provided message loop
is too restrictive, a set of more low-level message processing
functions is provided.
The netdriver API has been designed so as to allow alleviation of one
current protocol bottleneck: the fact that at most one send request
and one receive request may be pending at any time. Changing this
aspect will however require a significant rewrite of libnetdriver,
and possibly debugging of drivers that are not able to cope with (in
particular) queuing multiple packets for transmission at once.
Beyond that, the design of the new API is based on the current
protocol, and may be changed/extended later to allow for non-ethernet
network drivers, exposure of link status, multicast address
configuration, suspend and resume, and any other features that are in
fact long overdue.
Change-Id: I47ec47e05852c42f92af04549d41524f928efec2
This patch adds (very limited) support for memory-mapping pages on
file systems that are mounted on the special "none" device and that
do not implement PEEK support by themselves. This includes hgfs,
vbfs, and procfs.
The solution is implemented in libvtreefs, and consists of allocating
pages, filling them with content by calling the file system's READ
functionality, passing the pages to VM, and freeing them again. A new
VM flag is used to indicate that these pages should be mapped in only
once, and thus not cached beyond their single use. This prevents
stale data from getting mapped in without the involvement of the file
system, which would be problematic on file systems where file contents
may become outdated at any time. No VM caching means no sharing and
poor performance, but mmap no longer fails on these file systems.
Compared to a libc-based approach, this patch retains the on-demand
nature of mmap. Especially tail(1) is known to map in a large file
area only to use a small portion of it.
All file systems now need to be given permission for the SETCACHEPAGE
and CLEARCACHE calls to VM.
A very basic regression test is added to test74.
Change-Id: I17afc4cb97315b515cad1542521b98f293b6b559
This obviates the need for several file system implementations to
remember the device on which they are mounted.
Change-Id: Ida8325cf4bcf072e61761cfee34e3f7ed2d750b9
- rename start_vtreefs to run_vtreefs, since the function returns upon
termination these days;
- add get_inode_slots function to retrieve the number of indexed slots;
- add support for extra per-inode data for arbitrary storage.
Change-Id: If2d365d7b478a1cecc9e20fb2b3e70c1a1cf7243
The entire infrastructure relied on an ACK feature, and as such, it
has been broken for years now, with no easy way to repair it.
Change-Id: I783c2a21276967af115a642199f31fef0f14a572
- synchronize request type with ioctl by making it unsigned long;
- unbreak VFS requests, as they were being sent to PM;
- use proper ioctl direction flags (and new numbers) for requests;
- remove some needless header inclusions;
- svrctl is in libc, make its message name reflect this;
- keep backward compatibility: svrctl is part of the userland ABI.
Change-Id: I44902e8d0d11b8ebc1ef3bda94d2202481743c9b
UDS expects the device number of the actual socket, not the device on
which the socket happens to reside. The code worked only because PFS
returned the same value in the st_dev stat field, which it will have
to continue doing for a while now.
Change-Id: I426d38a86a96307ff6e6ed8099d37dae02d6bf2b
The new functionality aims to save each file system server from having
to implement its own block I/O routines just so that it can serve as a
root file system. The new source file (bio.c) lists the requirements
that file system servers have to fulfill in order to use the routines.
Change-Id: Ia0190fd5c30e8c2097ed8f4b0e3ccde1827e0b92
The file system may not be expecting these upcalls at arbitrary
moments, while they serve only as a performance optimization anyway.
Change-Id: I0748fd1f6c2645ddbb64466093ee36025aac45e0
This library provides new abstractions for the upper (VFS) side of
file system services, and should be used for all file system service
implementations from now on. It provides the following functionality:
- a function call table abstraction, hiding the details of the
VFS-FS protocol with simple parameters;
- a (currently limited) number of per-function steps required for
all file system implementations, such as copying in and out path
names and result buffers;
- a default implementation for multicomponent path lookups, such
that the file system merely has to implement resolution of single
components at a time;
- an abstraction for copying data from and to the file system, which
allows transparent intraprocess copying as required for the lookup
implementation;
- a set of functions to simplify getdents implementations.
The message loop provided by the library is currently for use by
single-threaded file system implementations only. Multithreaded file
system services may use the more low-level message processing
functionality.
Protocol-level optimizations such as including names in protocol
messages may be hidden entirely in this library. In addition, in the
future, the lookup implementation may be replaced by a single-
component lookup VFS/FS protocol request as part of a VFS name cache
implementation; this, too, can be hidden entirely in this library.
Change-Id: Ib34f0d0e021dfa3426ce8826efcf3eaa94d3ef3e
Known limitations:
- comment for now testisofs, as iso9660fs is known to be broken.
Benefits:
- near 3x speed improvement on C++ code compilation, bringing down
make build to from 44min down to 21min.
- Allows for X applications to work properly, which should be available
in near-term future through pkgsrc for 3.3.0.
Change-Id: I8f4179a7ea925ed381642add32cfd8c5822217e4
- replace a stray assert(0) with abort()
- remove unrequired copy-pasted #undef NDEBUG
- replace some #if NDEBUG by #if DEBUG as they protect debug printf()s.
Change-Id: Iff4c0331b06e860d32d91ce6b1d6c765ed065c8b
- assert() is macro which is defined as empty, while panic is always
present. I added an explicit abort() after the macro to make sure the
function never returns in case of wrong flags.
- Fixed gcc build with -NDEBUG, -Os for ARM.
* A few 'may be used uninitialized' messages
* A few new missing support library functions where added in libminc.
Change-Id: I69fcda2cd3888390b7ddeff4c0cd849105ce86ff
