USE_WALL feature can now be used.
Started implementing a USE_CHECKING_WALLS feature, that has a kernel
daemon running that periodically checks the walls of *all* memory
allocations - does not work yet (because of some initialization order
problems).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9396 a95241bf-73f2-0310-859d-f6bbb57e9c96
team and process group IDs as well.
Added a ToDo item that process groups do only work well as long as
their group leader is still active.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9394 a95241bf-73f2-0310-859d-f6bbb57e9c96
ignored completely (so that zero and below become valid).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9393 a95241bf-73f2-0310-859d-f6bbb57e9c96
Now uses waitpid() instead of wait_for_thread() - and surprise, it doesn't
work yet :)
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9372 a95241bf-73f2-0310-859d-f6bbb57e9c96
no longer pop %fs from the iframe in that case, too.
When we restore a custom stack, however, we'll now pop %fs from the iframe
as such strange threads may even do their own business with it.
Added some comments.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9369 a95241bf-73f2-0310-859d-f6bbb57e9c96
We also should disable interrupts in arch_restore_fork_frame().
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9368 a95241bf-73f2-0310-859d-f6bbb57e9c96
always copies the arguments and environment arrays.
No need for the "owns_arrays" in team_arg anymore - it now always owns
them.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9363 a95241bf-73f2-0310-859d-f6bbb57e9c96
Switched to load_image() instad of using the old team_create_team() to start "init".
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9361 a95241bf-73f2-0310-859d-f6bbb57e9c96
The path name is now always taken from the provided argument list (except for
_user_exec() which now replaces args[0] with the path argument).
The team_arg structure know nows if it must delete its arguments or not - this
also fixes a memory leak in case team_create_team() failed when invoked from
userland.
load_image() is now also exported for the kernel.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9360 a95241bf-73f2-0310-859d-f6bbb57e9c96
Enhanced exec_test.c to use some process functions as well.
Switched to load_image() instead of using the old _kern_create_team() syscall.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9359 a95241bf-73f2-0310-859d-f6bbb57e9c96
process ID (in the current session).
get_death_entry() could call free() with interrupts disabled; now follows
the same mechanism as chosen with the groups (ie. it returns a pointer
to a death_entry to be freed by the calling function).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9354 a95241bf-73f2-0310-859d-f6bbb57e9c96
the specified process group (by passing in a negative pid, as in BeOS).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9351 a95241bf-73f2-0310-859d-f6bbb57e9c96
- we now have a working process group/session model. Implemented related syscalls
and the kernel backend, some POSIX process functions are now available in the
kernel as well.
- dying teams now monitor their exitus by leaving death_entry structures in
their parent team.
- Implemented wait_for_child() for pid == -1 and pid > 0; IOW waiting for process
groups is missing yet. Hasn't yet been tested intensively.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9350 a95241bf-73f2-0310-859d-f6bbb57e9c96
may be renamed):
- B_RELEASE_ALL: the semaphore count is set to 0, all waiting threads are released
(the "count" argument of release_etc_sem() is ignored then)
- B_RELEASE_IF_WAITING_ONLY: the semaphore count is only decreased if there
are any waiting threads; ie. the semaphore is signaled
Together, they will make the pthread_cond_*() functions easy to implement, and
they come in handy at other places, too.
Removed release_threads variable in release_sem_etc(), as it was useless.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9330 a95241bf-73f2-0310-859d-f6bbb57e9c96
Also removed syscall for that function (shouldn't have been there in the
first place, anyway).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9325 a95241bf-73f2-0310-859d-f6bbb57e9c96
no longer abuse the semaphore mechanism to carry an extra status value,
instead, wait_for_thread() registers a death_entry with the thread that
will be filled upon exit.
Removed delete_sem_etc() as it's no longer of use or available.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9323 a95241bf-73f2-0310-859d-f6bbb57e9c96
flag set.
Fixes _user_wait_for_thread() which returned an invalid return code in case
of error.
Some minor cosmetics.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9321 a95241bf-73f2-0310-859d-f6bbb57e9c96
not be restored from the iframe; it will be set by the scheduling code.
This is only of concern on SMP machines and could prevent TLS from working.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9310 a95241bf-73f2-0310-859d-f6bbb57e9c96
Made it less vulnerable to thread-safety issues; it could occassionally return an
error when there was an error in a concurrent dlfcn function.
Renamed global static variables to have the "s" prefix.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9305 a95241bf-73f2-0310-859d-f6bbb57e9c96
same mechanism as NewOS, ie. using the iframe "stack" pointer (using the
new i386_get_current_iframe() function).
Implemented fork() support functions arch_{store|restore}_fork_frame().
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9292 a95241bf-73f2-0310-859d-f6bbb57e9c96
know about this specifier, it has to be replaced by B_EXACT_ADDRESS).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9281 a95241bf-73f2-0310-859d-f6bbb57e9c96
team's address space - for internal use, anyway.
Added a ToDo: item to exec_team() to remind me of alarms and signals.
Started implementing fork_team() based on team_create_team() - it's not
completed (or even works), though.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9279 a95241bf-73f2-0310-859d-f6bbb57e9c96
must not be accessed without it.
Implemented _get_next_area_info() and _user_get_next_area_info().
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9277 a95241bf-73f2-0310-859d-f6bbb57e9c96
reference to the address space, so that the calling team deletion function
doesn't have to do this.
delete_address_space() doesn't have to delete any areas anymore, since
at the point it's called, there is no area anymore, anyway: each area
holds a reference to the address space.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9276 a95241bf-73f2-0310-859d-f6bbb57e9c96
This actually reflects the only usage of this function anyway, and thus,
fixes the resize_area() function.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9265 a95241bf-73f2-0310-859d-f6bbb57e9c96
vm_page no longer keeps a pointer to the vm_cache_ref but to the vm_cache.
This cleans the page handling when having to make an area copy-on-write;
regions belong to vm_cache_ref, pages to vm_cache.
Fixed some return types.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9264 a95241bf-73f2-0310-859d-f6bbb57e9c96
It currently checks for temporary caches, but it should really only be used
with regions for anonymous memory.
Fixed some return types.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9259 a95241bf-73f2-0310-859d-f6bbb57e9c96
Changed the way a vm_address_space is deleted: instead of having to explicitly
call vm_delete_aspace(), the last vm_put_aspace() will remove it.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9255 a95241bf-73f2-0310-859d-f6bbb57e9c96
Changed the way a vm_address_space is deleted: instead of having to explicitly
call vm_delete_aspace(), the last vm_put_aspace() will remove it.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9254 a95241bf-73f2-0310-859d-f6bbb57e9c96
Should be tested and verified more intensively, though.
Moved the creation, initialization, and deletion of the team_arg structure out of
team_create_team(); it's now also used by exec_team().
Renamed kfree_strings_array() to free_strings_array().
Renamed the fields of the team_arg structure to be more in sync with the names used
at other places.
Fixed the routine that copied the environment/arguments into the userland space.
Improved and fixed the user_copy_strings_array() function.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9250 a95241bf-73f2-0310-859d-f6bbb57e9c96
of vm_delete_aspace() (which now calls it).
Fixed some return types.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9248 a95241bf-73f2-0310-859d-f6bbb57e9c96
file descriptors that have O_CLOEXEC set.
Minor cleanup.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9242 a95241bf-73f2-0310-859d-f6bbb57e9c96
shouldn't be used in anything that we control (like strncpy()).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9223 a95241bf-73f2-0310-859d-f6bbb57e9c96
back to load_container() so that map_image() has the chance to succeed.
The FATAL() macro now uses dprintf() instead of printf() when TRACE_RLD is defined.
Minor cleanup.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9218 a95241bf-73f2-0310-859d-f6bbb57e9c96
in dir_vnode_to_path() which does now continue until hitting "/".
* Refactored common volume root to mount point resolution code into a
separate function.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9210 a95241bf-73f2-0310-859d-f6bbb57e9c96
As we currently don't have these environment variables set at init time, their values
are hardcoded for the moment.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9207 a95241bf-73f2-0310-859d-f6bbb57e9c96
via B_SET_(NON)BLOCKING_IO to notify the device about any access mode changes.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9199 a95241bf-73f2-0310-859d-f6bbb57e9c96
and will check against those. This fixes the rare case an interrupt frame
could cause before.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9191 a95241bf-73f2-0310-859d-f6bbb57e9c96
potential bug if MAIN_STACK_SIZE and STACK_SIZE would have been different.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9190 a95241bf-73f2-0310-859d-f6bbb57e9c96
read over two blocks, some memory behind that memory could be trashed.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9188 a95241bf-73f2-0310-859d-f6bbb57e9c96
waiting on the write semaphore).
The write functions didn't change the buffer address on a partial write
which could lead to corruption in the pipe data.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9111 a95241bf-73f2-0310-859d-f6bbb57e9c96
Fixed a warning that popped up if TRACE_VM was defined.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9110 a95241bf-73f2-0310-859d-f6bbb57e9c96
This fixes the build as unistd.h defined sleep() as well.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9108 a95241bf-73f2-0310-859d-f6bbb57e9c96
It now prints a warning if the device driver doesn't handle iovecs correctly.
Again, this could lead to very strange behaviour.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9092 a95241bf-73f2-0310-859d-f6bbb57e9c96
name of the node.
devfs_read_stat() now (presumably temporary) sets the size of the device
for partitions and reports a block device rather than a character device
if the size unequals zero.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@9085 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Turned sMountOpMutex into a recursive lock.
* Re-inserted sMountOpLock locking in vnode_path_to_vnode(). It doesn't
deadlock on mounting anymore. :-P
* Added a check in fs_mount(), whether the mount point is a directory. If
it wasn't, in general it wouldn't be possible to get its parent
directory.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8993 a95241bf-73f2-0310-859d-f6bbb57e9c96
why that was there before. Thanks to Ingo for pointing this out.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8981 a95241bf-73f2-0310-859d-f6bbb57e9c96
extensions/ directory - which will contain glibc specific additions
to the POSIX standard (getopt() is here because of getopt_long() etc.).
Also added obstack functions.
Added wcsmbs to the build.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8969 a95241bf-73f2-0310-859d-f6bbb57e9c96
Now maintains uid/gid/mode, modification and creation time for all nodes.
Implemented rootfs_write_stat() completely. rootfs_read_stat() now fills
in all required values.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8962 a95241bf-73f2-0310-859d-f6bbb57e9c96
time in st_mtime.
rootfs_write_stat() now properly locks the node access.
Minor cleanup.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8960 a95241bf-73f2-0310-859d-f6bbb57e9c96
permissions are not tested, though.
Removed the STREAM_TYPE_ constants.
Now maintains a modification and creation time for all nodes.
Implemented rootfs_write_stat() completely. rootfs_read_stat() now fills
in all required values.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8959 a95241bf-73f2-0310-859d-f6bbb57e9c96
Added note to dirfd() that it should be excluded from the kernel build
once the build system allows this.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8958 a95241bf-73f2-0310-859d-f6bbb57e9c96
This normally resides in the posix/ directory in the glibc source repository.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8944 a95241bf-73f2-0310-859d-f6bbb57e9c96
console driver now clears the screen when it's initialized.
This improves the first impression of the standard boot process when
booting in text mode (which currently happens when you press "escape"
in the early boot stage).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8939 a95241bf-73f2-0310-859d-f6bbb57e9c96
IDE drivers :-)) don't respect the total amount of bytes to be read in the
former readv() (now read_pages()) call. This can potentially fix all sorts
of strange errors in userland :)
Fixed the debug build.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8938 a95241bf-73f2-0310-859d-f6bbb57e9c96
recreates a race condition, having it here breaks the kernel as fs_mount()
calls this function and already holds the mutex.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8936 a95241bf-73f2-0310-859d-f6bbb57e9c96
Now panics if devfs_publish_device() fails.
Tracing can now be turned on and off (it's now completely silent in normal operation).
Some cleanups.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8932 a95241bf-73f2-0310-859d-f6bbb57e9c96
unmounted.
* Clarified a comment in fs_mount().
* Added a bit of documentation regarding the mutexes and the basic vnode
functions.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8930 a95241bf-73f2-0310-859d-f6bbb57e9c96
available at /bin/sh (a symlink will do). Unfortunately there's some
problem with our shell that causes it to crash in most cases. :-(
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8925 a95241bf-73f2-0310-859d-f6bbb57e9c96
always do. More often than not one gets page faults. As it happens
sometimes even before anything meaty has been done at all
(e.g. in main()->init_statements()->shell_var_set_text()->
shell_value_init_text()->malloc()), I believe that's a problem not caused
by me. And I wasn't able to track it down.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8924 a95241bf-73f2-0310-859d-f6bbb57e9c96
Userland. It must always set the d_pdev and d_pino fields and, if the
Entry refers to a mount point even replace d_dev and d_ino.
* Added a TODO comment to entry_ref_to_vnode(). It should take care of
mount points, as vnode_path_to_vnode() does, I believe. Will verify
and fix that next.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8921 a95241bf-73f2-0310-859d-f6bbb57e9c96
this is the one thing most likely to fail - it now also handles this case
gracefully instead of dying.
Small cleanup, cleared some other ToDos: some user functions now delete
the area when they could not copy the target address.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8911 a95241bf-73f2-0310-859d-f6bbb57e9c96
leaving the tags Qt style though. That's a bit weird, but that's the
way Axel likes it, and we want to keep him happy, don't we? ;-P
* The common_[path_]read_stat() now fills in st_dev and st_ino in the
result stat structure after the FS had its go. The information is
readily available in the VFS, so why bother the FSs. In fact devfs
didn't (and still doesn't) fill in st_dev which gave dev_for_path()
some head ache.
* Modified _user_open_parent_dir() to accept a NULL buffer for the
entry name and added a doxygen comment for it.
df now works properly on Haiku (save that printf() doesn't support
floating point numbers at the moment).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8898 a95241bf-73f2-0310-859d-f6bbb57e9c96
always also access it through the devfs_vnode. This also solves an issue
in devfs_read() where an access of a cookie->stream would crash the
kernel, since cookie->stream was not set in devfs_open().
* devfs_open() now doesn't fail for directories and symlinks anymore.
That prevented BNode to work for devfs dirs and symlinks.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8897 a95241bf-73f2-0310-859d-f6bbb57e9c96
`.'. This fixes the _kern_open_parent_dir() syscall for the root directory.
I realized, that rootfs_read_dir() doesn't return the entries `.' and `..',
unlike Be's rootfs. Shall I add those, Axel?
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8895 a95241bf-73f2-0310-859d-f6bbb57e9c96
This has the advantage that the device can now be access in the right context.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8876 a95241bf-73f2-0310-859d-f6bbb57e9c96
Fixed the calculation of the amount of bytes read in bootfs_read_pages().
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8874 a95241bf-73f2-0310-859d-f6bbb57e9c96
Added an vfs_get_cookie_from_fd() call to at least temporarily support the
cookie as in the original devfs (not sure yet how this changes).
Implemented vfs_get_file_map() which calls the corresponding FS function.
Moved vfs_get_vnode_cache() around.
vfs_get_vnode() temporarily does its job unlocked - its only to be used (safely)
from within the file cache, but this should definitely be done better.
Fixed a bug in get_vnode_name() - it did not support getting the name of
a root directory; it didn't pass the call through to the parent file system.
Fixed a bug in _user_entry_ref_to_path() which would add another "/" for
files immediately under the root (ie. "df" showed "//boot" as mount point
for the boot partition).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@8873 a95241bf-73f2-0310-859d-f6bbb57e9c96
