Add SCHED_RESET_ON_FORK to <sys/sched.h>. If this flag is set, SCHED_FIFO
and SCHED_RR are reset to SCHED_OTHER and negative nice values are reset to
zero in each child process created with fork(2).
Signed-off-by: Christian Franke <christian.franke@t-online.de>
Add SCHED_BATCH to <sys/sched.h>. SCHED_BATCH is similar to SCHED_OTHER,
except that the nice value is mapped to a one step lower Windows priority.
Rework the mapping functions to ease the addition of this functionality.
Signed-off-by: Christian Franke <christian.franke@t-online.de>
init_reopen_attr() doesn't guard against a NULL handle. However,
there are scenarios calling functions deliberately with a NULL handle,
for instance, av::setup() calling check_file_access() only if opening
the file did NOT succeed.
So check for a NULL handle in init_reopen_attr() and if so, use the
name based approach filling the OBJECT_ATTRIBUTES struct, just as in
the has_buggy_reopen() case.
Fixes: 4c9d01fdad2a ("* mount.h (class fs_info): Add has_buggy_reopen flag and accessor methods.")
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
There seems to be no rationale reason for _cygtls::spinning to exist,
so it has been removed. _cygtls::spinning was introduced in the commit
edc4f86ad282A, and this flag means that another thread is waiting to
acquire _cygtls::stacklock. It is checked in the sig thread in _cygtls::
interrupt_now(), and if spinning is true, the interrupt will be delayed.
However, in this case, _cygtls::incyg is also set, so it is supposed to
be covered by _cygtls::incyg flag and does not seem to add any value.
If some problems happen in the signal handling, it might be a good idea
to revert this commit and check if the issue will be fixed.
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Currently, the signal queue is touched by the thread sig as well as
other threads that call sigaction_worker(). This potentially has
a possibility to destroy the signal queue chain. A possible worst
result may be a self-loop chain which causes infinite loop. With
this patch, lock()/unlock() are introduce to avoid such a situation.
Fixes: 474048c26edf ("* sigproc.cc (pending_signals::add): Just index directly into signal array rather than treating the array as a heap.")
Suggested-by: Corinna Vinschen <corinna@vinschen.de>
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
The queue is cleaned up by removing the entries having si_signo == 0
while processing the queued signals, however, sigpacket::process() may
set si_signo in the queue to 0 of the entry already processed but not
succeed by calling sig_clear(). This patch ensures the sig_clear()
to remove the entry from the queue chain. For this purpose, the pointer
prev has been added to the sigpacket. This is to handle the following
case appropriately.
Consider the queued signal chain of:
A->B->C->D
without pointer prev. Assume that the pointer 'q' and 'qnext' point to
C, and process() is processing C. If B is cleared in process(), A->next
should be set to to C in sigpacket::clear().
Then, if process() for C succeeds, C should be removed from the queue,
so A->next should be set to D. However, we cannot do that because we do
not have the pointer to A in the while loop in wait_sig().
With the pointer prev, we can easily access A and C in sigpacket::clear()
as well as A and D in the while loop in wait_sig() using the pointer prev
and next without pursuing the chain.
Addresses: https://cygwin.com/pipermail/cygwin/2024-November/256744.html
Fixes: 9d2155089e87 ("(wait_sig): Define variable q to be the start of the signal queue. Just iterate through sigq queue, deleting processed or zeroed signals")
Reported-by: Christian Franke <Christian.Franke@t-online.de>
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Given that spinning is only checked once at the start of a
_cygtls::interrupt_now() which is called in a loop, it's probably
not necessary to mark _cygtls::spinning as volatile.
However, spinning is changed from assembler code and we don't
want the compiler to make funny assumptions, so, better safe than
sorry.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
If SCHED_OTHER is selected, set the Windows priority according to the
nice value. If SCHED_FIFO or SCHED_RR is selected, preserve the nice
value and set the Windows priority according to the sched_priority
parameter.
Signed-off-by: Christian Franke <christian.franke@t-online.de>
This patch calls Sleep(0) in the wait loop in lock() to increase the
chance of being unlocked in other threads. The lock(), unlock() and
locked() are moved from sigfe.s to cygtls.h so that allows inline
expansion.
Addresses: https://cygwin.com/pipermail/cygwin/2024-November/256744.html
Fixes: 61522196c715 ("* Merge in cygwin-64bit-branch.")
Reported-by: Christian Franke <Christian.Franke@t-online.de>
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
This method has been removed when we stopped supporting older Windows
versions and only one supported FAST_CWD structure remained. Belatedly
drop the declaration as well.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
This was already used in the FAST_CWD check, and could be used in a
couple other places.
I found the "emulated"/process value returned from the function largely
useless, so I did not cache it. It is useless because, as the docs say,
it is set to IMAGE_FILE_MACHINE_UNKNOWN (0) if the process is not
running under WOW64, but Microsoft also doesn't consider x64-on-ARM64 to
be WOW64, so it is set to 0 regardless if the process is ARM64 or x64.
You can tell the difference via
GetProcessInformation(ProcessMachineTypeInfo), but for the current
process even that's overkill: what we really want to know is the
IMAGE_FILE_MACHINE_* constant for the Cygwin dll itself, which is
conveniently located in memory already, so cache that in wincap also for
easy comparisons.
Signed-off-by: Jeremy Drake <cygwin@jdrake.com>
In _cygtls::handle_SIGCONT(), the sig thread waits for the main thread
to process the signal without unlocking the TLS area. This causes a
deadlock if the main thread tries to acquire a lock for the TLS area
in the meantime. With this patch, unlock the TLS before calling yield()
in handle_SIGCONT().
Addresses: https://cygwin.com/pipermail/cygwin/2024-November/256744.html
Fixes: 26158dc3e9c2("* exceptions.cc (sigpacket::process): Lock _cygtls area of thread before accessing it.")
Reported-by: Christian Franke <Christian.Franke@t-online.de>
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Windows silently sets a lower priority than requested if the new priority
requires administrator privileges. Revert to previous priority and fail
with EACCES or EPERM in this case.
Signed-off-by: Christian Franke <christian.franke@t-online.de>
The currently handled signal in a thread is called _cygtls::sig.
The variable name "sig" is used pretty often in the Cygwin source.
This makes it tricky to distinguish the currently handled signal
from any other usage of "sig".
Therefore, rename _cygtls::sig to _cygtls::current_sig
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Previous fix (commit df0953aa298c) fixes only a part of the problem.
Since exec() overrides the cygwin pid of the caller process, it makes
console owner handling complex. This patch makes console use Windows
pid as the owner pid (con.owner) instead of cygwin pid to make the
handling simpler.
Fixes: df0953aa298c ("Cygwin: console: Fix open() failure when the console owner calls exec().")
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Previously, cygwin read pipe used non-blocking mode although non-
cygwin app uses blocking-mode by default. Despite this requirement,
if a cygwin app is executed from a non-cygwin app and the cygwin
app exits, read pipe remains on non-blocking mode because of the
commit fc691d0246b9. Due to this behaviour, the non-cygwin app
cannot read the pipe correctly after that. Similarly, if a non-
cygwin app is executed from a cygwin app and the non-cygwin app
exits, the read pipe mode remains on blocking mode although cygwin
read pipe should be non-blocking mode.
These bugs were provoked by pipe mode toggling between cygwin and
non-cygwin apps. To make management of pipe mode simpler, this
patch has re-designed the pipe implementation. In this new
implementation, both read and write pipe basically use only blocking
mode and the behaviour corresponding to the pipe mode is simulated
in raw_read() and raw_write(). Only when NtQueryInformationFile
(FilePipeLocalInformation) fails for some reasons, the raw_read()/
raw_write() cannot simulate non-blocking access. Therefore, the pipe
mode is temporarily changed to non-blocking mode.
Moreover, because the fact that NtSetInformationFile() in
set_pipe_non_blocking(true) fails with STATUS_PIPE_BUSY if the pipe
is not empty has been found, query handle is not necessary anymore.
This allows the implementation much simpler than before.
Addresses: https://github.com/git-for-windows/git/issues/5115
Fixes: fc691d0246b9 ("Cygwin: pipe: Make sure to set read pipe non-blocking for cygwin apps.");
Reported-by: isaacag, Johannes Schindelin <Johannes.Schindelin@gmx.de>
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>, Ken Brown <kbrown@cornell.edu>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
There was a long-standing issue that pseudo console ownership could
not hand over from the process whose ctty is /dev/cons* rather than
/dev/pty*. This problem happens when a cygwin app starts non-cygwin
app in a pty, then the non-cygwin app starts multiple cygwin apps,
and the non-cygwin app ends before the second cygwin apps end.
In this case, the stub process of the non-cygwin app hands over the
ownership of pcon to one of the second cygwin apps, however, this
app does not hand over the ownership of pcon to another second
cygwin app. This is due to the fact that the hand-over feature is
implemented only in fhandler_pty_slave but not in fhandler_console.
With this patch, the second cygwin apps check if their console device
is inside a pseudo console, and if so, it tries to hand over the
ownership of the pseudo console to anther process that is attached
to the same pseudo console.
Addresses: https://cygwin.com/pipermail/cygwin/2024-February/255388.html
Fixes: 253352e796ff ("Cygwin: pty: Allow multiple apps to enable pseudo console simultaneously.")
Reported-by: lmari Lauhakangas <ilmari.lauhakangas@libreoffice.org>, Hossein Nourikhah <hossein@libreoffice.org>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
When win32-input-mode (which is supported by Windows Termainal) is
set by "\033[?9001h", cons_master_thread does not work properly and
consumes larger and larger memory space. This is because sending
event by WriteConsoleInput() is translated into the sequence that
is used by win32-input-mode. Due to this behaviour, write-back
of the INPUT_RECORDs does not work as expected. With this patch,
cons_master_thread is disabled on win32-input-mode where the signal
keys such as Ctrl-C, Ctrl-Z etc. never comes.
Addresses: https://cygwin.com/pipermail/cygwin/2024-August/256380.html
Fixes: ff4440fcf768 ("Cygwin: console: Introduce new thread which handles input signal.")
Reported-by: Adamyg Mob <adamyg.mob@gmail.com>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
So far fhandler_pipe_fifo::raw_write always returns -1/EINTR
if a signal arrived. Linux does not do that if there's still
space left in the pipe buffer.
The Linux buffer handling can't be emulated by Cygwin, but we
can do something similar which makes it much more likely to still
write successfully even if the buffer is almost full.
Utilize pipe_data_available to return valid pipe buffer usage
to raw_write, allowing a more sophisticated way to fill the
buffer while maintaining comaptibility with non-Cygwin pipes.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Fix a gcc 12 warning about a narrowing conversion in case labels for
tape options.
> In file included from /wip/cygwin/src/winsup/cygwin/include/sys/mtio.h:14,
> from ../../../../src/winsup/cygwin/fhandler/tape.cc:13:
> ../../../../src/winsup/cygwin/fhandler/tape.cc: In member function ‘int mtinfo_drive::set_options(HANDLE, int32_t)’:
> ../../../../src/winsup/cygwin/fhandler/tape.cc:965:12: error: narrowing conversion of ‘4026531840’ from ‘unsigned int’ to ‘int’ [-Wnarrowing]
Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Fix warnings with gcc 12 about narrowing conversions of NTSTATUS
constants when used as case labels, e.g:
> ../../../../src/winsup/cygwin/exceptions.cc: In static member function ‘static int exception::handle(EXCEPTION_RECORD*, void*, CONTEXT*, PDISPATCHER_CONTEXT)’:
> ../../../../src/winsup/cygwin/exceptions.cc:670:10: error: narrowing conversion of ‘-1073741682’ from ‘NTSTATUS’ {aka ‘int’} to ‘unsigned int’ [-Wnarrowing]
See also: c5bdf60ac46401a51a7e974333d9622966e22d67
Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
This disables a warning seen with gcc 12 caused by intrinsics used by
the inline implementation of NtCurrentTeb() inside w32api headers.
> In function ‘long long unsigned int __readgsqword(unsigned int)’,
> inlined from ‘_TEB* NtCurrentTeb()’ at /usr/include/w32api/winnt.h:10020:86,
> [...]
> /usr/include/w32api/psdk_inc/intrin-impl.h:838:1: error: array subscript 0 is outside array bounds of ‘long long unsigned int [0]’ [-Werror=array-bounds]
See also: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105523#c6
Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Previously, if two or more processes request to allocate a console at
the same time, the same device number could be accidentally allocated.
This patch fixes the issue by creating a named mutex (input_mutex) and
checking if GetLastError() is ERROR_ALREADY_EXIST or not to ensure
the unit number will be allocated in an atomic way. Thanks to this,
EnumWindow() is not necessary anymore to scan console units.
This also makes minor device numbers unique and console devices visible
across sessions. This disallows duplicated device number for different
sessions, so the MAX_CONS_DEV has been increased from 64 to 128.
Additionally, the console allocation and scanning will be faster as
a side effect of eliminating EnumWindows().
Fixes: 3721a756b0d8 ("Cygwin: console: Make the console accessible from other terminals.")
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Previously, the following steps failed with error:
1) Open /dev/dsp with O_RDONLY
2) Open /dev/dsp with O_WRONLY
3) Issue SNDCTL_DSP_GETOSPACE ioctl() for 2)
This is because IS_WRITE() returns false for 2) due to incorrect
openflags handling in archetype instance. This patch fixes the
issue by adding open_setup() to fhandler_dev_dsp to set openflags
correctly for each instance.
Fixes: 92ddb7429065 ("* fhandler_dsp.cc (fhandler_dev_dsp::open): Remove archetype handling.")
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
To avoid race issues, pthread::once() uses pthread_mutex. This caused
the handle leak which was fixed by the commit 2c5433e5da82. However,
this fix introduced another race issue, i.e., the mutex may be used
after it is destroyed. This patch fixes the issue. Special thanks to
Bruno Haible for discussing how to fix this.
Addresses: https://cygwin.com/pipermail/cygwin/2024-May/255987.html
Reported-by: Bruno Haible <bruno@clisp.org>
Fixes: 2c5433e5da82 ("Cygwin: pthread: Fix handle leak in pthread_once.")
Reviewed-by: Ken Brown <kbrown@cornell.edu>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Commit 80f722e97cf7 ("Cygwin: opendir(3): move ENOTDIR check into main
function") introduced a bug in fhandler_virtual handling. While the
assertion that path_conv::check() already calls exists() and sets
FILE_ATTRIBUTE_DIRECTORY accordingly, the exists() function is called
on a fhandler_virtual object created for just this code snippet. The
side effect of this is that the fileid member in the calling
fhandler_virtual object is not set after path_conv::check().
Move the fhandler_virtual::fileid member to path_conv::_virt_fileid
and create matching path_conv::virt_fileid() and fhandler_virtual::fileid()
methods.
Let path_conv::check() propagate the fileid set in the local
fhandler_virtual::exists() call to its own _virt_fileid.
Use new fhandler_virtual::fileid() method throughout.
Fixes: 80f722e97cf7 ("Cygwin: opendir(3): move ENOTDIR check into main function")
Reported-by: Bruce Jerrick <bmj001@gmail.com>
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
SMBv1 is ultimately deprecated since Novemer 2023. It's also
not installed by default on latest Windows versions.
Drop using the WNet (SMBv1) API in favor of using Network Discovery.
Given there's no documented Win32 API for that, the code now uses
the Shell API.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
If pipe reader is a non-cygwin app first, and cygwin process reads
the same pipe after that, the pipe has been set to bclocking mode
for the cygwin app. However, the commit 9e4d308cd592 assumes the
pipe for cygwin process always is non-blocking mode. With this patch,
the pipe mode is reset to non-blocking when cygwin app is started.
Addresses: https://cygwin.com/pipermail/cygwin/2024-March/255644.html
Fixes: 9e4d308cd592 ("Cygwin: pipe: Adopt FILE_SYNCHRONOUS_IO_NONALERT flag for read pipe.")
Reported-by: wh <wh9692@protonmail.com>
Reviewed-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Since commit 15e82eef3a40b ("Cygwin: getgrent: fix local SAM enumeration
on domain member machines") we skip enumerating local BUILTIN accounts
if we also enumerate AD. However, there are two local accounts which
are only available in local SAM, not in AD. Don't skip enumerating
those.
Fixes: 15e82eef3a40b ("Cygwin: getgrent: fix local SAM enumeration on domain member machines")
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Chances are high that Cygwin recalls offline files from remote
storage, even if the file is only accessed during stat(2) or
readdir(3).
To avoid this
- make sure Cygwin is placeholder-aware,
- open files in path_conv handling, as well as in stat(2)/readdir(3)
scenarios with FILE_OPEN_NO_RECALL, and
- during symlink checking or testing for executablility, don't even
try to open the file if one of the OFFLINE attributes is set.
Reported-by: Marcin Wisnicki <mwisnicki@gmail.com>
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
This patch eliminates verbose NtQueryObject() calls in the procedure
to get query_hdl by storing pipe name into fhandler_base::pc when
the pipe is created. fhandler_pipe::temporary_query_hdl() uses the
storedpipe name rather than the name retrieved by NtQueryObject().
Suggested-by: Corinna Vinschen <corinna@vinschen.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
Start at index 0 and use 32 bytes as minimal allocation size.
Reduce NBUCKETS, we don't have such big objects on the
cygheap anyway.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Use constexpr instead of const for errmap[] to allow static_assert
checks on its values.
Signed-off-by: Christian Franke <christian.franke@t-online.de>
If a removable (USB) device is disconnected after opening its raw
device, R/W attempts fail with ERROR_NO_SUCH_DEVICE(433). If the
raw device of a partition is used, ERROR_MEDIA_CHANGED(1110) is
returned instead. Both are mapped to ENODEV(19) because <errno.h>
does not offer a value which better matches ERROR_MEDIA_CHANGED.
Signed-off-by: Christian Franke <christian.franke@t-online.de>
Previously, three similar callback fuctions were used in console
code. This patch unifies these functions to ease maintenance cost.
Fixes: 8aad3a7edeb2 ("Cygwin: console: Fix a problem that minor ID is incorrect in ConEmu.")
Suggested-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Takashi Yano <takashi.yano@nifty.ne.jp>
- Basically maintain a hidden DIR* inside fhandlers.
- lseek has to be tweaked to allow basic seeking on the directory
descriptor.
- the current implementation does not keep the dir positions
between duplicated descriptor in sync. In fact, every descriptor
keeps its own copy of the DIR* and after dup/fork/exec, the
directory position is reset to 0, i. e., to the start of the
directory, as if rewinddir() has been called on the new descriptors.
While this behaviour isn't yet covered by the Issue 8 draft,
a bug report along these lines exists and will probably be
picked up for TC1.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
The content of DIR should never have been exposed into userspace.
Move struct __DIR to local dirent.h and only keep forward declaration
in exported dirent.h.
This allows to tweak the structure in future, for instance, to
add thread-safety.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
- drop appending .exe.lnk to files
- drop exe_suffixes, it's the same as stat_suffixes for a long time
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
Provide the same debugging opportunities for api_fatal() as we do for a
core-dumping signal:
1) Break into any attached debugger
2) Start JIT debugger (if configured) (keeping these under DEBUGGING doesn't seem helpful)
3) Write a coredump (if rlim_core > 1MB)
4) Write a stackdump (if that failed, or 0 < rlim_core <= 1MB)
Pre-format a command to be executed on a fatal error to run 'dumper'
(using an absolute path).
Factor out executing a pre-formatted command, so we can use that for
invoking the JIT debugger in try_to_debug() (if error_start is present
in the CYGWIN env var) and to invoke dumper when a fatal error occurs.
On a fatal error, if the core file size limit is greater than 1MB,
invoke dumper to write a core dump. Otherwise, if that limit is greater
than 0, write a .stackdump file, as previously.
Adjust and clarify the associated documentation.
Also: Fix so that the error_start JIT debugger is now invoked, even when
ulimit -c is zero.
Also: Fix uses of console_printf() inside exec_prepared_command(). It's
output is written via the Windows console device, so needs to use
Windows-style line endings.
Also: consistently return non-zero from try_to_debug() if we debugged.
Future work: Truncate or remove the file written, if it exceeds the
maximum size set by the ulimit.
Future work: Using the words "fatal error" could probably be improved
on. This means exiting on one of the "certain signals whose default
action is to cause the process to terminate and produce a core dump
file".
Given that SSDs don't have a seek penalty, we can enable
automatic sparsifying of files on SSDs, even if the "sparse"
mount option is not set.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
In preparation of using the FileFsSectorSizeInformation info class,
add a couple of missing definitions.
Signed-off-by: Corinna Vinschen <corinna@vinschen.de>
