Remove all the epoll file descriptors after a fork since they are
no longer valid. The uv__signal_global_once_init function needs to
be run after uv__platform_loop_init so that the epoll pthread_atfork
handlers get run first.
PR-URL: https://github.com/libuv/libuv/pull/1496
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
The kqueue documentation states that the file descriptor is not
inherited by the child process. Hence, do not close it in uv_loop_fork
to avoid tampering with a possibly valid file descriptor already opened
by the child process.
PR-URL: https://github.com/libuv/libuv/pull/1583
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Fixes uv_fs_stat and uv_fs_lstat returning EINVAL when invoked on a
non-symlink reparse point.
1. Only tries to read symlinks when invoked via lstat (do_lstat == 1).
Rationale is that only lstat can set S_IFLNK because when a file is
tested by stat, symlinks are resolved by the OS and the returned file
must be real. Note that broken symlinks fail at CreateFile.
FILE_ATTRIBUTE_REPARSE_POINT is used by filesystem drivers for purposes
besides symlinks, and uv_fs_stat fails when invoked on these files
because fs__readlink_handle returns ERROR_SYMLINK_NOT_SUPPORTED. By
ignoring the attribute in uv_fs_stat, these files are now handled
correctly.
2. Modifies the logic added to fs__stat_handle to fix#995 as follows:
A failed fs__readlink_handle on a file with a reparse point indicates
that the file is not a symlink. The fix for #995 added code to fall
through and behave as with a normal file in this case. However, this is
not correct because lstat had opened the file with
FILE_FLAG_OPEN_REPARSE_POINT, preventing the filesystem from acting
based on the reparse point contents.
The fix makes fs__stat_handle fail back to the higher level
fs__stat_impl, which sets do_lstat to 0 and re-opens the file without
FILE_FLAG_OPEN_REPARSE_POINT, allowing normal filesystem processing to
take place.
This is also a slightly cleaner solution as symlink fallback is only
handled in one place (fs__stat_impl) instead of two (fs__stat_impl and
fs__stat_handle).
Note that the error tested in the fix for #995,
ERROR_NOT_A_REPARSE_POINT, is not actually returned by Windows in the
case of a non-symlink reparse point. I attempted to reproduce the error
by repeating the test steps in the issue but failed. However, the the
fix logic is preserved out of caution.
3. Adds tests to fs-test.c for the above two changes.
Thorough testing requires some non-trivial setup - like an OSX computer
on the LAN or a custom filesystem driver - so these tests are left
commented out for manual invocation.
PR-URL: https://github.com/libuv/libuv/pull/1522
Reviewed-By: Bartosz Sosnowski <bartosz@janeasystems.com>
Define stable cross-platform file open constants so that users can
pass `UV_FS_O_RDWR` rather than `_O_RDWR` (win) or `O_RDWR` (unix).
Map `UV_FS_O_DIRECT`, `UV_FS_O_DSYNC` and `UV_FS_O_SYNC` to
`FILE_FLAG_NO_BUFFERING` and `FILE_FLAG_WRITE_THROUGH` (win).
Fixes: https://github.com/libuv/libuv/issues/1550
PR-URL: https://github.com/libuv/libuv/pull/1567
Reviewed-By: Bartosz Sosnowski <bartosz@janeasystems.com>
`fs__stat_handle()` used to set st_blksize to `2048` regardless of the
underlying physical sector size of the disk.
`4096` is a better constant to avoid read-modify-write behavior on
advanced format drives.
Fixes: https://github.com/libuv/libuv/issues/1563
PR-URL: https://github.com/libuv/libuv/pull/1566
Reviewed-By: Bartosz Sosnowski <bartosz@janeasystems.com>
libuv retries when sendmsg(2) fails with EAGAIN or
EWOULDBLOCK. This commit adds similar functionality for
ENOBUFS.
PR-URL: https://github.com/libuv/libuv/pull/1573
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
On NetBSD, do not exclude PF_INET6 in uv__ifaddr_exclude().
PR-URL: https://github.com/libuv/libuv/pull/1576
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
This commit causes NetBSD, like other BSDs, to reuse
uv__cloexec_ioctl and uv__nonblock_ioctl. This fixes
poll_nested_kqueue.
PR-URL: https://github.com/libuv/libuv/pull/1575
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
PR-URL: https://github.com/libuv/libuv/pull/1578
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
On z/OS, fs mounts can have variables starting with '$'. The problem is
that following symlinks that contain these variables can lead to a loop
because readlink($SYSVAR) will return $SYSVAR. To work around this, we
must resolve the $SYSVAR to an actual path using realpath.
More information about z/OS symlink resolution is available here:
https://www.ibm.com/support/knowledgecenter/SSLTBW_2.1.0/
com.ibm.zos.v2r1.bpxb100/sym.htm
PR-URL: https://github.com/libuv/libuv/pull/1472
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
NetBSD as an extension returns with ptsname(3) and ptsname_r(3) the
slave device name for both descriptors, the master one and slave one.
Workaround this problem and verify the device major and compare it with
the pts driver. Major numbers for the master and the slave TTY are
machine-dependent. On amd64 they are respectively 6 (ptc) and 5 (pts).
PR-URL: https://github.com/libuv/libuv/pull/1533
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
The most critical bug is using IFLA_RTA instead of IFA_RTA on address
message. This causes android to not return any ipv6 address.
PR-URL: https://github.com/libuv/libuv/pull/1561
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
This will eliminate a race condition that occurs between
epoll_close_fd, epoll_ctl and epoll_create in the
signal_multiple_loops test.
PR-URL: https://github.com/libuv/libuv/pull/1560
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Setup proper sysctl(2) arguments in order to retrieve the executable
name.
Use an intermediate buffer beacause if it is too short, the sysctl(2)
call will return error. With this intermediate buffer populate the outer
one.
Note issue with too long file names on NetBSD-8(beta), this is caused by
the current implementation of vnode to path translator in the kernel. It
does not cache longer file names than 31 character ones.
PR-URL: https://github.com/libuv/libuv/pull/1532
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
musl has tiny default thread stack sizes compared to glibc (80 kB vs.
2048 kB or more) so set an explicit stack size to harmonize between
different libcs.
Fixes: https://github.com/libuv/libuv/issues/1507
PR-URL: https://github.com/libuv/libuv/pull/1526
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
NetBSD supports:
- O_CLOEXEC
- accept4(2) like API with paccept(2)
NetBSD 8.0 will get compat accept4(2)
- SOCK_NONBLOCK
- SOCK_CLOEXEC
- dup3(2)
PR-URL: https://github.com/libuv/libuv/pull/1534
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
This commit adds the O_TRUNC flag to the destination file in
the sendfile() fallback. This allows smaller source files to
be properly copied into larger existing destination files.
Refs: https://github.com/libuv/libuv/pull/1551
PR-URL: https://github.com/libuv/libuv/pull/1552
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Initialize the error variable in uv_fs_copyfile() to zero. If
the source file being copied has no data in it, then it's
possible to make it to the exit without ever setting error.
In this situation, the function's behavior is unpredictable.
Refs: https://github.com/libuv/libuv/pull/1551
PR-URL: https://github.com/libuv/libuv/pull/1552
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
The uv__has_forked_with_cfrunloop local variable is unused on
BSDs and causes compilation warnings. Observed on NetBSD/amd64
8.99.2 with GCC 5.4.0.
PR-URL: https://github.com/libuv/libuv/pull/1529
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
uv_fs_sendfile() is used as the fallback for uv_fs_copyfile().
However, sendfile() does not necessarily send all of the bytes
in one call. This commit adds a loop to ensure all data is
sent.
Fixes: https://github.com/libuv/libuv/issues/1517
PR-URL: https://github.com/libuv/libuv/pull/1494
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
This commit adds a POST macro to the Windows fs functions,
similar to the one used on Unix platforms.
PR-URL: https://github.com/libuv/libuv/pull/1509
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
This commit introduces an INIT macro to file system functions on
Windows, similar to the one used on Unix platforms. The macro
checks for NULL requests, and returns UV_EINVAL in such
scenarios. This commit also adds support for passing NULL to
uv_fs_req_cleanup(). In this scenario, the function is a
no-op.
Fixes: https://github.com/libuv/libuv/issues/1508
PR-URL: https://github.com/libuv/libuv/pull/1509
Reviewed-By: Bartosz Sosnowski <bartosz@janeasystems.com>
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
Prior to this change, several of the fs functions checked for
invalid arguments before initializing the fs request. If a
consumer received a UV_EINVAL from one of these functions, and
then called uv_fs_req_cleanup(), the application would crash, as
the pointer being freed was not allocated. This commit makes
sure that all fs functions initialize the request before returning.
Fixes: https://github.com/libuv/libuv/issues/1508
PR-URL: https://github.com/libuv/libuv/pull/1509
Reviewed-By: Bartosz Sosnowski <bartosz@janeasystems.com>
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
Some functions, such as uv_fs_scandir() can be called with
a NULL loop argument. In this case, it's possible that the
Windows API hasn't been initialized, leading to a crash.
This commit adds additional uv__once_init() calls to mitigate
this situation.
Fixes: https://github.com/libuv/libuv/issues/1488
PR-URL: https://github.com/libuv/libuv/pull/1512
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
This commit partly reverses libuv#1441.
It is true that pthread_cond_timedwait is available on older API levels,
but if we do not call pthread_condattr_setclock then we cannot use
timestamps from CLOCK_MONOTONIC with pthread_cond_timedwait.
PR-URL: https://github.com/libuv/libuv/pull/1511
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
After the poll call, the code updates all the event objects
in the array. Instead, use the integer value returned by poll
to limit the number of epoll_event objects that are updated.
e.g. If poll returns 5, we know that we only need to update
5 event objects.
PR-URL: https://github.com/libuv/libuv/pull/1391
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
To set an initial value to a semaphore, the semop call won't do it.
We need to use the semctl call with the SETVAL flag
PR-URL: https://github.com/libuv/libuv/pull/1473
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Add SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE to CreateSymbolicLink
which allows unelevated users to create symbolic links on supported
platforms.
Fixes: https://github.com/libuv/libuv/issues/1157
PR-URL: https://github.com/libuv/libuv/pull/1466
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Avoid repeated wake-ups in `uv__io_poll()` when the previously watched
file descriptor has been duplicated (with `dup(2)`) and then closed.
Because epoll reports events for _file descriptions_ rather than _file
descriptors_, events on the duplicated file descriptor cause the event
loop to wake up.
Libuv then tries to unregister the event listener for the original file
descriptor but that fails with EBADF because the file descriptor is
closed.
Since libuv uses epoll in level-triggered mode, it effectively results
in a busy loop because the event is re-reported as soon as libuv calls
epoll_wait() again.
I tried hard to write a test case for this but there is no directly
observable behavior, only increased CPU usuage.
Fixes: https://github.com/libuv/libuv/issues/1148
PR-URL: https://github.com/libuv/libuv/pull/1468
Reviewed-By: Colin Ihrig <cjihrig@gmail.com
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
On Unix you can pass bound and even connected socket to uv_tcp_open
and it will work as expected: you can run uv_listen or other io
functions (uv_read, uv_write) on it without problems.
On windows on the other hand the function always assumes to have clean
socket and for example uv_listen will try to bind it again, and
uv_read/write will return with errors about unreadable streams
(basically invalid internal flags).
To check if socket is already bound uv_tcp_getsockname is called which
on windows returns error when socket is unbound. To further
differentiate connected one from just bound, uv_tcp_getpeername also
returns error but when target socket is not connected.
PR-URL: https://github.com/libuv/libuv/pull/1447
Reviewed-By: Bartosz Sosnowski <bartosz@janeasystems.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
More accurate Resident Set Size (Central Storage size on Z)
is stored in the MVS Data Areas managed by the RSM (Real
Storage Manager).
PR-URL: https://github.com/libuv/libuv/pull/1244
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Code was assuming that the nl_pid was always equal to the process id.
As the netlink documentation says, "nl_pid is usually the PID of the
process owning the destination socket. However, nl_pid identifies a
netlink socket, not a process".
There are two cases in which this assumption breaks down. The first is
that some other part of the application has already bound a netlink
socket with a nl_pid equal to the PID of the process. The second is
Android version 5.0 where it appears that nl_pid is not always defaulted
to the PID of the process.
The result of this bad assumption is that netlink message responses to
our request are ignored due to having a bad nl_pid. We eventually run
out of messages on the netlink socket and block forever awaiting our
response.
PR-URL: https://github.com/libuv/libuv/pull/1451
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
One tick on the high resolution clock (TOD clock) on z/OS
is equivalent to (1/4.096) nanoseconds.
PR-URL: https://github.com/libuv/libuv/pull/1453
Reviewed-By: Colin Ihrig <cjihrig@gmail.com
Reviewed-By: Santiago Gimeno <santiago.gimeno@gmail.com>
Reviewed-By: Saúl Ibarra Corretgé <saghul@gmail.com>
The latest NDK (r15 at the time of this writing) no longer needs this.
`pthread_cond_timedwait` is available at any (reasonable) API level.
PR-URL: https://github.com/libuv/libuv/pull/1441
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
It's only available on API level >= 21.
PR-URL: https://github.com/libuv/libuv/pull/1441
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
