| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
This replaces the use of uint8_t to represent booleans in the
configuration type system with the nxt_bool_t type (which happens to be
a uint8_t).
This is part of the work to use nxt_bool_t in structures to represent
booleans rather than uint8_t's which is more intuitive.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This has been replaced by NXT_CONF_MAP_BOOL.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This replaces the use of NXT_CONF_MAP_INT8 for denoting boolean values.
It also makes NXT_CONF_MAP_BOOL an alias for NXT_CONF_MAP_INT8 in
nxt_conf_map_object(). A future commit will remove NXT_CONF_MAP_INT8.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
When adding a boolean config option, this is of type
NXT_CONF_VALUE_BOOLEAN, however when mapping such an entry to a
structure member via nxt_conf_map_object() this is done as a
NXT_CONF_MAP_INT8. It took a bit of head scratching to find out that it
should have been a NXT_CONF_MAP_INT8 and not a NXT_CONF_MAP_INT.
Introduce a new map type of NXT_CONF_MAP_BOOL that will eventually
replace NXT_CONF_MAP_INT8 for representing booleans...
We use the short form _BOOL as that matches the naming convention of the
other types in nxt_conf_map_type_t.
This is part of a set of patches and eventually NXT_CONF_MAP_BOOL will
simply replace NXT_CONF_MAP_INT8.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This commit complements the earlier one that changed a bunch of uint8_t
struct members to nxt_bool_t (aka uint8_t).
It's not entirely clear why these were specifically done as uint32_t's.
I've checked these structure layouts with pahole(1) after this change on
both 32 and 64bit and the layouts haven't changed.
We may get a 3 or 7 byte hole after the nxt_bool_t or 3 bytes of padding
at the end of the structure, but the structure members are still at the
same offsets and the structure sizes have remained the same.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Replace the usage of uint8_t in structures to represent boolean values
with our nxt_bool_t type.
This will result in no change in structure layout as the nxt_bool_t is
now a uint8_t, same as what it's replacing.
Even though it's essentially the same type, it makes it much clearer as
to what its purpose is.
This was largely done with the following script from Alex, with some
manual conversions
$ grep -rl 'uint8_t.*1 bit' src/ \
| xargs sed -i '/uint8_t.*1 bit/{s/uint8_t /nxt_bool_t /;s/; *\/\*.*/;/}'
This doesn't convert the non-uint8_t booleans, they will be handled
separately.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
We have a 'bool' type, nxt_bool_t which is defined as a nxt_uint_t, that
is used as local variables, and function return types and arguments.
However in structs we use a uint8_t, due to not wanting to waste memory
in structures, but the use of uint8_t in structs is confusing and
unintuitive.
We could just switch to using the C99 _Bool type, however there are
concerns about this possibly not always being 1-byte on platforms on
which Unit runs and might mess with structure layouts.
(Switching to _Bool as the underlying type is still an eventual goal)
The alternative is to just make our nxt_bool_t be a uint8_t.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
In nxt_kqueue_poll() error is declared as a nxt_bool_t aka unsigned int
(on x86-64 anyway).
It is used both as a boolean and as the return storage for a bitwise AND
operation.
This causes the following issue after we have changed nxt_bool_t to be a
uint8_t (which will happen in a subsequent commit)
gcc12 -c -pipe -fPIC -fvisibility=hidden -O -W -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wmissing-prototypes -Werror -g -O2 -I src -I build -I/usr/local/include -o build/src/nxt_kqueue_engine.o -MMD -MF build/src/nxt_kqueue_engine.dep -MT build/src/nxt_kqueue_engine.o src/nxt_kqueue_engine.c
src/nxt_kqueue_engine.c: In function 'nxt_kqueue_poll':
src/nxt_kqueue_engine.c:728:17: error: overflow in conversion from 'int' to 'nxt_bool_t' {aka 'unsigned char'} changes value from '(int)kev->flags & 16384' to '0' [-Werror=overflow]
728 | error = (kev->flags & EV_ERROR);
| ^
cc1: all warnings being treated as errors
EV_ERROR has the value 16384, after the AND operation error holds 16384,
however this overflows and wraps around (64 times) exactly to 0.
Rather than conflating the use of error, keep error as a boolean (it is
used further down the function) but do the AND operation inside the
if ().
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This is required for the next patch which will change nxt_bool_t from a
nxt_uint_t to a uint8_t, which produces the following compiler warning
cc -c -pipe -fPIC -fvisibility=hidden -O -W -Wall -Wextra -Wno-unused-parameter -Wwrite-strings -Wmissing-prototypes -Werror -g -I src -I build \
\
\
-o build/src/nxt_sprintf.o \
-MMD -MF build/src/nxt_sprintf.dep -MT build/src/nxt_sprintf.o \
src/nxt_sprintf.c
In file included from src/nxt_unix.h:160,
from src/nxt_main.h:31,
from src/nxt_sprintf.c:7:
src/nxt_sprintf.c: In function ‘nxt_vsprintf’:
src/nxt_sprintf.c:368:44: error: ‘nxt_bool_t’ {aka ‘unsigned char’} is promoted to ‘int’ when passed through ‘...’ [-Werror]
368 | ui64 = (uint64_t) va_arg(args, nxt_bool_t);
| ^
src/nxt_sprintf.c:368:44: note: (so you should pass ‘int’ not ‘nxt_bool_t’ {aka ‘unsigned char’} to ‘va_arg’)
src/nxt_sprintf.c:368:44: note: if this code is reached, the program will abort
cc1: all warnings being treated as errors
Lets use unsigned int as it matches the signedness of nxt_bool_t.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This removes an unused structure from src/nxt_var.c that would have been
changed to switch to using nxt_bool_t for boolean values in a subsequent
commit.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This removes a bunch of unused files that would have been touched by
subsequent commits that switch to using nxt_bool_t (AKA unit6_t) in
structures.
In auto/sources we have
NXT_LIB_SRC0=" \
src/nxt_buf_filter.c \
src/nxt_job_file.c \
src/nxt_stream_module.c \
src/nxt_stream_source.c \
src/nxt_upstream_source.c \
src/nxt_http_source.c \
src/nxt_fastcgi_source.c \
src/nxt_fastcgi_record_parse.c \
\
src/nxt_mem_pool_cleanup.h \
src/nxt_mem_pool_cleanup.c \
"
None of these seem to actually be used anywhere (other than within
themselves). That variable is _not_ referenced anywhere else.
Also remove the unused related header files: src/nxt_buf_filter.h,
src/nxt_fastcgi_source.h, src/nxt_http_source.h, src/nxt_job_file.h,
src/nxt_stream_source.h and src/nxt_upstream_source.h
Also, these files do not seem to be used, no mention under auto/ or build/
src/nxt_file_cache.c
src/nxt_cache.c
src/nxt_job_file_cache.c
src/nxt_cache.h is #included in src/nxt_main.h, but AFAICT is not
actually used.
With all the above removed
$ ./congfigure --openssl --debug --tests && make -j && make -j tests &&
make libnxt
all builds.
Buildbot passes.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
- Configuration: added "/config/settings/http/log_route".
Type: bool
Default: false
This adds configurability to the error log. It allows enabling and
disabling logs related to how the router performs selection of the
routes.
- HTTP: logging request line.
Log level: [notice]
The request line is essential to understand which logs correspond to
which request when reading the logs.
- HTTP: logging route that's been discarded.
Log level: [info]
- HTTP: logging route whose action is selected.
Log level: [notice]
- HTTP: logging when "fallback" action is taken.
Log level: [notice]
Closes: <https://github.com/nginx/unit/issues/758>
Link: <https://github.com/nginx/unit/pull/824>
Link: <https://github.com/nginx/unit/pull/839>
Suggested-by: Timo Stark <t.stark@nginx.com>
Suggested-by: Mark L Wood-Patrick <mwoodpatrick@gmail.com>
Suggested-by: Liam Crilly <liam@nginx.com>
Tested-by: Liam Crilly <liam@nginx.com>
Acked-by: Artem Konev <a.konev@f5.com>
Cc: Andrew Clayton <a.clayton@nginx.com>
Cc: Andrei Zeliankou <zelenkov@nginx.com>
Reviewed-by: Zhidao Hong <z.hong@f5.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
|
|
This considerably simplifies the function, and will also help log the
iteration in which we are, which corresponds to the route array element.
Link: <https://github.com/nginx/unit/pull/824>
Link: <https://github.com/nginx/unit/pull/839>
Reviewed-by: Andrew Clayton <a.clayton@nginx.com>
Tested-by: Liam Crilly <liam@nginx.com>
Reviewed-by: Zhidao Hong <z.hong@f5.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
|
|
This prctl(2) option was enabled in commit 0277d8f1 ("Isolation: Fix the
enablement of PR_SET_NO_NEW_PRIVS.") and this was being set by default.
This prctl(2) when enabled renders (amongst other things) the set-UID
and set-GID bits on executables ineffective after an execve(2).
This causes an issue for applications that want to execute the
sendmail(8) binary, this includes the PHP mail() function, which is
usually set-GID.
After some internal discussion it was decided to disable this option by
default.
Closes: <https://github.com/nginx/unit/issues/852>
Fixes: 0277d8f1 ("Isolation: Fix the enablement of PR_SET_NO_NEW_PRIVS.")
Fixes: e2b53e16 ("Added "rootfs" feature.")
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
When starting unit, if its Unix domain control socket was already active
you would get an error message like
2023/03/15 18:07:55 [alert] 53875#8669650 connect(5, unix:/tmp/control.sock) succeed, address already in use
which is confusing in a couple of regards, firstly we have the classic
success/failure message and secondly 'address already in use' is an
actual errno value, EADDRINUSE and we didn't get an error from this
connect(2).
Re-word this error message for greater clarity.
Reported-by: Liam Crilly <liam.crilly@nginx.com>
Cc: Liam Crilly <liam.crilly@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Currently when using Unix domain sockets for requests, if unit is
reconfigured then it will fail if it tries to bind(2) again to a Unix
domain socket with something like
2023/02/25 19:15:50 [alert] 35274#35274 bind(\"unix:/tmp/unit.sock\") failed (98: Address already in use)
When closing such a socket we really need to unlink(2) it. However that
presents a problem in that when running as root, while the main process
runs as root and creates the socket, it's the router process, that runs
as an unprivileged user, e.g nobody, that closes the socket and would
thus remove it, but couldn't due to not having permission, even if the
socket is mode 0666, you need write permissions on the containing
directory to remove a file.
There are several options to solve this, all with varying degrees of
complexity and utility.
1) Give the user who the router process runs as write permission to
the directory containing the listen sockets. These can then be
unlink(2)'d from the router process.
Simple and would work, but perhaps not the most elegant.
2) Using capabilities(7). The router process could temporarily attain
the CAP_DAC_OVERRIDE capability, unlink(7) the socket, then
relinquish the capability until required again.
These are Linux specific (other systems may have similar mechanisms
which would be extra work to support). There is also a, albeit
small, window where the router process is running with elevated
privileges.
3) Have the main process do the unlink(2), it is after all the process
that created the socket.
This is what this commit implements.
We create a new port IPC message type of NXT_PORT_MSG_SOCKET_UNLINK,
that is used by the router process to notify the main process about a
Unix domain socket to unlink(2).
Upon doing a reconfigure the router process will call
nxt_router_listen_socket_release() which will close the socket, we
extend this function in the case of non-abstract Unix domain sockets, so
that it will send a message to the main process containing a copy of the
nxt_sockaddr_t structure that will contain the filename of the socket.
In the main process the handler that we have defined,
nxt_main_port_socket_unlink_handler(), for this message type will run
and allow us to look for the socket in question in the listen_sockets
array and remove it and unlink(2) the socket.
This then allows the reconfigure to work if it tries to bind(2) again to
a socket that previously existed.
Link: <https://github.com/nginx/unit/issues/669>
Link: <https://github.com/nginx/unit/pull/735>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
In nxt_process_quit() there is a loop that iterates over the
task->thread->runtime->listen_sockets array and closes the connections.
This code has been there from the beginning
$ git log --pretty=oneline -S'if (rt->listen_sockets != NULL)'
e9e5ddd5a5d9ce99768833137eac2551a710becf Refactor of process management.
6f2c9acd1841ca20a1388b34aef64e9f00459090 Processes refactoring. The cycle has been renamed to the runtime.
$ git log --pretty=oneline -S'if (cycle->listen_sockets != NULL) {'
6f2c9acd1841ca20a1388b34aef64e9f00459090 Processes refactoring. The cycle has been renamed to the runtime.
16cbf3c076a0aca6d47adaf3f719493674cf2363 Initial version.
but never seems to have been used (AFAICT and certainly not recently,
confirmed by code inspection and running pytests with a bunch of
language modules enabled and the code in question was never executed) as
the listen_sockets array has never been populated... until now.
The previous commit now adds Unix domain sockets to this array so that
they can be unlink(2)'d upon exit and reconfiguration.
This has now caused this dormant code to become active as it now tries
to close these sockets (from at least the prototype processes), this
array is inherited via fork by other processes.
The file descriptor for these sockets is set to -1 when they are put
into this array. This then results in close(-1) calls which caused
multiple failures in the pytests such as
> assert not alerts, 'alert(s)'
E AssertionError: alert(s)
E assert not ['2023/03/09 23:26:14 [alert] 137673#137673 socket close(-1) failed (9: Bad file descriptor)']
I think the simplest thing is to just remove this code.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
If we don't remove the Unix domain listen socket file then when Unit
restarts it get an error like
2023/02/25 23:10:11 [alert] 36388#36388 bind(\"unix:/tmp/unit.sock\") failed (98: Address already in use)
This patch makes use of the listen_sockets array, that is already
allocated in the main process but never populated, to place the Unix
domain listen sockets into.
At shutdown we can then loop through this array and unlink(2) any Unix
domain sockets found therein.
Closes: <https://github.com/nginx/unit/issues/792>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
In nxt_router_prepare_msg() we create a buffer (nxt_unit_request_t *req)
that gets sent to an application process that contains details about a
client request.
This buffer was always a little larger than needed due to allocating space
for the remote address _and_ port and the local address _and_ port. We
also allocate space for the local port separately.
->{local,remote}->length includes the port number and ':' and also the
'[]' for IPv6. E.g [2001:db8::1]:8080
->{local,remote}->address_length represents the length of the unadorned
IP address. E.g 2001:db8::1
Update the buffer size so that we only allocate what is actually needed.
Suggested-by: Zhidao HONG <z.hong@f5.com>
Cc: Zhidao HONG <z.hong@f5.com>
Reviewed-by: Zhidao HONG <z.hong@f5.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
User @bes-internal reported a Perl module crasher on GitHub.
This was due to a Perl application sending back two responses, for each
response we would call down into XS_NGINX__Unit__Sandbox_cb(), the first
time pctx->req would point to a valid nxt_unit_request_info_t, the
second time pctx->req would be NULL.
Add an invalid responses check which covers this case.
Closes: <https://github.com/nginx/unit/issues/841>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This fixes an issue reported by @Peter2121 on GitHub.
In nxt_router_prepare_msg() we create a buffer (nxt_unit_request_t *req)
that gets sent to an application process that contains details about a
client request.
The req structure comprises various members with the final member being
an array (specified as a flexible array member, with its actual length
denoted by the req->fields_count member) of nxt_unit_field_t's. These
structures specify the length and offset for the various request headers
name/value pairs which are stored after some request metadata that is
stored immediately after this array of structs as individual nul
terminated strings.
After this we have the body content data (if any). So it looks a little
like
(gdb) x /64bs 0x7f38c976e060
0x7f38c976e060: "\353\346\244\t\006" <-- First nxt_unit_field_t
0x7f38c976e066: ""
0x7f38c976e067: ""
0x7f38c976e068: "T\001"
0x7f38c976e06b: ""
0x7f38c976e06c: "Z\001"
0x7f38c976e06f: ""
...
0x7f38c976e170: "\362#\244\v$" <-- Last nxt_unit_field_t
0x7f38c976e176: ""
0x7f38c976e177: ""
0x7f38c976e178: "\342\002"
0x7f38c976e17b: ""
0x7f38c976e17c: "\352\002"
0x7f38c976e17f: ""
0x7f38c976e180: "POST" <-- Start of request metadata
0x7f38c976e185: "HTTP/1.1"
0x7f38c976e18e: "unix:"
0x7f38c976e194: "unix:/dev/shm/842.sock"
0x7f38c976e1ab: ""
0x7f38c976e1ac: "fedora"
0x7f38c976e1b3: "/842.php"
0x7f38c976e1bc: "HTTP_HOST" <-- Start of header fields
0x7f38c976e1c6: "fedora"
0x7f38c976e1cd: "HTTP_X_FORWARDED_PROTO"
0x7f38c976e1e4: "https"
...
0x7f38c976e45a: "HTTP_COOKIE"
0x7f38c976e466: "PHPSESSID=8apkg25r9s9vju3pi085i21eh4"
0x7f38c976e48b: "public_form=sended" <-- Body content
Well that's how things are supposed to look! When using Unix domain
sockets what we actually got looked like
...
0x7f6141f3445a: "HTTP_COOKIE"
0x7f6141f34466: "PHPSESSID=uo5b2nu9buijkc89jotbgmd60vpublic_form=sended"
Here, the body content (from a POST for example) has been appended
straight onto the end of the last header field value. In this case
corrupting the PHP session cookie. The body content would still be
found by the application as its offset into this buffer is correct.
This problem was actually caused by a0327445 ("PHP: allowed to specify
URLs without a trailing '/'.") which added an extra item into this
request buffer specifying the port number that unit is listening on that
handled this request.
Unfortunately when I wrote that patch I didn't increase the size of this
request buffer to accommodate it.
When using normal TCP sockets we actually end up allocating more space
than required for this buffer, we track the end of this buffer up to
where the body content would go and so we have a few spare bytes between
the nul byte of the last field header value and the start of the body
content.
When using Unix domain sockets, they have no associated port number and
thus the port number has a length of 0 bytes, but we still write a '\0'
in there using up a byte that we didn't account for, this causes us to
loose the nul byte of the last header fields value causing the body data
to be appended to the last header field value.
The fix is simple, account for the local port length, we also add 1 to
it, this covers the nul byte, even if there is no port as with Unix
domain sockets.
Closes: <https://github.com/nginx/unit/issues/842>
Fixes: a0327445 ("PHP: allowed to specify URLs without a trailing '/'.")
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
When running as a daemon. unit currently sets umask(0), i.e no umask.
This is resulting in various directories being created with a mode of
0777, e.g
rwxrwxrwx
this is currently affecting cgroup and rootfs directories, which are
being created with a mode of 0777, and when running as a daemon as there
is no umask to restrict the permissions.
This also affects the language modules (the umask is inherited over
fork(2)) whereby unless something explicitly sets a umask, files and
directories will be created with full permissions, 0666 (rw-rw-rw-)/
0777 (rwxrwxrwx) respectively.
This could be an unwitting security issue.
My original idea was to just remove the umask(0) call and thus inherit
the umask from the executing shell/program.
However there was some concern about just inheriting whatever umask was
in effect.
Alex suggested that rather than simply removing the umask(0) call we
change it to a value of 022 (which is a common default), which will
result in directories and files with permissions at most of 0755
(rwxr-xr-x) & 0644 (rw-r--r--).
If applications need some other umask set, they can (as they always have
been able to) set their own umask(2).
Suggested-by: Alejandro Colomar <alx.manpages@gmail.com>
Reviewed-by: Liam Crilly <liam@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
When using the 'rootfs' isolation option, by default a tmpfs filesystem
is mounted on tmp/. Currently this is mounted with a mode of 0777, i.e
drwxrwxrwx. 3 root root 60 Feb 22 11:56 tmp
however this should really have the sticky bit[0] set (as is per-normal for
such directories) to prevent users from having free reign on the files
contained within.
What we really want is it mounted with a mode of 01777, i.e
drwxrwxrwt. 3 root root 60 Feb 22 11:57 tmp
[0]: To quote inode(7)
"The sticky bit (S_ISVTX) on a directory means that a file in that
directory can be renamed or deleted only by the owner of the file, by
the owner of the directory, and by a privileged process."
Reviewed-by: Liam Crilly <liam@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Since the previous commit, nxt_getpid() is only ever aliased to
getpid(2).
nxt_getpid() was only used once in the code, while there are multiple
direct uses of getpid(2)
$ grep -r "getpid()" src/
src/nxt_unit.c: nxt_unit_pid = getpid();
src/nxt_process.c: nxt_pid = nxt_getpid();
src/nxt_process.c: nxt_pid = getpid();
src/nxt_lib.c: nxt_pid = getpid();
src/nxt_process.h:#define nxt_getpid() \
src/nxt_process.h:#define nxt_getpid() \
src/nxt_process.h: getpid()
Just remove it and convert the _single_ instance of nxt_getpid() to
getpid(2).
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
When using SYS_clone we used the getpid kernel system call directly via
syscall(SYS_getpid) to avoid issues with cached pids.
However since we are now only using fork(2) (+ unshare(2) for
namespaces) we no longer need to call the kernel getpid directly as the
fork(2) will ensure the cached pid is invalidated.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Since the previous commit, this is no longer used.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
On GitHub, @razvanphp & @hbernaciak both reported issues running the
APCu PHP module under Unit.
When using this module they were seeing errors like
'apcu_fetch(): Failed to acquire read lock'
However when running APCu under php-fpm, everything was fine.
The issue turned out to be due to our use of SYS_clone breaking the
pthreads(7) API used by APCu. Even if we had been using glibc's
clone(2) wrapper we would still have run into problems due to a known
issue there.
Essentially the problem is when using clone, glibc doesn't update the
TID cache, so the child ends up having the same TID as the parent and
that is used in various parts of pthreads(7) such as in the various
locking primitives, so when APCu was grabbing a lock it ended up using
the TID of the main unit process (rather than that of the php
application processes that was grabbing the lock).
So due to the above what was happening was when one of the application
processes went to grab either a read or write lock, the lock was
actually being attributed to the main unit process. If a process had
acquired the write lock, then if a process tried to acquire a read or
write lock then glibc would return EDEADLK due to detecting a deadlock
situation due to thinking the process already held the write lock when
in fact it didn't.
It seems the right way to do this is via fork(2) and unshare(2). We
already use fork(2) on other platforms.
This requires a few tricks to keep the essence of the processes the same
as before when using clone
1) We use the prctl(2) PR_SET_CHILD_SUBREAPER option (if its
available, since Linux 3.4) to make the main unit process inherit
prototype processes after a double fork(2), rather than them being
reparented to 'init'.
This avoids needing to ^C twice to fully exit unit when running in
the foreground. It's probably also better if they maintain their
parent child relationship where possible.
2) We use a double fork(2) technique on the prototype processes to
ensure they themselves end up in a new PID namespace as PID 1 (when
CLONE_NEWPID is being used).
When using unshare(CLONE_NEWPID), the calling process is _not_
placed in the namespace (as discussed in pid_namespaces(7)). It
only sets things up so that subsequent children are placed in a PID
namespace.
Having the prototype processes as PID 1 in the new PID namespace is
probably a good thing and matches the behaviour of clone(2). Also,
some isolation tests break if the prototype process is not PID 1.
3) Due to the above double fork(2) the main unit process looses track
of the prototype process ID, which it needs to know.
To solve this, we employ a simple pipe(2) between the main unit and
prototype processes and pass the prototype grandchild PID from the
parent of the second fork(2) before exiting. This needs to be done
from the parent and not the grandchild, as the grandchild will see
itself having a PID of 1 while the main process needs its
externally visible PID.
Link: <https://www.php.net/manual/en/book.apcu.php>
Link: <https://sourceware.org/bugzilla/show_bug.cgi?id=21793>
Closes: <https://github.com/nginx/unit/issues/694>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Due to the need to replace our use of clone/__NR_clone on Linux with
fork(2)/unshare(2) for enabling Linux namespaces(7) to keep the
pthreads(7) API working. Let's rename NXT_HAVE_CLONE to
NXT_HAVE_LINUX_NS, i.e name it after the feature, not how it's
implemented, then in future if we change how we do namespaces again we
don't have to rename this.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This commit fixed the njs memory leak happened in the config validation, updating and http requests.
|
|
This adds a check to nxt_python_asgi_get_event_loop() on the
event_loop_func name in the case that running that function fails, and
if it's get_running_loop() that failed we skip printing an error message
as this is an often expected behaviour since the previous commit and we
don't want users reporting erroneous bugs.
This check will always happen regardless of Python version while it
really only applies to Python >= 3.7, there didn't seem much point
adding complexity to the code for this case and in what will be an ever
diminishing case of people running older Pythons.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Several users on GitHub reported issues with running Python ASGI apps on
Unit with Python 3.11.1 (this would also effect Python 3.10.9) with the
following error from Unit
2023/01/15 22:43:22 [alert] 0#77128 [unit] Python failed to call 'asyncio.get_event_loop'
TL;DR
asyncio.get_event_loop() is currently broken due to the process of
deprecating part or all of it.
First some history.
In Unit we had this commit
commit 8dcb0b9987033d0349a6ecf528014a9daa574787
Author: Max Romanov <max.romanov@nginx.com>
Date: Thu Nov 5 00:04:59 2020 +0300
Python: request processing in multiple threads.
One of things this did was to create a new asyncio event loop in each
thread using asyncio.new_event_loop().
It's perhaps worth noting that all these asyncio.* functions are Python
functions that we call from the C code in Unit.
Then we had this commit
commit f27fbd9b4d2bdaddf1e7001d0d0bc5586ba04cd4
Author: Max Romanov <max.romanov@nginx.com>
Date: Tue Jul 20 10:37:54 2021 +0300
Python: using default event_loop for main thread for ASGI.
This changed things so that Unit calls asyncio.get_event_loop() in the
_main_ thread (but still calls asyncio.new_event_loop() in the other
threads).
asyncio.get_event_loop() up until recently would either return an
already running event loop or return a newly created one.
This was done for $reasons that the commit message and GitHub issue #560
hint at. But the intimation is that there can already be an event loop
running from the application (I assume it's referring to the users
application) at this point and if there is we should use it.
Now for the Python side of things.
On the main branch we had
commit 172c0f2752d8708b6dda7b42e6c5a3519420a4e8
Author: Serhiy Storchaka <storchaka@gmail.com>
Date: Sun Apr 25 13:40:44 2021 +0300
bpo-39529: Deprecate creating new event loop in asyncio.get_event_loop() (GH-23554)
This commit began the deprecating of asyncio.get_event_loop().
commit fd38a2f0ec03b4eec5e3cfd41241d198b1ee555a
Author: Serhiy Storchaka <storchaka@gmail.com>
Date: Tue Dec 6 19:42:12 2022 +0200
gh-93453: No longer create an event loop in get_event_loop() (#98440)
This turned asyncio.get_event_loop() into a RuntimeError _if_ there
isn't a current event loop.
commit e5bd5ad70d9e549eeb80aadb4f3ccb0f2f23266d
Author: Serhiy Storchaka <storchaka@gmail.com>
Date: Fri Jan 13 14:40:29 2023 +0200
gh-100160: Restore and deprecate implicit creation of an event loop (GH-100410)
This re-creates the event loop if there wasn't one and emits a
deprecation warning.
After at least the last two commits Unit no longer works with the Python
_main_ branch.
Meanwhile on the 3.11 branch we had
commit 3fae04b10e2655a20a3aadb5e0d63e87206d0c67
Author: Serhiy Storchaka <storchaka@gmail.com>
Date: Tue Dec 6 17:15:44 2022 +0200
[3.11] gh-93453: Only emit deprecation warning in asyncio.get_event_loop when a new event loop is created (#99949)
which is what caused our breakage, though perhaps unintentionally as we
get the following traceback
Traceback (most recent call last):
File "/usr/lib64/python3.11/asyncio/events.py", line 676, in get_event_loop
f = sys._getframe(1)
^^^^^^^^^^^^^^^^
ValueError: call stack is not deep enough
2023/01/18 02:46:10 [alert] 0#180279 [unit] Python failed to call 'asyncio.get_event_loop'
However, regardless, it is clear we need to stop using
asyncio.get_event_loop().
One option is to switch to the higher level asyncio.run() API, however
that is a rather large change.
This commit takes the simpler approach of using
asyncio.get_running_loop() (which it seems get_event_loop() will
eventually be an alias of) in the _main_ thread to return the currently
running event loop, or if there is no current event loop, it will call
asyncio.new_event_loop() to return a newly created event loop.
I believe this mimics the current behaviour. In my testing
get_event_loop() seemed to always return a newly created loop, as when
just calling get_running_loop() it would return NULL and we would fail
out.
When running two processes each with 2 threads we would get the
following loops with Python 3.11.0 and unpatched Unit
<_UnixSelectorEventLoop running=False closed=False debug=False>
<_UnixSelectorEventLoop running=False closed=False debug=False>
<_UnixSelectorEventLoop running=False closed=False debug=False>
<_UnixSelectorEventLoop running=False closed=False debug=False>
and with Python 3.11.1 and a patched Unit we would get
<_UnixSelectorEventLoop running=False closed=False debug=False>
<_UnixSelectorEventLoop running=False closed=False debug=False>
<_UnixSelectorEventLoop running=False closed=False debug=False>
<_UnixSelectorEventLoop running=False closed=False debug=False>
Tested-by: Rafał Safin <rafal.safin12@gmail.com>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This is a preparatory patch that factors out the asyncio event loop
creation code from nxt_python_asgi_ctx_data_alloc() into its own
function, to facilitate being called multiple times.
This a part of the work to move away from using the
asyncio.get_event_loop() function due to it no longer creating event
loops if there wasn't one running.
See the following commit for the gory details.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This allows one to simply run `./configure` and expect it to
produce sane defaults for an install.
Previously, without specifying `--prefix=...`, `make install`
would simply fail, recommending to set `--prefix` or `DESTDIR`,
but that recommendation was incomplete at best, since it didn't
set many of the subdirs needed for a good organization.
Setting `DESTDIR` was even worse, since that shouldn't even affect
an installation (it is required to be transparent to the
installation).
/usr/local is the historic Unix standard path to use for
installations from source made manually by the admin of the
system. Some package managers (Homebrew, I'm looking specifically
at you) have abused that path to install their things, but 1) it's
not our fault that someone else incorrectly abuses that path (and
they seem to be fixing it for newer archs; e.g., they started
using /opt/homebrew for Apple Silicon), 2) there's no better path
than /usr/local, 3) we still allow changing it for systems where
this might not be the desired path (MacOS Intel with hombrew), and
4) it's _the standard_.
See a related conversation with Ingo (OpenBSD maintainer):
On 7/27/22 16:16, Ingo Schwarze wrote:
> Hi Alejandro,
[...]
>
> Alejandro Colomar wrote on Sun, Jul 24, 2022 at 07:07:18PM +0200:
>> On 7/24/22 16:57, Ingo Schwarze wrote:
>>> Alejandro Colomar wrote on Sun, Jul 24, 2022 at 01:20:46PM +0200:
>
>>>> /usr/local is for sysadmins to build from source;
>
>>> Doing that is *very* strongly discouraged on OpenBSD.
>
>> I guess that's why the directory was reused in the BSDs to install ports
>> (probably ports were installed by the sysadmin there, and by extension,
>> ports are now always installed there, but that's just a guess).
>
> Maybe. In any case, the practice of using /usr/local for packages
> created from ports is significantly older than the recommendation
> to refrain from using upstream "make install" outside the ports
> framework.
>
> * The FreeBSD ports framework was started by Jordan Hubbard in 1993.
> * The ports framework was ported from FreeBSD to OpenBSD
> by Niklas Hallqvist in 1996.
> * NetBSD pkgsrc was forked from FreeBSD ports by Alistair G. Crooks
> and Hubert Feyrer in 1997.
>
> I failed to quickly find Jordan's original version, but rev. 1.1
> of /usr/ports/infrastructure/mk/bsd.port.mk in OpenBSD (dated Jun 3
> 22:47:10 1996 UTC) already said
>
> LOCALBASE ?= /usr/local
> PREFIX ?= ${LOCALBASE}
>
[...]
>> I had a discussion in NGINX Unit about it, and
>> the decission for now has been: "support prefix=/usr/local for default
>> manual installation through the Makefile, and let BSD users adjust to
>> their preferred path".
>
> That's an *excellent* solution for the task, thanks for doing it
> the right way. By setting PREFIX=/usr/local by default in the
> upstream Makefile, you are minimizing the work for *BSD porters.
>
> The BSD ports frameworks will typically run the upstreak "make install"
> with the variable DESTDIR set to a custom value, for example
>
> DESTDIR=/usr/ports/pobj/groff-1.23.0/fake-amd64
>
> so if the upstream Makefile sets PREFIX=/usr/local ,
> that's perfect, everything gets installed to the right place
> without an intervention by the person doing the porting.
>
> Of course, if the upstream Makefile would use some other PREFIX,
> that would not be a huge obstacle. All we have to do in that case
> is pass the option --prefix=/usr/local to the ./configure script,
> or something equivalent if the software isn't using GNU configure.
>
>> We were concerned that we might get collisions
>> with the BSD port also installing in /usr/local, but that's the least
>> evil (and considering BSD users don't typically run `make install`, it's
>> not so bad).
>
> It's not bad at all. It's perfect.
>
> Of course, if a user wants to install *without* the ports framework,
> they have to provide their own --prefix. But that's not an issue
> because it is easy to do, and installing without a port is discouraged
> anyway.
===
Directory variables should never contain a trailing slash (I've
learned that the hard way, where some things would break
unexpectedly). Especially, make(1) is likely to have problems
when things have double slashes or a trailing slash, since it
treats filenames as text strings. I've removed the trailing slash
from the prefix, and added it to the derivate variables just after
the prefix. pkg-config(1) also expects directory variables to have
no trailing slash.
===
I also removed the code that would set variables as depending on
the prefix if they didn't start with a slash, because that is a
rather non-obvious behavior, and things should not always depend
on prefix, but other dirs such as $(runstatedir), so if we keep
a similar behavior it would be very unreliable. Better keep
variables intact if set, or use the default if unset.
===
Print the real defaults for ./configure --help, rather than the actual
values.
===
I used a subdirectory under the standard /var/lib for NXT_STATE,
instead of a homemade "state" dir that does the same thing.
===
Modified the Makefile to create some dirs that weren't being
created, and also remove those that weren't being removed in
uninstall, probably because someone forgot to add them.
===
Add new options for setting the new variables, and rename some to be
consistent with the standard names. Keep the old ones at configuration
time for compatibility, but mark them as deprecated. Don't keep the old
ones at exec time.
===
A summary of the default config is:
Unit configuration summary:
bin directory: ............. "/usr/local/bin"
sbin directory: ............ "/usr/local/sbin"
lib directory: ............. "/usr/local/lib"
include directory: ......... "/usr/local/include"
man pages directory: ....... "/usr/local/share/man"
modules directory: ......... "/usr/local/lib/unit/modules"
state directory: ........... "/usr/local/var/lib/unit"
tmp directory: ............. "/tmp"
pid file: .................. "/usr/local/var/run/unit/unit.pid"
log file: .................. "/usr/local/var/log/unit/unit.log"
control API socket: ........ "unix:/usr/local/var/run/unit/control.unit.sock"
Link: <https://www.gnu.org/prep/standards/html_node/Directory-Variables.html>
Link: <https://refspecs.linuxfoundation.org/FHS_3.0/fhs/index.html>
Reviewed-by: Artem Konev <a.konev@f5.com>
Reviewed-by: Andrew Clayton <a.clayton@nginx.com>
Tested-by: Andrew Clayton <a.clayton@nginx.com>
Reviewed-by: Konstantin Pavlov <thresh@nginx.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
|
|
Previously the PHP module would produce one of four status codes
200 OK
301 Moved Permanently
500 Internal Server Error
503 Service Unavailable
200 for successful requests, 301 for cases where the url was a directory
without a trailing '/', 500 for bad PHP or non-existing PHP file and 503
for all other errors.
With this commit we now handle missing files and directories, returning
404 Not Found and files and directories that don't allow access,
returning 403 Forbidden.
We do these checks in two places, when we check if we should do a
directory redirect (bar -> bar/) and in the nxt_php_execute() function.
One snag with the latter is that the php_execute_script() function only
returns success/failure (no reason). However while it took a
zend_file_handle structure with the filename of the script to run, we
can instead pass through an already opened file-pointer (FILE *) via
that structure. So we can try opening the script ourselves and do the
required checks before calling php_execute_script().
We also make use of the zend_stream_init_fp() function that initialises
the zend_file_handle structure if it's available otherwise we use our
own version. This is good because the zend_file_handle structure has
changed over time and the zend_stream_init_fp() function should change
with it.
Closes: <https://github.com/nginx/unit/issues/767>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Cc: Andrei Zeliankou <zelenkov@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Create a const char *filename variable to hold
ctx->script_filename.start, which is a much more manageable name and
will negate the need for any more casting in the following commit when
we switch to using a FILE * instead of a filename in
php_execute_script().
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Cc: Andrei Zeliankou <zelenkov@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Where possible make use of the zend_stream_init_filename() function
introduced in PHP 7.4.
This is essentially a preparatory patch for switching to using an
already opened file-pointer in nxt_php_execute(). While wrapping this
new code in a PHP version check with a fallback to our own function is
perhaps slightly overkill, it does reduce the diff of the commit that
switches to a FILE *.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Cc: Andrei Zeliankou <zelenkov@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
We're going to use zend_stream_init_filename in a following commit. To
reduce the diff of that change, move the current code that will be
replaced, to a function that has the same interface.
We use strlen(3) here to be able to use an interface without passing the
length, but we will remove that call in a following code, so it has no
performance issues.
Co-developed-by: Andrew Clayton <a.clayton@nginx.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
Reviewed-by: Andrew Clayton <a.clayton@nginx.com>
Cc: Andrei Zeliankou <zelenkov@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This commit is to loop through the request objects headers,
arguments, and cookies.
|
|
@dward on GitHub reported an issue with a URL like
http://foo.bar/test.php?blah=test.php/foo
where we would end up trying to run the script
test.php?blah=test.php
In the PHP module the format 'file.php/' is treated as a special case in
nxt_php_dynamic_request() where we check the _path_ part of the url for
the string '.php/'.
The problem is that the path actually also contains the query string,
thus we were finding 'test.php/' in the above URL and treating that
whole path as the script to run.
The fix is simple, replace the strstr(3) with a memmem(3), where we can
limit the amount of path we use for the check.
The trick here and what is not obvious from the code is that while
path.start points to the whole path including the query string,
path.length only contains the length of the _path_ part.
NOTE: memmem(3) is a GNU extension and is neither specified by POSIX or
ISO C, however it is available on a number of other systems, including:
FreeBSD, OpenBSD, NetBSD, illumos, and macOS.
If it comes to it we can implement a simple alternative for systems
which lack memmem(3).
This also adds a test case (provided by @dward) to cover this.
Closes: <https://github.com/nginx/unit/issues/781>
Cc: Andrei Zeliankou <zelenkov@nginx.com>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Reviewed-by: Andrei Zeliankou <zelenkov@nginx.com> [test]
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
The nxt_websocket_header_t structure defines the layout of a websocket
frame header. As the websocket frame is mapped directly onto this
structure its layout needs to match how it's coming off the network.
The network being big endian means on big endian systems the structure
layout can simply match that of the websocket frame header. On little
endian systems we need to reverse the two bytes.
This was done via the BYTE_ORDER, BIG_ENDIAN and LITTLE_ENDIAN macros,
however these are not universal, e.g they are _not_ defined on illumos
(OpenSolaris / OpenIndiana) and so we get the following compiler errors
In file included from src/nxt_h1proto.c:12:0:
src/nxt_websocket_header.h:25:13: error: duplicate member 'opcode'
uint8_t opcode:4;
^~~~~~
src/nxt_websocket_header.h:26:13: error: duplicate member 'rsv3'
uint8_t rsv3:1;
^~~~
src/nxt_websocket_header.h:27:13: error: duplicate member 'rsv2'
uint8_t rsv2:1;
^~~~
src/nxt_websocket_header.h:28:13: error: duplicate member 'rsv1'
uint8_t rsv1:1;
^~~~
src/nxt_websocket_header.h:29:13: error: duplicate member 'fin'
uint8_t fin:1;
^~~
src/nxt_websocket_header.h:31:13: error: duplicate member 'payload_len'
uint8_t payload_len:7;
^~~~~~~~~~~
src/nxt_websocket_header.h:32:13: error: duplicate member 'mask'
uint8_t mask:1;
^~~~
This commit fixes that by using the new NXT_HAVE_{BIG,LITTLE}_ENDIAN
macros introduced in the previous commit.
Closes: <https://github.com/nginx/unit/issues/297>
Fixes: e501c74 ("Introducing websocket support in router and libunit.")
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
There was a couple of reports of Python applications failing due to the
following type of error
File "/opt/netbox/netbox/netbox/configuration.py", line 25, in _import
print(f"\U0001f9ec loaded config '{path}'")
UnicodeEncodeError: 'ascii' codec can't encode character '\U0001f9ec' in
position 0: ordinal not in range(128)
due to the use of Unicode text in the print() statement.
This only happened for python 3.8+ when using the "home" configuration
option as this meant we were going through the new PyConfig
configuration.
When using this new configuration method with the 'isolated' specific
API (for embedded Python) UTF-8 is disabled by default,
PyPreConfig->utf8_mode = 0.
To fix this we need to setup the Python pre config and enable utf-8
mode. However rather than enable utf-8 unconditionally we can set to it
to -1 so that it will use the LC_CTYPE environment variable to determine
whether to enable utf-8 mode or not. utf-8 mode will be enabled if
LC_CTYPE is either: C, POSIX or some specific UTF-8 locale. This is the
default utf8_mode setting when using the non-isolated PyPreConfig API.
Reported-by: Tobias Genannt <tobias.genannt@kappa-velorum.net>
Tested-by: Tobias Genannt <tobias.genannt@kappa-velorum.net>
Link: <https://peps.python.org/pep-0587/>
Link: <https://docs.python.org/3/c-api/init_config.html#c.PyPreConfig.utf8_mode>
Fixes: 491d0f70 ("Python: Added support for Python 3.11.")
Closes: <https://github.com/nginx/unit/issues/817>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This is a preparatory patch for future work and cleans up the code a
little in the Python 3.8+ variant of nxt_python3_init_config().
The main advantage being we no longer have calls to PyConfig_Clear() in
two different paths.
The variables have a little extra space in their declarations to allow
for the next patch which introduces a variable with a longer type name,
which will help reduce the size of the diff.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This patch gives users the option to set a `"prefix"` attribute
for Python applications, either at the top level or for specific
`"target"`s. If the attribute is present, the value of `"prefix"`
must be a string beginning with `"/"`. If the value of the `"prefix"`
attribute is longer than 1 character and ends in `"/"`, the
trailing `"/"` is stripped.
The purpose of the `"prefix"` attribute is to set the `SCRIPT_NAME`
context value for WSGI applications and the `root_path` context
value for ASGI applications, allowing applications to properly route
requests regardless of the path that the server uses to expose the
application.
The context value is only set if the request's URL path begins with
the value of the `"prefix"` attribute. In all other cases, the
`SCRIPT_NAME` or `root_path` values are not set. In addition, for
WSGI applications, the value of `"prefix"` will be stripped from
the beginning of the request's URL path before it is sent to the
application.
Reviewed-by: Andrei Zeliankou <zelenkov@nginx.com>
Reviewed-by: Artem Konev <artem.konev@nginx.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
|
|
Signed-off-by: Alejandro Colomar <alx@nginx.com>
|
|
This is an extension to the previous commit, which made
large_header_buffer_size a valid configuration setting.
This commit makes a related value, large_header_buffers, a valid
configuration setting.
While large_header_buffer_size effectively limits the maximum size of
any single header (although unit will try to pack multiple headers into
a buffer if they wholly fit).
large_header_buffers limits how many of these 'large' buffers are
available. It makes sense to also allow this to be user set.
large_header_buffers is already set by the configuration system in
nxt_router.c it just isn't set as a valid config option in
nxt_conf_validation.c
With this change users can set this option in their config if required
by the following
"settings": {
"http": {
"large_header_buffers": 8
}
},
It retains its default value of 4 if this is not set.
NOTE: This is being released as undocumented and subject to change as it
exposes internal workings of unit.
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
@JanMikes and @tagur87 on GitHub both reported issues with long URLs
that were exceeding the 8192 byte large_header_buffer_size setting,
which resulted in a HTTP 431 error (Request Header Fields Too Large).
This can be resolved in the code by updating the following line in
src/nxt_router.c::nxt_router_conf_create()
skcf->large_header_buffer_size = 8192;
However, requiring users to modify unit and install custom versions is
less than ideal. We could increase the value, but to what?
This commit takes the option of allowing the user to set this option in
their config by making large_header_buffer_size a valid configuration
setting.
large_header_buffer_size is already set by the configuration system in
nxt_router.c it just isn't set as a valid config option in
nxt_conf_validation.c
With this change users can set this option in their config if required
by the following
"settings": {
"http": {
"large_header_buffer_size": 16384
}
},
It retains its default value of 8192 bytes if this is not set.
With this commit, without the above setting or too low a value, with a
long URL you get a 431 error. With the above setting set to a large
enough value, the request is successful.
NOTE: This setting really determines the maximum size of any single
header _value_. Also, unit will try and place multiple values
into a buffer _if_ they fully fit.
NOTE: This is being released as undocumented and subject to change as it
exposes internal workings of unit.
Closes: <https://github.com/nginx/unit/issues/521>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This hooks the cgroup support up to the config system so it can actually
be used.
To make use of this in unit a new "cgroup" section has been added to the
isolation configuration.
e.g
"applications": {
"python": {
"type": "python",
"processes": 5,
"path": "/opt/unit/unit-cgroup-test/",
"module": "app",
"isolation": {
"cgroup": {
"path": "app/python"
}
}
}
}
Now there are two ways to specify the path, relative, like the above
(without a leading '/') and absolute (with a leading '/').
In the above case the "python" application is placed into its own cgroup
under CGROUP_ROOT/<main unit process cgroup>/app/python. Whereas if you
specified say
"path": "/unit/app/python"
Then the python application would be placed under
CGROUP_ROOT/unit/app/python
The first option allows you to easily take advantage of any resource
limits that have already been configured for unit.
With the second method (absolute pathname) if you know of an already
existing cgroup where you'd like to place it, you can, e.g
"path": "/system.slice/unit/python"
Where system.slice has already been created by systemd and may already
have some overall system limits applied which would also apply to unit.
Limits apply down the hierarchy and lower groups can't exceed the
previous group limits.
So what does this actually look like? Lets take the unit-calculator
application[0] and have each of its applications placed into their own
cgroup. If we give each application a new section like
"isolation": {
"cgroup": {
"path": "/unit/unit-calculator/add"
}
}
changing the path for each one, we can visualise the result with the
systemd-cgls command, e.g
│ └─session-5.scope (#4561)
│ ├─ 6667 sshd: andrew [priv]
│ ├─ 6684 sshd: andrew@pts/0
│ ├─ 6685 -bash
│ ├─ 12632 unit: main v1.28.0 [/opt/unit/sbin/unitd --control 127.0.0.1:808>
│ ├─ 12634 unit: controller
│ ├─ 12635 unit: router
│ ├─ 13550 systemd-cgls
│ └─ 13551 less
├─unit (#4759)
│ └─unit-calculator (#5037)
│ ├─subtract (#5069)
│ │ ├─ 12650 unit: "subtract" prototype
│ │ └─ 12651 unit: "subtract" application
│ ├─multiply (#5085)
│ │ ├─ 12653 unit: "multiply" prototype
│ │ └─ 12654 unit: "multiply" application
│ ├─divide (#5101)
│ │ ├─ 12671 unit: "divide" prototype
│ │ └─ 12672 node divide.js
│ ├─sqroot (#5117)
│ │ ├─ 12679 unit: "sqroot" prototype
│ │ └─ 12680 /home/andrew/src/unit-calculator/sqroot/sqroot
│ └─add (#5053)
│ ├─ 12648 unit: "add" prototype
│ └─ 12649 unit: "add" application
We used an absolute path so the cgroups will be created relative to the
main cgroupfs mount, e.g /sys/fs/cgroup
We can see that the main unit processes are in the same cgroup as the
shell from where they were started, by default child process are placed
into the same cgroup as the parent.
Then we can see that each application has been placed into its own
cgroup under /sys/fs/cgroup
Taking another example of a simple 5 process python application, with
"isolation": {
"cgroup": {
"path": "app/python"
}
}
Here we have specified a relative path and thus the python application
will be placed below the existing cgroup that contains the main unit
process. E.g
│ │ │ ├─app-glib-cinnamon\x2dcustom\x2dlauncher\x2d3-43951.scope (#90951)
│ │ │ │ ├─ 988 unit: main v1.28.0 [/opt/unit/sbin/unitd --no-daemon]
│ │ │ │ ├─ 990 unit: controller
│ │ │ │ ├─ 991 unit: router
│ │ │ │ ├─ 43951 xterm -bg rgb:20/20/20 -fg white -fa DejaVu Sans Mono
│ │ │ │ ├─ 43956 bash
│ │ │ │ ├─ 58828 sudo -i
│ │ │ │ ├─ 58831 -bash
│ │ │ │ └─app (#107351)
│ │ │ │ └─python (#107367)
│ │ │ │ ├─ 992 unit: "python" prototype
│ │ │ │ ├─ 993 unit: "python" application
│ │ │ │ ├─ 994 unit: "python" application
│ │ │ │ ├─ 995 unit: "python" application
│ │ │ │ ├─ 996 unit: "python" application
│ │ │ │ └─ 997 unit: "python" application
[0]: <https://github.com/lcrilly/unit-calculator>
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
This commit hooks into the cgroup infrastructure added in the previous
commit to create per-application cgroups.
It does this by adding each "prototype process" into its own cgroup,
then each child process inherits its parents cgroup.
If we fail to create a cgroup we simply fail the process. This behaviour
may get enhanced in the future.
This won't actually do anything yet. Subsequent commits will hook this
up to the build and config systems.
Reviewed-by: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
|
Firstly, this is not to be confused with CLONE_NEWCGROUP which unit
already supports and is related to namespaces. To re-cap, namespaces
allow processes to have different views of various parts of the system
such as filesystem mounts, networking, hostname etc.
Whereas cgroup[0] is a Linux kernel facility for collecting a bunch of
processes together to perform some task on the group as a whole, for
example to implement resource limits.
There are two parts to cgroup, the core part of organising processes
into a hierarchy and the controllers which are responsible for enforcing
resource limits etc.
There are currently two versions of the cgroup sub-system, the original
cgroup and a version 2[1] introduced in 3.16 (August 2014) and marked
stable in 4.5 (March 2016).
This commit supports the cgroup V2 API and implements the ability to
place applications into their own cgroup on a per-application basis.
You can put them each into their own cgroup or you can group some
together. The ability to set resource limits can easily be added in
future.
The initial use case of this would be to aid in observability of unit
applications which becomes much easier if you can just monitor them on a
per cgroup basis.
One thing to note about cgroup, is that unlike namespaces which are
controlled via system calls such as clone(2) and unshare(2), cgroups are
setup and controlled through the cgroupfs pseudo-filesystem.
cgroup is Linux only and this support will only be enabled if configure
finds the cgroup2 filesystem mount, e.g
cgroup2 on /sys/fs/cgroup type cgroup2 (rw,nosuid,nodev,noexec,relatime,seclabel,nsdelegate,memory_recursiveprot)
The cgroups are removed on shutdown or as required on reconfiguration.
This commit just adds the basic infrastructure for using cgroups within
unit. Subsequent commits will wire up this support.
It supports creating cgroups relative to the main cgroup root and also
below the cgroup of the main unit process.
[0]: <https://man7.org/linux/man-pages/man7/cgroups.7.html>
[1]: <https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html>
Cc: Alejandro Colomar <alx@nginx.com>
Signed-off-by: Andrew Clayton <a.clayton@nginx.com>
|
