shaba/systemd
1
0
Fork 0
mirror of https://github.com/systemd/systemd.git synced 2024-12-23 21:35:11 +03:00
Code

core: add new logic for services to store file descriptors in PID 1

Browse Source 
With this change it is possible to send file descriptors to PID 1, via
sd_pid_notify_with_fds() which PID 1 will store individually for each
service, and pass via the usual fd passing logic on next invocation.
This is useful for enable daemon reload schemes where daemons serialize
their state to /run, push their fds into PID 1 and terminate, restoring
their state on next start from the data in /run and passed in from PID
1.

The fds are kept by PID 1 as long as no POLLHUP or POLLERR is seen on
them, and the service they belong to are either not dead or failed, or
have a job queued.
This commit is contained in:
Lennart Poettering 2015-01-06 00:26:25 +01:00
parent 7539904965
commit a354329f72
18 changed files with 598 additions and 104 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
Makefile-man.am
View File

@ -198,6 +198,9 @@ MANPAGES_ALIAS += \
man/sd_journal_wait.3 \
man/sd_machine_get_ifindices.3 \
man/sd_notifyf.3 \
man/sd_pid_notify.3 \
man/sd_pid_notify_with_fds.3 \
man/sd_pid_notifyf.3 \
man/sleep.conf.d.5 \
man/system.conf.d.5 \
man/systemd-ask-password-console.path.8 \
@ -308,6 +311,9 @@ man/sd_journal_test_cursor.3: man/sd_journal_get_cursor.3
man/sd_journal_wait.3: man/sd_journal_get_fd.3
man/sd_machine_get_ifindices.3: man/sd_machine_get_class.3
man/sd_notifyf.3: man/sd_notify.3
man/sd_pid_notify.3: man/sd_notify.3
man/sd_pid_notify_with_fds.3: man/sd_notify.3
man/sd_pid_notifyf.3: man/sd_notify.3
man/sleep.conf.d.5: man/systemd-sleep.conf.5
man/system.conf.d.5: man/systemd-system.conf.5
man/systemd-ask-password-console.path.8: man/systemd-ask-password-console.service.8
@ -566,6 +572,15 @@ man/sd_machine_get_ifindices.html: man/sd_machine_get_class.html
man/sd_notifyf.html: man/sd_notify.html
$(html-alias)
man/sd_pid_notify.html: man/sd_notify.html
$(html-alias)
man/sd_pid_notify_with_fds.html: man/sd_notify.html
$(html-alias)
man/sd_pid_notifyf.html: man/sd_notify.html
$(html-alias)
man/sleep.conf.d.html: man/systemd-sleep.conf.html
$(html-alias)
@ -674,12 +689,6 @@ man/systemd-user.conf.html: man/systemd-system.conf.html
man/user.conf.d.html: man/systemd-system.conf.html
$(html-alias)
if ENABLE_HWDB
MANPAGES += \
man/hwdb.7 \
man/systemd-hwdb.8
endif
if ENABLE_BACKLIGHT
MANPAGES += \
@ -750,6 +759,16 @@ man/systemd-hostnamed.html: man/systemd-hostnamed.service.html
endif
if ENABLE_HWDB
MANPAGES += \
man/hwdb.7 \
man/systemd-hwdb.8
MANPAGES_ALIAS += \
#
endif
if ENABLE_KDBUS
MANPAGES += \
man/sd_bus_creds_get_pid.3 \

21
man/sd_listen_fds.xml
View File

@ -73,7 +73,7 @@
<para>If the <parameter>unset_environment</parameter>
parameter is non-zero,
<function>sd_listen_fds()</function> will unset the
<varname>$LISTEN_FDS</varname>/<varname>$LISTEN_PID</varname>
<varname>$LISTEN_FDS</varname> and <varname>$LISTEN_PID</varname>
environment variables before returning (regardless of
whether the function call itself succeeded or
not). Further calls to
@ -83,10 +83,11 @@
<para>If a daemon receives more than one file
descriptor, they will be passed in the same order as
configured in the systemd socket definition
file. Nonetheless, it is recommended to verify the
correct socket types before using them. To simplify
this checking, the functions
configured in the systemd socket unit file (see
<citerefentry><refentrytitle>systemd.socket</refentrytitle><manvolnum>5</manvolnum></citerefentry>
for details). Nonetheless, it is recommended to verify
the correct socket types before using them. To
simplify this checking, the functions
<citerefentry><refentrytitle>sd_is_fifo</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
<citerefentry><refentrytitle>sd_is_socket</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
<citerefentry><refentrytitle>sd_is_socket_inet</refentrytitle><manvolnum>3</manvolnum></citerefentry>,
@ -103,6 +104,16 @@
<para>This function call will set the FD_CLOEXEC flag
for all passed file descriptors to avoid further
inheritance to children of the calling process.</para>
<para>If multiple socket units activate the same
service the order of the file descriptors passed to
its main process is undefined. If additional file
descriptors have been passed to the service manager
using
<citerefentry><refentrytitle>sd_pid_notify_with_fds</refentrytitle><manvolnum>3</manvolnum></citerefentry>'s
<literal>FDSTORE=1</literal> messages, these file
descriptors are passed last, in arbitrary order, and
with duplicates removed.</para>
</refsect1>
<refsect1>

122
man/sd_notify.xml
View File

@ -46,6 +46,9 @@
<refnamediv>
<refname>sd_notify</refname>
<refname>sd_notifyf</refname>
<refname>sd_pid_notify</refname>
<refname>sd_pid_notifyf</refname>
<refname>sd_pid_notify_with_fds</refname>
<refpurpose>Notify service manager about start-up completion and other service status changes</refpurpose>
</refnamediv>
@ -65,6 +68,30 @@
<paramdef>const char *<parameter>format</parameter></paramdef>
<paramdef>...</paramdef>
</funcprototype>
<funcprototype>
<funcdef>int <function>sd_pid_notify</function></funcdef>
<paramdef>pid_t <parameter>pid</parameter></paramdef>
<paramdef>int <parameter>unset_environment</parameter></paramdef>
<paramdef>const char *<parameter>state</parameter></paramdef>
</funcprototype>
<funcprototype>
<funcdef>int <function>sd_pid_notifyf</function></funcdef>
<paramdef>pid_t <parameter>pid</parameter></paramdef>
<paramdef>int <parameter>unset_environment</parameter></paramdef>
<paramdef>const char *<parameter>format</parameter></paramdef>
<paramdef>...</paramdef>
</funcprototype>
<funcprototype>
<funcdef>int <function>sd_pid_notify_with_fds</function></funcdef>
<paramdef>pid_t <parameter>pid</parameter></paramdef>
<paramdef>int <parameter>unset_environment</parameter></paramdef>
<paramdef>const char *<parameter>state</parameter></paramdef>
<paramdef>const int *<parameter>fds</parameter></paramdef>
<paramdef>unsigned <parameter>n_fds</parameter></paramdef>
</funcprototype>
</funcsynopsis>
</refsynopsisdiv>
@ -175,7 +202,7 @@
<varlistentry>
<term>MAINPID=...</term>
<listitem><para>The main pid of the
<listitem><para>The main process ID (PID) of the
service, in case the service manager did
not fork off the process
itself. Example:
@ -185,7 +212,7 @@
<varlistentry>
<term>WATCHDOG=1</term>
<listitem><para>Tells systemd to
<listitem><para>Tells the service manager to
update the watchdog timestamp. This is
the keep-alive ping that services need
to issue in regular intervals if
@ -199,12 +226,53 @@
check if the the watchdog is enabled.
</para></listitem>
</varlistentry>
<varlistentry>
<term>FDSTORE=1</term>
<listitem><para>Stores additional file
descriptors in the service
manager. File descriptors sent this
way will be maintained per-service by
the service manager and be passed
again using the usual file descriptor
passing logic on the next invocation
of the service (see
<citerefentry><refentrytitle>sd_listen_fds</refentrytitle><manvolnum>3</manvolnum></citerefentry>). This
is useful for implementing service
restart schemes where services
serialize their state to
<filename>/run</filename>, push their
file descriptors to the system
manager, and are then restarted,
retrieving their state again via
socket passing and
<filename>/run</filename>. Note that
the service manager will accept
messages for a service only if
<varname>FileDescriptorStoreMax=</varname>
is set to non-zero for it (defaults to
zero). See
<citerefentry><refentrytitle>systemd.service</refentrytitle><manvolnum>5</manvolnum></citerefentry>
for details. Multiple arrays of file
descriptors may be sent in seperate
messages, in which case the arrays are
combined. Note that the service
manager removes duplicate file
descriptors before passing them to the
service. Use
<function>sd_pid_notify_with_fds()</function>
to send messages with
<literal>FDSTORE=1</literal>, see
below.</para></listitem>
</varlistentry>
</variablelist>
<para>It is recommended to prefix variable names that
are not shown in the list above with
<varname>X_</varname> to avoid namespace
clashes.</para>
are not listed above with <varname>X_</varname> to
avoid namespace clashes.</para>
<para>Note that systemd will accept status data sent
from a service only if the
@ -217,6 +285,36 @@
<function>sd_notify()</function> but takes a
<function>printf()</function>-like format string plus
arguments.</para>
<para><function>sd_pid_notify()</function> and
<function>sd_pid_notifyf()</function> are similar to
<function>sd_notify()</function> and
<function>sd_notifyf()</function> but take a process
ID (PID) to use as originating PID for the message as
first argument. This is useful to send notification
messages on behalf of other processes, provided the
appropriate privileges are available. If the PID
argument is specified as 0 the process ID of the
calling process is used, in which case the calls are
fully equivalent to <function>sd_notify()</function>
and <function>sd_notifyf()</function>.</para>
<para><function>sd_pid_notify_with_fds()</function> is
similar to <function>sd_pid_notify()</function> but
takes an additional array of file descriptors. These
file descriptors are sent along the notification
message to the service manager. This is particularly
useful for sending <literal>FDSTORE=1</literal>
messages, as described above. The additional arguments
are a pointer to the file descriptor array plus the
number of file descriptors in the array. If the number
of file descriptors is passed as 0, the call is fully
equivalent to <function>sd_pid_notify()</function>,
i.e. no file descriptors are passed. Note that sending
file descriptors to the service manager on messages
that do not expect them (i.e. without
<literal>FDSTORE=1</literal>) they are immediately
closed on reception.</para>
</refsect1>
<refsect1>
@ -295,13 +393,25 @@
<example>
<title>Error Cause Notification</title>
<para>A service could send the following shortly before exiting, on failure</para>
<para>A service could send the following shortly before exiting, on failure:</para>
<programlisting>sd_notifyf(0, "STATUS=Failed to start up: %s\n"
"ERRNO=%i",
strerror(errno),
errno);</programlisting>
</example>
<example>
<title>Store a File Descriptor in the Service Manager</title>
<para>To store an open file descriptor in the
service manager, in order to continue
operation after a service restart without
losing state use
<literal>FDSTORE=1</literal>:</para>
<programlisting>sd_pid_notify_with_fds(0, 0, "FDSTORE=1", &amp;fd, 1);</programlisting>
</example>
</refsect1>
<refsect1>

29
man/systemd.service.xml
View File

@ -1117,6 +1117,35 @@
command.</para></listitem>
</varlistentry>
<varlistentry>
<term><varname>FileDescriptorStoreMax=</varname></term>
<listitem><para>Configure how many
file descriptors may be stored in the
service manager for the service using
<citerefentry><refentrytitle>sd_pid_notify_with_fds</refentrytitle><manvolnum>3</manvolnum></citerefentry>'s
<literal>FDSTORE=1</literal>
messages. This is useful for
implementing service restart schemes
where the state is serialized to
<filename>/run</filename> and the file
descriptors passed to the service
manager, to allow restarts without
losing state. Defaults to 0, i.e. no
file descriptors may be stored in the
service manager by default. All file
descriptors passed to the service
manager from a specific service are
passed back to the service's main
process on the next service
restart. Any file descriptors passed
to the service manager are
automatically closed when POLLHUP or
POLLERR is seen on them, or when the
service is fully stopped and no job
queued or being executed for
it.</para></listitem>
</varlistentry>
</variablelist>
<para>Check

1
src/core/dbus-service.c
View File

@ -59,6 +59,7 @@ const sd_bus_vtable bus_service_vtable[] = {
SD_BUS_PROPERTY("MainPID", "u", bus_property_get_pid, offsetof(Service, main_pid), SD_BUS_VTABLE_PROPERTY_EMITS_CHANGE),
SD_BUS_PROPERTY("ControlPID", "u", bus_property_get_pid, offsetof(Service, control_pid), SD_BUS_VTABLE_PROPERTY_EMITS_CHANGE),
SD_BUS_PROPERTY("BusName", "s", NULL, offsetof(Service, bus_name), SD_BUS_VTABLE_PROPERTY_CONST),
SD_BUS_PROPERTY("FileDescriptorStoreMax", "u", NULL, offsetof(Service, n_fd_store_max), SD_BUS_VTABLE_PROPERTY_CONST),
SD_BUS_PROPERTY("StatusText", "s", NULL, offsetof(Service, status_text), SD_BUS_VTABLE_PROPERTY_EMITS_CHANGE),
SD_BUS_PROPERTY("StatusErrno", "i", NULL, offsetof(Service, status_errno), SD_BUS_VTABLE_PROPERTY_EMITS_CHANGE),
SD_BUS_PROPERTY("Result", "s", property_get_result, offsetof(Service, result), SD_BUS_VTABLE_PROPERTY_EMITS_CHANGE),

1
src/core/load-fragment-gperf.gperf.m4
View File

@ -226,6 +226,7 @@ Service.SuccessExitStatus, config_parse_set_status, 0,
Service.SysVStartPriority, config_parse_warn_compat, DISABLED_LEGACY, 0
Service.NonBlocking, config_parse_bool, 0, offsetof(Service, exec_context.non_blocking)
Service.BusName, config_parse_unit_string_printf, 0, offsetof(Service, bus_name)
Service.FileDescriptorStoreMax, config_parse_unsigned, 0, offsetof(Service, n_fd_store_max)
Service.NotifyAccess, config_parse_notify_access, 0, offsetof(Service, notify_access)
Service.Sockets, config_parse_service_sockets, 0, 0
m4_ifdef(`ENABLE_KDBUS',

78
src/core/manager.c
View File

@ -84,6 +84,9 @@
#define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
#define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
#define NOTIFY_FD_MAX 768
#define NOTIFY_BUFFER_MAX PIPE_BUF
static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
@ -1449,7 +1452,7 @@ static unsigned manager_dispatch_dbus_queue(Manager *m) {
return n;
}
static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, char *buf, size_t n) {
static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, char *buf, size_t n, FDSet *fds) {
_cleanup_strv_free_ char **tags = NULL;
assert(m);
@ -1466,12 +1469,13 @@ static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, char *
log_unit_debug(u->id, "Got notification message for unit %s", u->id);
if (UNIT_VTABLE(u)->notify_message)
UNIT_VTABLE(u)->notify_message(u, pid, tags);
UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
}
static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
Manager *m = userdata;
ssize_t n;
int r;
assert(m);
assert(m->notify_fd == fd);
@ -1482,73 +1486,101 @@ static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t
}
for (;;) {
char buf[4096];
_cleanup_fdset_free_ FDSet *fds = NULL;
char buf[NOTIFY_BUFFER_MAX+1];
struct iovec iovec = {
.iov_base = buf,
.iov_len = sizeof(buf)-1,
};
bool found = false;
union {
struct cmsghdr cmsghdr;
uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
} control = {};
struct msghdr msghdr = {
.msg_iov = &iovec,
.msg_iovlen = 1,
.msg_control = &control,
.msg_controllen = sizeof(control),
};
struct ucred *ucred;
struct cmsghdr *cmsg;
struct ucred *ucred = NULL;
bool found = false;
Unit *u1, *u2, *u3;
int *fd_array = NULL;
unsigned n_fds = 0;
n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT);
if (n <= 0) {
if (n == 0)
return -EIO;
n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
if (n < 0) {
if (errno == EAGAIN || errno == EINTR)
break;
return -errno;
}
if (n == 0)
return -ECONNRESET;
if (msghdr.msg_controllen < CMSG_LEN(sizeof(struct ucred)) ||
control.cmsghdr.cmsg_level != SOL_SOCKET ||
control.cmsghdr.cmsg_type != SCM_CREDENTIALS ||
control.cmsghdr.cmsg_len != CMSG_LEN(sizeof(struct ucred))) {
log_warning("Received notify message without credentials. Ignoring.");
for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
fd_array = (int*) CMSG_DATA(cmsg);
n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
} else if (cmsg->cmsg_level == SOL_SOCKET &&
cmsg->cmsg_type == SCM_CREDENTIALS &&
cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
ucred = (struct ucred*) CMSG_DATA(cmsg);
}
}
if (n_fds > 0) {
assert(fd_array);
r = fdset_new_array(&fds, fd_array, n_fds);
if (r < 0) {
close_many(fd_array, n_fds);
return log_oom();
}
}
if (!ucred || ucred->pid <= 0) {
log_warning("Received notify message without valid credentials. Ignoring.");
continue;
}
ucred = (struct ucred*) CMSG_DATA(&control.cmsghdr);
if ((size_t) n >= sizeof(buf)) {
log_warning("Received notify message exceeded maximum size. Ignoring.");
continue;
}
assert((size_t) n < sizeof(buf));
buf[n] = 0;
/* Notify every unit that might be interested, but try
* to avoid notifying the same one multiple times. */
u1 = manager_get_unit_by_pid(m, ucred->pid);
if (u1) {
manager_invoke_notify_message(m, u1, ucred->pid, buf, n);
manager_invoke_notify_message(m, u1, ucred->pid, buf, n, fds);
found = true;
}
u2 = hashmap_get(m->watch_pids1, LONG_TO_PTR(ucred->pid));
if (u2 && u2 != u1) {
manager_invoke_notify_message(m, u2, ucred->pid, buf, n);
manager_invoke_notify_message(m, u2, ucred->pid, buf, n, fds);
found = true;
}
u3 = hashmap_get(m->watch_pids2, LONG_TO_PTR(ucred->pid));
if (u3 && u3 != u2 && u3 != u1) {
manager_invoke_notify_message(m, u3, ucred->pid, buf, n);
manager_invoke_notify_message(m, u3, ucred->pid, buf, n, fds);
found = true;
}
if (!found)
log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
if (fdset_size(fds) > 0)
log_warning("Got auxiliary fds with notification message, closing all.");
}
return 0;

166
src/core/service.c
View File

@ -242,6 +242,42 @@ static void service_reset_watchdog(Service *s) {
service_start_watchdog(s);
}
static void service_fd_store_unlink(ServiceFDStore *fs) {
if (!fs)
return;
if (fs->service) {
assert(fs->service->n_fd_store > 0);
LIST_REMOVE(fd_store, fs->service->fd_store, fs);
fs->service->n_fd_store--;
}
if (fs->event_source) {
sd_event_source_set_enabled(fs->event_source, SD_EVENT_OFF);
sd_event_source_unref(fs->event_source);
}
safe_close(fs->fd);
free(fs);
}
static void service_release_resources(Unit *u) {
Service *s = SERVICE(u);
assert(s);
if (!s->fd_store)
return;
log_debug("Releasing all resources for %s", u->id);
while (s->fd_store)
service_fd_store_unlink(s->fd_store);
assert(s->n_fd_store == 0);
}
static void service_done(Unit *u) {
Service *s = SERVICE(u);
@ -286,6 +322,8 @@ static void service_done(Unit *u) {
service_stop_watchdog(s);
s->timer_event_source = sd_event_source_unref(s->timer_event_source);
service_release_resources(u);
}
static int service_arm_timer(Service *s, usec_t usec) {
@ -549,6 +587,14 @@ static void service_dump(Unit *u, FILE *f, const char *prefix) {
if (s->status_text)
fprintf(f, "%sStatus Text: %s\n",
prefix, s->status_text);
if (s->n_fd_store_max > 0) {
fprintf(f,
"%sFile Descriptor Store Max: %u\n"
"%sFile Descriptor Store Current: %u\n",
prefix, s->n_fd_store_max,
prefix, s->n_fd_store);
}
}
static int service_load_pid_file(Service *s, bool may_warn) {
@ -806,10 +852,10 @@ static int service_coldplug(Unit *u) {
}
static int service_collect_fds(Service *s, int **fds, unsigned *n_fds) {
_cleanup_free_ int *rfds = NULL;
unsigned rn_fds = 0;
Iterator i;
int r;
int *rfds = NULL;
unsigned rn_fds = 0;
Unit *u;
assert(s);
@ -831,10 +877,12 @@ static int service_collect_fds(Service *s, int **fds, unsigned *n_fds) {
r = socket_collect_fds(sock, &cfds, &cn_fds);
if (r < 0)
goto fail;
return r;
if (!cfds)
if (cn_fds <= 0) {
free(cfds);
continue;
}
if (!rfds) {
rfds = cfds;
@ -842,32 +890,39 @@ static int service_collect_fds(Service *s, int **fds, unsigned *n_fds) {
} else {
int *t;
t = new(int, rn_fds+cn_fds);
t = realloc(rfds, (rn_fds + cn_fds) * sizeof(int));
if (!t) {
free(cfds);
r = -ENOMEM;
goto fail;
return -ENOMEM;
}
memcpy(t, rfds, rn_fds * sizeof(int));
memcpy(t+rn_fds, cfds, cn_fds * sizeof(int));
free(rfds);
memcpy(t + rn_fds, cfds, cn_fds * sizeof(int));
rfds = t;
rn_fds += cn_fds;
free(cfds);
rfds = t;
rn_fds = rn_fds+cn_fds;
}
}
if (s->n_fd_store > 0) {
ServiceFDStore *fs;
int *t;
t = realloc(rfds, (rn_fds + s->n_fd_store) * sizeof(int));
if (!t)
return -ENOMEM;
rfds = t;
LIST_FOREACH(fd_store, fs, s->fd_store)
rfds[rn_fds++] = fs->fd;
}
*fds = rfds;
*n_fds = rn_fds;
rfds = NULL;
return 0;
fail:
free(rfds);
return r;
}
static int service_spawn(
@ -2543,7 +2598,75 @@ static int service_dispatch_watchdog(sd_event_source *source, usec_t usec, void
return 0;
}
static void service_notify_message(Unit *u, pid_t pid, char **tags) {
static int on_fd_store_io(sd_event_source *e, int fd, uint32_t revents, void *userdata) {
ServiceFDStore *fs = userdata;
assert(e);
assert(fs);
/* If we get either EPOLLHUP or EPOLLERR, it's time to remove this entry from the fd store */
service_fd_store_unlink(fs);
return 0;
}
static int service_add_fd_set(Service *s, FDSet *fds) {
int r;
assert(s);
if (fdset_size(fds) <= 0)
return 0;
while (s->n_fd_store < s->n_fd_store_max) {
_cleanup_close_ int fd = -1;
ServiceFDStore *fs;
bool same = false;
fd = fdset_steal_first(fds);
if (fd < 0)
break;
LIST_FOREACH(fd_store, fs, s->fd_store) {
r = same_fd(fs->fd, fd);
if (r < 0)
return log_unit_error_errno(UNIT(s)->id, r, "%s: Couldn't check if same fd: %m", UNIT(s)->id);
if (r > 0) {
same = true;
break;
}
}
if (same)
continue;
fs = new0(ServiceFDStore, 1);
if (!fs)
return log_oom();
fs->fd = fd;
fs->service = s;
r = sd_event_add_io(UNIT(s)->manager->event, &fs->event_source, fd, 0, on_fd_store_io, fs);
if (r < 0) {
free(fs);
return log_unit_error_errno(UNIT(s)->id, r, "%s: Failed to add even source: %m", UNIT(s)->id);
}
LIST_PREPEND(fd_store, s->fd_store, fs);
s->n_fd_store++;
fd = -1;
log_unit_debug(UNIT(s)->id, "%s: added fd to fd store.", UNIT(s)->id);
}
if (fdset_size(fds) > 0)
log_unit_warning(UNIT(s)->id, "%s: tried to store more fds than FDStoreMax=%u allows, closing remaining.", UNIT(s)->id, s->n_fd_store_max);
return 0;
}
static void service_notify_message(Unit *u, pid_t pid, char **tags, FDSet *fds) {
Service *s = SERVICE(u);
_cleanup_free_ char *cc = NULL;
bool notify_dbus = false;
@ -2675,6 +2798,12 @@ static void service_notify_message(Unit *u, pid_t pid, char **tags) {
service_reset_watchdog(s);
}
/* Add the passed fds to the fd store */
if (strv_find(tags, "FDSTORE=1")) {
log_unit_debug(u->id, "%s: got FDSTORE=1", u->id);
service_add_fd_set(s, fds);
}
/* Notify clients about changed status or main pid */
if (notify_dbus)
unit_add_to_dbus_queue(u);
@ -2917,6 +3046,7 @@ const UnitVTable service_vtable = {
.init = service_init,
.done = service_done,
.load = service_load,
.release_resources = service_release_resources,
.coldplug = service_coldplug,

14
src/core/service.h
View File

@ -22,6 +22,7 @@
***/
typedef struct Service Service;
typedef struct ServiceFDStore ServiceFDStore;
#include "unit.h"
#include "path.h"
@ -115,6 +116,15 @@ typedef enum ServiceResult {
_SERVICE_RESULT_INVALID = -1
} ServiceResult;
struct ServiceFDStore {
Service *service;
int fd;
sd_event_source *event_source;
LIST_FIELDS(ServiceFDStore, fd_store);
};
struct Service {
Unit meta;
@ -198,6 +208,10 @@ struct Service {
NotifyAccess notify_access;
NotifyState notify_state;
ServiceFDStore *fd_store;
unsigned n_fd_store;
unsigned n_fd_store_max;
};
extern const UnitVTable service_vtable;

25
src/core/unit.c
View File

@ -278,21 +278,32 @@ int unit_set_description(Unit *u, const char *description) {
}
bool unit_check_gc(Unit *u) {
UnitActiveState state;
assert(u);
if (UNIT_VTABLE(u)->no_gc)
return true;
if (u->no_gc)
return true;
if (u->job)
return true;
if (u->nop_job)
return true;
if (unit_active_state(u) != UNIT_INACTIVE)
state = unit_active_state(u);
/* If the unit is inactive and failed and no job is queued for
* it, then release its runtime resources */
if (UNIT_IS_INACTIVE_OR_FAILED(state) &&
UNIT_VTABLE(u)->release_resources)
UNIT_VTABLE(u)->release_resources(u);
/* But we keep the unit object around for longer when it is
* referenced or configured to not be gc'ed */
if (state != UNIT_INACTIVE)
return true;
if (UNIT_VTABLE(u)->no_gc)
return true;
if (u->no_gc)
return true;
if (u->refs)

6
src/core/unit.h
View File

@ -345,6 +345,10 @@ struct UnitVTable {
* way */
bool (*check_gc)(Unit *u);
/* When the unit is not running and no job for it queued we
* shall release its runtime resources */
void (*release_resources)(Unit *u);
/* Return true when this unit is suitable for snapshotting */
bool (*check_snapshot)(Unit *u);
@ -359,7 +363,7 @@ struct UnitVTable {
void (*notify_cgroup_empty)(Unit *u);
/* Called whenever a process of this unit sends us a message */
void (*notify_message)(Unit *u, pid_t pid, char **tags);
void (*notify_message)(Unit *u, pid_t pid, char **tags, FDSet *fds);
/* Called whenever a name this Unit registered for comes or
* goes away. */

5
src/libsystemd/libsystemd.sym.m4
View File

@ -158,6 +158,11 @@ global:
sd_session_get_desktop;
} LIBSYSTEMD_216;
LIBSYSTEMD_219 {
global:
sd_pid_notify_with_fds;
} LIBSYSTEMD_217;
m4_ifdef(`ENABLE_KDBUS',
LIBSYSTEMD_FUTURE {
global:

98
src/libsystemd/sd-daemon/sd-daemon.c
View File

@ -340,16 +340,28 @@ _public_ int sd_is_mq(int fd, const char *path) {
return 1;
}
_public_ int sd_pid_notify(pid_t pid, int unset_environment, const char *state) {
union sockaddr_union sockaddr = {};
_cleanup_close_ int fd = -1;
struct msghdr msghdr = {};
struct iovec iovec = {};
const char *e;
_public_ int sd_pid_notify_with_fds(pid_t pid, int unset_environment, const char *state, const int *fds, unsigned n_fds) {
union sockaddr_union sockaddr = {
.sa.sa_family = AF_UNIX,
};
struct iovec iovec = {
.iov_base = (char*) state,
};
struct msghdr msghdr = {
.msg_iov = &iovec,
.msg_iovlen = 1,
.msg_name = &sockaddr,
};
union {
struct cmsghdr cmsghdr;
uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
} control = {};
uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
CMSG_SPACE(sizeof(int) * n_fds)];
} control;
_cleanup_close_ int fd = -1;
struct cmsghdr *cmsg = NULL;
const char *e;
size_t controllen_without_ucred = 0;
bool try_without_ucred = false;
int r;
if (!state) {
@ -357,6 +369,11 @@ _public_ int sd_pid_notify(pid_t pid, int unset_environment, const char *state)
goto finish;
}
if (n_fds > 0 && !fds) {
r = -EINVAL;
goto finish;
}
e = getenv("NOTIFY_SOCKET");
if (!e)
return 0;
@ -373,42 +390,50 @@ _public_ int sd_pid_notify(pid_t pid, int unset_environment, const char *state)
goto finish;
}
sockaddr.sa.sa_family = AF_UNIX;
strncpy(sockaddr.un.sun_path, e, sizeof(sockaddr.un.sun_path));
iovec.iov_len = strlen(state);
strncpy(sockaddr.un.sun_path, e, sizeof(sockaddr.un.sun_path));
if (sockaddr.un.sun_path[0] == '@')
sockaddr.un.sun_path[0] = 0;
iovec.iov_base = (char*) state;
iovec.iov_len = strlen(state);
msghdr.msg_name = &sockaddr;
msghdr.msg_namelen = offsetof(struct sockaddr_un, sun_path) + strlen(e);
if (msghdr.msg_namelen > sizeof(struct sockaddr_un))
msghdr.msg_namelen = sizeof(struct sockaddr_un);
msghdr.msg_iov = &iovec;
msghdr.msg_iovlen = 1;
if (pid != 0 && pid != getpid()) {
struct cmsghdr *cmsg;
struct ucred ucred = {};
if (n_fds > 0) {
msghdr.msg_control = &control;
msghdr.msg_controllen = sizeof(control);
msghdr.msg_controllen = CMSG_LEN(sizeof(int) * n_fds);
cmsg = CMSG_FIRSTHDR(&msghdr);
cmsg->cmsg_level = SOL_SOCKET;
cmsg->cmsg_type = SCM_RIGHTS;
cmsg->cmsg_len = CMSG_LEN(sizeof(int) * n_fds);
memcpy(CMSG_DATA(cmsg), fds, sizeof(int) * n_fds);
}
if (pid != 0 && pid != getpid()) {
struct ucred *ucred;
try_without_ucred = true;
controllen_without_ucred = msghdr.msg_controllen;
msghdr.msg_control = &control;
msghdr.msg_controllen += CMSG_LEN(sizeof(struct ucred));
if (cmsg)
cmsg = CMSG_NXTHDR(&msghdr, cmsg);
else
cmsg = CMSG_FIRSTHDR(&msghdr);
cmsg->cmsg_level = SOL_SOCKET;
cmsg->cmsg_type = SCM_CREDENTIALS;
cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
ucred.pid = pid;
ucred.uid = getuid();
ucred.gid = getgid();
memcpy(CMSG_DATA(cmsg), &ucred, sizeof(struct ucred));
msghdr.msg_controllen = cmsg->cmsg_len;
ucred = (struct ucred*) CMSG_DATA(cmsg);
ucred->pid = pid;
ucred->uid = getuid();
ucred->gid = getgid();
}
/* First try with fake ucred data, as requested */
@ -417,10 +442,11 @@ _public_ int sd_pid_notify(pid_t pid, int unset_environment, const char *state)
goto finish;
}
/* If that failed, try with our own instead */
if (msghdr.msg_control) {
msghdr.msg_control = NULL;
msghdr.msg_controllen = 0;
/* If that failed, try with our own ucred instead */
if (try_without_ucred) {
if (controllen_without_ucred <= 0)
msghdr.msg_control = NULL;
msghdr.msg_controllen = controllen_without_ucred;
if (sendmsg(fd, &msghdr, MSG_NOSIGNAL) >= 0) {
r = 1;
@ -437,8 +463,12 @@ finish:
return r;
}
_public_ int sd_pid_notify(pid_t pid, int unset_environment, const char *state) {
return sd_pid_notify_with_fds(pid, unset_environment, state, NULL, 0);
}
_public_ int sd_notify(int unset_environment, const char *state) {
return sd_pid_notify(0, unset_environment, state);
return sd_pid_notify_with_fds(0, unset_environment, state, NULL, 0);
}
_public_ int sd_pid_notifyf(pid_t pid, int unset_environment, const char *format, ...) {

54
src/shared/fdset.c
View File

@ -41,7 +41,31 @@ FDSet *fdset_new(void) {
return MAKE_FDSET(set_new(NULL));
}
void fdset_free(FDSet *s) {
int fdset_new_array(FDSet **ret, int *fds, unsigned n_fds) {
unsigned i;
FDSet *s;
int r;
assert(ret);
s = fdset_new();
if (!s)
return -ENOMEM;
for (i = 0; i < n_fds; i++) {
r = fdset_put(s, fds[i]);
if (r < 0) {
set_free(MAKE_SET(s));
return r;
}
}
*ret = s;
return 0;
}
FDSet* fdset_free(FDSet *s) {
void *p;
while ((p = set_steal_first(MAKE_SET(s)))) {
@ -61,6 +85,7 @@ void fdset_free(FDSet *s) {
}
set_free(MAKE_SET(s));
return NULL;
}
int fdset_put(FDSet *s, int fd) {
@ -70,6 +95,19 @@ int fdset_put(FDSet *s, int fd) {
return set_put(MAKE_SET(s), FD_TO_PTR(fd));
}
int fdset_consume(FDSet *s, int fd) {
int r;
assert(s);
assert(fd >= 0);
r = fdset_put(s, fd);
if (r <= 0)
safe_close(fd);
return r;
}
int fdset_put_dup(FDSet *s, int fd) {
int copy, r;
@ -223,6 +261,10 @@ unsigned fdset_size(FDSet *fds) {
return set_size(MAKE_SET(fds));
}
bool fdset_isempty(FDSet *fds) {
return set_isempty(MAKE_SET(fds));
}
int fdset_iterate(FDSet *s, Iterator *i) {
void *p;
@ -232,3 +274,13 @@ int fdset_iterate(FDSet *s, Iterator *i) {
return PTR_TO_FD(p);
}
int fdset_steal_first(FDSet *fds) {
void *p;
p = set_steal_first(MAKE_SET(fds));
if (!p)
return -ENOENT;
return PTR_TO_FD(p);
}

11
src/shared/fdset.h
View File

@ -27,25 +27,30 @@
typedef struct FDSet FDSet;
FDSet* fdset_new(void);
void fdset_free(FDSet *s);
FDSet* fdset_free(FDSet *s);
int fdset_put(FDSet *s, int fd);
int fdset_put_dup(FDSet *s, int fd);
int fdset_consume(FDSet *s, int fd);
bool fdset_contains(FDSet *s, int fd);
int fdset_remove(FDSet *s, int fd);
int fdset_new_fill(FDSet **_s);
int fdset_new_listen_fds(FDSet **_s, bool unset);
int fdset_new_array(FDSet **ret, int *fds, unsigned n_fds);
int fdset_new_fill(FDSet **ret);
int fdset_new_listen_fds(FDSet **ret, bool unset);
int fdset_cloexec(FDSet *fds, bool b);
int fdset_close_others(FDSet *fds);
unsigned fdset_size(FDSet *fds);
bool fdset_isempty(FDSet *fds);
int fdset_iterate(FDSet *s, Iterator *i);
int fdset_steal_first(FDSet *fds);
#define FDSET_FOREACH(fd, fds, i) \
for ((i) = ITERATOR_FIRST, (fd) = fdset_iterate((fds), &(i)); (fd) >= 0; (fd) = fdset_iterate((fds), &(i)))

25
src/shared/util.c
View File

@ -7672,3 +7672,28 @@ int fd_setcrtime(int fd, usec_t usec) {
return 0;
}
int same_fd(int a, int b) {
struct stat sta, stb;
assert(a >= 0);
assert(b >= 0);
if (a == b)
return true;
if (fstat(a, &sta) < 0)
return -errno;
if (fstat(b, &stb) < 0)
return -errno;
if ((sta.st_mode & S_IFMT) != (stb.st_mode & S_IFMT))
return false;
if (S_ISREG(sta.st_mode) || S_ISDIR(sta.st_mode) || S_ISFIFO(sta.st_mode) || S_ISSOCK(sta.st_mode) || S_ISLNK(sta.st_mode))
return (sta.st_dev == stb.st_dev) && (sta.st_ino == stb.st_ino);
/* We consider all device fds different... */
return false;
}

2
src/shared/util.h
View File

@ -1069,3 +1069,5 @@ int fd_setcrtime(int fd, usec_t usec);
int fd_getcrtime(int fd, usec_t *usec);
int path_getcrtime(const char *p, usec_t *usec);
int fd_getcrtime_at(int dirfd, const char *name, usec_t *usec, int flags);
int same_fd(int a, int b);

13
src/systemd/sd-daemon.h
View File

@ -190,6 +190,12 @@ int sd_is_mq(int fd, const char *path);
timestamps to detect failed services. Also see
sd_watchdog_enabled() below.
FDSTORE=1 Store the file descriptors passed along with the
message in the per-service file descriptor store,
and pass them to the main process again on next
invocation. This variable is only supported with
sd_pid_notify_with_fds().
Daemons can choose to send additional variables. However, it is
recommended to prefix variable names not listed above with X_.
@ -242,6 +248,13 @@ int sd_pid_notify(pid_t pid, int unset_environment, const char *state);
*/
int sd_pid_notifyf(pid_t pid, int unset_environment, const char *format, ...) _sd_printf_(3,4);
/*
Similar to sd_pid_notify(), but also passes the specified fd array
to the service manager for storage. This is particularly useful for
FDSTORE=1 messages.
*/
int sd_pid_notify_with_fds(pid_t pid, int unset_environment, const char *state, const int *fds, unsigned n_fds);
/*
Returns > 0 if the system was booted with systemd. Returns < 0 on
error. Returns 0 if the system was not booted with systemd. Note