One of the downsides of epoll is that if you have a hung epoll-based
program, you can't track down which fds it's waiting for; those fds are
hidden inside the epoll strutures in the kernel and you can't see them.
At all.
This patch exposes this information in /proc/<pid>/fdinfo/<fd>, so lsof can
show it, or coders can just cat it during debugging.
It has a few stylistic issues and the minor detail of not having been
tested; but other than that, what do you think? All those event-driven
programs out there need debugging. If lsof is good, so is this...
Signed-Off-By: Chip Salzenberg <chip@xxxxxxxxx>
fs/eventpoll.c | 46 ++++++++++++++++++++++++++++++++-------------
fs/proc/fd.c | 7 ++++++-
include/linux/eventpoll.h | 19 +++++++++++++++++++
3 files changed, 58 insertions(+), 14 deletions(-)
diff --git a/fs/eventpoll.c b/fs/eventpoll.c
index cd96649..9212503 100644
--- a/fs/eventpoll.c
+++ b/fs/eventpoll.c
@@ -34,6 +34,7 @@
#include <linux/mutex.h>
#include <linux/anon_inodes.h>
#include <linux/device.h>
+#include <linux/seq_file.h>
#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/mman.h>
@@ -297,13 +298,6 @@ ctl_table epoll_table[] = {
};
#endif /* CONFIG_SYSCTL */
-static const struct file_operations eventpoll_fops;
-
-static inline int is_file_epoll(struct file *f)
-{
- return f->f_op == &eventpoll_fops;
-}
-
/* Setup the structure that is used as key for the RB tree */
static inline void ep_set_ffd(struct epoll_filefd *ffd,
struct file *file, int fd)
@@ -784,7 +778,7 @@ static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
}
/* File callbacks that implement the eventpoll file behaviour */
-static const struct file_operations eventpoll_fops = {
+const struct file_operations eventpoll_fops = {
.release = ep_eventpoll_release,
.poll = ep_eventpoll_poll,
.llseek = noop_llseek,
@@ -1065,7 +1059,7 @@ static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
list_for_each_entry(epi, &file->f_ep_links, fllink) {
child_file = epi->ep->file;
- if (is_file_epoll(child_file)) {
+ if (is_file_eventpoll(child_file)) {
if (list_empty(&child_file->f_ep_links)) {
if (path_count_inc(call_nests)) {
error = -1;
@@ -1551,7 +1545,7 @@ static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
list_add(&ep->visited_list_link, &visited_list);
for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
epi = rb_entry(rbp, struct epitem, rbn);
- if (unlikely(is_file_epoll(epi->ffd.file))) {
+ if (unlikely(is_file_eventpoll(epi->ffd.file))) {
ep_tovisit = epi->ffd.file->private_data;
if (ep_tovisit->visited)
continue;
@@ -1719,7 +1713,7 @@ SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
* adding an epoll file descriptor inside itself.
*/
error = -EINVAL;
- if (file == tfile || !is_file_epoll(file))
+ if (file == tfile || !is_file_eventpoll(file))
goto error_tgt_fput;
/*
@@ -1745,7 +1739,7 @@ SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
did_lock_epmutex = 1;
}
if (op == EPOLL_CTL_ADD) {
- if (is_file_epoll(tfile)) {
+ if (is_file_eventpoll(tfile)) {
error = -ELOOP;
if (ep_loop_check(ep, tfile) != 0) {
clear_tfile_check_list();
@@ -1831,7 +1825,7 @@ SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
* the user passed to us _is_ an eventpoll file.
*/
error = -EINVAL;
- if (!is_file_epoll(f.file))
+ if (!is_file_eventpoll(f.file))
goto error_fput;
/*
@@ -1892,6 +1886,32 @@ SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
return error;
}
+#ifdef CONFIG_PROC_FS
+
+/* return a printable representation of the eventpoll state for /proc/#/fdinfo/# */
+void eventpoll_proc_fdinfo(struct seq_file *m, struct file *file) {
+ struct eventpoll *ep;
+ struct epitem *epi = NULL;
+ struct rb_node *rbp;
+
+ BUG_ON(!is_file_eventpoll(file));
+ ep = file->private_data;
+ mutex_lock_nested(&ep->mtx, 0);
+
+ for (rbp = ep->rbr.rb_node; rbp; ) {
+ epi = rb_entry(rbp, struct epitem, rbn);
+ if (rbp == ep->rbr.rb_node)
+ seq_puts(m, "eventpoll:");
+ seq_printf(m, " %d:%x:%llx", epi->ffd.fd, epi->event.events, epi->event.data);
+ }
+ if (epi)
+ seq_putc(m, '\n');
+
+ mutex_unlock(&ep->mtx);
+}
+
+#endif /* CONFIG_PROC_FS */
+
static int __init eventpoll_init(void)
{
struct sysinfo si;
diff --git a/fs/proc/fd.c b/fs/proc/fd.c
index f28a875..ba2d6dd 100644
--- a/fs/proc/fd.c
+++ b/fs/proc/fd.c
@@ -8,6 +8,9 @@
#include <linux/security.h>
#include <linux/file.h>
#include <linux/seq_file.h>
+#ifdef CONFIG_EPOLL
+#include <linux/eventpoll.h>
+#endif
#include <linux/proc_fs.h>
@@ -48,8 +51,10 @@ static int seq_show(struct seq_file *m, void *v)
}
if (!ret) {
- seq_printf(m, "pos:\t%lli\nflags:\t0%o\n",
+ seq_printf(m, "pos:\t%lli\nflags:\t0%o\n",
(long long)file->f_pos, f_flags);
+ if (unlikely(is_file_eventpoll(file)))
+ eventpoll_proc_fdinfo(m, file);
fput(file);
}
diff --git a/include/linux/eventpoll.h b/include/linux/eventpoll.h
index 6daf6d4..b179ab4 100644
--- a/include/linux/eventpoll.h
+++ b/include/linux/eventpoll.h
@@ -18,10 +18,13 @@
/* Forward declarations to avoid compiler errors */
struct file;
+struct seq_file;
#ifdef CONFIG_EPOLL
+#include <linux/fs.h>
+
/* Used to initialize the epoll bits inside the "struct file" */
static inline void eventpoll_init_file(struct file *file)
{
@@ -61,11 +64,27 @@ static inline void eventpoll_release(struct file *file)
eventpoll_release_file(file);
}
+extern const struct file_operations eventpoll_fops;
+
+static inline bool is_file_eventpoll(struct file *f)
+{
+ return f->f_op == &eventpoll_fops;
+}
+
#else
static inline void eventpoll_init_file(struct file *file) {}
static inline void eventpoll_release(struct file *file) {}
+static inline bool is_file_eventpoll(struct file *f) { return false; }
#endif
+
+#if defined(CONFIG_EPOLL) && defined(CONFIG_PROC_FS)
+void eventpoll_proc_fdinfo(struct seq_file *m, struct file *file);
+#else
+static inline void eventpoll_proc_fdinfo(struct seq_file *m, struct file *file) {}
+#endif
+
+
#endif /* #ifndef _LINUX_EVENTPOLL_H */
--
Chip Salzenberg
Signed-Off-By: Chip Salzenberg <chip@xxxxxxxxx>
fs/eventpoll.c | 46 ++++++++++++++++++++++++++++++++-------------
fs/proc/fd.c | 7 ++++++-
include/linux/eventpoll.h | 19 +++++++++++++++++++
3 files changed, 58 insertions(+), 14 deletions(-)
diff --git a/fs/eventpoll.c b/fs/eventpoll.c
index cd96649..9212503 100644
--- a/fs/eventpoll.c
+++ b/fs/eventpoll.c
@@ -34,6 +34,7 @@
#include <linux/mutex.h>
#include <linux/anon_inodes.h>
#include <linux/device.h>
+#include <linux/seq_file.h>
#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/mman.h>
@@ -297,13 +298,6 @@ ctl_table epoll_table[] = {
};
#endif /* CONFIG_SYSCTL */
-static const struct file_operations eventpoll_fops;
-
-static inline int is_file_epoll(struct file *f)
-{
- return f->f_op == &eventpoll_fops;
-}
-
/* Setup the structure that is used as key for the RB tree */
static inline void ep_set_ffd(struct epoll_filefd *ffd,
struct file *file, int fd)
@@ -784,7 +778,7 @@ static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait)
}
/* File callbacks that implement the eventpoll file behaviour */
-static const struct file_operations eventpoll_fops = {
+const struct file_operations eventpoll_fops = {
.release = ep_eventpoll_release,
.poll = ep_eventpoll_poll,
.llseek = noop_llseek,
@@ -1065,7 +1059,7 @@ static int reverse_path_check_proc(void *priv, void *cookie, int call_nests)
list_for_each_entry(epi, &file->f_ep_links, fllink) {
child_file = epi->ep->file;
- if (is_file_epoll(child_file)) {
+ if (is_file_eventpoll(child_file)) {
if (list_empty(&child_file->f_ep_links)) {
if (path_count_inc(call_nests)) {
error = -1;
@@ -1551,7 +1545,7 @@ static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
list_add(&ep->visited_list_link, &visited_list);
for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) {
epi = rb_entry(rbp, struct epitem, rbn);
- if (unlikely(is_file_epoll(epi->ffd.file))) {
+ if (unlikely(is_file_eventpoll(epi->ffd.file))) {
ep_tovisit = epi->ffd.file->private_data;
if (ep_tovisit->visited)
continue;
@@ -1719,7 +1713,7 @@ SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
* adding an epoll file descriptor inside itself.
*/
error = -EINVAL;
- if (file == tfile || !is_file_epoll(file))
+ if (file == tfile || !is_file_eventpoll(file))
goto error_tgt_fput;
/*
@@ -1745,7 +1739,7 @@ SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd,
did_lock_epmutex = 1;
}
if (op == EPOLL_CTL_ADD) {
- if (is_file_epoll(tfile)) {
+ if (is_file_eventpoll(tfile)) {
error = -ELOOP;
if (ep_loop_check(ep, tfile) != 0) {
clear_tfile_check_list();
@@ -1831,7 +1825,7 @@ SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events,
* the user passed to us _is_ an eventpoll file.
*/
error = -EINVAL;
- if (!is_file_epoll(f.file))
+ if (!is_file_eventpoll(f.file))
goto error_fput;
/*
@@ -1892,6 +1886,32 @@ SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events,
return error;
}
+#ifdef CONFIG_PROC_FS
+
+/* return a printable representation of the eventpoll state for /proc/#/fdinfo/# */
+void eventpoll_proc_fdinfo(struct seq_file *m, struct file *file) {
+ struct eventpoll *ep;
+ struct epitem *epi = NULL;
+ struct rb_node *rbp;
+
+ BUG_ON(!is_file_eventpoll(file));
+ ep = file->private_data;
+ mutex_lock_nested(&ep->mtx, 0);
+
+ for (rbp = ep->rbr.rb_node; rbp; ) {
+ epi = rb_entry(rbp, struct epitem, rbn);
+ if (rbp == ep->rbr.rb_node)
+ seq_puts(m, "eventpoll:");
+ seq_printf(m, " %d:%x:%llx", epi->ffd.fd, epi->event.events, epi->event.data);
+ }
+ if (epi)
+ seq_putc(m, '\n');
+
+ mutex_unlock(&ep->mtx);
+}
+
+#endif /* CONFIG_PROC_FS */
+
static int __init eventpoll_init(void)
{
struct sysinfo si;
diff --git a/fs/proc/fd.c b/fs/proc/fd.c
index f28a875..ba2d6dd 100644
--- a/fs/proc/fd.c
+++ b/fs/proc/fd.c
@@ -8,6 +8,9 @@
#include <linux/security.h>
#include <linux/file.h>
#include <linux/seq_file.h>
+#ifdef CONFIG_EPOLL
+#include <linux/eventpoll.h>
+#endif
#include <linux/proc_fs.h>
@@ -48,8 +51,10 @@ static int seq_show(struct seq_file *m, void *v)
}
if (!ret) {
- seq_printf(m, "pos:\t%lli\nflags:\t0%o\n",
+ seq_printf(m, "pos:\t%lli\nflags:\t0%o\n",
(long long)file->f_pos, f_flags);
+ if (unlikely(is_file_eventpoll(file)))
+ eventpoll_proc_fdinfo(m, file);
fput(file);
}
diff --git a/include/linux/eventpoll.h b/include/linux/eventpoll.h
index 6daf6d4..b179ab4 100644
--- a/include/linux/eventpoll.h
+++ b/include/linux/eventpoll.h
@@ -18,10 +18,13 @@
/* Forward declarations to avoid compiler errors */
struct file;
+struct seq_file;
#ifdef CONFIG_EPOLL
+#include <linux/fs.h>
+
/* Used to initialize the epoll bits inside the "struct file" */
static inline void eventpoll_init_file(struct file *file)
{
@@ -61,11 +64,27 @@ static inline void eventpoll_release(struct file *file)
eventpoll_release_file(file);
}
+extern const struct file_operations eventpoll_fops;
+
+static inline bool is_file_eventpoll(struct file *f)
+{
+ return f->f_op == &eventpoll_fops;
+}
+
#else
static inline void eventpoll_init_file(struct file *file) {}
static inline void eventpoll_release(struct file *file) {}
+static inline bool is_file_eventpoll(struct file *f) { return false; }
#endif
+
+#if defined(CONFIG_EPOLL) && defined(CONFIG_PROC_FS)
+void eventpoll_proc_fdinfo(struct seq_file *m, struct file *file);
+#else
+static inline void eventpoll_proc_fdinfo(struct seq_file *m, struct file *file) {}
+#endif
+
+
#endif /* #ifndef _LINUX_EVENTPOLL_H */
