aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--fs/eventpoll.c48
1 files changed, 27 insertions, 21 deletions
diff --git a/fs/eventpoll.c b/fs/eventpoll.c
index 4c16127c96be..d4255be29bc6 100644
--- a/fs/eventpoll.c
+++ b/fs/eventpoll.c
@@ -134,7 +134,7 @@ struct poll_safewake {
134 * have an entry of this type linked to the "rbr" RB tree. 134 * have an entry of this type linked to the "rbr" RB tree.
135 */ 135 */
136struct epitem { 136struct epitem {
137 /* RB-Tree node used to link this structure to the eventpoll rb-tree */ 137 /* RB tree node used to link this structure to the eventpoll RB tree */
138 struct rb_node rbn; 138 struct rb_node rbn;
139 139
140 /* List header used to link this structure to the eventpoll ready list */ 140 /* List header used to link this structure to the eventpoll ready list */
@@ -191,7 +191,7 @@ struct eventpoll {
191 /* List of ready file descriptors */ 191 /* List of ready file descriptors */
192 struct list_head rdllist; 192 struct list_head rdllist;
193 193
194 /* RB-Tree root used to store monitored fd structs */ 194 /* RB tree root used to store monitored fd structs */
195 struct rb_root rbr; 195 struct rb_root rbr;
196 196
197 /* 197 /*
@@ -241,7 +241,7 @@ static struct kmem_cache *epi_cache __read_mostly;
241static struct kmem_cache *pwq_cache __read_mostly; 241static struct kmem_cache *pwq_cache __read_mostly;
242 242
243 243
244/* Setup the structure that is used as key for the rb-tree */ 244/* Setup the structure that is used as key for the RB tree */
245static inline void ep_set_ffd(struct epoll_filefd *ffd, 245static inline void ep_set_ffd(struct epoll_filefd *ffd,
246 struct file *file, int fd) 246 struct file *file, int fd)
247{ 247{
@@ -249,7 +249,7 @@ static inline void ep_set_ffd(struct epoll_filefd *ffd,
249 ffd->fd = fd; 249 ffd->fd = fd;
250} 250}
251 251
252/* Compare rb-tree keys */ 252/* Compare RB tree keys */
253static inline int ep_cmp_ffd(struct epoll_filefd *p1, 253static inline int ep_cmp_ffd(struct epoll_filefd *p1,
254 struct epoll_filefd *p2) 254 struct epoll_filefd *p2)
255{ 255{
@@ -257,20 +257,20 @@ static inline int ep_cmp_ffd(struct epoll_filefd *p1,
257 (p1->file < p2->file ? -1 : p1->fd - p2->fd)); 257 (p1->file < p2->file ? -1 : p1->fd - p2->fd));
258} 258}
259 259
260/* Special initialization for the rb-tree node to detect linkage */ 260/* Special initialization for the RB tree node to detect linkage */
261static inline void ep_rb_initnode(struct rb_node *n) 261static inline void ep_rb_initnode(struct rb_node *n)
262{ 262{
263 rb_set_parent(n, n); 263 rb_set_parent(n, n);
264} 264}
265 265
266/* Removes a node from the rb-tree and marks it for a fast is-linked check */ 266/* Removes a node from the RB tree and marks it for a fast is-linked check */
267static inline void ep_rb_erase(struct rb_node *n, struct rb_root *r) 267static inline void ep_rb_erase(struct rb_node *n, struct rb_root *r)
268{ 268{
269 rb_erase(n, r); 269 rb_erase(n, r);
270 rb_set_parent(n, n); 270 rb_set_parent(n, n);
271} 271}
272 272
273/* Fast check to verify that the item is linked to the main rb-tree */ 273/* Fast check to verify that the item is linked to the main RB tree */
274static inline int ep_rb_linked(struct rb_node *n) 274static inline int ep_rb_linked(struct rb_node *n)
275{ 275{
276 return rb_parent(n) != n; 276 return rb_parent(n) != n;
@@ -531,6 +531,8 @@ void eventpoll_release_file(struct file *file)
531 * We don't want to get "file->f_ep_lock" because it is not 531 * We don't want to get "file->f_ep_lock" because it is not
532 * necessary. It is not necessary because we're in the "struct file" 532 * necessary. It is not necessary because we're in the "struct file"
533 * cleanup path, and this means that noone is using this file anymore. 533 * cleanup path, and this means that noone is using this file anymore.
534 * So, for example, epoll_ctl() cannot hit here sicne if we reach this
535 * point, the file counter already went to zero and fget() would fail.
534 * The only hit might come from ep_free() but by holding the mutex 536 * The only hit might come from ep_free() but by holding the mutex
535 * will correctly serialize the operation. We do need to acquire 537 * will correctly serialize the operation. We do need to acquire
536 * "ep->mtx" after "epmutex" because ep_remove() requires it when called 538 * "ep->mtx" after "epmutex" because ep_remove() requires it when called
@@ -802,7 +804,9 @@ error_unregister:
802 804
803 /* 805 /*
804 * We need to do this because an event could have been arrived on some 806 * We need to do this because an event could have been arrived on some
805 * allocated wait queue. 807 * allocated wait queue. Note that we don't care about the ep->ovflist
808 * list, since that is used/cleaned only inside a section bound by "mtx".
809 * And ep_insert() is called with "mtx" held.
806 */ 810 */
807 spin_lock_irqsave(&ep->lock, flags); 811 spin_lock_irqsave(&ep->lock, flags);
808 if (ep_is_linked(&epi->rdllink)) 812 if (ep_is_linked(&epi->rdllink))
@@ -845,8 +849,7 @@ static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_even
845 849
846 /* 850 /*
847 * If the item is "hot" and it is not registered inside the ready 851 * If the item is "hot" and it is not registered inside the ready
848 * list, push it inside. If the item is not "hot" and it is currently 852 * list, push it inside.
849 * registered inside the ready list, unlink it.
850 */ 853 */
851 if (revents & event->events) { 854 if (revents & event->events) {
852 if (!ep_is_linked(&epi->rdllink)) { 855 if (!ep_is_linked(&epi->rdllink)) {
@@ -966,15 +969,16 @@ errxit:
966 ep->ovflist = EP_UNACTIVE_PTR; 969 ep->ovflist = EP_UNACTIVE_PTR;
967 970
968 /* 971 /*
969 * In case of error in the event-send loop, we might still have items 972 * In case of error in the event-send loop, or in case the number of
970 * inside the "txlist". We need to splice them back inside ep->rdllist. 973 * ready events exceeds the userspace limit, we need to splice the
974 * "txlist" back inside ep->rdllist.
971 */ 975 */
972 list_splice(&txlist, &ep->rdllist); 976 list_splice(&txlist, &ep->rdllist);
973 977
974 if (!list_empty(&ep->rdllist)) { 978 if (!list_empty(&ep->rdllist)) {
975 /* 979 /*
976 * Wake up (if active) both the eventpoll wait list and the ->poll() 980 * Wake up (if active) both the eventpoll wait list and the ->poll()
977 * wait list. 981 * wait list (delayed after we release the lock).
978 */ 982 */
979 if (waitqueue_active(&ep->wq)) 983 if (waitqueue_active(&ep->wq))
980 __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE | 984 __wake_up_locked(&ep->wq, TASK_UNINTERRUPTIBLE |
@@ -1064,11 +1068,10 @@ retry:
1064} 1068}
1065 1069
1066/* 1070/*
1067 * It opens an eventpoll file descriptor by suggesting a storage of "size" 1071 * It opens an eventpoll file descriptor. The "size" parameter is there
1068 * file descriptors. The size parameter is just an hint about how to size 1072 * for historical reasons, when epoll was using an hash instead of an
1069 * data structures. It won't prevent the user to store more than "size" 1073 * RB tree. With the current implementation, the "size" parameter is ignored
1070 * file descriptors inside the epoll interface. It is the kernel part of 1074 * (besides sanity checks).
1071 * the userspace epoll_create(2).
1072 */ 1075 */
1073asmlinkage long sys_epoll_create(int size) 1076asmlinkage long sys_epoll_create(int size)
1074{ 1077{
@@ -1114,8 +1117,7 @@ error_return:
1114/* 1117/*
1115 * The following function implements the controller interface for 1118 * The following function implements the controller interface for
1116 * the eventpoll file that enables the insertion/removal/change of 1119 * the eventpoll file that enables the insertion/removal/change of
1117 * file descriptors inside the interest set. It represents 1120 * file descriptors inside the interest set.
1118 * the kernel part of the user space epoll_ctl(2).
1119 */ 1121 */
1120asmlinkage long sys_epoll_ctl(int epfd, int op, int fd, 1122asmlinkage long sys_epoll_ctl(int epfd, int op, int fd,
1121 struct epoll_event __user *event) 1123 struct epoll_event __user *event)
@@ -1167,7 +1169,11 @@ asmlinkage long sys_epoll_ctl(int epfd, int op, int fd,
1167 1169
1168 mutex_lock(&ep->mtx); 1170 mutex_lock(&ep->mtx);
1169 1171
1170 /* Try to lookup the file inside our RB tree */ 1172 /*
1173 * Try to lookup the file inside our RB tree, Since we grabbed "mtx"
1174 * above, we can be sure to be able to use the item looked up by
1175 * ep_find() till we release the mutex.
1176 */
1171 epi = ep_find(ep, tfile, fd); 1177 epi = ep_find(ep, tfile, fd);
1172 1178
1173 error = -EINVAL; 1179 error = -EINVAL;