Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 2 | * fs/eventpoll.c (Efficient event retrieval implementation) |
| 3 | * Copyright (C) 2001,...,2009 Davide Libenzi |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify |
| 6 | * it under the terms of the GNU General Public License as published by |
| 7 | * the Free Software Foundation; either version 2 of the License, or |
| 8 | * (at your option) any later version. |
| 9 | * |
| 10 | * Davide Libenzi <davidel@xmailserver.org> |
| 11 | * |
| 12 | */ |
| 13 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/init.h> |
| 15 | #include <linux/kernel.h> |
| 16 | #include <linux/sched.h> |
| 17 | #include <linux/fs.h> |
| 18 | #include <linux/file.h> |
| 19 | #include <linux/signal.h> |
| 20 | #include <linux/errno.h> |
| 21 | #include <linux/mm.h> |
| 22 | #include <linux/slab.h> |
| 23 | #include <linux/poll.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/string.h> |
| 25 | #include <linux/list.h> |
| 26 | #include <linux/hash.h> |
| 27 | #include <linux/spinlock.h> |
| 28 | #include <linux/syscalls.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/rbtree.h> |
| 30 | #include <linux/wait.h> |
| 31 | #include <linux/eventpoll.h> |
| 32 | #include <linux/mount.h> |
| 33 | #include <linux/bitops.h> |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 34 | #include <linux/mutex.h> |
Davide Libenzi | da66f7c | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 35 | #include <linux/anon_inodes.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | #include <asm/uaccess.h> |
| 37 | #include <asm/system.h> |
| 38 | #include <asm/io.h> |
| 39 | #include <asm/mman.h> |
| 40 | #include <asm/atomic.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | /* |
| 43 | * LOCKING: |
| 44 | * There are three level of locking required by epoll : |
| 45 | * |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 46 | * 1) epmutex (mutex) |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 47 | * 2) ep->mtx (mutex) |
| 48 | * 3) ep->lock (spinlock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | * |
| 50 | * The acquire order is the one listed above, from 1 to 3. |
| 51 | * We need a spinlock (ep->lock) because we manipulate objects |
| 52 | * from inside the poll callback, that might be triggered from |
| 53 | * a wake_up() that in turn might be called from IRQ context. |
| 54 | * So we can't sleep inside the poll callback and hence we need |
| 55 | * a spinlock. During the event transfer loop (from kernel to |
| 56 | * user space) we could end up sleeping due a copy_to_user(), so |
| 57 | * we need a lock that will allow us to sleep. This lock is a |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 58 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
| 59 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). |
| 60 | * Then we also need a global mutex to serialize eventpoll_release_file() |
| 61 | * and ep_free(). |
| 62 | * This mutex is acquired by ep_free() during the epoll file |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | * cleanup path and it is also acquired by eventpoll_release_file() |
| 64 | * if a file has been pushed inside an epoll set and it is then |
| 65 | * close()d without a previous call toepoll_ctl(EPOLL_CTL_DEL). |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 66 | * It is possible to drop the "ep->mtx" and to use the global |
| 67 | * mutex "epmutex" (together with "ep->lock") to have it working, |
| 68 | * but having "ep->mtx" will make the interface more scalable. |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 69 | * Events that require holding "epmutex" are very rare, while for |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 70 | * normal operations the epoll private "ep->mtx" will guarantee |
| 71 | * a better scalability. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | */ |
| 73 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | /* Epoll private bits inside the event mask */ |
| 75 | #define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET) |
| 76 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 77 | /* Maximum number of nesting allowed inside epoll sets */ |
| 78 | #define EP_MAX_NESTS 4 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | |
Davide Libenzi | e3306dd | 2005-09-27 21:45:33 -0700 | [diff] [blame] | 80 | /* Maximum msec timeout value storeable in a long int */ |
| 81 | #define EP_MAX_MSTIMEO min(1000ULL * MAX_SCHEDULE_TIMEOUT / HZ, (LONG_MAX - 999ULL) / HZ) |
| 82 | |
Davide Libenzi | b611967 | 2006-10-11 01:21:44 -0700 | [diff] [blame] | 83 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
| 84 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 85 | #define EP_UNACTIVE_PTR ((void *) -1L) |
| 86 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 87 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
| 88 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | struct epoll_filefd { |
| 90 | struct file *file; |
| 91 | int fd; |
| 92 | }; |
| 93 | |
| 94 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 95 | * Structure used to track possible nested calls, for too deep recursions |
| 96 | * and loop cycles. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 98 | struct nested_call_node { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | struct list_head llink; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 100 | void *cookie; |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame^] | 101 | int cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | }; |
| 103 | |
| 104 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 105 | * This structure is used as collector for nested calls, to check for |
| 106 | * maximum recursion dept and loop cycles. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 108 | struct nested_calls { |
| 109 | struct list_head tasks_call_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | spinlock_t lock; |
| 111 | }; |
| 112 | |
| 113 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 114 | * Each file descriptor added to the eventpoll interface will |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 115 | * have an entry of this type linked to the "rbr" RB tree. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | */ |
| 117 | struct epitem { |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 118 | /* RB tree node used to link this structure to the eventpoll RB tree */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | struct rb_node rbn; |
| 120 | |
| 121 | /* List header used to link this structure to the eventpoll ready list */ |
| 122 | struct list_head rdllink; |
| 123 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 124 | /* |
| 125 | * Works together "struct eventpoll"->ovflist in keeping the |
| 126 | * single linked chain of items. |
| 127 | */ |
| 128 | struct epitem *next; |
| 129 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | /* The file descriptor information this item refers to */ |
| 131 | struct epoll_filefd ffd; |
| 132 | |
| 133 | /* Number of active wait queue attached to poll operations */ |
| 134 | int nwait; |
| 135 | |
| 136 | /* List containing poll wait queues */ |
| 137 | struct list_head pwqlist; |
| 138 | |
| 139 | /* The "container" of this item */ |
| 140 | struct eventpoll *ep; |
| 141 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | /* List header used to link this item to the "struct file" items list */ |
| 143 | struct list_head fllink; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 144 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 145 | /* The structure that describe the interested events and the source fd */ |
| 146 | struct epoll_event event; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 147 | }; |
| 148 | |
| 149 | /* |
| 150 | * This structure is stored inside the "private_data" member of the file |
| 151 | * structure and rapresent the main data sructure for the eventpoll |
| 152 | * interface. |
| 153 | */ |
| 154 | struct eventpoll { |
| 155 | /* Protect the this structure access */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 156 | spinlock_t lock; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 157 | |
| 158 | /* |
| 159 | * This mutex is used to ensure that files are not removed |
| 160 | * while epoll is using them. This is held during the event |
| 161 | * collection loop, the file cleanup path, the epoll file exit |
| 162 | * code and the ctl operations. |
| 163 | */ |
| 164 | struct mutex mtx; |
| 165 | |
| 166 | /* Wait queue used by sys_epoll_wait() */ |
| 167 | wait_queue_head_t wq; |
| 168 | |
| 169 | /* Wait queue used by file->poll() */ |
| 170 | wait_queue_head_t poll_wait; |
| 171 | |
| 172 | /* List of ready file descriptors */ |
| 173 | struct list_head rdllist; |
| 174 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 175 | /* RB tree root used to store monitored fd structs */ |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 176 | struct rb_root rbr; |
| 177 | |
| 178 | /* |
| 179 | * This is a single linked list that chains all the "struct epitem" that |
| 180 | * happened while transfering ready events to userspace w/out |
| 181 | * holding ->lock. |
| 182 | */ |
| 183 | struct epitem *ovflist; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 184 | |
| 185 | /* The user that created the eventpoll descriptor */ |
| 186 | struct user_struct *user; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 187 | }; |
| 188 | |
| 189 | /* Wait structure used by the poll hooks */ |
| 190 | struct eppoll_entry { |
| 191 | /* List header used to link this structure to the "struct epitem" */ |
| 192 | struct list_head llink; |
| 193 | |
| 194 | /* The "base" pointer is set to the container "struct epitem" */ |
| 195 | void *base; |
| 196 | |
| 197 | /* |
| 198 | * Wait queue item that will be linked to the target file wait |
| 199 | * queue head. |
| 200 | */ |
| 201 | wait_queue_t wait; |
| 202 | |
| 203 | /* The wait queue head that linked the "wait" wait queue item */ |
| 204 | wait_queue_head_t *whead; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | }; |
| 206 | |
| 207 | /* Wrapper struct used by poll queueing */ |
| 208 | struct ep_pqueue { |
| 209 | poll_table pt; |
| 210 | struct epitem *epi; |
| 211 | }; |
| 212 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 213 | /* Used by the ep_send_events() function as callback private data */ |
| 214 | struct ep_send_events_data { |
| 215 | int maxevents; |
| 216 | struct epoll_event __user *events; |
| 217 | }; |
| 218 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | /* |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 220 | * Configuration options available inside /proc/sys/fs/epoll/ |
| 221 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 222 | /* Maximum number of epoll watched descriptors, per user */ |
| 223 | static int max_user_watches __read_mostly; |
| 224 | |
| 225 | /* |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 226 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 228 | static DEFINE_MUTEX(epmutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 229 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 230 | /* Used for safe wake up implementation */ |
| 231 | static struct nested_calls poll_safewake_ncalls; |
| 232 | |
| 233 | /* Used to call file's f_op->poll() under the nested calls boundaries */ |
| 234 | static struct nested_calls poll_readywalk_ncalls; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | |
| 236 | /* Slab cache used to allocate "struct epitem" */ |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 237 | static struct kmem_cache *epi_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | |
| 239 | /* Slab cache used to allocate "struct eppoll_entry" */ |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 240 | static struct kmem_cache *pwq_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 242 | #ifdef CONFIG_SYSCTL |
| 243 | |
| 244 | #include <linux/sysctl.h> |
| 245 | |
| 246 | static int zero; |
| 247 | |
| 248 | ctl_table epoll_table[] = { |
| 249 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 250 | .procname = "max_user_watches", |
| 251 | .data = &max_user_watches, |
| 252 | .maxlen = sizeof(int), |
| 253 | .mode = 0644, |
| 254 | .proc_handler = &proc_dointvec_minmax, |
| 255 | .extra1 = &zero, |
| 256 | }, |
| 257 | { .ctl_name = 0 } |
| 258 | }; |
| 259 | #endif /* CONFIG_SYSCTL */ |
| 260 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 261 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 262 | /* Setup the structure that is used as key for the RB tree */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 263 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
| 264 | struct file *file, int fd) |
| 265 | { |
| 266 | ffd->file = file; |
| 267 | ffd->fd = fd; |
| 268 | } |
| 269 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 270 | /* Compare RB tree keys */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 271 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
| 272 | struct epoll_filefd *p2) |
| 273 | { |
| 274 | return (p1->file > p2->file ? +1: |
| 275 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); |
| 276 | } |
| 277 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 278 | /* Tells us if the item is currently linked */ |
| 279 | static inline int ep_is_linked(struct list_head *p) |
| 280 | { |
| 281 | return !list_empty(p); |
| 282 | } |
| 283 | |
| 284 | /* Get the "struct epitem" from a wait queue pointer */ |
Davide Libenzi | cdac75e | 2008-04-29 00:58:34 -0700 | [diff] [blame] | 285 | static inline struct epitem *ep_item_from_wait(wait_queue_t *p) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 286 | { |
| 287 | return container_of(p, struct eppoll_entry, wait)->base; |
| 288 | } |
| 289 | |
| 290 | /* Get the "struct epitem" from an epoll queue wrapper */ |
Davide Libenzi | cdac75e | 2008-04-29 00:58:34 -0700 | [diff] [blame] | 291 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 292 | { |
| 293 | return container_of(p, struct ep_pqueue, pt)->epi; |
| 294 | } |
| 295 | |
| 296 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 297 | static inline int ep_op_has_event(int op) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 298 | { |
| 299 | return op != EPOLL_CTL_DEL; |
| 300 | } |
| 301 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 302 | /* Initialize the poll safe wake up structure */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 303 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 304 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 305 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
| 306 | spin_lock_init(&ncalls->lock); |
| 307 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 309 | /** |
| 310 | * ep_call_nested - Perform a bound (possibly) nested call, by checking |
| 311 | * that the recursion limit is not exceeded, and that |
| 312 | * the same nested call (by the meaning of same cookie) is |
| 313 | * no re-entered. |
| 314 | * |
| 315 | * @ncalls: Pointer to the nested_calls structure to be used for this call. |
| 316 | * @max_nests: Maximum number of allowed nesting calls. |
| 317 | * @nproc: Nested call core function pointer. |
| 318 | * @priv: Opaque data to be passed to the @nproc callback. |
| 319 | * @cookie: Cookie to be used to identify this nested call. |
| 320 | * |
| 321 | * Returns: Returns the code returned by the @nproc callback, or -1 if |
| 322 | * the maximum recursion limit has been exceeded. |
| 323 | */ |
| 324 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
| 325 | int (*nproc)(void *, void *, int), void *priv, |
| 326 | void *cookie) |
| 327 | { |
| 328 | int error, call_nests = 0; |
| 329 | unsigned long flags; |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame^] | 330 | int this_cpu = get_cpu(); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 331 | struct list_head *lsthead = &ncalls->tasks_call_list; |
| 332 | struct nested_call_node *tncur; |
| 333 | struct nested_call_node tnode; |
| 334 | |
| 335 | spin_lock_irqsave(&ncalls->lock, flags); |
| 336 | |
| 337 | /* |
| 338 | * Try to see if the current task is already inside this wakeup call. |
| 339 | * We use a list here, since the population inside this set is always |
| 340 | * very much limited. |
| 341 | */ |
| 342 | list_for_each_entry(tncur, lsthead, llink) { |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame^] | 343 | if (tncur->cpu == this_cpu && |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 344 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
| 345 | /* |
| 346 | * Ops ... loop detected or maximum nest level reached. |
| 347 | * We abort this wake by breaking the cycle itself. |
| 348 | */ |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame^] | 349 | error = -1; |
| 350 | goto out_unlock; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 351 | } |
| 352 | } |
| 353 | |
| 354 | /* Add the current task and cookie to the list */ |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame^] | 355 | tnode.cpu = this_cpu; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 356 | tnode.cookie = cookie; |
| 357 | list_add(&tnode.llink, lsthead); |
| 358 | |
| 359 | spin_unlock_irqrestore(&ncalls->lock, flags); |
| 360 | |
| 361 | /* Call the nested function */ |
| 362 | error = (*nproc)(priv, cookie, call_nests); |
| 363 | |
| 364 | /* Remove the current task from the list */ |
| 365 | spin_lock_irqsave(&ncalls->lock, flags); |
| 366 | list_del(&tnode.llink); |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame^] | 367 | out_unlock: |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 368 | spin_unlock_irqrestore(&ncalls->lock, flags); |
| 369 | |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame^] | 370 | put_cpu(); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 371 | return error; |
| 372 | } |
| 373 | |
| 374 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) |
| 375 | { |
| 376 | wake_up_nested((wait_queue_head_t *) cookie, 1 + call_nests); |
| 377 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | } |
| 379 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 380 | /* |
| 381 | * Perform a safe wake up of the poll wait list. The problem is that |
| 382 | * with the new callback'd wake up system, it is possible that the |
| 383 | * poll callback is reentered from inside the call to wake_up() done |
| 384 | * on the poll wait queue head. The rule is that we cannot reenter the |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 385 | * wake up code from the same task more than EP_MAX_NESTS times, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | * and we cannot reenter the same wait queue head at all. This will |
| 387 | * enable to have a hierarchy of epoll file descriptor of no more than |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 388 | * EP_MAX_NESTS deep. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 390 | static void ep_poll_safewake(wait_queue_head_t *wq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 392 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, |
| 393 | ep_poll_wakeup_proc, NULL, wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 394 | } |
| 395 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 396 | /* |
| 397 | * This function unregister poll callbacks from the associated file descriptor. |
| 398 | * Since this must be called without holding "ep->lock" the atomic exchange trick |
| 399 | * will protect us from multiple unregister. |
| 400 | */ |
| 401 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
| 402 | { |
| 403 | int nwait; |
| 404 | struct list_head *lsthead = &epi->pwqlist; |
| 405 | struct eppoll_entry *pwq; |
| 406 | |
| 407 | /* This is called without locks, so we need the atomic exchange */ |
| 408 | nwait = xchg(&epi->nwait, 0); |
| 409 | |
| 410 | if (nwait) { |
| 411 | while (!list_empty(lsthead)) { |
| 412 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); |
| 413 | |
| 414 | list_del_init(&pwq->llink); |
| 415 | remove_wait_queue(pwq->whead, &pwq->wait); |
| 416 | kmem_cache_free(pwq_cache, pwq); |
| 417 | } |
| 418 | } |
| 419 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 420 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 421 | /** |
| 422 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for |
| 423 | * the scan code, to call f_op->poll(). Also allows for |
| 424 | * O(NumReady) performance. |
| 425 | * |
| 426 | * @ep: Pointer to the epoll private data structure. |
| 427 | * @sproc: Pointer to the scan callback. |
| 428 | * @priv: Private opaque data passed to the @sproc callback. |
| 429 | * |
| 430 | * Returns: The same integer error code returned by the @sproc callback. |
| 431 | */ |
| 432 | static int ep_scan_ready_list(struct eventpoll *ep, |
| 433 | int (*sproc)(struct eventpoll *, |
| 434 | struct list_head *, void *), |
| 435 | void *priv) |
| 436 | { |
| 437 | int error, pwake = 0; |
| 438 | unsigned long flags; |
| 439 | struct epitem *epi, *nepi; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 440 | LIST_HEAD(txlist); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 441 | |
| 442 | /* |
| 443 | * We need to lock this because we could be hit by |
| 444 | * eventpoll_release_file() and epoll_ctl(EPOLL_CTL_DEL). |
| 445 | */ |
| 446 | mutex_lock(&ep->mtx); |
| 447 | |
| 448 | /* |
| 449 | * Steal the ready list, and re-init the original one to the |
| 450 | * empty list. Also, set ep->ovflist to NULL so that events |
| 451 | * happening while looping w/out locks, are not lost. We cannot |
| 452 | * have the poll callback to queue directly on ep->rdllist, |
| 453 | * because we want the "sproc" callback to be able to do it |
| 454 | * in a lockless way. |
| 455 | */ |
| 456 | spin_lock_irqsave(&ep->lock, flags); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 457 | list_splice_init(&ep->rdllist, &txlist); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 458 | ep->ovflist = NULL; |
| 459 | spin_unlock_irqrestore(&ep->lock, flags); |
| 460 | |
| 461 | /* |
| 462 | * Now call the callback function. |
| 463 | */ |
| 464 | error = (*sproc)(ep, &txlist, priv); |
| 465 | |
| 466 | spin_lock_irqsave(&ep->lock, flags); |
| 467 | /* |
| 468 | * During the time we spent inside the "sproc" callback, some |
| 469 | * other events might have been queued by the poll callback. |
| 470 | * We re-insert them inside the main ready-list here. |
| 471 | */ |
| 472 | for (nepi = ep->ovflist; (epi = nepi) != NULL; |
| 473 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
| 474 | /* |
| 475 | * We need to check if the item is already in the list. |
| 476 | * During the "sproc" callback execution time, items are |
| 477 | * queued into ->ovflist but the "txlist" might already |
| 478 | * contain them, and the list_splice() below takes care of them. |
| 479 | */ |
| 480 | if (!ep_is_linked(&epi->rdllink)) |
| 481 | list_add_tail(&epi->rdllink, &ep->rdllist); |
| 482 | } |
| 483 | /* |
| 484 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after |
| 485 | * releasing the lock, events will be queued in the normal way inside |
| 486 | * ep->rdllist. |
| 487 | */ |
| 488 | ep->ovflist = EP_UNACTIVE_PTR; |
| 489 | |
| 490 | /* |
| 491 | * Quickly re-inject items left on "txlist". |
| 492 | */ |
| 493 | list_splice(&txlist, &ep->rdllist); |
| 494 | |
| 495 | if (!list_empty(&ep->rdllist)) { |
| 496 | /* |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 497 | * Wake up (if active) both the eventpoll wait list and |
| 498 | * the ->poll() wait list (delayed after we release the lock). |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 499 | */ |
| 500 | if (waitqueue_active(&ep->wq)) |
| 501 | wake_up_locked(&ep->wq); |
| 502 | if (waitqueue_active(&ep->poll_wait)) |
| 503 | pwake++; |
| 504 | } |
| 505 | spin_unlock_irqrestore(&ep->lock, flags); |
| 506 | |
| 507 | mutex_unlock(&ep->mtx); |
| 508 | |
| 509 | /* We have to call this outside the lock */ |
| 510 | if (pwake) |
| 511 | ep_poll_safewake(&ep->poll_wait); |
| 512 | |
| 513 | return error; |
| 514 | } |
| 515 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 516 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 517 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 518 | * all the associated resources. Must be called with "mtx" held. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 519 | */ |
| 520 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) |
| 521 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 522 | unsigned long flags; |
| 523 | struct file *file = epi->ffd.file; |
| 524 | |
| 525 | /* |
| 526 | * Removes poll wait queue hooks. We _have_ to do this without holding |
| 527 | * the "ep->lock" otherwise a deadlock might occur. This because of the |
| 528 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait |
| 529 | * queue head lock when unregistering the wait queue. The wakeup callback |
| 530 | * will run by holding the wait queue head lock and will call our callback |
| 531 | * that will try to get "ep->lock". |
| 532 | */ |
| 533 | ep_unregister_pollwait(ep, epi); |
| 534 | |
| 535 | /* Remove the current item from the list of epoll hooks */ |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 536 | spin_lock(&file->f_lock); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 537 | if (ep_is_linked(&epi->fllink)) |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 538 | list_del_init(&epi->fllink); |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 539 | spin_unlock(&file->f_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | |
Davide Libenzi | cdac75e | 2008-04-29 00:58:34 -0700 | [diff] [blame] | 541 | rb_erase(&epi->rbn, &ep->rbr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 543 | spin_lock_irqsave(&ep->lock, flags); |
| 544 | if (ep_is_linked(&epi->rdllink)) |
| 545 | list_del_init(&epi->rdllink); |
| 546 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 548 | /* At this point it is safe to free the eventpoll item */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 549 | kmem_cache_free(epi_cache, epi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 551 | atomic_dec(&ep->user->epoll_watches); |
| 552 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 553 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | } |
| 555 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | static void ep_free(struct eventpoll *ep) |
| 557 | { |
| 558 | struct rb_node *rbp; |
| 559 | struct epitem *epi; |
| 560 | |
| 561 | /* We need to release all tasks waiting for these file */ |
| 562 | if (waitqueue_active(&ep->poll_wait)) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 563 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 564 | |
| 565 | /* |
| 566 | * We need to lock this because we could be hit by |
| 567 | * eventpoll_release_file() while we're freeing the "struct eventpoll". |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 568 | * We do not need to hold "ep->mtx" here because the epoll file |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | * is on the way to be removed and no one has references to it |
| 570 | * anymore. The only hit might come from eventpoll_release_file() but |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 571 | * holding "epmutex" is sufficent here. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 572 | */ |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 573 | mutex_lock(&epmutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | |
| 575 | /* |
| 576 | * Walks through the whole tree by unregistering poll callbacks. |
| 577 | */ |
| 578 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
| 579 | epi = rb_entry(rbp, struct epitem, rbn); |
| 580 | |
| 581 | ep_unregister_pollwait(ep, epi); |
| 582 | } |
| 583 | |
| 584 | /* |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 585 | * Walks through the whole tree by freeing each "struct epitem". At this |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 586 | * point we are sure no poll callbacks will be lingering around, and also by |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 587 | * holding "epmutex" we can be sure that no file cleanup code will hit |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 588 | * us during this operation. So we can avoid the lock on "ep->lock". |
| 589 | */ |
Stephen Hemminger | c80544d | 2007-10-18 03:07:05 -0700 | [diff] [blame] | 590 | while ((rbp = rb_first(&ep->rbr)) != NULL) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 591 | epi = rb_entry(rbp, struct epitem, rbn); |
| 592 | ep_remove(ep, epi); |
| 593 | } |
| 594 | |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 595 | mutex_unlock(&epmutex); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 596 | mutex_destroy(&ep->mtx); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 597 | free_uid(ep->user); |
Davide Libenzi | f0ee9aa | 2007-05-15 01:40:57 -0700 | [diff] [blame] | 598 | kfree(ep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 599 | } |
| 600 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 601 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
| 602 | { |
| 603 | struct eventpoll *ep = file->private_data; |
| 604 | |
Davide Libenzi | f0ee9aa | 2007-05-15 01:40:57 -0700 | [diff] [blame] | 605 | if (ep) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 606 | ep_free(ep); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 607 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 608 | return 0; |
| 609 | } |
| 610 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 611 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
| 612 | void *priv) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 613 | { |
| 614 | struct epitem *epi, *tmp; |
| 615 | |
| 616 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
| 617 | if (epi->ffd.file->f_op->poll(epi->ffd.file, NULL) & |
| 618 | epi->event.events) |
| 619 | return POLLIN | POLLRDNORM; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 620 | else { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 621 | /* |
| 622 | * Item has been dropped into the ready list by the poll |
| 623 | * callback, but it's not actually ready, as far as |
| 624 | * caller requested events goes. We can remove it here. |
| 625 | */ |
| 626 | list_del_init(&epi->rdllink); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 627 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 628 | } |
| 629 | |
| 630 | return 0; |
| 631 | } |
| 632 | |
| 633 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) |
| 634 | { |
| 635 | return ep_scan_ready_list(priv, ep_read_events_proc, NULL); |
| 636 | } |
| 637 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 638 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
| 639 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 640 | int pollflags; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 641 | struct eventpoll *ep = file->private_data; |
| 642 | |
| 643 | /* Insert inside our poll wait queue */ |
| 644 | poll_wait(file, &ep->poll_wait, wait); |
| 645 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 646 | /* |
| 647 | * Proceed to find out if wanted events are really available inside |
| 648 | * the ready list. This need to be done under ep_call_nested() |
| 649 | * supervision, since the call to f_op->poll() done on listed files |
| 650 | * could re-enter here. |
| 651 | */ |
| 652 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, |
| 653 | ep_poll_readyevents_proc, ep, ep); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 654 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 655 | return pollflags != -1 ? pollflags : 0; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 656 | } |
| 657 | |
| 658 | /* File callbacks that implement the eventpoll file behaviour */ |
| 659 | static const struct file_operations eventpoll_fops = { |
| 660 | .release = ep_eventpoll_release, |
| 661 | .poll = ep_eventpoll_poll |
| 662 | }; |
| 663 | |
| 664 | /* Fast test to see if the file is an evenpoll file */ |
| 665 | static inline int is_file_epoll(struct file *f) |
| 666 | { |
| 667 | return f->f_op == &eventpoll_fops; |
| 668 | } |
| 669 | |
| 670 | /* |
| 671 | * This is called from eventpoll_release() to unlink files from the eventpoll |
| 672 | * interface. We need to have this facility to cleanup correctly files that are |
| 673 | * closed without being removed from the eventpoll interface. |
| 674 | */ |
| 675 | void eventpoll_release_file(struct file *file) |
| 676 | { |
| 677 | struct list_head *lsthead = &file->f_ep_links; |
| 678 | struct eventpoll *ep; |
| 679 | struct epitem *epi; |
| 680 | |
| 681 | /* |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 682 | * We don't want to get "file->f_lock" because it is not |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 683 | * necessary. It is not necessary because we're in the "struct file" |
| 684 | * cleanup path, and this means that noone is using this file anymore. |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 685 | * So, for example, epoll_ctl() cannot hit here since if we reach this |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 686 | * point, the file counter already went to zero and fget() would fail. |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 687 | * The only hit might come from ep_free() but by holding the mutex |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 688 | * will correctly serialize the operation. We do need to acquire |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 689 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 690 | * from anywhere but ep_free(). |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 691 | * |
| 692 | * Besides, ep_remove() acquires the lock, so we can't hold it here. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 693 | */ |
| 694 | mutex_lock(&epmutex); |
| 695 | |
| 696 | while (!list_empty(lsthead)) { |
| 697 | epi = list_first_entry(lsthead, struct epitem, fllink); |
| 698 | |
| 699 | ep = epi->ep; |
| 700 | list_del_init(&epi->fllink); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 701 | mutex_lock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 702 | ep_remove(ep, epi); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 703 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 704 | } |
| 705 | |
| 706 | mutex_unlock(&epmutex); |
| 707 | } |
| 708 | |
| 709 | static int ep_alloc(struct eventpoll **pep) |
| 710 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 711 | int error; |
| 712 | struct user_struct *user; |
| 713 | struct eventpoll *ep; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 714 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 715 | user = get_current_user(); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 716 | error = -ENOMEM; |
| 717 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); |
| 718 | if (unlikely(!ep)) |
| 719 | goto free_uid; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 720 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 721 | spin_lock_init(&ep->lock); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 722 | mutex_init(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 723 | init_waitqueue_head(&ep->wq); |
| 724 | init_waitqueue_head(&ep->poll_wait); |
| 725 | INIT_LIST_HEAD(&ep->rdllist); |
| 726 | ep->rbr = RB_ROOT; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 727 | ep->ovflist = EP_UNACTIVE_PTR; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 728 | ep->user = user; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 729 | |
| 730 | *pep = ep; |
| 731 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 732 | return 0; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 733 | |
| 734 | free_uid: |
| 735 | free_uid(user); |
| 736 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 737 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 738 | |
| 739 | /* |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 740 | * Search the file inside the eventpoll tree. The RB tree operations |
| 741 | * are protected by the "mtx" mutex, and ep_find() must be called with |
| 742 | * "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | */ |
| 744 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) |
| 745 | { |
| 746 | int kcmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 747 | struct rb_node *rbp; |
| 748 | struct epitem *epi, *epir = NULL; |
| 749 | struct epoll_filefd ffd; |
| 750 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 751 | ep_set_ffd(&ffd, file, fd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 752 | for (rbp = ep->rbr.rb_node; rbp; ) { |
| 753 | epi = rb_entry(rbp, struct epitem, rbn); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 754 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 755 | if (kcmp > 0) |
| 756 | rbp = rbp->rb_right; |
| 757 | else if (kcmp < 0) |
| 758 | rbp = rbp->rb_left; |
| 759 | else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 760 | epir = epi; |
| 761 | break; |
| 762 | } |
| 763 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 764 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 765 | return epir; |
| 766 | } |
| 767 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 768 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 769 | * This is the callback that is passed to the wait queue wakeup |
| 770 | * machanism. It is called by the stored file descriptors when they |
| 771 | * have events to report. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 772 | */ |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 773 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 775 | int pwake = 0; |
| 776 | unsigned long flags; |
| 777 | struct epitem *epi = ep_item_from_wait(wait); |
| 778 | struct eventpoll *ep = epi->ep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 780 | spin_lock_irqsave(&ep->lock, flags); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 781 | |
| 782 | /* |
| 783 | * If the event mask does not contain any poll(2) event, we consider the |
| 784 | * descriptor to be disabled. This condition is likely the effect of the |
| 785 | * EPOLLONESHOT bit that disables the descriptor when an event is received, |
| 786 | * until the next EPOLL_CTL_MOD will be issued. |
| 787 | */ |
| 788 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 789 | goto out_unlock; |
| 790 | |
| 791 | /* |
| 792 | * If we are trasfering events to userspace, we can hold no locks |
| 793 | * (because we're accessing user memory, and because of linux f_op->poll() |
| 794 | * semantics). All the events that happens during that period of time are |
| 795 | * chained in ep->ovflist and requeued later on. |
| 796 | */ |
| 797 | if (unlikely(ep->ovflist != EP_UNACTIVE_PTR)) { |
| 798 | if (epi->next == EP_UNACTIVE_PTR) { |
| 799 | epi->next = ep->ovflist; |
| 800 | ep->ovflist = epi; |
| 801 | } |
| 802 | goto out_unlock; |
| 803 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 804 | |
| 805 | /* If this file is already in the ready list we exit soon */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 806 | if (!ep_is_linked(&epi->rdllink)) |
| 807 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 808 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 809 | /* |
| 810 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() |
| 811 | * wait list. |
| 812 | */ |
| 813 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 814 | wake_up_locked(&ep->wq); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 815 | if (waitqueue_active(&ep->poll_wait)) |
| 816 | pwake++; |
| 817 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 818 | out_unlock: |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 819 | spin_unlock_irqrestore(&ep->lock, flags); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 820 | |
| 821 | /* We have to call this outside the lock */ |
| 822 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 823 | ep_poll_safewake(&ep->poll_wait); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 824 | |
| 825 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 826 | } |
| 827 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | /* |
| 829 | * This is the callback that is used to add our wait queue to the |
| 830 | * target file wakeup lists. |
| 831 | */ |
| 832 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
| 833 | poll_table *pt) |
| 834 | { |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 835 | struct epitem *epi = ep_item_from_epqueue(pt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 836 | struct eppoll_entry *pwq; |
| 837 | |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 838 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
| 840 | pwq->whead = whead; |
| 841 | pwq->base = epi; |
| 842 | add_wait_queue(whead, &pwq->wait); |
| 843 | list_add_tail(&pwq->llink, &epi->pwqlist); |
| 844 | epi->nwait++; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 845 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | /* We have to signal that an error occurred */ |
| 847 | epi->nwait = -1; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 848 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 849 | } |
| 850 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 851 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
| 852 | { |
| 853 | int kcmp; |
| 854 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; |
| 855 | struct epitem *epic; |
| 856 | |
| 857 | while (*p) { |
| 858 | parent = *p; |
| 859 | epic = rb_entry(parent, struct epitem, rbn); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 860 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 861 | if (kcmp > 0) |
| 862 | p = &parent->rb_right; |
| 863 | else |
| 864 | p = &parent->rb_left; |
| 865 | } |
| 866 | rb_link_node(&epi->rbn, parent, p); |
| 867 | rb_insert_color(&epi->rbn, &ep->rbr); |
| 868 | } |
| 869 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 870 | /* |
| 871 | * Must be called with "mtx" held. |
| 872 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 873 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
| 874 | struct file *tfile, int fd) |
| 875 | { |
| 876 | int error, revents, pwake = 0; |
| 877 | unsigned long flags; |
| 878 | struct epitem *epi; |
| 879 | struct ep_pqueue epq; |
| 880 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 881 | if (unlikely(atomic_read(&ep->user->epoll_watches) >= |
| 882 | max_user_watches)) |
| 883 | return -ENOSPC; |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 884 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 885 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | |
| 887 | /* Item initialization follow here ... */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 888 | INIT_LIST_HEAD(&epi->rdllink); |
| 889 | INIT_LIST_HEAD(&epi->fllink); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 890 | INIT_LIST_HEAD(&epi->pwqlist); |
| 891 | epi->ep = ep; |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 892 | ep_set_ffd(&epi->ffd, tfile, fd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | epi->event = *event; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | epi->nwait = 0; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 895 | epi->next = EP_UNACTIVE_PTR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 896 | |
| 897 | /* Initialize the poll table using the queue callback */ |
| 898 | epq.epi = epi; |
| 899 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); |
| 900 | |
| 901 | /* |
| 902 | * Attach the item to the poll hooks and get current event bits. |
| 903 | * We can safely use the file* here because its usage count has |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 904 | * been increased by the caller of this function. Note that after |
| 905 | * this operation completes, the poll callback can start hitting |
| 906 | * the new item. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 907 | */ |
| 908 | revents = tfile->f_op->poll(tfile, &epq.pt); |
| 909 | |
| 910 | /* |
| 911 | * We have to check if something went wrong during the poll wait queue |
| 912 | * install process. Namely an allocation for a wait queue failed due |
| 913 | * high memory pressure. |
| 914 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 915 | error = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 916 | if (epi->nwait < 0) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 917 | goto error_unregister; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 918 | |
| 919 | /* Add the current item to the list of active epoll hook for this file */ |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 920 | spin_lock(&tfile->f_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | list_add_tail(&epi->fllink, &tfile->f_ep_links); |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 922 | spin_unlock(&tfile->f_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 924 | /* |
| 925 | * Add the current item to the RB tree. All RB tree operations are |
| 926 | * protected by "mtx", and ep_insert() is called with "mtx" held. |
| 927 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | ep_rbtree_insert(ep, epi); |
| 929 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 930 | /* We have to drop the new item inside our item list to keep track of it */ |
| 931 | spin_lock_irqsave(&ep->lock, flags); |
| 932 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 933 | /* If the file is already "ready" we drop it inside the ready list */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 934 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 935 | list_add_tail(&epi->rdllink, &ep->rdllist); |
| 936 | |
| 937 | /* Notify waiting tasks that events are available */ |
| 938 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 939 | wake_up_locked(&ep->wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 940 | if (waitqueue_active(&ep->poll_wait)) |
| 941 | pwake++; |
| 942 | } |
| 943 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 944 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 945 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 946 | atomic_inc(&ep->user->epoll_watches); |
| 947 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 948 | /* We have to call this outside the lock */ |
| 949 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 950 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 951 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 952 | return 0; |
| 953 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 954 | error_unregister: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 955 | ep_unregister_pollwait(ep, epi); |
| 956 | |
| 957 | /* |
| 958 | * We need to do this because an event could have been arrived on some |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 959 | * allocated wait queue. Note that we don't care about the ep->ovflist |
| 960 | * list, since that is used/cleaned only inside a section bound by "mtx". |
| 961 | * And ep_insert() is called with "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 962 | */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 963 | spin_lock_irqsave(&ep->lock, flags); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 964 | if (ep_is_linked(&epi->rdllink)) |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 965 | list_del_init(&epi->rdllink); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 966 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 967 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 968 | kmem_cache_free(epi_cache, epi); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 969 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 970 | return error; |
| 971 | } |
| 972 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | /* |
| 974 | * Modify the interest event mask by dropping an event if the new mask |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 975 | * has a match in the current file status. Must be called with "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 976 | */ |
| 977 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) |
| 978 | { |
| 979 | int pwake = 0; |
| 980 | unsigned int revents; |
| 981 | unsigned long flags; |
| 982 | |
| 983 | /* |
| 984 | * Set the new event interest mask before calling f_op->poll(), otherwise |
| 985 | * a potential race might occur. In fact if we do this operation inside |
| 986 | * the lock, an event might happen between the f_op->poll() call and the |
| 987 | * new event set registering. |
| 988 | */ |
| 989 | epi->event.events = event->events; |
| 990 | |
| 991 | /* |
| 992 | * Get current event bits. We can safely use the file* here because |
| 993 | * its usage count has been increased by the caller of this function. |
| 994 | */ |
| 995 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); |
| 996 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 997 | spin_lock_irqsave(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | |
| 999 | /* Copy the data member from inside the lock */ |
| 1000 | epi->event.data = event->data; |
| 1001 | |
| 1002 | /* |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1003 | * If the item is "hot" and it is not registered inside the ready |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 1004 | * list, push it inside. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1006 | if (revents & event->events) { |
| 1007 | if (!ep_is_linked(&epi->rdllink)) { |
| 1008 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1010 | /* Notify waiting tasks that events are available */ |
| 1011 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1012 | wake_up_locked(&ep->wq); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1013 | if (waitqueue_active(&ep->poll_wait)) |
| 1014 | pwake++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | } |
| 1016 | } |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1017 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1018 | |
| 1019 | /* We have to call this outside the lock */ |
| 1020 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1021 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | |
| 1023 | return 0; |
| 1024 | } |
| 1025 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1026 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
| 1027 | void *priv) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1028 | { |
| 1029 | struct ep_send_events_data *esed = priv; |
| 1030 | int eventcnt; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1031 | unsigned int revents; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1032 | struct epitem *epi; |
| 1033 | struct epoll_event __user *uevent; |
| 1034 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1035 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1036 | * We can loop without lock because we are passed a task private list. |
| 1037 | * Items cannot vanish during the loop because ep_scan_ready_list() is |
| 1038 | * holding "mtx" during this call. |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1039 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1040 | for (eventcnt = 0, uevent = esed->events; |
| 1041 | !list_empty(head) && eventcnt < esed->maxevents;) { |
| 1042 | epi = list_first_entry(head, struct epitem, rdllink); |
| 1043 | |
| 1044 | list_del_init(&epi->rdllink); |
| 1045 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1046 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL) & |
| 1047 | epi->event.events; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1048 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1049 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1050 | * If the event mask intersect the caller-requested one, |
| 1051 | * deliver the event to userspace. Again, ep_scan_ready_list() |
| 1052 | * is holding "mtx", so no operations coming from userspace |
| 1053 | * can change the item. |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1054 | */ |
| 1055 | if (revents) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1056 | if (__put_user(revents, &uevent->events) || |
| 1057 | __put_user(epi->event.data, &uevent->data)) |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1058 | return eventcnt ? eventcnt : -EFAULT; |
| 1059 | eventcnt++; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1060 | uevent++; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1061 | if (epi->event.events & EPOLLONESHOT) |
| 1062 | epi->event.events &= EP_PRIVATE_BITS; |
| 1063 | else if (!(epi->event.events & EPOLLET)) { |
| 1064 | /* |
| 1065 | * If this file has been added with Level |
| 1066 | * Trigger mode, we need to insert back inside |
| 1067 | * the ready list, so that the next call to |
| 1068 | * epoll_wait() will check again the events |
| 1069 | * availability. At this point, noone can insert |
| 1070 | * into ep->rdllist besides us. The epoll_ctl() |
| 1071 | * callers are locked out by |
| 1072 | * ep_scan_ready_list() holding "mtx" and the |
| 1073 | * poll callback will queue them in ep->ovflist. |
| 1074 | */ |
| 1075 | list_add_tail(&epi->rdllink, &ep->rdllist); |
| 1076 | } |
| 1077 | } |
| 1078 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1079 | |
| 1080 | return eventcnt; |
| 1081 | } |
| 1082 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1083 | static int ep_send_events(struct eventpoll *ep, |
| 1084 | struct epoll_event __user *events, int maxevents) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1085 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1086 | struct ep_send_events_data esed; |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 1087 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1088 | esed.maxevents = maxevents; |
| 1089 | esed.events = events; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1090 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1091 | return ep_scan_ready_list(ep, ep_send_events_proc, &esed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1092 | } |
| 1093 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1094 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
| 1095 | int maxevents, long timeout) |
| 1096 | { |
| 1097 | int res, eavail; |
| 1098 | unsigned long flags; |
| 1099 | long jtimeout; |
| 1100 | wait_queue_t wait; |
| 1101 | |
| 1102 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1103 | * Calculate the timeout by checking for the "infinite" value (-1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1104 | * and the overflow condition. The passed timeout is in milliseconds, |
| 1105 | * that why (t * HZ) / 1000. |
| 1106 | */ |
Davide Libenzi | e3306dd | 2005-09-27 21:45:33 -0700 | [diff] [blame] | 1107 | jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ? |
| 1108 | MAX_SCHEDULE_TIMEOUT : (timeout * HZ + 999) / 1000; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1109 | |
| 1110 | retry: |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1111 | spin_lock_irqsave(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1112 | |
| 1113 | res = 0; |
| 1114 | if (list_empty(&ep->rdllist)) { |
| 1115 | /* |
| 1116 | * We don't have any available event to return to the caller. |
| 1117 | * We need to sleep here, and we will be wake up by |
| 1118 | * ep_poll_callback() when events will become available. |
| 1119 | */ |
| 1120 | init_waitqueue_entry(&wait, current); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1121 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
Davide Libenzi | 3419b23 | 2006-06-25 05:48:14 -0700 | [diff] [blame] | 1122 | __add_wait_queue(&ep->wq, &wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1123 | |
| 1124 | for (;;) { |
| 1125 | /* |
| 1126 | * We don't want to sleep if the ep_poll_callback() sends us |
| 1127 | * a wakeup in between. That's why we set the task state |
| 1128 | * to TASK_INTERRUPTIBLE before doing the checks. |
| 1129 | */ |
| 1130 | set_current_state(TASK_INTERRUPTIBLE); |
| 1131 | if (!list_empty(&ep->rdllist) || !jtimeout) |
| 1132 | break; |
| 1133 | if (signal_pending(current)) { |
| 1134 | res = -EINTR; |
| 1135 | break; |
| 1136 | } |
| 1137 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1138 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | jtimeout = schedule_timeout(jtimeout); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1140 | spin_lock_irqsave(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1141 | } |
Davide Libenzi | 3419b23 | 2006-06-25 05:48:14 -0700 | [diff] [blame] | 1142 | __remove_wait_queue(&ep->wq, &wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | |
| 1144 | set_current_state(TASK_RUNNING); |
| 1145 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1146 | /* Is it worth to try to dig for events ? */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1147 | eavail = !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1148 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1149 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1150 | |
| 1151 | /* |
| 1152 | * Try to transfer events to user space. In case we get 0 events and |
| 1153 | * there's still timeout left over, we go trying again in search of |
| 1154 | * more luck. |
| 1155 | */ |
| 1156 | if (!res && eavail && |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1157 | !(res = ep_send_events(ep, events, maxevents)) && jtimeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1158 | goto retry; |
| 1159 | |
| 1160 | return res; |
| 1161 | } |
| 1162 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1163 | /* |
Andrew Morton | 523723b | 2008-08-12 15:09:01 -0700 | [diff] [blame] | 1164 | * Open an eventpoll file descriptor. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1165 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1166 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1167 | { |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1168 | int error; |
| 1169 | struct eventpoll *ep = NULL; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1170 | |
Ulrich Drepper | e38b36f | 2008-07-23 21:29:42 -0700 | [diff] [blame] | 1171 | /* Check the EPOLL_* constant for consistency. */ |
| 1172 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); |
| 1173 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1174 | if (flags & ~EPOLL_CLOEXEC) |
| 1175 | return -EINVAL; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1176 | /* |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1177 | * Create the internal data structure ("struct eventpoll"). |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1178 | */ |
Ulrich Drepper | 9fe5ad9 | 2008-07-23 21:29:43 -0700 | [diff] [blame] | 1179 | error = ep_alloc(&ep); |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1180 | if (error < 0) |
| 1181 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1182 | /* |
| 1183 | * Creates all the items needed to setup an eventpoll file. That is, |
Al Viro | 2030a42 | 2008-02-23 06:46:49 -0500 | [diff] [blame] | 1184 | * a file structure and a free file descriptor. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1185 | */ |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1186 | error = anon_inode_getfd("[eventpoll]", &eventpoll_fops, ep, |
| 1187 | flags & O_CLOEXEC); |
| 1188 | if (error < 0) |
Al Viro | 2030a42 | 2008-02-23 06:46:49 -0500 | [diff] [blame] | 1189 | ep_free(ep); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1190 | |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1191 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1192 | } |
| 1193 | |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1194 | SYSCALL_DEFINE1(epoll_create, int, size) |
Ulrich Drepper | a0998b5 | 2008-07-23 21:29:27 -0700 | [diff] [blame] | 1195 | { |
Ulrich Drepper | 9fe5ad9 | 2008-07-23 21:29:43 -0700 | [diff] [blame] | 1196 | if (size < 0) |
| 1197 | return -EINVAL; |
| 1198 | |
| 1199 | return sys_epoll_create1(0); |
Ulrich Drepper | a0998b5 | 2008-07-23 21:29:27 -0700 | [diff] [blame] | 1200 | } |
| 1201 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1202 | /* |
| 1203 | * The following function implements the controller interface for |
| 1204 | * the eventpoll file that enables the insertion/removal/change of |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 1205 | * file descriptors inside the interest set. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1206 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1207 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
| 1208 | struct epoll_event __user *, event) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1209 | { |
| 1210 | int error; |
| 1211 | struct file *file, *tfile; |
| 1212 | struct eventpoll *ep; |
| 1213 | struct epitem *epi; |
| 1214 | struct epoll_event epds; |
| 1215 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1216 | error = -EFAULT; |
| 1217 | if (ep_op_has_event(op) && |
| 1218 | copy_from_user(&epds, event, sizeof(struct epoll_event))) |
| 1219 | goto error_return; |
| 1220 | |
| 1221 | /* Get the "struct file *" for the eventpoll file */ |
| 1222 | error = -EBADF; |
| 1223 | file = fget(epfd); |
| 1224 | if (!file) |
| 1225 | goto error_return; |
| 1226 | |
| 1227 | /* Get the "struct file *" for the target file */ |
| 1228 | tfile = fget(fd); |
| 1229 | if (!tfile) |
| 1230 | goto error_fput; |
| 1231 | |
| 1232 | /* The target file descriptor must support poll */ |
| 1233 | error = -EPERM; |
| 1234 | if (!tfile->f_op || !tfile->f_op->poll) |
| 1235 | goto error_tgt_fput; |
| 1236 | |
| 1237 | /* |
| 1238 | * We have to check that the file structure underneath the file descriptor |
| 1239 | * the user passed to us _is_ an eventpoll file. And also we do not permit |
| 1240 | * adding an epoll file descriptor inside itself. |
| 1241 | */ |
| 1242 | error = -EINVAL; |
| 1243 | if (file == tfile || !is_file_epoll(file)) |
| 1244 | goto error_tgt_fput; |
| 1245 | |
| 1246 | /* |
| 1247 | * At this point it is safe to assume that the "private_data" contains |
| 1248 | * our own data structure. |
| 1249 | */ |
| 1250 | ep = file->private_data; |
| 1251 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1252 | mutex_lock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1253 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 1254 | /* |
| 1255 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" |
| 1256 | * above, we can be sure to be able to use the item looked up by |
| 1257 | * ep_find() till we release the mutex. |
| 1258 | */ |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1259 | epi = ep_find(ep, tfile, fd); |
| 1260 | |
| 1261 | error = -EINVAL; |
| 1262 | switch (op) { |
| 1263 | case EPOLL_CTL_ADD: |
| 1264 | if (!epi) { |
| 1265 | epds.events |= POLLERR | POLLHUP; |
| 1266 | |
| 1267 | error = ep_insert(ep, &epds, tfile, fd); |
| 1268 | } else |
| 1269 | error = -EEXIST; |
| 1270 | break; |
| 1271 | case EPOLL_CTL_DEL: |
| 1272 | if (epi) |
| 1273 | error = ep_remove(ep, epi); |
| 1274 | else |
| 1275 | error = -ENOENT; |
| 1276 | break; |
| 1277 | case EPOLL_CTL_MOD: |
| 1278 | if (epi) { |
| 1279 | epds.events |= POLLERR | POLLHUP; |
| 1280 | error = ep_modify(ep, epi, &epds); |
| 1281 | } else |
| 1282 | error = -ENOENT; |
| 1283 | break; |
| 1284 | } |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1285 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1286 | |
| 1287 | error_tgt_fput: |
| 1288 | fput(tfile); |
| 1289 | error_fput: |
| 1290 | fput(file); |
| 1291 | error_return: |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1292 | |
| 1293 | return error; |
| 1294 | } |
| 1295 | |
| 1296 | /* |
| 1297 | * Implement the event wait interface for the eventpoll file. It is the kernel |
| 1298 | * part of the user space epoll_wait(2). |
| 1299 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1300 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
| 1301 | int, maxevents, int, timeout) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1302 | { |
| 1303 | int error; |
| 1304 | struct file *file; |
| 1305 | struct eventpoll *ep; |
| 1306 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1307 | /* The maximum number of event must be greater than zero */ |
| 1308 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) |
| 1309 | return -EINVAL; |
| 1310 | |
| 1311 | /* Verify that the area passed by the user is writeable */ |
| 1312 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) { |
| 1313 | error = -EFAULT; |
| 1314 | goto error_return; |
| 1315 | } |
| 1316 | |
| 1317 | /* Get the "struct file *" for the eventpoll file */ |
| 1318 | error = -EBADF; |
| 1319 | file = fget(epfd); |
| 1320 | if (!file) |
| 1321 | goto error_return; |
| 1322 | |
| 1323 | /* |
| 1324 | * We have to check that the file structure underneath the fd |
| 1325 | * the user passed to us _is_ an eventpoll file. |
| 1326 | */ |
| 1327 | error = -EINVAL; |
| 1328 | if (!is_file_epoll(file)) |
| 1329 | goto error_fput; |
| 1330 | |
| 1331 | /* |
| 1332 | * At this point it is safe to assume that the "private_data" contains |
| 1333 | * our own data structure. |
| 1334 | */ |
| 1335 | ep = file->private_data; |
| 1336 | |
| 1337 | /* Time to fish for events ... */ |
| 1338 | error = ep_poll(ep, events, maxevents, timeout); |
| 1339 | |
| 1340 | error_fput: |
| 1341 | fput(file); |
| 1342 | error_return: |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1343 | |
| 1344 | return error; |
| 1345 | } |
| 1346 | |
Roland McGrath | f3de272 | 2008-04-30 00:53:09 -0700 | [diff] [blame] | 1347 | #ifdef HAVE_SET_RESTORE_SIGMASK |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1348 | |
| 1349 | /* |
| 1350 | * Implement the event wait interface for the eventpoll file. It is the kernel |
| 1351 | * part of the user space epoll_pwait(2). |
| 1352 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1353 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
| 1354 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, |
| 1355 | size_t, sigsetsize) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1356 | { |
| 1357 | int error; |
| 1358 | sigset_t ksigmask, sigsaved; |
| 1359 | |
| 1360 | /* |
| 1361 | * If the caller wants a certain signal mask to be set during the wait, |
| 1362 | * we apply it here. |
| 1363 | */ |
| 1364 | if (sigmask) { |
| 1365 | if (sigsetsize != sizeof(sigset_t)) |
| 1366 | return -EINVAL; |
| 1367 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) |
| 1368 | return -EFAULT; |
| 1369 | sigdelsetmask(&ksigmask, sigmask(SIGKILL) | sigmask(SIGSTOP)); |
| 1370 | sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); |
| 1371 | } |
| 1372 | |
| 1373 | error = sys_epoll_wait(epfd, events, maxevents, timeout); |
| 1374 | |
| 1375 | /* |
| 1376 | * If we changed the signal mask, we need to restore the original one. |
| 1377 | * In case we've got a signal while waiting, we do not restore the |
| 1378 | * signal mask yet, and we allow do_signal() to deliver the signal on |
| 1379 | * the way back to userspace, before the signal mask is restored. |
| 1380 | */ |
| 1381 | if (sigmask) { |
| 1382 | if (error == -EINTR) { |
| 1383 | memcpy(¤t->saved_sigmask, &sigsaved, |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1384 | sizeof(sigsaved)); |
Roland McGrath | 4e4c22c | 2008-04-30 00:53:06 -0700 | [diff] [blame] | 1385 | set_restore_sigmask(); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1386 | } else |
| 1387 | sigprocmask(SIG_SETMASK, &sigsaved, NULL); |
| 1388 | } |
| 1389 | |
| 1390 | return error; |
| 1391 | } |
| 1392 | |
Roland McGrath | f3de272 | 2008-04-30 00:53:09 -0700 | [diff] [blame] | 1393 | #endif /* HAVE_SET_RESTORE_SIGMASK */ |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1394 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1395 | static int __init eventpoll_init(void) |
| 1396 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1397 | struct sysinfo si; |
| 1398 | |
| 1399 | si_meminfo(&si); |
Davide Libenzi | 9df04e1 | 2009-01-29 14:25:26 -0800 | [diff] [blame] | 1400 | /* |
| 1401 | * Allows top 4% of lomem to be allocated for epoll watches (per user). |
| 1402 | */ |
| 1403 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1404 | EP_ITEM_COST; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1405 | |
| 1406 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1407 | ep_nested_calls_init(&poll_safewake_ncalls); |
| 1408 | |
| 1409 | /* Initialize the structure used to perform file's f_op->poll() calls */ |
| 1410 | ep_nested_calls_init(&poll_readywalk_ncalls); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1411 | |
| 1412 | /* Allocates slab cache used to allocate "struct epitem" items */ |
| 1413 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1414 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1415 | |
| 1416 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ |
| 1417 | pwq_cache = kmem_cache_create("eventpoll_pwq", |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1418 | sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1419 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1420 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1421 | } |
Davide Libenzi | cea6924 | 2007-05-10 22:23:22 -0700 | [diff] [blame] | 1422 | fs_initcall(eventpoll_init); |