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> |
Ingo Molnar | 174cd4b | 2017-02-02 19:15:33 +0100 | [diff] [blame] | 16 | #include <linux/sched/signal.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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> |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 36 | #include <linux/device.h> |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 37 | #include <linux/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | #include <asm/io.h> |
| 39 | #include <asm/mman.h> |
Arun Sharma | 60063497 | 2011-07-26 16:09:06 -0700 | [diff] [blame] | 40 | #include <linux/atomic.h> |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 41 | #include <linux/proc_fs.h> |
| 42 | #include <linux/seq_file.h> |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 43 | #include <linux/compat.h> |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 44 | #include <linux/rculist.h> |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 45 | #include <net/busy_poll.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | /* |
| 48 | * LOCKING: |
| 49 | * There are three level of locking required by epoll : |
| 50 | * |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 51 | * 1) epmutex (mutex) |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 52 | * 2) ep->mtx (mutex) |
| 53 | * 3) ep->lock (spinlock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | * |
| 55 | * The acquire order is the one listed above, from 1 to 3. |
| 56 | * We need a spinlock (ep->lock) because we manipulate objects |
| 57 | * from inside the poll callback, that might be triggered from |
| 58 | * a wake_up() that in turn might be called from IRQ context. |
| 59 | * So we can't sleep inside the poll callback and hence we need |
| 60 | * a spinlock. During the event transfer loop (from kernel to |
| 61 | * user space) we could end up sleeping due a copy_to_user(), so |
| 62 | * 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] | 63 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
| 64 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). |
| 65 | * Then we also need a global mutex to serialize eventpoll_release_file() |
| 66 | * and ep_free(). |
| 67 | * This mutex is acquired by ep_free() during the epoll file |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | * cleanup path and it is also acquired by eventpoll_release_file() |
| 69 | * if a file has been pushed inside an epoll set and it is then |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 70 | * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL). |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 71 | * It is also acquired when inserting an epoll fd onto another epoll |
| 72 | * fd. We do this so that we walk the epoll tree and ensure that this |
| 73 | * insertion does not create a cycle of epoll file descriptors, which |
| 74 | * could lead to deadlock. We need a global mutex to prevent two |
| 75 | * simultaneous inserts (A into B and B into A) from racing and |
| 76 | * constructing a cycle without either insert observing that it is |
| 77 | * going to. |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 78 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
| 79 | * case when one epoll fd is added to another. In this case, we |
| 80 | * always acquire the locks in the order of nesting (i.e. after |
| 81 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired |
| 82 | * before e2->mtx). Since we disallow cycles of epoll file |
| 83 | * descriptors, this ensures that the mutexes are well-ordered. In |
| 84 | * order to communicate this nesting to lockdep, when walking a tree |
| 85 | * of epoll file descriptors, we use the current recursion depth as |
| 86 | * the lockdep subkey. |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 87 | * It is possible to drop the "ep->mtx" and to use the global |
| 88 | * mutex "epmutex" (together with "ep->lock") to have it working, |
| 89 | * but having "ep->mtx" will make the interface more scalable. |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 90 | * Events that require holding "epmutex" are very rare, while for |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 91 | * normal operations the epoll private "ep->mtx" will guarantee |
| 92 | * a better scalability. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | */ |
| 94 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | /* Epoll private bits inside the event mask */ |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 96 | #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 98 | #define EPOLLINOUT_BITS (POLLIN | POLLOUT) |
| 99 | |
| 100 | #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | POLLERR | POLLHUP | \ |
| 101 | EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE) |
| 102 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 103 | /* Maximum number of nesting allowed inside epoll sets */ |
| 104 | #define EP_MAX_NESTS 4 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | |
Davide Libenzi | b611967 | 2006-10-11 01:21:44 -0700 | [diff] [blame] | 106 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
| 107 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 108 | #define EP_UNACTIVE_PTR ((void *) -1L) |
| 109 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 110 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
| 111 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | struct epoll_filefd { |
| 113 | struct file *file; |
| 114 | int fd; |
Eric Wong | 39732ca | 2013-04-30 15:27:38 -0700 | [diff] [blame] | 115 | } __packed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | |
| 117 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 118 | * Structure used to track possible nested calls, for too deep recursions |
| 119 | * and loop cycles. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 121 | struct nested_call_node { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | struct list_head llink; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 123 | void *cookie; |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 124 | void *ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | }; |
| 126 | |
| 127 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 128 | * This structure is used as collector for nested calls, to check for |
| 129 | * maximum recursion dept and loop cycles. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 131 | struct nested_calls { |
| 132 | struct list_head tasks_call_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | spinlock_t lock; |
| 134 | }; |
| 135 | |
| 136 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | * Each file descriptor added to the eventpoll interface will |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 138 | * have an entry of this type linked to the "rbr" RB tree. |
Eric Wong | 39732ca | 2013-04-30 15:27:38 -0700 | [diff] [blame] | 139 | * Avoid increasing the size of this struct, there can be many thousands |
| 140 | * of these on a server and we do not want this to take another cache line. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | */ |
| 142 | struct epitem { |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 143 | union { |
| 144 | /* RB tree node links this structure to the eventpoll RB tree */ |
| 145 | struct rb_node rbn; |
| 146 | /* Used to free the struct epitem */ |
| 147 | struct rcu_head rcu; |
| 148 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | |
| 150 | /* List header used to link this structure to the eventpoll ready list */ |
| 151 | struct list_head rdllink; |
| 152 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 153 | /* |
| 154 | * Works together "struct eventpoll"->ovflist in keeping the |
| 155 | * single linked chain of items. |
| 156 | */ |
| 157 | struct epitem *next; |
| 158 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | /* The file descriptor information this item refers to */ |
| 160 | struct epoll_filefd ffd; |
| 161 | |
| 162 | /* Number of active wait queue attached to poll operations */ |
| 163 | int nwait; |
| 164 | |
| 165 | /* List containing poll wait queues */ |
| 166 | struct list_head pwqlist; |
| 167 | |
| 168 | /* The "container" of this item */ |
| 169 | struct eventpoll *ep; |
| 170 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 171 | /* List header used to link this item to the "struct file" items list */ |
| 172 | struct list_head fllink; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 173 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 174 | /* wakeup_source used when EPOLLWAKEUP is set */ |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 175 | struct wakeup_source __rcu *ws; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 176 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 177 | /* The structure that describe the interested events and the source fd */ |
| 178 | struct epoll_event event; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 179 | }; |
| 180 | |
| 181 | /* |
| 182 | * This structure is stored inside the "private_data" member of the file |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 183 | * structure and represents the main data structure for the eventpoll |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 184 | * interface. |
| 185 | */ |
| 186 | struct eventpoll { |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 187 | /* Protect the access to this structure */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 188 | spinlock_t lock; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 189 | |
| 190 | /* |
| 191 | * This mutex is used to ensure that files are not removed |
| 192 | * while epoll is using them. This is held during the event |
| 193 | * collection loop, the file cleanup path, the epoll file exit |
| 194 | * code and the ctl operations. |
| 195 | */ |
| 196 | struct mutex mtx; |
| 197 | |
| 198 | /* Wait queue used by sys_epoll_wait() */ |
| 199 | wait_queue_head_t wq; |
| 200 | |
| 201 | /* Wait queue used by file->poll() */ |
| 202 | wait_queue_head_t poll_wait; |
| 203 | |
| 204 | /* List of ready file descriptors */ |
| 205 | struct list_head rdllist; |
| 206 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 207 | /* RB tree root used to store monitored fd structs */ |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 208 | struct rb_root_cached rbr; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 209 | |
| 210 | /* |
| 211 | * This is a single linked list that chains all the "struct epitem" that |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 212 | * happened while transferring ready events to userspace w/out |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 213 | * holding ->lock. |
| 214 | */ |
| 215 | struct epitem *ovflist; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 216 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 217 | /* wakeup_source used when ep_scan_ready_list is running */ |
| 218 | struct wakeup_source *ws; |
| 219 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 220 | /* The user that created the eventpoll descriptor */ |
| 221 | struct user_struct *user; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 222 | |
| 223 | struct file *file; |
| 224 | |
| 225 | /* used to optimize loop detection check */ |
| 226 | int visited; |
| 227 | struct list_head visited_list_link; |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 228 | |
| 229 | #ifdef CONFIG_NET_RX_BUSY_POLL |
| 230 | /* used to track busy poll napi_id */ |
| 231 | unsigned int napi_id; |
| 232 | #endif |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 233 | }; |
| 234 | |
| 235 | /* Wait structure used by the poll hooks */ |
| 236 | struct eppoll_entry { |
| 237 | /* List header used to link this structure to the "struct epitem" */ |
| 238 | struct list_head llink; |
| 239 | |
| 240 | /* The "base" pointer is set to the container "struct epitem" */ |
Tony Battersby | 4f0989d | 2009-03-31 15:24:16 -0700 | [diff] [blame] | 241 | struct epitem *base; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 242 | |
| 243 | /* |
| 244 | * Wait queue item that will be linked to the target file wait |
| 245 | * queue head. |
| 246 | */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 247 | wait_queue_entry_t wait; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 248 | |
| 249 | /* The wait queue head that linked the "wait" wait queue item */ |
| 250 | wait_queue_head_t *whead; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | }; |
| 252 | |
| 253 | /* Wrapper struct used by poll queueing */ |
| 254 | struct ep_pqueue { |
| 255 | poll_table pt; |
| 256 | struct epitem *epi; |
| 257 | }; |
| 258 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 259 | /* Used by the ep_send_events() function as callback private data */ |
| 260 | struct ep_send_events_data { |
| 261 | int maxevents; |
| 262 | struct epoll_event __user *events; |
| 263 | }; |
| 264 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | /* |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 266 | * Configuration options available inside /proc/sys/fs/epoll/ |
| 267 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 268 | /* Maximum number of epoll watched descriptors, per user */ |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 269 | static long max_user_watches __read_mostly; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 270 | |
| 271 | /* |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 272 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 274 | static DEFINE_MUTEX(epmutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 276 | /* Used to check for epoll file descriptor inclusion loops */ |
| 277 | static struct nested_calls poll_loop_ncalls; |
| 278 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 279 | /* Used for safe wake up implementation */ |
| 280 | static struct nested_calls poll_safewake_ncalls; |
| 281 | |
| 282 | /* Used to call file's f_op->poll() under the nested calls boundaries */ |
| 283 | static struct nested_calls poll_readywalk_ncalls; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 284 | |
| 285 | /* Slab cache used to allocate "struct epitem" */ |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 286 | static struct kmem_cache *epi_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | |
| 288 | /* Slab cache used to allocate "struct eppoll_entry" */ |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 289 | static struct kmem_cache *pwq_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 291 | /* Visited nodes during ep_loop_check(), so we can unset them when we finish */ |
| 292 | static LIST_HEAD(visited_list); |
| 293 | |
| 294 | /* |
| 295 | * List of files with newly added links, where we may need to limit the number |
| 296 | * of emanating paths. Protected by the epmutex. |
| 297 | */ |
| 298 | static LIST_HEAD(tfile_check_list); |
| 299 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 300 | #ifdef CONFIG_SYSCTL |
| 301 | |
| 302 | #include <linux/sysctl.h> |
| 303 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 304 | static long zero; |
| 305 | static long long_max = LONG_MAX; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 306 | |
Joe Perches | 1f7e061 | 2014-06-06 14:38:05 -0700 | [diff] [blame] | 307 | struct ctl_table epoll_table[] = { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 308 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 309 | .procname = "max_user_watches", |
| 310 | .data = &max_user_watches, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 311 | .maxlen = sizeof(max_user_watches), |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 312 | .mode = 0644, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 313 | .proc_handler = proc_doulongvec_minmax, |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 314 | .extra1 = &zero, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 315 | .extra2 = &long_max, |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 316 | }, |
Eric W. Biederman | ab09203 | 2009-11-05 14:25:10 -0800 | [diff] [blame] | 317 | { } |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 318 | }; |
| 319 | #endif /* CONFIG_SYSCTL */ |
| 320 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 321 | static const struct file_operations eventpoll_fops; |
| 322 | |
| 323 | static inline int is_file_epoll(struct file *f) |
| 324 | { |
| 325 | return f->f_op == &eventpoll_fops; |
| 326 | } |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 327 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 328 | /* Setup the structure that is used as key for the RB tree */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 329 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
| 330 | struct file *file, int fd) |
| 331 | { |
| 332 | ffd->file = file; |
| 333 | ffd->fd = fd; |
| 334 | } |
| 335 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 336 | /* Compare RB tree keys */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 337 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
| 338 | struct epoll_filefd *p2) |
| 339 | { |
| 340 | return (p1->file > p2->file ? +1: |
| 341 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); |
| 342 | } |
| 343 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 344 | /* Tells us if the item is currently linked */ |
| 345 | static inline int ep_is_linked(struct list_head *p) |
| 346 | { |
| 347 | return !list_empty(p); |
| 348 | } |
| 349 | |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 350 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 351 | { |
| 352 | return container_of(p, struct eppoll_entry, wait); |
| 353 | } |
| 354 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 355 | /* Get the "struct epitem" from a wait queue pointer */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 356 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 357 | { |
| 358 | return container_of(p, struct eppoll_entry, wait)->base; |
| 359 | } |
| 360 | |
| 361 | /* Get the "struct epitem" from an epoll queue wrapper */ |
Davide Libenzi | cdac75e | 2008-04-29 00:58:34 -0700 | [diff] [blame] | 362 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 363 | { |
| 364 | return container_of(p, struct ep_pqueue, pt)->epi; |
| 365 | } |
| 366 | |
| 367 | /* 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] | 368 | static inline int ep_op_has_event(int op) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 369 | { |
Andrew Morton | a80a6b8 | 2012-11-08 15:53:35 -0800 | [diff] [blame] | 370 | return op != EPOLL_CTL_DEL; |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 371 | } |
| 372 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | /* Initialize the poll safe wake up structure */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 374 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 376 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
| 377 | spin_lock_init(&ncalls->lock); |
| 378 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 380 | /** |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 381 | * ep_events_available - Checks if ready events might be available. |
| 382 | * |
| 383 | * @ep: Pointer to the eventpoll context. |
| 384 | * |
| 385 | * Returns: Returns a value different than zero if ready events are available, |
| 386 | * or zero otherwise. |
| 387 | */ |
| 388 | static inline int ep_events_available(struct eventpoll *ep) |
| 389 | { |
| 390 | return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; |
| 391 | } |
| 392 | |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 393 | #ifdef CONFIG_NET_RX_BUSY_POLL |
| 394 | static bool ep_busy_loop_end(void *p, unsigned long start_time) |
| 395 | { |
| 396 | struct eventpoll *ep = p; |
| 397 | |
| 398 | return ep_events_available(ep) || busy_loop_timeout(start_time); |
| 399 | } |
| 400 | #endif /* CONFIG_NET_RX_BUSY_POLL */ |
| 401 | |
| 402 | /* |
| 403 | * Busy poll if globally on and supporting sockets found && no events, |
| 404 | * busy loop will return if need_resched or ep_events_available. |
| 405 | * |
| 406 | * we must do our busy polling with irqs enabled |
| 407 | */ |
| 408 | static void ep_busy_loop(struct eventpoll *ep, int nonblock) |
| 409 | { |
| 410 | #ifdef CONFIG_NET_RX_BUSY_POLL |
| 411 | unsigned int napi_id = READ_ONCE(ep->napi_id); |
| 412 | |
| 413 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) |
| 414 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep); |
| 415 | #endif |
| 416 | } |
| 417 | |
| 418 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) |
| 419 | { |
| 420 | #ifdef CONFIG_NET_RX_BUSY_POLL |
| 421 | if (ep->napi_id) |
| 422 | ep->napi_id = 0; |
| 423 | #endif |
| 424 | } |
| 425 | |
| 426 | /* |
| 427 | * Set epoll busy poll NAPI ID from sk. |
| 428 | */ |
| 429 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) |
| 430 | { |
| 431 | #ifdef CONFIG_NET_RX_BUSY_POLL |
| 432 | struct eventpoll *ep; |
| 433 | unsigned int napi_id; |
| 434 | struct socket *sock; |
| 435 | struct sock *sk; |
| 436 | int err; |
| 437 | |
| 438 | if (!net_busy_loop_on()) |
| 439 | return; |
| 440 | |
| 441 | sock = sock_from_file(epi->ffd.file, &err); |
| 442 | if (!sock) |
| 443 | return; |
| 444 | |
| 445 | sk = sock->sk; |
| 446 | if (!sk) |
| 447 | return; |
| 448 | |
| 449 | napi_id = READ_ONCE(sk->sk_napi_id); |
| 450 | ep = epi->ep; |
| 451 | |
| 452 | /* Non-NAPI IDs can be rejected |
| 453 | * or |
| 454 | * Nothing to do if we already have this ID |
| 455 | */ |
| 456 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) |
| 457 | return; |
| 458 | |
| 459 | /* record NAPI ID for use in next busy poll */ |
| 460 | ep->napi_id = napi_id; |
| 461 | #endif |
| 462 | } |
| 463 | |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 464 | /** |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 465 | * ep_call_nested - Perform a bound (possibly) nested call, by checking |
| 466 | * that the recursion limit is not exceeded, and that |
| 467 | * the same nested call (by the meaning of same cookie) is |
| 468 | * no re-entered. |
| 469 | * |
| 470 | * @ncalls: Pointer to the nested_calls structure to be used for this call. |
| 471 | * @max_nests: Maximum number of allowed nesting calls. |
| 472 | * @nproc: Nested call core function pointer. |
| 473 | * @priv: Opaque data to be passed to the @nproc callback. |
| 474 | * @cookie: Cookie to be used to identify this nested call. |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 475 | * @ctx: This instance context. |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 476 | * |
| 477 | * Returns: Returns the code returned by the @nproc callback, or -1 if |
| 478 | * the maximum recursion limit has been exceeded. |
| 479 | */ |
| 480 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
| 481 | int (*nproc)(void *, void *, int), void *priv, |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 482 | void *cookie, void *ctx) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 483 | { |
| 484 | int error, call_nests = 0; |
| 485 | unsigned long flags; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 486 | struct list_head *lsthead = &ncalls->tasks_call_list; |
| 487 | struct nested_call_node *tncur; |
| 488 | struct nested_call_node tnode; |
| 489 | |
| 490 | spin_lock_irqsave(&ncalls->lock, flags); |
| 491 | |
| 492 | /* |
| 493 | * Try to see if the current task is already inside this wakeup call. |
| 494 | * We use a list here, since the population inside this set is always |
| 495 | * very much limited. |
| 496 | */ |
| 497 | list_for_each_entry(tncur, lsthead, llink) { |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 498 | if (tncur->ctx == ctx && |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 499 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
| 500 | /* |
| 501 | * Ops ... loop detected or maximum nest level reached. |
| 502 | * We abort this wake by breaking the cycle itself. |
| 503 | */ |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame] | 504 | error = -1; |
| 505 | goto out_unlock; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 506 | } |
| 507 | } |
| 508 | |
| 509 | /* Add the current task and cookie to the list */ |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 510 | tnode.ctx = ctx; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 511 | tnode.cookie = cookie; |
| 512 | list_add(&tnode.llink, lsthead); |
| 513 | |
| 514 | spin_unlock_irqrestore(&ncalls->lock, flags); |
| 515 | |
| 516 | /* Call the nested function */ |
| 517 | error = (*nproc)(priv, cookie, call_nests); |
| 518 | |
| 519 | /* Remove the current task from the list */ |
| 520 | spin_lock_irqsave(&ncalls->lock, flags); |
| 521 | list_del(&tnode.llink); |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 522 | out_unlock: |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 523 | spin_unlock_irqrestore(&ncalls->lock, flags); |
| 524 | |
| 525 | return error; |
| 526 | } |
| 527 | |
Steven Rostedt | 02edc6f | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 528 | /* |
| 529 | * As described in commit 0ccf831cb lockdep: annotate epoll |
| 530 | * the use of wait queues used by epoll is done in a very controlled |
| 531 | * manner. Wake ups can nest inside each other, but are never done |
| 532 | * with the same locking. For example: |
| 533 | * |
| 534 | * dfd = socket(...); |
| 535 | * efd1 = epoll_create(); |
| 536 | * efd2 = epoll_create(); |
| 537 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); |
| 538 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); |
| 539 | * |
| 540 | * When a packet arrives to the device underneath "dfd", the net code will |
| 541 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a |
| 542 | * callback wakeup entry on that queue, and the wake_up() performed by the |
| 543 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll |
| 544 | * (efd1) notices that it may have some event ready, so it needs to wake up |
| 545 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() |
| 546 | * that ends up in another wake_up(), after having checked about the |
| 547 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to |
| 548 | * avoid stack blasting. |
| 549 | * |
| 550 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle |
| 551 | * this special case of epoll. |
| 552 | */ |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 553 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 554 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, |
| 555 | unsigned long events, int subclass) |
| 556 | { |
| 557 | unsigned long flags; |
| 558 | |
| 559 | spin_lock_irqsave_nested(&wqueue->lock, flags, subclass); |
| 560 | wake_up_locked_poll(wqueue, events); |
| 561 | spin_unlock_irqrestore(&wqueue->lock, flags); |
| 562 | } |
| 563 | #else |
| 564 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, |
| 565 | unsigned long events, int subclass) |
| 566 | { |
| 567 | wake_up_poll(wqueue, events); |
| 568 | } |
| 569 | #endif |
| 570 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 571 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) |
| 572 | { |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 573 | ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN, |
| 574 | 1 + call_nests); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 575 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 576 | } |
| 577 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 578 | /* |
| 579 | * Perform a safe wake up of the poll wait list. The problem is that |
| 580 | * with the new callback'd wake up system, it is possible that the |
| 581 | * poll callback is reentered from inside the call to wake_up() done |
| 582 | * 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] | 583 | * 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] | 584 | * and we cannot reenter the same wait queue head at all. This will |
| 585 | * 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] | 586 | * EP_MAX_NESTS deep. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 587 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 588 | static void ep_poll_safewake(wait_queue_head_t *wq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | { |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 590 | int this_cpu = get_cpu(); |
| 591 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 592 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 593 | ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu); |
| 594 | |
| 595 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 596 | } |
| 597 | |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 598 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
| 599 | { |
| 600 | wait_queue_head_t *whead; |
| 601 | |
| 602 | rcu_read_lock(); |
Oleg Nesterov | 138e4ad | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 603 | /* |
| 604 | * If it is cleared by POLLFREE, it should be rcu-safe. |
| 605 | * If we read NULL we need a barrier paired with |
| 606 | * smp_store_release() in ep_poll_callback(), otherwise |
| 607 | * we rely on whead->lock. |
| 608 | */ |
| 609 | whead = smp_load_acquire(&pwq->whead); |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 610 | if (whead) |
| 611 | remove_wait_queue(whead, &pwq->wait); |
| 612 | rcu_read_unlock(); |
| 613 | } |
| 614 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 615 | /* |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 616 | * This function unregisters poll callbacks from the associated file |
| 617 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from |
| 618 | * ep_free). |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 619 | */ |
| 620 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
| 621 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 622 | struct list_head *lsthead = &epi->pwqlist; |
| 623 | struct eppoll_entry *pwq; |
| 624 | |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 625 | while (!list_empty(lsthead)) { |
| 626 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 627 | |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 628 | list_del(&pwq->llink); |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 629 | ep_remove_wait_queue(pwq); |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 630 | kmem_cache_free(pwq_cache, pwq); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 631 | } |
| 632 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 633 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 634 | /* call only when ep->mtx is held */ |
| 635 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) |
| 636 | { |
| 637 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); |
| 638 | } |
| 639 | |
| 640 | /* call only when ep->mtx is held */ |
| 641 | static inline void ep_pm_stay_awake(struct epitem *epi) |
| 642 | { |
| 643 | struct wakeup_source *ws = ep_wakeup_source(epi); |
| 644 | |
| 645 | if (ws) |
| 646 | __pm_stay_awake(ws); |
| 647 | } |
| 648 | |
| 649 | static inline bool ep_has_wakeup_source(struct epitem *epi) |
| 650 | { |
| 651 | return rcu_access_pointer(epi->ws) ? true : false; |
| 652 | } |
| 653 | |
| 654 | /* call when ep->mtx cannot be held (ep_poll_callback) */ |
| 655 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) |
| 656 | { |
| 657 | struct wakeup_source *ws; |
| 658 | |
| 659 | rcu_read_lock(); |
| 660 | ws = rcu_dereference(epi->ws); |
| 661 | if (ws) |
| 662 | __pm_stay_awake(ws); |
| 663 | rcu_read_unlock(); |
| 664 | } |
| 665 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 666 | /** |
| 667 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for |
| 668 | * the scan code, to call f_op->poll(). Also allows for |
| 669 | * O(NumReady) performance. |
| 670 | * |
| 671 | * @ep: Pointer to the epoll private data structure. |
| 672 | * @sproc: Pointer to the scan callback. |
| 673 | * @priv: Private opaque data passed to the @sproc callback. |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 674 | * @depth: The current depth of recursive f_op->poll calls. |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 675 | * @ep_locked: caller already holds ep->mtx |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 676 | * |
| 677 | * Returns: The same integer error code returned by the @sproc callback. |
| 678 | */ |
| 679 | static int ep_scan_ready_list(struct eventpoll *ep, |
| 680 | int (*sproc)(struct eventpoll *, |
| 681 | struct list_head *, void *), |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 682 | void *priv, int depth, bool ep_locked) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 683 | { |
| 684 | int error, pwake = 0; |
| 685 | unsigned long flags; |
| 686 | struct epitem *epi, *nepi; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 687 | LIST_HEAD(txlist); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 688 | |
| 689 | /* |
| 690 | * We need to lock this because we could be hit by |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 691 | * eventpoll_release_file() and epoll_ctl(). |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 692 | */ |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 693 | |
| 694 | if (!ep_locked) |
| 695 | mutex_lock_nested(&ep->mtx, depth); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 696 | |
| 697 | /* |
| 698 | * Steal the ready list, and re-init the original one to the |
| 699 | * empty list. Also, set ep->ovflist to NULL so that events |
| 700 | * happening while looping w/out locks, are not lost. We cannot |
| 701 | * have the poll callback to queue directly on ep->rdllist, |
| 702 | * because we want the "sproc" callback to be able to do it |
| 703 | * in a lockless way. |
| 704 | */ |
| 705 | spin_lock_irqsave(&ep->lock, flags); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 706 | list_splice_init(&ep->rdllist, &txlist); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 707 | ep->ovflist = NULL; |
| 708 | spin_unlock_irqrestore(&ep->lock, flags); |
| 709 | |
| 710 | /* |
| 711 | * Now call the callback function. |
| 712 | */ |
| 713 | error = (*sproc)(ep, &txlist, priv); |
| 714 | |
| 715 | spin_lock_irqsave(&ep->lock, flags); |
| 716 | /* |
| 717 | * During the time we spent inside the "sproc" callback, some |
| 718 | * other events might have been queued by the poll callback. |
| 719 | * We re-insert them inside the main ready-list here. |
| 720 | */ |
| 721 | for (nepi = ep->ovflist; (epi = nepi) != NULL; |
| 722 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
| 723 | /* |
| 724 | * We need to check if the item is already in the list. |
| 725 | * During the "sproc" callback execution time, items are |
| 726 | * queued into ->ovflist but the "txlist" might already |
| 727 | * contain them, and the list_splice() below takes care of them. |
| 728 | */ |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 729 | if (!ep_is_linked(&epi->rdllink)) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 730 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 731 | ep_pm_stay_awake(epi); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 732 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 733 | } |
| 734 | /* |
| 735 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after |
| 736 | * releasing the lock, events will be queued in the normal way inside |
| 737 | * ep->rdllist. |
| 738 | */ |
| 739 | ep->ovflist = EP_UNACTIVE_PTR; |
| 740 | |
| 741 | /* |
| 742 | * Quickly re-inject items left on "txlist". |
| 743 | */ |
| 744 | list_splice(&txlist, &ep->rdllist); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 745 | __pm_relax(ep->ws); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 746 | |
| 747 | if (!list_empty(&ep->rdllist)) { |
| 748 | /* |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 749 | * Wake up (if active) both the eventpoll wait list and |
| 750 | * the ->poll() wait list (delayed after we release the lock). |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 751 | */ |
| 752 | if (waitqueue_active(&ep->wq)) |
| 753 | wake_up_locked(&ep->wq); |
| 754 | if (waitqueue_active(&ep->poll_wait)) |
| 755 | pwake++; |
| 756 | } |
| 757 | spin_unlock_irqrestore(&ep->lock, flags); |
| 758 | |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 759 | if (!ep_locked) |
| 760 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 761 | |
| 762 | /* We have to call this outside the lock */ |
| 763 | if (pwake) |
| 764 | ep_poll_safewake(&ep->poll_wait); |
| 765 | |
| 766 | return error; |
| 767 | } |
| 768 | |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 769 | static void epi_rcu_free(struct rcu_head *head) |
| 770 | { |
| 771 | struct epitem *epi = container_of(head, struct epitem, rcu); |
| 772 | kmem_cache_free(epi_cache, epi); |
| 773 | } |
| 774 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 776 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 777 | * all the associated resources. Must be called with "mtx" held. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 778 | */ |
| 779 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) |
| 780 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 781 | unsigned long flags; |
| 782 | struct file *file = epi->ffd.file; |
| 783 | |
| 784 | /* |
| 785 | * Removes poll wait queue hooks. We _have_ to do this without holding |
| 786 | * the "ep->lock" otherwise a deadlock might occur. This because of the |
| 787 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait |
| 788 | * queue head lock when unregistering the wait queue. The wakeup callback |
| 789 | * will run by holding the wait queue head lock and will call our callback |
| 790 | * that will try to get "ep->lock". |
| 791 | */ |
| 792 | ep_unregister_pollwait(ep, epi); |
| 793 | |
| 794 | /* Remove the current item from the list of epoll hooks */ |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 795 | spin_lock(&file->f_lock); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 796 | list_del_rcu(&epi->fllink); |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 797 | spin_unlock(&file->f_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 799 | rb_erase_cached(&epi->rbn, &ep->rbr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 801 | spin_lock_irqsave(&ep->lock, flags); |
| 802 | if (ep_is_linked(&epi->rdllink)) |
| 803 | list_del_init(&epi->rdllink); |
| 804 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 805 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 806 | wakeup_source_unregister(ep_wakeup_source(epi)); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 807 | /* |
| 808 | * At this point it is safe to free the eventpoll item. Use the union |
| 809 | * field epi->rcu, since we are trying to minimize the size of |
| 810 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by |
| 811 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make |
| 812 | * use of the rbn field. |
| 813 | */ |
| 814 | call_rcu(&epi->rcu, epi_rcu_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 816 | atomic_long_dec(&ep->user->epoll_watches); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 817 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 818 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 819 | } |
| 820 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 821 | static void ep_free(struct eventpoll *ep) |
| 822 | { |
| 823 | struct rb_node *rbp; |
| 824 | struct epitem *epi; |
| 825 | |
| 826 | /* We need to release all tasks waiting for these file */ |
| 827 | if (waitqueue_active(&ep->poll_wait)) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 828 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 829 | |
| 830 | /* |
| 831 | * We need to lock this because we could be hit by |
| 832 | * eventpoll_release_file() while we're freeing the "struct eventpoll". |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 833 | * 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] | 834 | * is on the way to be removed and no one has references to it |
| 835 | * anymore. The only hit might come from eventpoll_release_file() but |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 836 | * holding "epmutex" is sufficient here. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 837 | */ |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 838 | mutex_lock(&epmutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | |
| 840 | /* |
| 841 | * Walks through the whole tree by unregistering poll callbacks. |
| 842 | */ |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 843 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | epi = rb_entry(rbp, struct epitem, rbn); |
| 845 | |
| 846 | ep_unregister_pollwait(ep, epi); |
Eric Dumazet | 91cf5ab | 2013-09-11 14:24:06 -0700 | [diff] [blame] | 847 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | } |
| 849 | |
| 850 | /* |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 851 | * Walks through the whole tree by freeing each "struct epitem". At this |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | * 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] | 853 | * 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] | 854 | * us during this operation. So we can avoid the lock on "ep->lock". |
Eric Wong | ddf676c | 2013-04-30 15:27:40 -0700 | [diff] [blame] | 855 | * We do not need to lock ep->mtx, either, we only do it to prevent |
| 856 | * a lockdep warning. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 857 | */ |
Eric Wong | ddf676c | 2013-04-30 15:27:40 -0700 | [diff] [blame] | 858 | mutex_lock(&ep->mtx); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 859 | while ((rbp = rb_first_cached(&ep->rbr)) != NULL) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 860 | epi = rb_entry(rbp, struct epitem, rbn); |
| 861 | ep_remove(ep, epi); |
Eric Dumazet | 91cf5ab | 2013-09-11 14:24:06 -0700 | [diff] [blame] | 862 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | } |
Eric Wong | ddf676c | 2013-04-30 15:27:40 -0700 | [diff] [blame] | 864 | mutex_unlock(&ep->mtx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 865 | |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 866 | mutex_unlock(&epmutex); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 867 | mutex_destroy(&ep->mtx); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 868 | free_uid(ep->user); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 869 | wakeup_source_unregister(ep->ws); |
Davide Libenzi | f0ee9aa | 2007-05-15 01:40:57 -0700 | [diff] [blame] | 870 | kfree(ep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 871 | } |
| 872 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 873 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
| 874 | { |
| 875 | struct eventpoll *ep = file->private_data; |
| 876 | |
Davide Libenzi | f0ee9aa | 2007-05-15 01:40:57 -0700 | [diff] [blame] | 877 | if (ep) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 878 | ep_free(ep); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 879 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 880 | return 0; |
| 881 | } |
| 882 | |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 883 | static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt) |
| 884 | { |
| 885 | pt->_key = epi->event.events; |
| 886 | |
| 887 | return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events; |
| 888 | } |
| 889 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 890 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
| 891 | void *priv) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 892 | { |
| 893 | struct epitem *epi, *tmp; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 894 | poll_table pt; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 895 | |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 896 | init_poll_funcptr(&pt, NULL); |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 897 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 898 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 899 | if (ep_item_poll(epi, &pt)) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 900 | return POLLIN | POLLRDNORM; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 901 | else { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 902 | /* |
| 903 | * Item has been dropped into the ready list by the poll |
| 904 | * callback, but it's not actually ready, as far as |
| 905 | * caller requested events goes. We can remove it here. |
| 906 | */ |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 907 | __pm_relax(ep_wakeup_source(epi)); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 908 | list_del_init(&epi->rdllink); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 909 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 910 | } |
| 911 | |
| 912 | return 0; |
| 913 | } |
| 914 | |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 915 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
| 916 | poll_table *pt); |
| 917 | |
| 918 | struct readyevents_arg { |
| 919 | struct eventpoll *ep; |
| 920 | bool locked; |
| 921 | }; |
| 922 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 923 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) |
| 924 | { |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 925 | struct readyevents_arg *arg = priv; |
| 926 | |
| 927 | return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL, |
| 928 | call_nests + 1, arg->locked); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 929 | } |
| 930 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 931 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
| 932 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 933 | int pollflags; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 934 | struct eventpoll *ep = file->private_data; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 935 | struct readyevents_arg arg; |
| 936 | |
| 937 | /* |
| 938 | * During ep_insert() we already hold the ep->mtx for the tfile. |
| 939 | * Prevent re-aquisition. |
| 940 | */ |
| 941 | arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc); |
| 942 | arg.ep = ep; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 943 | |
| 944 | /* Insert inside our poll wait queue */ |
| 945 | poll_wait(file, &ep->poll_wait, wait); |
| 946 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 947 | /* |
| 948 | * Proceed to find out if wanted events are really available inside |
| 949 | * the ready list. This need to be done under ep_call_nested() |
| 950 | * supervision, since the call to f_op->poll() done on listed files |
| 951 | * could re-enter here. |
| 952 | */ |
| 953 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 954 | ep_poll_readyevents_proc, &arg, ep, current); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 955 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 956 | return pollflags != -1 ? pollflags : 0; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 957 | } |
| 958 | |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 959 | #ifdef CONFIG_PROC_FS |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 960 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 961 | { |
| 962 | struct eventpoll *ep = f->private_data; |
| 963 | struct rb_node *rbp; |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 964 | |
| 965 | mutex_lock(&ep->mtx); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 966 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 967 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
Cyrill Gorcunov | 77493f0 | 2017-07-12 14:34:25 -0700 | [diff] [blame] | 968 | struct inode *inode = file_inode(epi->ffd.file); |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 969 | |
Cyrill Gorcunov | 77493f0 | 2017-07-12 14:34:25 -0700 | [diff] [blame] | 970 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
| 971 | " pos:%lli ino:%lx sdev:%x\n", |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 972 | epi->ffd.fd, epi->event.events, |
Cyrill Gorcunov | 77493f0 | 2017-07-12 14:34:25 -0700 | [diff] [blame] | 973 | (long long)epi->event.data, |
| 974 | (long long)epi->ffd.file->f_pos, |
| 975 | inode->i_ino, inode->i_sb->s_dev); |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 976 | if (seq_has_overflowed(m)) |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 977 | break; |
| 978 | } |
| 979 | mutex_unlock(&ep->mtx); |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 980 | } |
| 981 | #endif |
| 982 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 983 | /* File callbacks that implement the eventpoll file behaviour */ |
| 984 | static const struct file_operations eventpoll_fops = { |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 985 | #ifdef CONFIG_PROC_FS |
| 986 | .show_fdinfo = ep_show_fdinfo, |
| 987 | #endif |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 988 | .release = ep_eventpoll_release, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 989 | .poll = ep_eventpoll_poll, |
| 990 | .llseek = noop_llseek, |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 991 | }; |
| 992 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 993 | /* |
| 994 | * This is called from eventpoll_release() to unlink files from the eventpoll |
| 995 | * interface. We need to have this facility to cleanup correctly files that are |
| 996 | * closed without being removed from the eventpoll interface. |
| 997 | */ |
| 998 | void eventpoll_release_file(struct file *file) |
| 999 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1000 | struct eventpoll *ep; |
Konstantin Khlebnikov | ebe0618 | 2014-06-17 06:58:05 +0400 | [diff] [blame] | 1001 | struct epitem *epi, *next; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1002 | |
| 1003 | /* |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 1004 | * 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] | 1005 | * necessary. It is not necessary because we're in the "struct file" |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1006 | * cleanup path, and this means that no one is using this file anymore. |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1007 | * 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] | 1008 | * point, the file counter already went to zero and fget() would fail. |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1009 | * 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] | 1010 | * will correctly serialize the operation. We do need to acquire |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1011 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1012 | * from anywhere but ep_free(). |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 1013 | * |
| 1014 | * 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] | 1015 | */ |
| 1016 | mutex_lock(&epmutex); |
Konstantin Khlebnikov | ebe0618 | 2014-06-17 06:58:05 +0400 | [diff] [blame] | 1017 | list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1018 | ep = epi->ep; |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 1019 | mutex_lock_nested(&ep->mtx, 0); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1020 | ep_remove(ep, epi); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1021 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1022 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1023 | mutex_unlock(&epmutex); |
| 1024 | } |
| 1025 | |
| 1026 | static int ep_alloc(struct eventpoll **pep) |
| 1027 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1028 | int error; |
| 1029 | struct user_struct *user; |
| 1030 | struct eventpoll *ep; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1031 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1032 | user = get_current_user(); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1033 | error = -ENOMEM; |
| 1034 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); |
| 1035 | if (unlikely(!ep)) |
| 1036 | goto free_uid; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1037 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1038 | spin_lock_init(&ep->lock); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1039 | mutex_init(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1040 | init_waitqueue_head(&ep->wq); |
| 1041 | init_waitqueue_head(&ep->poll_wait); |
| 1042 | INIT_LIST_HEAD(&ep->rdllist); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1043 | ep->rbr = RB_ROOT_CACHED; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1044 | ep->ovflist = EP_UNACTIVE_PTR; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1045 | ep->user = user; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1046 | |
| 1047 | *pep = ep; |
| 1048 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1049 | return 0; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1050 | |
| 1051 | free_uid: |
| 1052 | free_uid(user); |
| 1053 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1054 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1055 | |
| 1056 | /* |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1057 | * Search the file inside the eventpoll tree. The RB tree operations |
| 1058 | * are protected by the "mtx" mutex, and ep_find() must be called with |
| 1059 | * "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | */ |
| 1061 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) |
| 1062 | { |
| 1063 | int kcmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | struct rb_node *rbp; |
| 1065 | struct epitem *epi, *epir = NULL; |
| 1066 | struct epoll_filefd ffd; |
| 1067 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1068 | ep_set_ffd(&ffd, file, fd); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1069 | for (rbp = ep->rbr.rb_root.rb_node; rbp; ) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1070 | epi = rb_entry(rbp, struct epitem, rbn); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1071 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | if (kcmp > 0) |
| 1073 | rbp = rbp->rb_right; |
| 1074 | else if (kcmp < 0) |
| 1075 | rbp = rbp->rb_left; |
| 1076 | else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | epir = epi; |
| 1078 | break; |
| 1079 | } |
| 1080 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1081 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1082 | return epir; |
| 1083 | } |
| 1084 | |
Cyrill Gorcunov | 92ef6da | 2017-07-12 14:34:31 -0700 | [diff] [blame] | 1085 | #ifdef CONFIG_CHECKPOINT_RESTORE |
Cyrill Gorcunov | 0791e36 | 2017-07-12 14:34:28 -0700 | [diff] [blame] | 1086 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
| 1087 | { |
| 1088 | struct rb_node *rbp; |
| 1089 | struct epitem *epi; |
| 1090 | |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1091 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
Cyrill Gorcunov | 0791e36 | 2017-07-12 14:34:28 -0700 | [diff] [blame] | 1092 | epi = rb_entry(rbp, struct epitem, rbn); |
| 1093 | if (epi->ffd.fd == tfd) { |
| 1094 | if (toff == 0) |
| 1095 | return epi; |
| 1096 | else |
| 1097 | toff--; |
| 1098 | } |
| 1099 | cond_resched(); |
| 1100 | } |
| 1101 | |
| 1102 | return NULL; |
| 1103 | } |
| 1104 | |
| 1105 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, |
| 1106 | unsigned long toff) |
| 1107 | { |
| 1108 | struct file *file_raw; |
| 1109 | struct eventpoll *ep; |
| 1110 | struct epitem *epi; |
| 1111 | |
| 1112 | if (!is_file_epoll(file)) |
| 1113 | return ERR_PTR(-EINVAL); |
| 1114 | |
| 1115 | ep = file->private_data; |
| 1116 | |
| 1117 | mutex_lock(&ep->mtx); |
| 1118 | epi = ep_find_tfd(ep, tfd, toff); |
| 1119 | if (epi) |
| 1120 | file_raw = epi->ffd.file; |
| 1121 | else |
| 1122 | file_raw = ERR_PTR(-ENOENT); |
| 1123 | mutex_unlock(&ep->mtx); |
| 1124 | |
| 1125 | return file_raw; |
| 1126 | } |
Cyrill Gorcunov | 92ef6da | 2017-07-12 14:34:31 -0700 | [diff] [blame] | 1127 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
Cyrill Gorcunov | 0791e36 | 2017-07-12 14:34:28 -0700 | [diff] [blame] | 1128 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1130 | * This is the callback that is passed to the wait queue wakeup |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 1131 | * mechanism. It is called by the stored file descriptors when they |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1132 | * have events to report. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1134 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1135 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1136 | int pwake = 0; |
| 1137 | unsigned long flags; |
| 1138 | struct epitem *epi = ep_item_from_wait(wait); |
| 1139 | struct eventpoll *ep = epi->ep; |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1140 | int ewake = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1141 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1142 | spin_lock_irqsave(&ep->lock, flags); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1143 | |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 1144 | ep_set_busy_poll_napi_id(epi); |
| 1145 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1146 | /* |
| 1147 | * If the event mask does not contain any poll(2) event, we consider the |
| 1148 | * descriptor to be disabled. This condition is likely the effect of the |
| 1149 | * EPOLLONESHOT bit that disables the descriptor when an event is received, |
| 1150 | * until the next EPOLL_CTL_MOD will be issued. |
| 1151 | */ |
| 1152 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1153 | goto out_unlock; |
| 1154 | |
| 1155 | /* |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 1156 | * Check the events coming with the callback. At this stage, not |
| 1157 | * every device reports the events in the "key" parameter of the |
| 1158 | * callback. We need to be able to handle both cases here, hence the |
| 1159 | * test for "key" != NULL before the event match test. |
| 1160 | */ |
| 1161 | if (key && !((unsigned long) key & epi->event.events)) |
| 1162 | goto out_unlock; |
| 1163 | |
| 1164 | /* |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 1165 | * If we are transferring events to userspace, we can hold no locks |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1166 | * (because we're accessing user memory, and because of linux f_op->poll() |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 1167 | * semantics). All the events that happen during that period of time are |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1168 | * chained in ep->ovflist and requeued later on. |
| 1169 | */ |
Davidlohr Bueso | fad3ec7 | 2019-01-03 15:27:09 -0800 | [diff] [blame] | 1170 | if (ep->ovflist != EP_UNACTIVE_PTR) { |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1171 | if (epi->next == EP_UNACTIVE_PTR) { |
| 1172 | epi->next = ep->ovflist; |
| 1173 | ep->ovflist = epi; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1174 | if (epi->ws) { |
| 1175 | /* |
| 1176 | * Activate ep->ws since epi->ws may get |
| 1177 | * deactivated at any time. |
| 1178 | */ |
| 1179 | __pm_stay_awake(ep->ws); |
| 1180 | } |
| 1181 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1182 | } |
| 1183 | goto out_unlock; |
| 1184 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1185 | |
| 1186 | /* If this file is already in the ready list we exit soon */ |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1187 | if (!ep_is_linked(&epi->rdllink)) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1188 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1189 | ep_pm_stay_awake_rcu(epi); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1190 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1191 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1192 | /* |
| 1193 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() |
| 1194 | * wait list. |
| 1195 | */ |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1196 | if (waitqueue_active(&ep->wq)) { |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 1197 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
| 1198 | !((unsigned long)key & POLLFREE)) { |
| 1199 | switch ((unsigned long)key & EPOLLINOUT_BITS) { |
| 1200 | case POLLIN: |
| 1201 | if (epi->event.events & POLLIN) |
| 1202 | ewake = 1; |
| 1203 | break; |
| 1204 | case POLLOUT: |
| 1205 | if (epi->event.events & POLLOUT) |
| 1206 | ewake = 1; |
| 1207 | break; |
| 1208 | case 0: |
| 1209 | ewake = 1; |
| 1210 | break; |
| 1211 | } |
| 1212 | } |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1213 | wake_up_locked(&ep->wq); |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1214 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1215 | if (waitqueue_active(&ep->poll_wait)) |
| 1216 | pwake++; |
| 1217 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1218 | out_unlock: |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1219 | spin_unlock_irqrestore(&ep->lock, flags); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1220 | |
| 1221 | /* We have to call this outside the lock */ |
| 1222 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1223 | ep_poll_safewake(&ep->poll_wait); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1224 | |
Oleg Nesterov | 138e4ad | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 1225 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
| 1226 | ewake = 1; |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1227 | |
Oleg Nesterov | 138e4ad | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 1228 | if ((unsigned long)key & POLLFREE) { |
| 1229 | /* |
| 1230 | * If we race with ep_remove_wait_queue() it can miss |
| 1231 | * ->whead = NULL and do another remove_wait_queue() after |
| 1232 | * us, so we can't use __remove_wait_queue(). |
| 1233 | */ |
| 1234 | list_del_init(&wait->entry); |
| 1235 | /* |
| 1236 | * ->whead != NULL protects us from the race with ep_free() |
| 1237 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock |
| 1238 | * held by the caller. Once we nullify it, nothing protects |
| 1239 | * ep/epi or even wait. |
| 1240 | */ |
| 1241 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); |
| 1242 | } |
| 1243 | |
| 1244 | return ewake; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | } |
| 1246 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1247 | /* |
| 1248 | * This is the callback that is used to add our wait queue to the |
| 1249 | * target file wakeup lists. |
| 1250 | */ |
| 1251 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
| 1252 | poll_table *pt) |
| 1253 | { |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1254 | struct epitem *epi = ep_item_from_epqueue(pt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1255 | struct eppoll_entry *pwq; |
| 1256 | |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 1257 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1258 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
| 1259 | pwq->whead = whead; |
| 1260 | pwq->base = epi; |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1261 | if (epi->event.events & EPOLLEXCLUSIVE) |
| 1262 | add_wait_queue_exclusive(whead, &pwq->wait); |
| 1263 | else |
| 1264 | add_wait_queue(whead, &pwq->wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1265 | list_add_tail(&pwq->llink, &epi->pwqlist); |
| 1266 | epi->nwait++; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1267 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1268 | /* We have to signal that an error occurred */ |
| 1269 | epi->nwait = -1; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1270 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1271 | } |
| 1272 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1273 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
| 1274 | { |
| 1275 | int kcmp; |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1276 | struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | struct epitem *epic; |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1278 | bool leftmost = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1279 | |
| 1280 | while (*p) { |
| 1281 | parent = *p; |
| 1282 | epic = rb_entry(parent, struct epitem, rbn); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1283 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1284 | if (kcmp > 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1285 | p = &parent->rb_right; |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1286 | leftmost = false; |
| 1287 | } else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1288 | p = &parent->rb_left; |
| 1289 | } |
| 1290 | rb_link_node(&epi->rbn, parent, p); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1291 | rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1292 | } |
| 1293 | |
Andrew Morton | a80a6b8 | 2012-11-08 15:53:35 -0800 | [diff] [blame] | 1294 | |
| 1295 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1296 | #define PATH_ARR_SIZE 5 |
| 1297 | /* |
| 1298 | * These are the number paths of length 1 to 5, that we are allowing to emanate |
| 1299 | * from a single file of interest. For example, we allow 1000 paths of length |
| 1300 | * 1, to emanate from each file of interest. This essentially represents the |
| 1301 | * potential wakeup paths, which need to be limited in order to avoid massive |
| 1302 | * uncontrolled wakeup storms. The common use case should be a single ep which |
| 1303 | * is connected to n file sources. In this case each file source has 1 path |
| 1304 | * of length 1. Thus, the numbers below should be more than sufficient. These |
| 1305 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify |
| 1306 | * and delete can't add additional paths. Protected by the epmutex. |
| 1307 | */ |
| 1308 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; |
| 1309 | static int path_count[PATH_ARR_SIZE]; |
| 1310 | |
| 1311 | static int path_count_inc(int nests) |
| 1312 | { |
Jason Baron | 93dc610 | 2012-03-16 16:34:03 -0400 | [diff] [blame] | 1313 | /* Allow an arbitrary number of depth 1 paths */ |
| 1314 | if (nests == 0) |
| 1315 | return 0; |
| 1316 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1317 | if (++path_count[nests] > path_limits[nests]) |
| 1318 | return -1; |
| 1319 | return 0; |
| 1320 | } |
| 1321 | |
| 1322 | static void path_count_init(void) |
| 1323 | { |
| 1324 | int i; |
| 1325 | |
| 1326 | for (i = 0; i < PATH_ARR_SIZE; i++) |
| 1327 | path_count[i] = 0; |
| 1328 | } |
| 1329 | |
| 1330 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) |
| 1331 | { |
| 1332 | int error = 0; |
| 1333 | struct file *file = priv; |
| 1334 | struct file *child_file; |
| 1335 | struct epitem *epi; |
| 1336 | |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1337 | /* CTL_DEL can remove links here, but that can't increase our count */ |
| 1338 | rcu_read_lock(); |
| 1339 | list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1340 | child_file = epi->ep->file; |
| 1341 | if (is_file_epoll(child_file)) { |
| 1342 | if (list_empty(&child_file->f_ep_links)) { |
| 1343 | if (path_count_inc(call_nests)) { |
| 1344 | error = -1; |
| 1345 | break; |
| 1346 | } |
| 1347 | } else { |
| 1348 | error = ep_call_nested(&poll_loop_ncalls, |
| 1349 | EP_MAX_NESTS, |
| 1350 | reverse_path_check_proc, |
| 1351 | child_file, child_file, |
| 1352 | current); |
| 1353 | } |
| 1354 | if (error != 0) |
| 1355 | break; |
| 1356 | } else { |
| 1357 | printk(KERN_ERR "reverse_path_check_proc: " |
| 1358 | "file is not an ep!\n"); |
| 1359 | } |
| 1360 | } |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1361 | rcu_read_unlock(); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1362 | return error; |
| 1363 | } |
| 1364 | |
| 1365 | /** |
| 1366 | * reverse_path_check - The tfile_check_list is list of file *, which have |
| 1367 | * links that are proposed to be newly added. We need to |
| 1368 | * make sure that those added links don't add too many |
| 1369 | * paths such that we will spend all our time waking up |
| 1370 | * eventpoll objects. |
| 1371 | * |
| 1372 | * Returns: Returns zero if the proposed links don't create too many paths, |
| 1373 | * -1 otherwise. |
| 1374 | */ |
| 1375 | static int reverse_path_check(void) |
| 1376 | { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1377 | int error = 0; |
| 1378 | struct file *current_file; |
| 1379 | |
| 1380 | /* let's call this for all tfiles */ |
| 1381 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1382 | path_count_init(); |
| 1383 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
| 1384 | reverse_path_check_proc, current_file, |
| 1385 | current_file, current); |
| 1386 | if (error) |
| 1387 | break; |
| 1388 | } |
| 1389 | return error; |
| 1390 | } |
| 1391 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1392 | static int ep_create_wakeup_source(struct epitem *epi) |
| 1393 | { |
| 1394 | const char *name; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1395 | struct wakeup_source *ws; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1396 | |
| 1397 | if (!epi->ep->ws) { |
| 1398 | epi->ep->ws = wakeup_source_register("eventpoll"); |
| 1399 | if (!epi->ep->ws) |
| 1400 | return -ENOMEM; |
| 1401 | } |
| 1402 | |
| 1403 | name = epi->ffd.file->f_path.dentry->d_name.name; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1404 | ws = wakeup_source_register(name); |
| 1405 | |
| 1406 | if (!ws) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1407 | return -ENOMEM; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1408 | rcu_assign_pointer(epi->ws, ws); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1409 | |
| 1410 | return 0; |
| 1411 | } |
| 1412 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1413 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
| 1414 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1415 | { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1416 | struct wakeup_source *ws = ep_wakeup_source(epi); |
| 1417 | |
Eric Wong | d6d67e7 | 2013-04-30 15:27:43 -0700 | [diff] [blame] | 1418 | RCU_INIT_POINTER(epi->ws, NULL); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1419 | |
| 1420 | /* |
| 1421 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is |
| 1422 | * used internally by wakeup_source_remove, too (called by |
| 1423 | * wakeup_source_unregister), so we cannot use call_rcu |
| 1424 | */ |
| 1425 | synchronize_rcu(); |
| 1426 | wakeup_source_unregister(ws); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1427 | } |
| 1428 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1429 | /* |
| 1430 | * Must be called with "mtx" held. |
| 1431 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1432 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1433 | struct file *tfile, int fd, int full_check) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1434 | { |
| 1435 | int error, revents, pwake = 0; |
| 1436 | unsigned long flags; |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1437 | long user_watches; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1438 | struct epitem *epi; |
| 1439 | struct ep_pqueue epq; |
| 1440 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1441 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
| 1442 | if (unlikely(user_watches >= max_user_watches)) |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1443 | return -ENOSPC; |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 1444 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1445 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1446 | |
| 1447 | /* Item initialization follow here ... */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1448 | INIT_LIST_HEAD(&epi->rdllink); |
| 1449 | INIT_LIST_HEAD(&epi->fllink); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1450 | INIT_LIST_HEAD(&epi->pwqlist); |
| 1451 | epi->ep = ep; |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1452 | ep_set_ffd(&epi->ffd, tfile, fd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | epi->event = *event; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | epi->nwait = 0; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1455 | epi->next = EP_UNACTIVE_PTR; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1456 | if (epi->event.events & EPOLLWAKEUP) { |
| 1457 | error = ep_create_wakeup_source(epi); |
| 1458 | if (error) |
| 1459 | goto error_create_wakeup_source; |
| 1460 | } else { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1461 | RCU_INIT_POINTER(epi->ws, NULL); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1462 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1463 | |
| 1464 | /* Initialize the poll table using the queue callback */ |
| 1465 | epq.epi = epi; |
| 1466 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); |
| 1467 | |
| 1468 | /* |
| 1469 | * Attach the item to the poll hooks and get current event bits. |
| 1470 | * We can safely use the file* here because its usage count has |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1471 | * been increased by the caller of this function. Note that after |
| 1472 | * this operation completes, the poll callback can start hitting |
| 1473 | * the new item. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1474 | */ |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 1475 | revents = ep_item_poll(epi, &epq.pt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1476 | |
| 1477 | /* |
| 1478 | * We have to check if something went wrong during the poll wait queue |
| 1479 | * install process. Namely an allocation for a wait queue failed due |
| 1480 | * high memory pressure. |
| 1481 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1482 | error = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1483 | if (epi->nwait < 0) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1484 | goto error_unregister; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1485 | |
| 1486 | /* 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] | 1487 | spin_lock(&tfile->f_lock); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1488 | list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links); |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 1489 | spin_unlock(&tfile->f_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1490 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1491 | /* |
| 1492 | * Add the current item to the RB tree. All RB tree operations are |
| 1493 | * protected by "mtx", and ep_insert() is called with "mtx" held. |
| 1494 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | ep_rbtree_insert(ep, epi); |
| 1496 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1497 | /* now check if we've created too many backpaths */ |
| 1498 | error = -EINVAL; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1499 | if (full_check && reverse_path_check()) |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1500 | goto error_remove_epi; |
| 1501 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1502 | /* We have to drop the new item inside our item list to keep track of it */ |
| 1503 | spin_lock_irqsave(&ep->lock, flags); |
| 1504 | |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 1505 | /* record NAPI ID of new item if present */ |
| 1506 | ep_set_busy_poll_napi_id(epi); |
| 1507 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | /* 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] | 1509 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1511 | ep_pm_stay_awake(epi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1512 | |
| 1513 | /* Notify waiting tasks that events are available */ |
| 1514 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1515 | wake_up_locked(&ep->wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1516 | if (waitqueue_active(&ep->poll_wait)) |
| 1517 | pwake++; |
| 1518 | } |
| 1519 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1520 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1521 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1522 | atomic_long_inc(&ep->user->epoll_watches); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1523 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1524 | /* We have to call this outside the lock */ |
| 1525 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1526 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1528 | return 0; |
| 1529 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1530 | error_remove_epi: |
| 1531 | spin_lock(&tfile->f_lock); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1532 | list_del_rcu(&epi->fllink); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1533 | spin_unlock(&tfile->f_lock); |
| 1534 | |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1535 | rb_erase_cached(&epi->rbn, &ep->rbr); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1536 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1537 | error_unregister: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1538 | ep_unregister_pollwait(ep, epi); |
| 1539 | |
| 1540 | /* |
| 1541 | * 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] | 1542 | * allocated wait queue. Note that we don't care about the ep->ovflist |
| 1543 | * list, since that is used/cleaned only inside a section bound by "mtx". |
| 1544 | * And ep_insert() is called with "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1545 | */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1546 | spin_lock_irqsave(&ep->lock, flags); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1547 | if (ep_is_linked(&epi->rdllink)) |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 1548 | list_del_init(&epi->rdllink); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1549 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1550 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1551 | wakeup_source_unregister(ep_wakeup_source(epi)); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1552 | |
| 1553 | error_create_wakeup_source: |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1554 | kmem_cache_free(epi_cache, epi); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1555 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1556 | return error; |
| 1557 | } |
| 1558 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1559 | /* |
| 1560 | * 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] | 1561 | * 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] | 1562 | */ |
| 1563 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) |
| 1564 | { |
| 1565 | int pwake = 0; |
| 1566 | unsigned int revents; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 1567 | poll_table pt; |
| 1568 | |
| 1569 | init_poll_funcptr(&pt, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1570 | |
| 1571 | /* |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1572 | * Set the new event interest mask before calling f_op->poll(); |
| 1573 | * otherwise we might miss an event that happens between the |
| 1574 | * f_op->poll() call and the new event set registering. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1575 | */ |
Eric Wong | 128dd17 | 2013-01-01 21:20:27 +0000 | [diff] [blame] | 1576 | epi->event.events = event->events; /* need barrier below */ |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1577 | epi->event.data = event->data; /* protected by mtx */ |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1578 | if (epi->event.events & EPOLLWAKEUP) { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1579 | if (!ep_has_wakeup_source(epi)) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1580 | ep_create_wakeup_source(epi); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1581 | } else if (ep_has_wakeup_source(epi)) { |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1582 | ep_destroy_wakeup_source(epi); |
| 1583 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1584 | |
| 1585 | /* |
Eric Wong | 128dd17 | 2013-01-01 21:20:27 +0000 | [diff] [blame] | 1586 | * The following barrier has two effects: |
| 1587 | * |
| 1588 | * 1) Flush epi changes above to other CPUs. This ensures |
| 1589 | * we do not miss events from ep_poll_callback if an |
| 1590 | * event occurs immediately after we call f_op->poll(). |
| 1591 | * We need this because we did not take ep->lock while |
| 1592 | * changing epi above (but ep_poll_callback does take |
| 1593 | * ep->lock). |
| 1594 | * |
| 1595 | * 2) We also need to ensure we do not miss _past_ events |
| 1596 | * when calling f_op->poll(). This barrier also |
| 1597 | * pairs with the barrier in wq_has_sleeper (see |
| 1598 | * comments for wq_has_sleeper). |
| 1599 | * |
| 1600 | * This barrier will now guarantee ep_poll_callback or f_op->poll |
| 1601 | * (or both) will notice the readiness of an item. |
| 1602 | */ |
| 1603 | smp_mb(); |
| 1604 | |
| 1605 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1606 | * Get current event bits. We can safely use the file* here because |
| 1607 | * its usage count has been increased by the caller of this function. |
| 1608 | */ |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 1609 | revents = ep_item_poll(epi, &pt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1610 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1611 | /* |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1612 | * 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] | 1613 | * list, push it inside. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1615 | if (revents & event->events) { |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1616 | spin_lock_irq(&ep->lock); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1617 | if (!ep_is_linked(&epi->rdllink)) { |
| 1618 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1619 | ep_pm_stay_awake(epi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1620 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1621 | /* Notify waiting tasks that events are available */ |
| 1622 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1623 | wake_up_locked(&ep->wq); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1624 | if (waitqueue_active(&ep->poll_wait)) |
| 1625 | pwake++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | } |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1627 | spin_unlock_irq(&ep->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1628 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1629 | |
| 1630 | /* We have to call this outside the lock */ |
| 1631 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1632 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1633 | |
| 1634 | return 0; |
| 1635 | } |
| 1636 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1637 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
| 1638 | void *priv) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1639 | { |
| 1640 | struct ep_send_events_data *esed = priv; |
| 1641 | int eventcnt; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1642 | unsigned int revents; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1643 | struct epitem *epi; |
| 1644 | struct epoll_event __user *uevent; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1645 | struct wakeup_source *ws; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 1646 | poll_table pt; |
| 1647 | |
| 1648 | init_poll_funcptr(&pt, NULL); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1649 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1650 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1651 | * We can loop without lock because we are passed a task private list. |
| 1652 | * Items cannot vanish during the loop because ep_scan_ready_list() is |
| 1653 | * holding "mtx" during this call. |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1654 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1655 | for (eventcnt = 0, uevent = esed->events; |
| 1656 | !list_empty(head) && eventcnt < esed->maxevents;) { |
| 1657 | epi = list_first_entry(head, struct epitem, rdllink); |
| 1658 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1659 | /* |
| 1660 | * Activate ep->ws before deactivating epi->ws to prevent |
| 1661 | * triggering auto-suspend here (in case we reactive epi->ws |
| 1662 | * below). |
| 1663 | * |
| 1664 | * This could be rearranged to delay the deactivation of epi->ws |
| 1665 | * instead, but then epi->ws would temporarily be out of sync |
| 1666 | * with ep_is_linked(). |
| 1667 | */ |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1668 | ws = ep_wakeup_source(epi); |
| 1669 | if (ws) { |
| 1670 | if (ws->active) |
| 1671 | __pm_stay_awake(ep->ws); |
| 1672 | __pm_relax(ws); |
| 1673 | } |
| 1674 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1675 | list_del_init(&epi->rdllink); |
| 1676 | |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 1677 | revents = ep_item_poll(epi, &pt); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1678 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1679 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1680 | * If the event mask intersect the caller-requested one, |
| 1681 | * deliver the event to userspace. Again, ep_scan_ready_list() |
| 1682 | * is holding "mtx", so no operations coming from userspace |
| 1683 | * can change the item. |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1684 | */ |
| 1685 | if (revents) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1686 | if (__put_user(revents, &uevent->events) || |
Tony Battersby | d030588 | 2009-03-31 15:24:14 -0700 | [diff] [blame] | 1687 | __put_user(epi->event.data, &uevent->data)) { |
| 1688 | list_add(&epi->rdllink, head); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1689 | ep_pm_stay_awake(epi); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1690 | return eventcnt ? eventcnt : -EFAULT; |
Tony Battersby | d030588 | 2009-03-31 15:24:14 -0700 | [diff] [blame] | 1691 | } |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1692 | eventcnt++; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1693 | uevent++; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1694 | if (epi->event.events & EPOLLONESHOT) |
| 1695 | epi->event.events &= EP_PRIVATE_BITS; |
| 1696 | else if (!(epi->event.events & EPOLLET)) { |
| 1697 | /* |
| 1698 | * If this file has been added with Level |
| 1699 | * Trigger mode, we need to insert back inside |
| 1700 | * the ready list, so that the next call to |
| 1701 | * epoll_wait() will check again the events |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1702 | * availability. At this point, no one can insert |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1703 | * into ep->rdllist besides us. The epoll_ctl() |
| 1704 | * callers are locked out by |
| 1705 | * ep_scan_ready_list() holding "mtx" and the |
| 1706 | * poll callback will queue them in ep->ovflist. |
| 1707 | */ |
| 1708 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1709 | ep_pm_stay_awake(epi); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1710 | } |
| 1711 | } |
| 1712 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1713 | |
| 1714 | return eventcnt; |
| 1715 | } |
| 1716 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1717 | static int ep_send_events(struct eventpoll *ep, |
| 1718 | struct epoll_event __user *events, int maxevents) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1719 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1720 | struct ep_send_events_data esed; |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 1721 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1722 | esed.maxevents = maxevents; |
| 1723 | esed.events = events; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1724 | |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1725 | return ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1726 | } |
| 1727 | |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1728 | static inline struct timespec64 ep_set_mstimeout(long ms) |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1729 | { |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1730 | struct timespec64 now, ts = { |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1731 | .tv_sec = ms / MSEC_PER_SEC, |
| 1732 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), |
| 1733 | }; |
| 1734 | |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1735 | ktime_get_ts64(&now); |
| 1736 | return timespec64_add_safe(now, ts); |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1737 | } |
| 1738 | |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1739 | /** |
| 1740 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied |
| 1741 | * event buffer. |
| 1742 | * |
| 1743 | * @ep: Pointer to the eventpoll context. |
| 1744 | * @events: Pointer to the userspace buffer where the ready events should be |
| 1745 | * stored. |
| 1746 | * @maxevents: Size (in terms of number of events) of the caller event buffer. |
| 1747 | * @timeout: Maximum timeout for the ready events fetch operation, in |
| 1748 | * milliseconds. If the @timeout is zero, the function will not block, |
| 1749 | * while if the @timeout is less than zero, the function will block |
| 1750 | * until at least one event has been retrieved (or an error |
| 1751 | * occurred). |
| 1752 | * |
| 1753 | * Returns: Returns the number of ready events which have been fetched, or an |
| 1754 | * error code, in case of error. |
| 1755 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1756 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
| 1757 | int maxevents, long timeout) |
| 1758 | { |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1759 | int res = 0, eavail, timed_out = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1760 | unsigned long flags; |
John Stultz | da8b44d | 2016-03-17 14:20:51 -0700 | [diff] [blame] | 1761 | u64 slack = 0; |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1762 | wait_queue_entry_t wait; |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1763 | ktime_t expires, *to = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1764 | |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1765 | if (timeout > 0) { |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1766 | struct timespec64 end_time = ep_set_mstimeout(timeout); |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1767 | |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1768 | slack = select_estimate_accuracy(&end_time); |
| 1769 | to = &expires; |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1770 | *to = timespec64_to_ktime(end_time); |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1771 | } else if (timeout == 0) { |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1772 | /* |
| 1773 | * Avoid the unnecessary trip to the wait queue loop, if the |
| 1774 | * caller specified a non blocking operation. |
| 1775 | */ |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1776 | timed_out = 1; |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1777 | spin_lock_irqsave(&ep->lock, flags); |
| 1778 | goto check_events; |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1779 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1780 | |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1781 | fetch_events: |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 1782 | |
| 1783 | if (!ep_events_available(ep)) |
| 1784 | ep_busy_loop(ep, timed_out); |
| 1785 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1786 | spin_lock_irqsave(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1787 | |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1788 | if (!ep_events_available(ep)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1789 | /* |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 1790 | * Busy poll timed out. Drop NAPI ID for now, we can add |
| 1791 | * it back in when we have moved a socket with a valid NAPI |
| 1792 | * ID onto the ready list. |
| 1793 | */ |
| 1794 | ep_reset_busy_poll_napi_id(ep); |
| 1795 | |
| 1796 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1797 | * We don't have any available event to return to the caller. |
| 1798 | * We need to sleep here, and we will be wake up by |
| 1799 | * ep_poll_callback() when events will become available. |
| 1800 | */ |
| 1801 | init_waitqueue_entry(&wait, current); |
Changli Gao | a93d2f1 | 2010-05-07 14:33:26 +0800 | [diff] [blame] | 1802 | __add_wait_queue_exclusive(&ep->wq, &wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1803 | |
| 1804 | for (;;) { |
| 1805 | /* |
| 1806 | * We don't want to sleep if the ep_poll_callback() sends us |
| 1807 | * a wakeup in between. That's why we set the task state |
| 1808 | * to TASK_INTERRUPTIBLE before doing the checks. |
| 1809 | */ |
| 1810 | set_current_state(TASK_INTERRUPTIBLE); |
David Rientjes | c257a34 | 2017-07-10 15:52:33 -0700 | [diff] [blame] | 1811 | /* |
| 1812 | * Always short-circuit for fatal signals to allow |
| 1813 | * threads to make a timely exit without the chance of |
| 1814 | * finding more events available and fetching |
| 1815 | * repeatedly. |
| 1816 | */ |
| 1817 | if (fatal_signal_pending(current)) { |
| 1818 | res = -EINTR; |
| 1819 | break; |
| 1820 | } |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1821 | if (ep_events_available(ep) || timed_out) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1822 | break; |
| 1823 | if (signal_pending(current)) { |
| 1824 | res = -EINTR; |
| 1825 | break; |
| 1826 | } |
| 1827 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1828 | spin_unlock_irqrestore(&ep->lock, flags); |
Rafael J. Wysocki | c511851 | 2013-10-29 13:12:56 +0100 | [diff] [blame] | 1829 | if (!schedule_hrtimeout_range(to, slack, HRTIMER_MODE_ABS)) |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1830 | timed_out = 1; |
| 1831 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1832 | spin_lock_irqsave(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1833 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1834 | |
Davidlohr Bueso | 4d5755b | 2015-02-13 14:39:08 -0800 | [diff] [blame] | 1835 | __remove_wait_queue(&ep->wq, &wait); |
| 1836 | __set_current_state(TASK_RUNNING); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1837 | } |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1838 | check_events: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1839 | /* Is it worth to try to dig for events ? */ |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1840 | eavail = ep_events_available(ep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1841 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1842 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1843 | |
| 1844 | /* |
| 1845 | * Try to transfer events to user space. In case we get 0 events and |
| 1846 | * there's still timeout left over, we go trying again in search of |
| 1847 | * more luck. |
| 1848 | */ |
| 1849 | if (!res && eavail && |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1850 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1851 | goto fetch_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1852 | |
| 1853 | return res; |
| 1854 | } |
| 1855 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1856 | /** |
| 1857 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() |
| 1858 | * API, to verify that adding an epoll file inside another |
| 1859 | * epoll structure, does not violate the constraints, in |
| 1860 | * terms of closed loops, or too deep chains (which can |
| 1861 | * result in excessive stack usage). |
| 1862 | * |
| 1863 | * @priv: Pointer to the epoll file to be currently checked. |
| 1864 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll |
| 1865 | * data structure pointer. |
| 1866 | * @call_nests: Current dept of the @ep_call_nested() call stack. |
| 1867 | * |
| 1868 | * Returns: Returns zero if adding the epoll @file inside current epoll |
| 1869 | * structure @ep does not violate the constraints, or -1 otherwise. |
| 1870 | */ |
| 1871 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) |
| 1872 | { |
| 1873 | int error = 0; |
| 1874 | struct file *file = priv; |
| 1875 | struct eventpoll *ep = file->private_data; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1876 | struct eventpoll *ep_tovisit; |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1877 | struct rb_node *rbp; |
| 1878 | struct epitem *epi; |
| 1879 | |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 1880 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1881 | ep->visited = 1; |
| 1882 | list_add(&ep->visited_list_link, &visited_list); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1883 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1884 | epi = rb_entry(rbp, struct epitem, rbn); |
| 1885 | if (unlikely(is_file_epoll(epi->ffd.file))) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1886 | ep_tovisit = epi->ffd.file->private_data; |
| 1887 | if (ep_tovisit->visited) |
| 1888 | continue; |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1889 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1890 | ep_loop_check_proc, epi->ffd.file, |
| 1891 | ep_tovisit, current); |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1892 | if (error != 0) |
| 1893 | break; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1894 | } else { |
| 1895 | /* |
| 1896 | * If we've reached a file that is not associated with |
| 1897 | * an ep, then we need to check if the newly added |
| 1898 | * links are going to add too many wakeup paths. We do |
| 1899 | * this by adding it to the tfile_check_list, if it's |
| 1900 | * not already there, and calling reverse_path_check() |
| 1901 | * during ep_insert(). |
| 1902 | */ |
| 1903 | if (list_empty(&epi->ffd.file->f_tfile_llink)) |
| 1904 | list_add(&epi->ffd.file->f_tfile_llink, |
| 1905 | &tfile_check_list); |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1906 | } |
| 1907 | } |
| 1908 | mutex_unlock(&ep->mtx); |
| 1909 | |
| 1910 | return error; |
| 1911 | } |
| 1912 | |
| 1913 | /** |
| 1914 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) |
| 1915 | * another epoll file (represented by @ep) does not create |
| 1916 | * closed loops or too deep chains. |
| 1917 | * |
| 1918 | * @ep: Pointer to the epoll private data structure. |
| 1919 | * @file: Pointer to the epoll file to be checked. |
| 1920 | * |
| 1921 | * Returns: Returns zero if adding the epoll @file inside current epoll |
| 1922 | * structure @ep does not violate the constraints, or -1 otherwise. |
| 1923 | */ |
| 1924 | static int ep_loop_check(struct eventpoll *ep, struct file *file) |
| 1925 | { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1926 | int ret; |
| 1927 | struct eventpoll *ep_cur, *ep_next; |
| 1928 | |
| 1929 | ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1930 | ep_loop_check_proc, file, ep, current); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1931 | /* clear visited list */ |
| 1932 | list_for_each_entry_safe(ep_cur, ep_next, &visited_list, |
| 1933 | visited_list_link) { |
| 1934 | ep_cur->visited = 0; |
| 1935 | list_del(&ep_cur->visited_list_link); |
| 1936 | } |
| 1937 | return ret; |
| 1938 | } |
| 1939 | |
| 1940 | static void clear_tfile_check_list(void) |
| 1941 | { |
| 1942 | struct file *file; |
| 1943 | |
| 1944 | /* first clear the tfile_check_list */ |
| 1945 | while (!list_empty(&tfile_check_list)) { |
| 1946 | file = list_first_entry(&tfile_check_list, struct file, |
| 1947 | f_tfile_llink); |
| 1948 | list_del_init(&file->f_tfile_llink); |
| 1949 | } |
| 1950 | INIT_LIST_HEAD(&tfile_check_list); |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1951 | } |
| 1952 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1953 | /* |
Andrew Morton | 523723b | 2008-08-12 15:09:01 -0700 | [diff] [blame] | 1954 | * Open an eventpoll file descriptor. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1955 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1956 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1957 | { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1958 | int error, fd; |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1959 | struct eventpoll *ep = NULL; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1960 | struct file *file; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1961 | |
Ulrich Drepper | e38b36f | 2008-07-23 21:29:42 -0700 | [diff] [blame] | 1962 | /* Check the EPOLL_* constant for consistency. */ |
| 1963 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); |
| 1964 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1965 | if (flags & ~EPOLL_CLOEXEC) |
| 1966 | return -EINVAL; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1967 | /* |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1968 | * Create the internal data structure ("struct eventpoll"). |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1969 | */ |
Ulrich Drepper | 9fe5ad9 | 2008-07-23 21:29:43 -0700 | [diff] [blame] | 1970 | error = ep_alloc(&ep); |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1971 | if (error < 0) |
| 1972 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1973 | /* |
| 1974 | * Creates all the items needed to setup an eventpoll file. That is, |
Al Viro | 2030a42 | 2008-02-23 06:46:49 -0500 | [diff] [blame] | 1975 | * a file structure and a free file descriptor. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1976 | */ |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1977 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
| 1978 | if (fd < 0) { |
| 1979 | error = fd; |
| 1980 | goto out_free_ep; |
| 1981 | } |
| 1982 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, |
Roland Dreier | 628ff7c | 2009-12-18 09:41:24 -0800 | [diff] [blame] | 1983 | O_RDWR | (flags & O_CLOEXEC)); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1984 | if (IS_ERR(file)) { |
| 1985 | error = PTR_ERR(file); |
| 1986 | goto out_free_fd; |
| 1987 | } |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1988 | ep->file = file; |
Al Viro | 9802274 | 2012-08-17 22:42:36 -0400 | [diff] [blame] | 1989 | fd_install(fd, file); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1990 | return fd; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1991 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1992 | out_free_fd: |
| 1993 | put_unused_fd(fd); |
| 1994 | out_free_ep: |
| 1995 | ep_free(ep); |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1996 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1997 | } |
| 1998 | |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1999 | SYSCALL_DEFINE1(epoll_create, int, size) |
Ulrich Drepper | a0998b5 | 2008-07-23 21:29:27 -0700 | [diff] [blame] | 2000 | { |
Davide Libenzi | bfe3891 | 2009-05-12 13:19:44 -0700 | [diff] [blame] | 2001 | if (size <= 0) |
Ulrich Drepper | 9fe5ad9 | 2008-07-23 21:29:43 -0700 | [diff] [blame] | 2002 | return -EINVAL; |
| 2003 | |
| 2004 | return sys_epoll_create1(0); |
Ulrich Drepper | a0998b5 | 2008-07-23 21:29:27 -0700 | [diff] [blame] | 2005 | } |
| 2006 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2007 | /* |
| 2008 | * The following function implements the controller interface for |
| 2009 | * the eventpoll file that enables the insertion/removal/change of |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 2010 | * file descriptors inside the interest set. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2011 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 2012 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
| 2013 | struct epoll_event __user *, event) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2014 | { |
| 2015 | int error; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2016 | int full_check = 0; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2017 | struct fd f, tf; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2018 | struct eventpoll *ep; |
| 2019 | struct epitem *epi; |
| 2020 | struct epoll_event epds; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2021 | struct eventpoll *tep = NULL; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2022 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2023 | error = -EFAULT; |
| 2024 | if (ep_op_has_event(op) && |
| 2025 | copy_from_user(&epds, event, sizeof(struct epoll_event))) |
| 2026 | goto error_return; |
| 2027 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2028 | error = -EBADF; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2029 | f = fdget(epfd); |
| 2030 | if (!f.file) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2031 | goto error_return; |
| 2032 | |
| 2033 | /* Get the "struct file *" for the target file */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2034 | tf = fdget(fd); |
| 2035 | if (!tf.file) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2036 | goto error_fput; |
| 2037 | |
| 2038 | /* The target file descriptor must support poll */ |
| 2039 | error = -EPERM; |
Al Viro | 72c2d53 | 2013-09-22 16:27:52 -0400 | [diff] [blame] | 2040 | if (!tf.file->f_op->poll) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2041 | goto error_tgt_fput; |
| 2042 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 2043 | /* Check if EPOLLWAKEUP is allowed */ |
Nicolas Iooss | c680e41 | 2014-09-09 14:50:51 -0700 | [diff] [blame] | 2044 | if (ep_op_has_event(op)) |
| 2045 | ep_take_care_of_epollwakeup(&epds); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 2046 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2047 | /* |
| 2048 | * We have to check that the file structure underneath the file descriptor |
| 2049 | * the user passed to us _is_ an eventpoll file. And also we do not permit |
| 2050 | * adding an epoll file descriptor inside itself. |
| 2051 | */ |
| 2052 | error = -EINVAL; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2053 | if (f.file == tf.file || !is_file_epoll(f.file)) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2054 | goto error_tgt_fput; |
| 2055 | |
| 2056 | /* |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 2057 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, |
| 2058 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. |
| 2059 | * Also, we do not currently supported nested exclusive wakeups. |
| 2060 | */ |
Cyrill Gorcunov | c857ab6 | 2017-02-27 14:27:40 -0800 | [diff] [blame] | 2061 | if (ep_op_has_event(op) && (epds.events & EPOLLEXCLUSIVE)) { |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 2062 | if (op == EPOLL_CTL_MOD) |
| 2063 | goto error_tgt_fput; |
| 2064 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || |
| 2065 | (epds.events & ~EPOLLEXCLUSIVE_OK_BITS))) |
| 2066 | goto error_tgt_fput; |
| 2067 | } |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 2068 | |
| 2069 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2070 | * At this point it is safe to assume that the "private_data" contains |
| 2071 | * our own data structure. |
| 2072 | */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2073 | ep = f.file->private_data; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2074 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2075 | /* |
| 2076 | * When we insert an epoll file descriptor, inside another epoll file |
| 2077 | * descriptor, there is the change of creating closed loops, which are |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 2078 | * better be handled here, than in more critical paths. While we are |
| 2079 | * checking for loops we also determine the list of files reachable |
| 2080 | * and hang them on the tfile_check_list, so we can check that we |
| 2081 | * haven't created too many possible wakeup paths. |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2082 | * |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2083 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
| 2084 | * the epoll file descriptor is attaching directly to a wakeup source, |
| 2085 | * unless the epoll file descriptor is nested. The purpose of taking the |
| 2086 | * 'epmutex' on add is to prevent complex toplogies such as loops and |
| 2087 | * deep wakeup paths from forming in parallel through multiple |
| 2088 | * EPOLL_CTL_ADD operations. |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2089 | */ |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 2090 | mutex_lock_nested(&ep->mtx, 0); |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2091 | if (op == EPOLL_CTL_ADD) { |
| 2092 | if (!list_empty(&f.file->f_ep_links) || |
| 2093 | is_file_epoll(tf.file)) { |
| 2094 | full_check = 1; |
| 2095 | mutex_unlock(&ep->mtx); |
| 2096 | mutex_lock(&epmutex); |
| 2097 | if (is_file_epoll(tf.file)) { |
| 2098 | error = -ELOOP; |
| 2099 | if (ep_loop_check(ep, tf.file) != 0) { |
| 2100 | clear_tfile_check_list(); |
| 2101 | goto error_tgt_fput; |
| 2102 | } |
| 2103 | } else |
| 2104 | list_add(&tf.file->f_tfile_llink, |
| 2105 | &tfile_check_list); |
| 2106 | mutex_lock_nested(&ep->mtx, 0); |
| 2107 | if (is_file_epoll(tf.file)) { |
| 2108 | tep = tf.file->private_data; |
| 2109 | mutex_lock_nested(&tep->mtx, 1); |
| 2110 | } |
| 2111 | } |
| 2112 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2113 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 2114 | /* |
| 2115 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" |
| 2116 | * above, we can be sure to be able to use the item looked up by |
| 2117 | * ep_find() till we release the mutex. |
| 2118 | */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2119 | epi = ep_find(ep, tf.file, fd); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2120 | |
| 2121 | error = -EINVAL; |
| 2122 | switch (op) { |
| 2123 | case EPOLL_CTL_ADD: |
| 2124 | if (!epi) { |
| 2125 | epds.events |= POLLERR | POLLHUP; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2126 | error = ep_insert(ep, &epds, tf.file, fd, full_check); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2127 | } else |
| 2128 | error = -EEXIST; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2129 | if (full_check) |
| 2130 | clear_tfile_check_list(); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2131 | break; |
| 2132 | case EPOLL_CTL_DEL: |
| 2133 | if (epi) |
| 2134 | error = ep_remove(ep, epi); |
| 2135 | else |
| 2136 | error = -ENOENT; |
| 2137 | break; |
| 2138 | case EPOLL_CTL_MOD: |
| 2139 | if (epi) { |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 2140 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
| 2141 | epds.events |= POLLERR | POLLHUP; |
| 2142 | error = ep_modify(ep, epi, &epds); |
| 2143 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2144 | } else |
| 2145 | error = -ENOENT; |
| 2146 | break; |
| 2147 | } |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2148 | if (tep != NULL) |
| 2149 | mutex_unlock(&tep->mtx); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 2150 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2151 | |
| 2152 | error_tgt_fput: |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2153 | if (full_check) |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2154 | mutex_unlock(&epmutex); |
| 2155 | |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2156 | fdput(tf); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2157 | error_fput: |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2158 | fdput(f); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2159 | error_return: |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2160 | |
| 2161 | return error; |
| 2162 | } |
| 2163 | |
| 2164 | /* |
| 2165 | * Implement the event wait interface for the eventpoll file. It is the kernel |
| 2166 | * part of the user space epoll_wait(2). |
| 2167 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 2168 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
| 2169 | int, maxevents, int, timeout) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2170 | { |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2171 | int error; |
| 2172 | struct fd f; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2173 | struct eventpoll *ep; |
| 2174 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2175 | /* The maximum number of event must be greater than zero */ |
| 2176 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) |
| 2177 | return -EINVAL; |
| 2178 | |
| 2179 | /* Verify that the area passed by the user is writeable */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2180 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) |
| 2181 | return -EFAULT; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2182 | |
| 2183 | /* Get the "struct file *" for the eventpoll file */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2184 | f = fdget(epfd); |
| 2185 | if (!f.file) |
| 2186 | return -EBADF; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2187 | |
| 2188 | /* |
| 2189 | * We have to check that the file structure underneath the fd |
| 2190 | * the user passed to us _is_ an eventpoll file. |
| 2191 | */ |
| 2192 | error = -EINVAL; |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2193 | if (!is_file_epoll(f.file)) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2194 | goto error_fput; |
| 2195 | |
| 2196 | /* |
| 2197 | * At this point it is safe to assume that the "private_data" contains |
| 2198 | * our own data structure. |
| 2199 | */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2200 | ep = f.file->private_data; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2201 | |
| 2202 | /* Time to fish for events ... */ |
| 2203 | error = ep_poll(ep, events, maxevents, timeout); |
| 2204 | |
| 2205 | error_fput: |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2206 | fdput(f); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2207 | return error; |
| 2208 | } |
| 2209 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2210 | /* |
| 2211 | * Implement the event wait interface for the eventpoll file. It is the kernel |
| 2212 | * part of the user space epoll_pwait(2). |
| 2213 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 2214 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
| 2215 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, |
| 2216 | size_t, sigsetsize) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2217 | { |
| 2218 | int error; |
| 2219 | sigset_t ksigmask, sigsaved; |
| 2220 | |
| 2221 | /* |
| 2222 | * If the caller wants a certain signal mask to be set during the wait, |
| 2223 | * we apply it here. |
| 2224 | */ |
| 2225 | if (sigmask) { |
| 2226 | if (sigsetsize != sizeof(sigset_t)) |
| 2227 | return -EINVAL; |
| 2228 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) |
| 2229 | return -EFAULT; |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2230 | sigsaved = current->blocked; |
| 2231 | set_current_blocked(&ksigmask); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2232 | } |
| 2233 | |
| 2234 | error = sys_epoll_wait(epfd, events, maxevents, timeout); |
| 2235 | |
| 2236 | /* |
| 2237 | * If we changed the signal mask, we need to restore the original one. |
| 2238 | * In case we've got a signal while waiting, we do not restore the |
| 2239 | * signal mask yet, and we allow do_signal() to deliver the signal on |
| 2240 | * the way back to userspace, before the signal mask is restored. |
| 2241 | */ |
| 2242 | if (sigmask) { |
| 2243 | if (error == -EINTR) { |
| 2244 | memcpy(¤t->saved_sigmask, &sigsaved, |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 2245 | sizeof(sigsaved)); |
Roland McGrath | 4e4c22c | 2008-04-30 00:53:06 -0700 | [diff] [blame] | 2246 | set_restore_sigmask(); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2247 | } else |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2248 | set_current_blocked(&sigsaved); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2249 | } |
| 2250 | |
| 2251 | return error; |
| 2252 | } |
| 2253 | |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2254 | #ifdef CONFIG_COMPAT |
| 2255 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, |
| 2256 | struct epoll_event __user *, events, |
| 2257 | int, maxevents, int, timeout, |
| 2258 | const compat_sigset_t __user *, sigmask, |
| 2259 | compat_size_t, sigsetsize) |
| 2260 | { |
| 2261 | long err; |
| 2262 | compat_sigset_t csigmask; |
| 2263 | sigset_t ksigmask, sigsaved; |
| 2264 | |
| 2265 | /* |
| 2266 | * If the caller wants a certain signal mask to be set during the wait, |
| 2267 | * we apply it here. |
| 2268 | */ |
| 2269 | if (sigmask) { |
| 2270 | if (sigsetsize != sizeof(compat_sigset_t)) |
| 2271 | return -EINVAL; |
| 2272 | if (copy_from_user(&csigmask, sigmask, sizeof(csigmask))) |
| 2273 | return -EFAULT; |
| 2274 | sigset_from_compat(&ksigmask, &csigmask); |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2275 | sigsaved = current->blocked; |
| 2276 | set_current_blocked(&ksigmask); |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2277 | } |
| 2278 | |
| 2279 | err = sys_epoll_wait(epfd, events, maxevents, timeout); |
| 2280 | |
| 2281 | /* |
| 2282 | * If we changed the signal mask, we need to restore the original one. |
| 2283 | * In case we've got a signal while waiting, we do not restore the |
| 2284 | * signal mask yet, and we allow do_signal() to deliver the signal on |
| 2285 | * the way back to userspace, before the signal mask is restored. |
| 2286 | */ |
| 2287 | if (sigmask) { |
| 2288 | if (err == -EINTR) { |
| 2289 | memcpy(¤t->saved_sigmask, &sigsaved, |
| 2290 | sizeof(sigsaved)); |
| 2291 | set_restore_sigmask(); |
| 2292 | } else |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2293 | set_current_blocked(&sigsaved); |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2294 | } |
| 2295 | |
| 2296 | return err; |
| 2297 | } |
| 2298 | #endif |
| 2299 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2300 | static int __init eventpoll_init(void) |
| 2301 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 2302 | struct sysinfo si; |
| 2303 | |
| 2304 | si_meminfo(&si); |
Davide Libenzi | 9df04e1 | 2009-01-29 14:25:26 -0800 | [diff] [blame] | 2305 | /* |
| 2306 | * Allows top 4% of lomem to be allocated for epoll watches (per user). |
| 2307 | */ |
| 2308 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 2309 | EP_ITEM_COST; |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 2310 | BUG_ON(max_user_watches < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2311 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2312 | /* |
| 2313 | * Initialize the structure used to perform epoll file descriptor |
| 2314 | * inclusion loops checks. |
| 2315 | */ |
| 2316 | ep_nested_calls_init(&poll_loop_ncalls); |
| 2317 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2318 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 2319 | ep_nested_calls_init(&poll_safewake_ncalls); |
| 2320 | |
| 2321 | /* Initialize the structure used to perform file's f_op->poll() calls */ |
| 2322 | ep_nested_calls_init(&poll_readywalk_ncalls); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2323 | |
Eric Wong | 39732ca | 2013-04-30 15:27:38 -0700 | [diff] [blame] | 2324 | /* |
| 2325 | * We can have many thousands of epitems, so prevent this from |
| 2326 | * using an extra cache line on 64-bit (and smaller) CPUs |
| 2327 | */ |
| 2328 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); |
| 2329 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2330 | /* Allocates slab cache used to allocate "struct epitem" items */ |
| 2331 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 2332 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2333 | |
| 2334 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ |
| 2335 | pwq_cache = kmem_cache_create("eventpoll_pwq", |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 2336 | sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2337 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2338 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2339 | } |
Davide Libenzi | cea6924 | 2007-05-10 22:23:22 -0700 | [diff] [blame] | 2340 | fs_initcall(eventpoll_init); |