Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 2 | * fs/eventpoll.c (Efficient event retrieval implementation) |
| 3 | * Copyright (C) 2001,...,2009 Davide Libenzi |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify |
| 6 | * it under the terms of the GNU General Public License as published by |
| 7 | * the Free Software Foundation; either version 2 of the License, or |
| 8 | * (at your option) any later version. |
| 9 | * |
| 10 | * Davide Libenzi <davidel@xmailserver.org> |
| 11 | * |
| 12 | */ |
| 13 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/init.h> |
| 15 | #include <linux/kernel.h> |
| 16 | #include <linux/sched.h> |
| 17 | #include <linux/fs.h> |
| 18 | #include <linux/file.h> |
| 19 | #include <linux/signal.h> |
| 20 | #include <linux/errno.h> |
| 21 | #include <linux/mm.h> |
| 22 | #include <linux/slab.h> |
| 23 | #include <linux/poll.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/string.h> |
| 25 | #include <linux/list.h> |
| 26 | #include <linux/hash.h> |
| 27 | #include <linux/spinlock.h> |
| 28 | #include <linux/syscalls.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/rbtree.h> |
| 30 | #include <linux/wait.h> |
| 31 | #include <linux/eventpoll.h> |
| 32 | #include <linux/mount.h> |
| 33 | #include <linux/bitops.h> |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 34 | #include <linux/mutex.h> |
Davide Libenzi | da66f7c | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 35 | #include <linux/anon_inodes.h> |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 36 | #include <linux/device.h> |
Colin Cross | eb5d550 | 2013-05-06 23:50:16 +0000 | [diff] [blame] | 37 | #include <linux/freezer.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | #include <asm/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include <asm/io.h> |
| 40 | #include <asm/mman.h> |
Arun Sharma | 60063497 | 2011-07-26 16:09:06 -0700 | [diff] [blame] | 41 | #include <linux/atomic.h> |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 42 | #include <linux/proc_fs.h> |
| 43 | #include <linux/seq_file.h> |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 44 | #include <linux/compat.h> |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 45 | #include <linux/rculist.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 */ |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 208 | struct rb_root rbr; |
| 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 */ |
Al Viro | 0ece241 | 2020-09-10 08:30:05 -0400 | [diff] [blame] | 226 | u64 gen; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 227 | }; |
| 228 | |
| 229 | /* Wait structure used by the poll hooks */ |
| 230 | struct eppoll_entry { |
| 231 | /* List header used to link this structure to the "struct epitem" */ |
| 232 | struct list_head llink; |
| 233 | |
| 234 | /* The "base" pointer is set to the container "struct epitem" */ |
Tony Battersby | 4f0989d | 2009-03-31 15:24:16 -0700 | [diff] [blame] | 235 | struct epitem *base; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 236 | |
| 237 | /* |
| 238 | * Wait queue item that will be linked to the target file wait |
| 239 | * queue head. |
| 240 | */ |
| 241 | wait_queue_t wait; |
| 242 | |
| 243 | /* The wait queue head that linked the "wait" wait queue item */ |
| 244 | wait_queue_head_t *whead; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | }; |
| 246 | |
| 247 | /* Wrapper struct used by poll queueing */ |
| 248 | struct ep_pqueue { |
| 249 | poll_table pt; |
| 250 | struct epitem *epi; |
| 251 | }; |
| 252 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 253 | /* Used by the ep_send_events() function as callback private data */ |
| 254 | struct ep_send_events_data { |
| 255 | int maxevents; |
| 256 | struct epoll_event __user *events; |
| 257 | }; |
| 258 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 259 | /* |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 260 | * Configuration options available inside /proc/sys/fs/epoll/ |
| 261 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 262 | /* Maximum number of epoll watched descriptors, per user */ |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 263 | static long max_user_watches __read_mostly; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 264 | |
| 265 | /* |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 266 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 268 | static DEFINE_MUTEX(epmutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | |
Al Viro | 0ece241 | 2020-09-10 08:30:05 -0400 | [diff] [blame] | 270 | static u64 loop_check_gen = 0; |
| 271 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 272 | /* Used to check for epoll file descriptor inclusion loops */ |
| 273 | static struct nested_calls poll_loop_ncalls; |
| 274 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 275 | /* Used for safe wake up implementation */ |
| 276 | static struct nested_calls poll_safewake_ncalls; |
| 277 | |
| 278 | /* Used to call file's f_op->poll() under the nested calls boundaries */ |
| 279 | static struct nested_calls poll_readywalk_ncalls; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | |
| 281 | /* Slab cache used to allocate "struct epitem" */ |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 282 | static struct kmem_cache *epi_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 283 | |
| 284 | /* Slab cache used to allocate "struct eppoll_entry" */ |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 285 | static struct kmem_cache *pwq_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 287 | /* |
| 288 | * List of files with newly added links, where we may need to limit the number |
| 289 | * of emanating paths. Protected by the epmutex. |
| 290 | */ |
| 291 | static LIST_HEAD(tfile_check_list); |
| 292 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 293 | #ifdef CONFIG_SYSCTL |
| 294 | |
| 295 | #include <linux/sysctl.h> |
| 296 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 297 | static long zero; |
| 298 | static long long_max = LONG_MAX; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 299 | |
Joe Perches | 1f7e061 | 2014-06-06 14:38:05 -0700 | [diff] [blame] | 300 | struct ctl_table epoll_table[] = { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 301 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 302 | .procname = "max_user_watches", |
| 303 | .data = &max_user_watches, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 304 | .maxlen = sizeof(max_user_watches), |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 305 | .mode = 0644, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 306 | .proc_handler = proc_doulongvec_minmax, |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 307 | .extra1 = &zero, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 308 | .extra2 = &long_max, |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 309 | }, |
Eric W. Biederman | ab09203 | 2009-11-05 14:25:10 -0800 | [diff] [blame] | 310 | { } |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 311 | }; |
| 312 | #endif /* CONFIG_SYSCTL */ |
| 313 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 314 | static const struct file_operations eventpoll_fops; |
| 315 | |
| 316 | static inline int is_file_epoll(struct file *f) |
| 317 | { |
| 318 | return f->f_op == &eventpoll_fops; |
| 319 | } |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 320 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 321 | /* Setup the structure that is used as key for the RB tree */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 322 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
| 323 | struct file *file, int fd) |
| 324 | { |
| 325 | ffd->file = file; |
| 326 | ffd->fd = fd; |
| 327 | } |
| 328 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 329 | /* Compare RB tree keys */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 330 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
| 331 | struct epoll_filefd *p2) |
| 332 | { |
| 333 | return (p1->file > p2->file ? +1: |
| 334 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); |
| 335 | } |
| 336 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 337 | /* Tells us if the item is currently linked */ |
| 338 | static inline int ep_is_linked(struct list_head *p) |
| 339 | { |
| 340 | return !list_empty(p); |
| 341 | } |
| 342 | |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 343 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_t *p) |
| 344 | { |
| 345 | return container_of(p, struct eppoll_entry, wait); |
| 346 | } |
| 347 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 348 | /* Get the "struct epitem" from a wait queue pointer */ |
Davide Libenzi | cdac75e | 2008-04-29 00:58:34 -0700 | [diff] [blame] | 349 | static inline struct epitem *ep_item_from_wait(wait_queue_t *p) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 350 | { |
| 351 | return container_of(p, struct eppoll_entry, wait)->base; |
| 352 | } |
| 353 | |
| 354 | /* Get the "struct epitem" from an epoll queue wrapper */ |
Davide Libenzi | cdac75e | 2008-04-29 00:58:34 -0700 | [diff] [blame] | 355 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 356 | { |
| 357 | return container_of(p, struct ep_pqueue, pt)->epi; |
| 358 | } |
| 359 | |
| 360 | /* 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] | 361 | static inline int ep_op_has_event(int op) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 362 | { |
Andrew Morton | a80a6b8 | 2012-11-08 15:53:35 -0800 | [diff] [blame] | 363 | return op != EPOLL_CTL_DEL; |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 364 | } |
| 365 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 366 | /* Initialize the poll safe wake up structure */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 367 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 369 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
| 370 | spin_lock_init(&ncalls->lock); |
| 371 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 373 | /** |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 374 | * ep_events_available - Checks if ready events might be available. |
| 375 | * |
| 376 | * @ep: Pointer to the eventpoll context. |
| 377 | * |
| 378 | * Returns: Returns a value different than zero if ready events are available, |
| 379 | * or zero otherwise. |
| 380 | */ |
| 381 | static inline int ep_events_available(struct eventpoll *ep) |
| 382 | { |
| 383 | return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; |
| 384 | } |
| 385 | |
| 386 | /** |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 387 | * ep_call_nested - Perform a bound (possibly) nested call, by checking |
| 388 | * that the recursion limit is not exceeded, and that |
| 389 | * the same nested call (by the meaning of same cookie) is |
| 390 | * no re-entered. |
| 391 | * |
| 392 | * @ncalls: Pointer to the nested_calls structure to be used for this call. |
| 393 | * @max_nests: Maximum number of allowed nesting calls. |
| 394 | * @nproc: Nested call core function pointer. |
| 395 | * @priv: Opaque data to be passed to the @nproc callback. |
| 396 | * @cookie: Cookie to be used to identify this nested call. |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 397 | * @ctx: This instance context. |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 398 | * |
| 399 | * Returns: Returns the code returned by the @nproc callback, or -1 if |
| 400 | * the maximum recursion limit has been exceeded. |
| 401 | */ |
| 402 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
| 403 | int (*nproc)(void *, void *, int), void *priv, |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 404 | void *cookie, void *ctx) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 405 | { |
| 406 | int error, call_nests = 0; |
| 407 | unsigned long flags; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 408 | struct list_head *lsthead = &ncalls->tasks_call_list; |
| 409 | struct nested_call_node *tncur; |
| 410 | struct nested_call_node tnode; |
| 411 | |
| 412 | spin_lock_irqsave(&ncalls->lock, flags); |
| 413 | |
| 414 | /* |
| 415 | * Try to see if the current task is already inside this wakeup call. |
| 416 | * We use a list here, since the population inside this set is always |
| 417 | * very much limited. |
| 418 | */ |
| 419 | list_for_each_entry(tncur, lsthead, llink) { |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 420 | if (tncur->ctx == ctx && |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 421 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
| 422 | /* |
| 423 | * Ops ... loop detected or maximum nest level reached. |
| 424 | * We abort this wake by breaking the cycle itself. |
| 425 | */ |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame] | 426 | error = -1; |
| 427 | goto out_unlock; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 428 | } |
| 429 | } |
| 430 | |
| 431 | /* Add the current task and cookie to the list */ |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 432 | tnode.ctx = ctx; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 433 | tnode.cookie = cookie; |
| 434 | list_add(&tnode.llink, lsthead); |
| 435 | |
| 436 | spin_unlock_irqrestore(&ncalls->lock, flags); |
| 437 | |
| 438 | /* Call the nested function */ |
| 439 | error = (*nproc)(priv, cookie, call_nests); |
| 440 | |
| 441 | /* Remove the current task from the list */ |
| 442 | spin_lock_irqsave(&ncalls->lock, flags); |
| 443 | list_del(&tnode.llink); |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 444 | out_unlock: |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 445 | spin_unlock_irqrestore(&ncalls->lock, flags); |
| 446 | |
| 447 | return error; |
| 448 | } |
| 449 | |
Steven Rostedt | 02edc6f | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 450 | /* |
| 451 | * As described in commit 0ccf831cb lockdep: annotate epoll |
| 452 | * the use of wait queues used by epoll is done in a very controlled |
| 453 | * manner. Wake ups can nest inside each other, but are never done |
| 454 | * with the same locking. For example: |
| 455 | * |
| 456 | * dfd = socket(...); |
| 457 | * efd1 = epoll_create(); |
| 458 | * efd2 = epoll_create(); |
| 459 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); |
| 460 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); |
| 461 | * |
| 462 | * When a packet arrives to the device underneath "dfd", the net code will |
| 463 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a |
| 464 | * callback wakeup entry on that queue, and the wake_up() performed by the |
| 465 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll |
| 466 | * (efd1) notices that it may have some event ready, so it needs to wake up |
| 467 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() |
| 468 | * that ends up in another wake_up(), after having checked about the |
| 469 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to |
| 470 | * avoid stack blasting. |
| 471 | * |
| 472 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle |
| 473 | * this special case of epoll. |
| 474 | */ |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 475 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 476 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, |
| 477 | unsigned long events, int subclass) |
| 478 | { |
| 479 | unsigned long flags; |
| 480 | |
| 481 | spin_lock_irqsave_nested(&wqueue->lock, flags, subclass); |
| 482 | wake_up_locked_poll(wqueue, events); |
| 483 | spin_unlock_irqrestore(&wqueue->lock, flags); |
| 484 | } |
| 485 | #else |
| 486 | static inline void ep_wake_up_nested(wait_queue_head_t *wqueue, |
| 487 | unsigned long events, int subclass) |
| 488 | { |
| 489 | wake_up_poll(wqueue, events); |
| 490 | } |
| 491 | #endif |
| 492 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 493 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) |
| 494 | { |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 495 | ep_wake_up_nested((wait_queue_head_t *) cookie, POLLIN, |
| 496 | 1 + call_nests); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 497 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | } |
| 499 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | /* |
| 501 | * Perform a safe wake up of the poll wait list. The problem is that |
| 502 | * with the new callback'd wake up system, it is possible that the |
| 503 | * poll callback is reentered from inside the call to wake_up() done |
| 504 | * 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] | 505 | * 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] | 506 | * and we cannot reenter the same wait queue head at all. This will |
| 507 | * 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] | 508 | * EP_MAX_NESTS deep. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 510 | static void ep_poll_safewake(wait_queue_head_t *wq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | { |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 512 | int this_cpu = get_cpu(); |
| 513 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 514 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 515 | ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu); |
| 516 | |
| 517 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | } |
| 519 | |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 520 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
| 521 | { |
| 522 | wait_queue_head_t *whead; |
| 523 | |
| 524 | rcu_read_lock(); |
Oleg Nesterov | d325f1f | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 525 | /* |
| 526 | * If it is cleared by POLLFREE, it should be rcu-safe. |
| 527 | * If we read NULL we need a barrier paired with |
| 528 | * smp_store_release() in ep_poll_callback(), otherwise |
| 529 | * we rely on whead->lock. |
| 530 | */ |
| 531 | whead = smp_load_acquire(&pwq->whead); |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 532 | if (whead) |
| 533 | remove_wait_queue(whead, &pwq->wait); |
| 534 | rcu_read_unlock(); |
| 535 | } |
| 536 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 537 | /* |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 538 | * This function unregisters poll callbacks from the associated file |
| 539 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from |
| 540 | * ep_free). |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 541 | */ |
| 542 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
| 543 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 544 | struct list_head *lsthead = &epi->pwqlist; |
| 545 | struct eppoll_entry *pwq; |
| 546 | |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 547 | while (!list_empty(lsthead)) { |
| 548 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 549 | |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 550 | list_del(&pwq->llink); |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 551 | ep_remove_wait_queue(pwq); |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 552 | kmem_cache_free(pwq_cache, pwq); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 553 | } |
| 554 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 556 | /* call only when ep->mtx is held */ |
| 557 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) |
| 558 | { |
| 559 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); |
| 560 | } |
| 561 | |
| 562 | /* call only when ep->mtx is held */ |
| 563 | static inline void ep_pm_stay_awake(struct epitem *epi) |
| 564 | { |
| 565 | struct wakeup_source *ws = ep_wakeup_source(epi); |
| 566 | |
| 567 | if (ws) |
| 568 | __pm_stay_awake(ws); |
| 569 | } |
| 570 | |
| 571 | static inline bool ep_has_wakeup_source(struct epitem *epi) |
| 572 | { |
| 573 | return rcu_access_pointer(epi->ws) ? true : false; |
| 574 | } |
| 575 | |
| 576 | /* call when ep->mtx cannot be held (ep_poll_callback) */ |
| 577 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) |
| 578 | { |
| 579 | struct wakeup_source *ws; |
| 580 | |
| 581 | rcu_read_lock(); |
| 582 | ws = rcu_dereference(epi->ws); |
| 583 | if (ws) |
| 584 | __pm_stay_awake(ws); |
| 585 | rcu_read_unlock(); |
| 586 | } |
| 587 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 588 | /** |
| 589 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for |
| 590 | * the scan code, to call f_op->poll(). Also allows for |
| 591 | * O(NumReady) performance. |
| 592 | * |
| 593 | * @ep: Pointer to the epoll private data structure. |
| 594 | * @sproc: Pointer to the scan callback. |
| 595 | * @priv: Private opaque data passed to the @sproc callback. |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 596 | * @depth: The current depth of recursive f_op->poll calls. |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 597 | * @ep_locked: caller already holds ep->mtx |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 598 | * |
| 599 | * Returns: The same integer error code returned by the @sproc callback. |
| 600 | */ |
| 601 | static int ep_scan_ready_list(struct eventpoll *ep, |
| 602 | int (*sproc)(struct eventpoll *, |
| 603 | struct list_head *, void *), |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 604 | void *priv, int depth, bool ep_locked) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 605 | { |
| 606 | int error, pwake = 0; |
| 607 | unsigned long flags; |
| 608 | struct epitem *epi, *nepi; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 609 | LIST_HEAD(txlist); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 610 | |
| 611 | /* |
| 612 | * We need to lock this because we could be hit by |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 613 | * eventpoll_release_file() and epoll_ctl(). |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 614 | */ |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 615 | |
| 616 | if (!ep_locked) |
| 617 | mutex_lock_nested(&ep->mtx, depth); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 618 | |
| 619 | /* |
| 620 | * Steal the ready list, and re-init the original one to the |
| 621 | * empty list. Also, set ep->ovflist to NULL so that events |
| 622 | * happening while looping w/out locks, are not lost. We cannot |
| 623 | * have the poll callback to queue directly on ep->rdllist, |
| 624 | * because we want the "sproc" callback to be able to do it |
| 625 | * in a lockless way. |
| 626 | */ |
| 627 | spin_lock_irqsave(&ep->lock, flags); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 628 | list_splice_init(&ep->rdllist, &txlist); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 629 | ep->ovflist = NULL; |
| 630 | spin_unlock_irqrestore(&ep->lock, flags); |
| 631 | |
| 632 | /* |
| 633 | * Now call the callback function. |
| 634 | */ |
| 635 | error = (*sproc)(ep, &txlist, priv); |
| 636 | |
| 637 | spin_lock_irqsave(&ep->lock, flags); |
| 638 | /* |
| 639 | * During the time we spent inside the "sproc" callback, some |
| 640 | * other events might have been queued by the poll callback. |
| 641 | * We re-insert them inside the main ready-list here. |
| 642 | */ |
| 643 | for (nepi = ep->ovflist; (epi = nepi) != NULL; |
| 644 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
| 645 | /* |
| 646 | * We need to check if the item is already in the list. |
| 647 | * During the "sproc" callback execution time, items are |
| 648 | * queued into ->ovflist but the "txlist" might already |
| 649 | * contain them, and the list_splice() below takes care of them. |
| 650 | */ |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 651 | if (!ep_is_linked(&epi->rdllink)) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 652 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 653 | ep_pm_stay_awake(epi); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 654 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 655 | } |
| 656 | /* |
| 657 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after |
| 658 | * releasing the lock, events will be queued in the normal way inside |
| 659 | * ep->rdllist. |
| 660 | */ |
| 661 | ep->ovflist = EP_UNACTIVE_PTR; |
| 662 | |
| 663 | /* |
| 664 | * Quickly re-inject items left on "txlist". |
| 665 | */ |
| 666 | list_splice(&txlist, &ep->rdllist); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 667 | __pm_relax(ep->ws); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 668 | |
| 669 | if (!list_empty(&ep->rdllist)) { |
| 670 | /* |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 671 | * Wake up (if active) both the eventpoll wait list and |
| 672 | * the ->poll() wait list (delayed after we release the lock). |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 673 | */ |
| 674 | if (waitqueue_active(&ep->wq)) |
| 675 | wake_up_locked(&ep->wq); |
| 676 | if (waitqueue_active(&ep->poll_wait)) |
| 677 | pwake++; |
| 678 | } |
| 679 | spin_unlock_irqrestore(&ep->lock, flags); |
| 680 | |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 681 | if (!ep_locked) |
| 682 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 683 | |
| 684 | /* We have to call this outside the lock */ |
| 685 | if (pwake) |
| 686 | ep_poll_safewake(&ep->poll_wait); |
| 687 | |
| 688 | return error; |
| 689 | } |
| 690 | |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 691 | static void epi_rcu_free(struct rcu_head *head) |
| 692 | { |
| 693 | struct epitem *epi = container_of(head, struct epitem, rcu); |
| 694 | kmem_cache_free(epi_cache, epi); |
| 695 | } |
| 696 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 698 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 699 | * all the associated resources. Must be called with "mtx" held. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 700 | */ |
| 701 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) |
| 702 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 703 | unsigned long flags; |
| 704 | struct file *file = epi->ffd.file; |
| 705 | |
| 706 | /* |
| 707 | * Removes poll wait queue hooks. We _have_ to do this without holding |
| 708 | * the "ep->lock" otherwise a deadlock might occur. This because of the |
| 709 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait |
| 710 | * queue head lock when unregistering the wait queue. The wakeup callback |
| 711 | * will run by holding the wait queue head lock and will call our callback |
| 712 | * that will try to get "ep->lock". |
| 713 | */ |
| 714 | ep_unregister_pollwait(ep, epi); |
| 715 | |
| 716 | /* Remove the current item from the list of epoll hooks */ |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 717 | spin_lock(&file->f_lock); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 718 | list_del_rcu(&epi->fllink); |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 719 | spin_unlock(&file->f_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 720 | |
Davide Libenzi | cdac75e | 2008-04-29 00:58:34 -0700 | [diff] [blame] | 721 | rb_erase(&epi->rbn, &ep->rbr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 722 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 723 | spin_lock_irqsave(&ep->lock, flags); |
| 724 | if (ep_is_linked(&epi->rdllink)) |
| 725 | list_del_init(&epi->rdllink); |
| 726 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 727 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 728 | wakeup_source_unregister(ep_wakeup_source(epi)); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 729 | /* |
| 730 | * At this point it is safe to free the eventpoll item. Use the union |
| 731 | * field epi->rcu, since we are trying to minimize the size of |
| 732 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by |
| 733 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make |
| 734 | * use of the rbn field. |
| 735 | */ |
| 736 | call_rcu(&epi->rcu, epi_rcu_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 737 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 738 | atomic_long_dec(&ep->user->epoll_watches); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 739 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 740 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 741 | } |
| 742 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | static void ep_free(struct eventpoll *ep) |
| 744 | { |
| 745 | struct rb_node *rbp; |
| 746 | struct epitem *epi; |
| 747 | |
| 748 | /* We need to release all tasks waiting for these file */ |
| 749 | if (waitqueue_active(&ep->poll_wait)) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 750 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 751 | |
| 752 | /* |
| 753 | * We need to lock this because we could be hit by |
| 754 | * eventpoll_release_file() while we're freeing the "struct eventpoll". |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 755 | * 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] | 756 | * is on the way to be removed and no one has references to it |
| 757 | * anymore. The only hit might come from eventpoll_release_file() but |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 758 | * holding "epmutex" is sufficient here. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | */ |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 760 | mutex_lock(&epmutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 761 | |
| 762 | /* |
| 763 | * Walks through the whole tree by unregistering poll callbacks. |
| 764 | */ |
| 765 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
| 766 | epi = rb_entry(rbp, struct epitem, rbn); |
| 767 | |
| 768 | ep_unregister_pollwait(ep, epi); |
Eric Dumazet | 91cf5ab | 2013-09-11 14:24:06 -0700 | [diff] [blame] | 769 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | } |
| 771 | |
| 772 | /* |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 773 | * Walks through the whole tree by freeing each "struct epitem". At this |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | * 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] | 775 | * 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] | 776 | * 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] | 777 | * We do not need to lock ep->mtx, either, we only do it to prevent |
| 778 | * a lockdep warning. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | */ |
Eric Wong | ddf676c | 2013-04-30 15:27:40 -0700 | [diff] [blame] | 780 | mutex_lock(&ep->mtx); |
Stephen Hemminger | c80544d | 2007-10-18 03:07:05 -0700 | [diff] [blame] | 781 | while ((rbp = rb_first(&ep->rbr)) != NULL) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | epi = rb_entry(rbp, struct epitem, rbn); |
| 783 | ep_remove(ep, epi); |
Eric Dumazet | 91cf5ab | 2013-09-11 14:24:06 -0700 | [diff] [blame] | 784 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 785 | } |
Eric Wong | ddf676c | 2013-04-30 15:27:40 -0700 | [diff] [blame] | 786 | mutex_unlock(&ep->mtx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 787 | |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 788 | mutex_unlock(&epmutex); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 789 | mutex_destroy(&ep->mtx); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 790 | free_uid(ep->user); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 791 | wakeup_source_unregister(ep->ws); |
Davide Libenzi | f0ee9aa | 2007-05-15 01:40:57 -0700 | [diff] [blame] | 792 | kfree(ep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 793 | } |
| 794 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 795 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
| 796 | { |
| 797 | struct eventpoll *ep = file->private_data; |
| 798 | |
Davide Libenzi | f0ee9aa | 2007-05-15 01:40:57 -0700 | [diff] [blame] | 799 | if (ep) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 800 | ep_free(ep); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 801 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 802 | return 0; |
| 803 | } |
| 804 | |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 805 | static inline unsigned int ep_item_poll(struct epitem *epi, poll_table *pt) |
| 806 | { |
| 807 | pt->_key = epi->event.events; |
| 808 | |
| 809 | return epi->ffd.file->f_op->poll(epi->ffd.file, pt) & epi->event.events; |
| 810 | } |
| 811 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 812 | static int ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
| 813 | void *priv) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 814 | { |
| 815 | struct epitem *epi, *tmp; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 816 | poll_table pt; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 817 | |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 818 | init_poll_funcptr(&pt, NULL); |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 819 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 820 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 821 | if (ep_item_poll(epi, &pt)) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 822 | return POLLIN | POLLRDNORM; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 823 | else { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 824 | /* |
| 825 | * Item has been dropped into the ready list by the poll |
| 826 | * callback, but it's not actually ready, as far as |
| 827 | * caller requested events goes. We can remove it here. |
| 828 | */ |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 829 | __pm_relax(ep_wakeup_source(epi)); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 830 | list_del_init(&epi->rdllink); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 831 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 832 | } |
| 833 | |
| 834 | return 0; |
| 835 | } |
| 836 | |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 837 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
| 838 | poll_table *pt); |
| 839 | |
| 840 | struct readyevents_arg { |
| 841 | struct eventpoll *ep; |
| 842 | bool locked; |
| 843 | }; |
| 844 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 845 | static int ep_poll_readyevents_proc(void *priv, void *cookie, int call_nests) |
| 846 | { |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 847 | struct readyevents_arg *arg = priv; |
| 848 | |
| 849 | return ep_scan_ready_list(arg->ep, ep_read_events_proc, NULL, |
| 850 | call_nests + 1, arg->locked); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 851 | } |
| 852 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 853 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) |
| 854 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 855 | int pollflags; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 856 | struct eventpoll *ep = file->private_data; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 857 | struct readyevents_arg arg; |
| 858 | |
| 859 | /* |
| 860 | * During ep_insert() we already hold the ep->mtx for the tfile. |
| 861 | * Prevent re-aquisition. |
| 862 | */ |
| 863 | arg.locked = wait && (wait->_qproc == ep_ptable_queue_proc); |
| 864 | arg.ep = ep; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 865 | |
| 866 | /* Insert inside our poll wait queue */ |
| 867 | poll_wait(file, &ep->poll_wait, wait); |
| 868 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 869 | /* |
| 870 | * Proceed to find out if wanted events are really available inside |
| 871 | * the ready list. This need to be done under ep_call_nested() |
| 872 | * supervision, since the call to f_op->poll() done on listed files |
| 873 | * could re-enter here. |
| 874 | */ |
| 875 | pollflags = ep_call_nested(&poll_readywalk_ncalls, EP_MAX_NESTS, |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 876 | ep_poll_readyevents_proc, &arg, ep, current); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 877 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 878 | return pollflags != -1 ? pollflags : 0; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 879 | } |
| 880 | |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 881 | #ifdef CONFIG_PROC_FS |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 882 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 883 | { |
| 884 | struct eventpoll *ep = f->private_data; |
| 885 | struct rb_node *rbp; |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 886 | |
| 887 | mutex_lock(&ep->mtx); |
| 888 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
| 889 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
| 890 | |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 891 | seq_printf(m, "tfd: %8d events: %8x data: %16llx\n", |
| 892 | epi->ffd.fd, epi->event.events, |
| 893 | (long long)epi->event.data); |
| 894 | if (seq_has_overflowed(m)) |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 895 | break; |
| 896 | } |
| 897 | mutex_unlock(&ep->mtx); |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 898 | } |
| 899 | #endif |
| 900 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 901 | /* File callbacks that implement the eventpoll file behaviour */ |
| 902 | static const struct file_operations eventpoll_fops = { |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 903 | #ifdef CONFIG_PROC_FS |
| 904 | .show_fdinfo = ep_show_fdinfo, |
| 905 | #endif |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 906 | .release = ep_eventpoll_release, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 907 | .poll = ep_eventpoll_poll, |
| 908 | .llseek = noop_llseek, |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 909 | }; |
| 910 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 911 | /* |
| 912 | * This is called from eventpoll_release() to unlink files from the eventpoll |
| 913 | * interface. We need to have this facility to cleanup correctly files that are |
| 914 | * closed without being removed from the eventpoll interface. |
| 915 | */ |
| 916 | void eventpoll_release_file(struct file *file) |
| 917 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 918 | struct eventpoll *ep; |
Konstantin Khlebnikov | ebe0618 | 2014-06-17 06:58:05 +0400 | [diff] [blame] | 919 | struct epitem *epi, *next; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 920 | |
| 921 | /* |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 922 | * 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] | 923 | * 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] | 924 | * 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] | 925 | * 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] | 926 | * point, the file counter already went to zero and fget() would fail. |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 927 | * 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] | 928 | * will correctly serialize the operation. We do need to acquire |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 929 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 930 | * from anywhere but ep_free(). |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 931 | * |
| 932 | * 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] | 933 | */ |
| 934 | mutex_lock(&epmutex); |
Konstantin Khlebnikov | ebe0618 | 2014-06-17 06:58:05 +0400 | [diff] [blame] | 935 | list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 936 | ep = epi->ep; |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 937 | mutex_lock_nested(&ep->mtx, 0); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 938 | ep_remove(ep, epi); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 939 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 940 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 941 | mutex_unlock(&epmutex); |
| 942 | } |
| 943 | |
| 944 | static int ep_alloc(struct eventpoll **pep) |
| 945 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 946 | int error; |
| 947 | struct user_struct *user; |
| 948 | struct eventpoll *ep; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 949 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 950 | user = get_current_user(); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 951 | error = -ENOMEM; |
| 952 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); |
| 953 | if (unlikely(!ep)) |
| 954 | goto free_uid; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 955 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 956 | spin_lock_init(&ep->lock); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 957 | mutex_init(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 958 | init_waitqueue_head(&ep->wq); |
| 959 | init_waitqueue_head(&ep->poll_wait); |
| 960 | INIT_LIST_HEAD(&ep->rdllist); |
| 961 | ep->rbr = RB_ROOT; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 962 | ep->ovflist = EP_UNACTIVE_PTR; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 963 | ep->user = user; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 964 | |
| 965 | *pep = ep; |
| 966 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 967 | return 0; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 968 | |
| 969 | free_uid: |
| 970 | free_uid(user); |
| 971 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 972 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | |
| 974 | /* |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 975 | * Search the file inside the eventpoll tree. The RB tree operations |
| 976 | * are protected by the "mtx" mutex, and ep_find() must be called with |
| 977 | * "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 978 | */ |
| 979 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) |
| 980 | { |
| 981 | int kcmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | struct rb_node *rbp; |
| 983 | struct epitem *epi, *epir = NULL; |
| 984 | struct epoll_filefd ffd; |
| 985 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 986 | ep_set_ffd(&ffd, file, fd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 987 | for (rbp = ep->rbr.rb_node; rbp; ) { |
| 988 | epi = rb_entry(rbp, struct epitem, rbn); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 989 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 990 | if (kcmp > 0) |
| 991 | rbp = rbp->rb_right; |
| 992 | else if (kcmp < 0) |
| 993 | rbp = rbp->rb_left; |
| 994 | else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 995 | epir = epi; |
| 996 | break; |
| 997 | } |
| 998 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 999 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1000 | return epir; |
| 1001 | } |
| 1002 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1003 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1004 | * This is the callback that is passed to the wait queue wakeup |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 1005 | * mechanism. It is called by the stored file descriptors when they |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1006 | * have events to report. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | */ |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1008 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1010 | int pwake = 0; |
| 1011 | unsigned long flags; |
| 1012 | struct epitem *epi = ep_item_from_wait(wait); |
| 1013 | struct eventpoll *ep = epi->ep; |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1014 | int ewake = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1016 | spin_lock_irqsave(&ep->lock, flags); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1017 | |
| 1018 | /* |
| 1019 | * If the event mask does not contain any poll(2) event, we consider the |
| 1020 | * descriptor to be disabled. This condition is likely the effect of the |
| 1021 | * EPOLLONESHOT bit that disables the descriptor when an event is received, |
| 1022 | * until the next EPOLL_CTL_MOD will be issued. |
| 1023 | */ |
| 1024 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1025 | goto out_unlock; |
| 1026 | |
| 1027 | /* |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 1028 | * Check the events coming with the callback. At this stage, not |
| 1029 | * every device reports the events in the "key" parameter of the |
| 1030 | * callback. We need to be able to handle both cases here, hence the |
| 1031 | * test for "key" != NULL before the event match test. |
| 1032 | */ |
| 1033 | if (key && !((unsigned long) key & epi->event.events)) |
| 1034 | goto out_unlock; |
| 1035 | |
| 1036 | /* |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 1037 | * If we are transferring events to userspace, we can hold no locks |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1038 | * (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] | 1039 | * semantics). All the events that happen during that period of time are |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1040 | * chained in ep->ovflist and requeued later on. |
| 1041 | */ |
Davidlohr Bueso | cf96e963 | 2019-01-03 15:27:09 -0800 | [diff] [blame] | 1042 | if (ep->ovflist != EP_UNACTIVE_PTR) { |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1043 | if (epi->next == EP_UNACTIVE_PTR) { |
| 1044 | epi->next = ep->ovflist; |
| 1045 | ep->ovflist = epi; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1046 | if (epi->ws) { |
| 1047 | /* |
| 1048 | * Activate ep->ws since epi->ws may get |
| 1049 | * deactivated at any time. |
| 1050 | */ |
| 1051 | __pm_stay_awake(ep->ws); |
| 1052 | } |
| 1053 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1054 | } |
| 1055 | goto out_unlock; |
| 1056 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1057 | |
| 1058 | /* If this file is already in the ready list we exit soon */ |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1059 | if (!ep_is_linked(&epi->rdllink)) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1060 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1061 | ep_pm_stay_awake_rcu(epi); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1062 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1063 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1064 | /* |
| 1065 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() |
| 1066 | * wait list. |
| 1067 | */ |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1068 | if (waitqueue_active(&ep->wq)) { |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 1069 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
| 1070 | !((unsigned long)key & POLLFREE)) { |
| 1071 | switch ((unsigned long)key & EPOLLINOUT_BITS) { |
| 1072 | case POLLIN: |
| 1073 | if (epi->event.events & POLLIN) |
| 1074 | ewake = 1; |
| 1075 | break; |
| 1076 | case POLLOUT: |
| 1077 | if (epi->event.events & POLLOUT) |
| 1078 | ewake = 1; |
| 1079 | break; |
| 1080 | case 0: |
| 1081 | ewake = 1; |
| 1082 | break; |
| 1083 | } |
| 1084 | } |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1085 | wake_up_locked(&ep->wq); |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1086 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1087 | if (waitqueue_active(&ep->poll_wait)) |
| 1088 | pwake++; |
| 1089 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1090 | out_unlock: |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1091 | spin_unlock_irqrestore(&ep->lock, flags); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1092 | |
| 1093 | /* We have to call this outside the lock */ |
| 1094 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1095 | ep_poll_safewake(&ep->poll_wait); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1096 | |
Oleg Nesterov | d325f1f | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 1097 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
| 1098 | ewake = 1; |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1099 | |
Oleg Nesterov | d325f1f | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 1100 | if ((unsigned long)key & POLLFREE) { |
| 1101 | /* |
| 1102 | * If we race with ep_remove_wait_queue() it can miss |
| 1103 | * ->whead = NULL and do another remove_wait_queue() after |
| 1104 | * us, so we can't use __remove_wait_queue(). |
| 1105 | */ |
| 1106 | list_del_init(&wait->task_list); |
| 1107 | /* |
| 1108 | * ->whead != NULL protects us from the race with ep_free() |
| 1109 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock |
| 1110 | * held by the caller. Once we nullify it, nothing protects |
| 1111 | * ep/epi or even wait. |
| 1112 | */ |
| 1113 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); |
| 1114 | } |
| 1115 | |
| 1116 | return ewake; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1117 | } |
| 1118 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | /* |
| 1120 | * This is the callback that is used to add our wait queue to the |
| 1121 | * target file wakeup lists. |
| 1122 | */ |
| 1123 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
| 1124 | poll_table *pt) |
| 1125 | { |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1126 | struct epitem *epi = ep_item_from_epqueue(pt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | struct eppoll_entry *pwq; |
| 1128 | |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 1129 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1130 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
| 1131 | pwq->whead = whead; |
| 1132 | pwq->base = epi; |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1133 | if (epi->event.events & EPOLLEXCLUSIVE) |
| 1134 | add_wait_queue_exclusive(whead, &pwq->wait); |
| 1135 | else |
| 1136 | add_wait_queue(whead, &pwq->wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1137 | list_add_tail(&pwq->llink, &epi->pwqlist); |
| 1138 | epi->nwait++; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1139 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1140 | /* We have to signal that an error occurred */ |
| 1141 | epi->nwait = -1; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1142 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | } |
| 1144 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1145 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
| 1146 | { |
| 1147 | int kcmp; |
| 1148 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; |
| 1149 | struct epitem *epic; |
| 1150 | |
| 1151 | while (*p) { |
| 1152 | parent = *p; |
| 1153 | epic = rb_entry(parent, struct epitem, rbn); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1154 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1155 | if (kcmp > 0) |
| 1156 | p = &parent->rb_right; |
| 1157 | else |
| 1158 | p = &parent->rb_left; |
| 1159 | } |
| 1160 | rb_link_node(&epi->rbn, parent, p); |
| 1161 | rb_insert_color(&epi->rbn, &ep->rbr); |
| 1162 | } |
| 1163 | |
Andrew Morton | a80a6b8 | 2012-11-08 15:53:35 -0800 | [diff] [blame] | 1164 | |
| 1165 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1166 | #define PATH_ARR_SIZE 5 |
| 1167 | /* |
| 1168 | * These are the number paths of length 1 to 5, that we are allowing to emanate |
| 1169 | * from a single file of interest. For example, we allow 1000 paths of length |
| 1170 | * 1, to emanate from each file of interest. This essentially represents the |
| 1171 | * potential wakeup paths, which need to be limited in order to avoid massive |
| 1172 | * uncontrolled wakeup storms. The common use case should be a single ep which |
| 1173 | * is connected to n file sources. In this case each file source has 1 path |
| 1174 | * of length 1. Thus, the numbers below should be more than sufficient. These |
| 1175 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify |
| 1176 | * and delete can't add additional paths. Protected by the epmutex. |
| 1177 | */ |
| 1178 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; |
| 1179 | static int path_count[PATH_ARR_SIZE]; |
| 1180 | |
| 1181 | static int path_count_inc(int nests) |
| 1182 | { |
Jason Baron | 93dc610 | 2012-03-16 16:34:03 -0400 | [diff] [blame] | 1183 | /* Allow an arbitrary number of depth 1 paths */ |
| 1184 | if (nests == 0) |
| 1185 | return 0; |
| 1186 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1187 | if (++path_count[nests] > path_limits[nests]) |
| 1188 | return -1; |
| 1189 | return 0; |
| 1190 | } |
| 1191 | |
| 1192 | static void path_count_init(void) |
| 1193 | { |
| 1194 | int i; |
| 1195 | |
| 1196 | for (i = 0; i < PATH_ARR_SIZE; i++) |
| 1197 | path_count[i] = 0; |
| 1198 | } |
| 1199 | |
| 1200 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) |
| 1201 | { |
| 1202 | int error = 0; |
| 1203 | struct file *file = priv; |
| 1204 | struct file *child_file; |
| 1205 | struct epitem *epi; |
| 1206 | |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1207 | /* CTL_DEL can remove links here, but that can't increase our count */ |
| 1208 | rcu_read_lock(); |
| 1209 | list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1210 | child_file = epi->ep->file; |
| 1211 | if (is_file_epoll(child_file)) { |
| 1212 | if (list_empty(&child_file->f_ep_links)) { |
| 1213 | if (path_count_inc(call_nests)) { |
| 1214 | error = -1; |
| 1215 | break; |
| 1216 | } |
| 1217 | } else { |
| 1218 | error = ep_call_nested(&poll_loop_ncalls, |
| 1219 | EP_MAX_NESTS, |
| 1220 | reverse_path_check_proc, |
| 1221 | child_file, child_file, |
| 1222 | current); |
| 1223 | } |
| 1224 | if (error != 0) |
| 1225 | break; |
| 1226 | } else { |
| 1227 | printk(KERN_ERR "reverse_path_check_proc: " |
| 1228 | "file is not an ep!\n"); |
| 1229 | } |
| 1230 | } |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1231 | rcu_read_unlock(); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1232 | return error; |
| 1233 | } |
| 1234 | |
| 1235 | /** |
| 1236 | * reverse_path_check - The tfile_check_list is list of file *, which have |
| 1237 | * links that are proposed to be newly added. We need to |
| 1238 | * make sure that those added links don't add too many |
| 1239 | * paths such that we will spend all our time waking up |
| 1240 | * eventpoll objects. |
| 1241 | * |
| 1242 | * Returns: Returns zero if the proposed links don't create too many paths, |
| 1243 | * -1 otherwise. |
| 1244 | */ |
| 1245 | static int reverse_path_check(void) |
| 1246 | { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1247 | int error = 0; |
| 1248 | struct file *current_file; |
| 1249 | |
| 1250 | /* let's call this for all tfiles */ |
| 1251 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1252 | path_count_init(); |
| 1253 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
| 1254 | reverse_path_check_proc, current_file, |
| 1255 | current_file, current); |
| 1256 | if (error) |
| 1257 | break; |
| 1258 | } |
| 1259 | return error; |
| 1260 | } |
| 1261 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1262 | static int ep_create_wakeup_source(struct epitem *epi) |
| 1263 | { |
Al Viro | f2b4a58 | 2020-09-24 19:41:58 -0400 | [diff] [blame] | 1264 | struct name_snapshot n; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1265 | struct wakeup_source *ws; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1266 | |
| 1267 | if (!epi->ep->ws) { |
| 1268 | epi->ep->ws = wakeup_source_register("eventpoll"); |
| 1269 | if (!epi->ep->ws) |
| 1270 | return -ENOMEM; |
| 1271 | } |
| 1272 | |
Al Viro | f2b4a58 | 2020-09-24 19:41:58 -0400 | [diff] [blame] | 1273 | take_dentry_name_snapshot(&n, epi->ffd.file->f_path.dentry); |
| 1274 | ws = wakeup_source_register(n.name); |
| 1275 | release_dentry_name_snapshot(&n); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1276 | |
| 1277 | if (!ws) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1278 | return -ENOMEM; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1279 | rcu_assign_pointer(epi->ws, ws); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1280 | |
| 1281 | return 0; |
| 1282 | } |
| 1283 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1284 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
| 1285 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1286 | { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1287 | struct wakeup_source *ws = ep_wakeup_source(epi); |
| 1288 | |
Eric Wong | d6d67e7 | 2013-04-30 15:27:43 -0700 | [diff] [blame] | 1289 | RCU_INIT_POINTER(epi->ws, NULL); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1290 | |
| 1291 | /* |
| 1292 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is |
| 1293 | * used internally by wakeup_source_remove, too (called by |
| 1294 | * wakeup_source_unregister), so we cannot use call_rcu |
| 1295 | */ |
| 1296 | synchronize_rcu(); |
| 1297 | wakeup_source_unregister(ws); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1298 | } |
| 1299 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1300 | /* |
| 1301 | * Must be called with "mtx" held. |
| 1302 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1303 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1304 | struct file *tfile, int fd, int full_check) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1305 | { |
| 1306 | int error, revents, pwake = 0; |
| 1307 | unsigned long flags; |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1308 | long user_watches; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1309 | struct epitem *epi; |
| 1310 | struct ep_pqueue epq; |
| 1311 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1312 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
| 1313 | if (unlikely(user_watches >= max_user_watches)) |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1314 | return -ENOSPC; |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 1315 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1316 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1317 | |
| 1318 | /* Item initialization follow here ... */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1319 | INIT_LIST_HEAD(&epi->rdllink); |
| 1320 | INIT_LIST_HEAD(&epi->fllink); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1321 | INIT_LIST_HEAD(&epi->pwqlist); |
| 1322 | epi->ep = ep; |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1323 | ep_set_ffd(&epi->ffd, tfile, fd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1324 | epi->event = *event; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1325 | epi->nwait = 0; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1326 | epi->next = EP_UNACTIVE_PTR; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1327 | if (epi->event.events & EPOLLWAKEUP) { |
| 1328 | error = ep_create_wakeup_source(epi); |
| 1329 | if (error) |
| 1330 | goto error_create_wakeup_source; |
| 1331 | } else { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1332 | RCU_INIT_POINTER(epi->ws, NULL); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1333 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1334 | |
Al Viro | a16d314 | 2020-09-09 22:25:06 -0400 | [diff] [blame] | 1335 | /* Add the current item to the list of active epoll hook for this file */ |
| 1336 | spin_lock(&tfile->f_lock); |
| 1337 | list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links); |
| 1338 | spin_unlock(&tfile->f_lock); |
| 1339 | |
| 1340 | /* |
| 1341 | * Add the current item to the RB tree. All RB tree operations are |
| 1342 | * protected by "mtx", and ep_insert() is called with "mtx" held. |
| 1343 | */ |
| 1344 | ep_rbtree_insert(ep, epi); |
| 1345 | |
| 1346 | /* now check if we've created too many backpaths */ |
| 1347 | error = -EINVAL; |
| 1348 | if (full_check && reverse_path_check()) |
| 1349 | goto error_remove_epi; |
| 1350 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1351 | /* Initialize the poll table using the queue callback */ |
| 1352 | epq.epi = epi; |
| 1353 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); |
| 1354 | |
| 1355 | /* |
| 1356 | * Attach the item to the poll hooks and get current event bits. |
| 1357 | * We can safely use the file* here because its usage count has |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1358 | * been increased by the caller of this function. Note that after |
| 1359 | * this operation completes, the poll callback can start hitting |
| 1360 | * the new item. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1361 | */ |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 1362 | revents = ep_item_poll(epi, &epq.pt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1363 | |
| 1364 | /* |
| 1365 | * We have to check if something went wrong during the poll wait queue |
| 1366 | * install process. Namely an allocation for a wait queue failed due |
| 1367 | * high memory pressure. |
| 1368 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1369 | error = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | if (epi->nwait < 0) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1371 | goto error_unregister; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1373 | /* We have to drop the new item inside our item list to keep track of it */ |
| 1374 | spin_lock_irqsave(&ep->lock, flags); |
| 1375 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1376 | /* 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] | 1377 | if ((revents & event->events) && !ep_is_linked(&epi->rdllink)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1378 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1379 | ep_pm_stay_awake(epi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1380 | |
| 1381 | /* Notify waiting tasks that events are available */ |
| 1382 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1383 | wake_up_locked(&ep->wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1384 | if (waitqueue_active(&ep->poll_wait)) |
| 1385 | pwake++; |
| 1386 | } |
| 1387 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1388 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1389 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1390 | atomic_long_inc(&ep->user->epoll_watches); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1391 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | /* We have to call this outside the lock */ |
| 1393 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1394 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1395 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1396 | return 0; |
| 1397 | |
Al Viro | a16d314 | 2020-09-09 22:25:06 -0400 | [diff] [blame] | 1398 | error_unregister: |
| 1399 | ep_unregister_pollwait(ep, epi); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1400 | error_remove_epi: |
| 1401 | spin_lock(&tfile->f_lock); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1402 | list_del_rcu(&epi->fllink); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1403 | spin_unlock(&tfile->f_lock); |
| 1404 | |
| 1405 | rb_erase(&epi->rbn, &ep->rbr); |
| 1406 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1407 | /* |
| 1408 | * 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] | 1409 | * allocated wait queue. Note that we don't care about the ep->ovflist |
| 1410 | * list, since that is used/cleaned only inside a section bound by "mtx". |
| 1411 | * And ep_insert() is called with "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1412 | */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1413 | spin_lock_irqsave(&ep->lock, flags); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1414 | if (ep_is_linked(&epi->rdllink)) |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 1415 | list_del_init(&epi->rdllink); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1416 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1417 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1418 | wakeup_source_unregister(ep_wakeup_source(epi)); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1419 | |
| 1420 | error_create_wakeup_source: |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1421 | kmem_cache_free(epi_cache, epi); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1422 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1423 | return error; |
| 1424 | } |
| 1425 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1426 | /* |
| 1427 | * 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] | 1428 | * 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] | 1429 | */ |
| 1430 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) |
| 1431 | { |
| 1432 | int pwake = 0; |
| 1433 | unsigned int revents; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 1434 | poll_table pt; |
| 1435 | |
| 1436 | init_poll_funcptr(&pt, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1437 | |
| 1438 | /* |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1439 | * Set the new event interest mask before calling f_op->poll(); |
| 1440 | * otherwise we might miss an event that happens between the |
| 1441 | * f_op->poll() call and the new event set registering. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1442 | */ |
Eric Wong | 128dd17 | 2013-01-01 21:20:27 +0000 | [diff] [blame] | 1443 | epi->event.events = event->events; /* need barrier below */ |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1444 | epi->event.data = event->data; /* protected by mtx */ |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1445 | if (epi->event.events & EPOLLWAKEUP) { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1446 | if (!ep_has_wakeup_source(epi)) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1447 | ep_create_wakeup_source(epi); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1448 | } else if (ep_has_wakeup_source(epi)) { |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1449 | ep_destroy_wakeup_source(epi); |
| 1450 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1451 | |
| 1452 | /* |
Eric Wong | 128dd17 | 2013-01-01 21:20:27 +0000 | [diff] [blame] | 1453 | * The following barrier has two effects: |
| 1454 | * |
| 1455 | * 1) Flush epi changes above to other CPUs. This ensures |
| 1456 | * we do not miss events from ep_poll_callback if an |
| 1457 | * event occurs immediately after we call f_op->poll(). |
| 1458 | * We need this because we did not take ep->lock while |
| 1459 | * changing epi above (but ep_poll_callback does take |
| 1460 | * ep->lock). |
| 1461 | * |
| 1462 | * 2) We also need to ensure we do not miss _past_ events |
| 1463 | * when calling f_op->poll(). This barrier also |
| 1464 | * pairs with the barrier in wq_has_sleeper (see |
| 1465 | * comments for wq_has_sleeper). |
| 1466 | * |
| 1467 | * This barrier will now guarantee ep_poll_callback or f_op->poll |
| 1468 | * (or both) will notice the readiness of an item. |
| 1469 | */ |
| 1470 | smp_mb(); |
| 1471 | |
| 1472 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | * Get current event bits. We can safely use the file* here because |
| 1474 | * its usage count has been increased by the caller of this function. |
| 1475 | */ |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 1476 | revents = ep_item_poll(epi, &pt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1478 | /* |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1479 | * 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] | 1480 | * list, push it inside. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1481 | */ |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1482 | if (revents & event->events) { |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1483 | spin_lock_irq(&ep->lock); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1484 | if (!ep_is_linked(&epi->rdllink)) { |
| 1485 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1486 | ep_pm_stay_awake(epi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1487 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1488 | /* Notify waiting tasks that events are available */ |
| 1489 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2 | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1490 | wake_up_locked(&ep->wq); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1491 | if (waitqueue_active(&ep->poll_wait)) |
| 1492 | pwake++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1493 | } |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1494 | spin_unlock_irq(&ep->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1496 | |
| 1497 | /* We have to call this outside the lock */ |
| 1498 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1499 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | |
| 1501 | return 0; |
| 1502 | } |
| 1503 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1504 | static int ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
| 1505 | void *priv) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1506 | { |
| 1507 | struct ep_send_events_data *esed = priv; |
| 1508 | int eventcnt; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1509 | unsigned int revents; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1510 | struct epitem *epi; |
| 1511 | struct epoll_event __user *uevent; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1512 | struct wakeup_source *ws; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 1513 | poll_table pt; |
| 1514 | |
| 1515 | init_poll_funcptr(&pt, NULL); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1516 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1517 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1518 | * We can loop without lock because we are passed a task private list. |
| 1519 | * Items cannot vanish during the loop because ep_scan_ready_list() is |
| 1520 | * holding "mtx" during this call. |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1521 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1522 | for (eventcnt = 0, uevent = esed->events; |
| 1523 | !list_empty(head) && eventcnt < esed->maxevents;) { |
| 1524 | epi = list_first_entry(head, struct epitem, rdllink); |
| 1525 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1526 | /* |
| 1527 | * Activate ep->ws before deactivating epi->ws to prevent |
| 1528 | * triggering auto-suspend here (in case we reactive epi->ws |
| 1529 | * below). |
| 1530 | * |
| 1531 | * This could be rearranged to delay the deactivation of epi->ws |
| 1532 | * instead, but then epi->ws would temporarily be out of sync |
| 1533 | * with ep_is_linked(). |
| 1534 | */ |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1535 | ws = ep_wakeup_source(epi); |
| 1536 | if (ws) { |
| 1537 | if (ws->active) |
| 1538 | __pm_stay_awake(ep->ws); |
| 1539 | __pm_relax(ws); |
| 1540 | } |
| 1541 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1542 | list_del_init(&epi->rdllink); |
| 1543 | |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 1544 | revents = ep_item_poll(epi, &pt); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1545 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1546 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1547 | * If the event mask intersect the caller-requested one, |
| 1548 | * deliver the event to userspace. Again, ep_scan_ready_list() |
| 1549 | * is holding "mtx", so no operations coming from userspace |
| 1550 | * can change the item. |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1551 | */ |
| 1552 | if (revents) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1553 | if (__put_user(revents, &uevent->events) || |
Tony Battersby | d030588 | 2009-03-31 15:24:14 -0700 | [diff] [blame] | 1554 | __put_user(epi->event.data, &uevent->data)) { |
| 1555 | list_add(&epi->rdllink, head); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1556 | ep_pm_stay_awake(epi); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1557 | return eventcnt ? eventcnt : -EFAULT; |
Tony Battersby | d030588 | 2009-03-31 15:24:14 -0700 | [diff] [blame] | 1558 | } |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1559 | eventcnt++; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1560 | uevent++; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1561 | if (epi->event.events & EPOLLONESHOT) |
| 1562 | epi->event.events &= EP_PRIVATE_BITS; |
| 1563 | else if (!(epi->event.events & EPOLLET)) { |
| 1564 | /* |
| 1565 | * If this file has been added with Level |
| 1566 | * Trigger mode, we need to insert back inside |
| 1567 | * the ready list, so that the next call to |
| 1568 | * epoll_wait() will check again the events |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1569 | * availability. At this point, no one can insert |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1570 | * into ep->rdllist besides us. The epoll_ctl() |
| 1571 | * callers are locked out by |
| 1572 | * ep_scan_ready_list() holding "mtx" and the |
| 1573 | * poll callback will queue them in ep->ovflist. |
| 1574 | */ |
| 1575 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1576 | ep_pm_stay_awake(epi); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1577 | } |
| 1578 | } |
| 1579 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1580 | |
| 1581 | return eventcnt; |
| 1582 | } |
| 1583 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1584 | static int ep_send_events(struct eventpoll *ep, |
| 1585 | struct epoll_event __user *events, int maxevents) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1587 | struct ep_send_events_data esed; |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 1588 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1589 | esed.maxevents = maxevents; |
| 1590 | esed.events = events; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1591 | |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1592 | 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] | 1593 | } |
| 1594 | |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1595 | static inline struct timespec64 ep_set_mstimeout(long ms) |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1596 | { |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1597 | struct timespec64 now, ts = { |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1598 | .tv_sec = ms / MSEC_PER_SEC, |
| 1599 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), |
| 1600 | }; |
| 1601 | |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1602 | ktime_get_ts64(&now); |
| 1603 | return timespec64_add_safe(now, ts); |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1604 | } |
| 1605 | |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1606 | /** |
| 1607 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied |
| 1608 | * event buffer. |
| 1609 | * |
| 1610 | * @ep: Pointer to the eventpoll context. |
| 1611 | * @events: Pointer to the userspace buffer where the ready events should be |
| 1612 | * stored. |
| 1613 | * @maxevents: Size (in terms of number of events) of the caller event buffer. |
| 1614 | * @timeout: Maximum timeout for the ready events fetch operation, in |
| 1615 | * milliseconds. If the @timeout is zero, the function will not block, |
| 1616 | * while if the @timeout is less than zero, the function will block |
| 1617 | * until at least one event has been retrieved (or an error |
| 1618 | * occurred). |
| 1619 | * |
| 1620 | * Returns: Returns the number of ready events which have been fetched, or an |
| 1621 | * error code, in case of error. |
| 1622 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1623 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
| 1624 | int maxevents, long timeout) |
| 1625 | { |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1626 | int res = 0, eavail, timed_out = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1627 | unsigned long flags; |
John Stultz | da8b44d | 2016-03-17 14:20:51 -0700 | [diff] [blame] | 1628 | u64 slack = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1629 | wait_queue_t wait; |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1630 | ktime_t expires, *to = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1631 | |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1632 | if (timeout > 0) { |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1633 | struct timespec64 end_time = ep_set_mstimeout(timeout); |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1634 | |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1635 | slack = select_estimate_accuracy(&end_time); |
| 1636 | to = &expires; |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1637 | *to = timespec64_to_ktime(end_time); |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1638 | } else if (timeout == 0) { |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1639 | /* |
| 1640 | * Avoid the unnecessary trip to the wait queue loop, if the |
| 1641 | * caller specified a non blocking operation. |
| 1642 | */ |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1643 | timed_out = 1; |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1644 | spin_lock_irqsave(&ep->lock, flags); |
| 1645 | goto check_events; |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1646 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1647 | |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1648 | fetch_events: |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1649 | spin_lock_irqsave(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1650 | |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1651 | if (!ep_events_available(ep)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1652 | /* |
| 1653 | * We don't have any available event to return to the caller. |
| 1654 | * We need to sleep here, and we will be wake up by |
| 1655 | * ep_poll_callback() when events will become available. |
| 1656 | */ |
| 1657 | init_waitqueue_entry(&wait, current); |
Changli Gao | a93d2f1 | 2010-05-07 14:33:26 +0800 | [diff] [blame] | 1658 | __add_wait_queue_exclusive(&ep->wq, &wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1659 | |
| 1660 | for (;;) { |
| 1661 | /* |
| 1662 | * We don't want to sleep if the ep_poll_callback() sends us |
| 1663 | * a wakeup in between. That's why we set the task state |
| 1664 | * to TASK_INTERRUPTIBLE before doing the checks. |
| 1665 | */ |
| 1666 | set_current_state(TASK_INTERRUPTIBLE); |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1667 | if (ep_events_available(ep) || timed_out) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1668 | break; |
| 1669 | if (signal_pending(current)) { |
| 1670 | res = -EINTR; |
| 1671 | break; |
| 1672 | } |
| 1673 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1674 | spin_unlock_irqrestore(&ep->lock, flags); |
Colin Cross | eb5d550 | 2013-05-06 23:50:16 +0000 | [diff] [blame] | 1675 | if (!freezable_schedule_hrtimeout_range(to, slack, |
| 1676 | HRTIMER_MODE_ABS)) |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1677 | timed_out = 1; |
| 1678 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1679 | spin_lock_irqsave(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1680 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1681 | |
Davidlohr Bueso | 4d5755b | 2015-02-13 14:39:08 -0800 | [diff] [blame] | 1682 | __remove_wait_queue(&ep->wq, &wait); |
| 1683 | __set_current_state(TASK_RUNNING); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1684 | } |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1685 | check_events: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1686 | /* Is it worth to try to dig for events ? */ |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1687 | eavail = ep_events_available(ep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1688 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1689 | spin_unlock_irqrestore(&ep->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1690 | |
| 1691 | /* |
| 1692 | * Try to transfer events to user space. In case we get 0 events and |
| 1693 | * there's still timeout left over, we go trying again in search of |
| 1694 | * more luck. |
| 1695 | */ |
| 1696 | if (!res && eavail && |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1697 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1698 | goto fetch_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | |
| 1700 | return res; |
| 1701 | } |
| 1702 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1703 | /** |
| 1704 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() |
| 1705 | * API, to verify that adding an epoll file inside another |
| 1706 | * epoll structure, does not violate the constraints, in |
| 1707 | * terms of closed loops, or too deep chains (which can |
| 1708 | * result in excessive stack usage). |
| 1709 | * |
| 1710 | * @priv: Pointer to the epoll file to be currently checked. |
| 1711 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll |
| 1712 | * data structure pointer. |
| 1713 | * @call_nests: Current dept of the @ep_call_nested() call stack. |
| 1714 | * |
| 1715 | * Returns: Returns zero if adding the epoll @file inside current epoll |
| 1716 | * structure @ep does not violate the constraints, or -1 otherwise. |
| 1717 | */ |
| 1718 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) |
| 1719 | { |
| 1720 | int error = 0; |
| 1721 | struct file *file = priv; |
| 1722 | struct eventpoll *ep = file->private_data; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1723 | struct eventpoll *ep_tovisit; |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1724 | struct rb_node *rbp; |
| 1725 | struct epitem *epi; |
| 1726 | |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 1727 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
Al Viro | 0ece241 | 2020-09-10 08:30:05 -0400 | [diff] [blame] | 1728 | ep->gen = loop_check_gen; |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1729 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
| 1730 | epi = rb_entry(rbp, struct epitem, rbn); |
| 1731 | if (unlikely(is_file_epoll(epi->ffd.file))) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1732 | ep_tovisit = epi->ffd.file->private_data; |
Al Viro | 0ece241 | 2020-09-10 08:30:05 -0400 | [diff] [blame] | 1733 | if (ep_tovisit->gen == loop_check_gen) |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1734 | continue; |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1735 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1736 | ep_loop_check_proc, epi->ffd.file, |
| 1737 | ep_tovisit, current); |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1738 | if (error != 0) |
| 1739 | break; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1740 | } else { |
| 1741 | /* |
| 1742 | * If we've reached a file that is not associated with |
| 1743 | * an ep, then we need to check if the newly added |
| 1744 | * links are going to add too many wakeup paths. We do |
| 1745 | * this by adding it to the tfile_check_list, if it's |
| 1746 | * not already there, and calling reverse_path_check() |
| 1747 | * during ep_insert(). |
| 1748 | */ |
Marc Zyngier | 9bbd203 | 2020-08-19 17:12:17 +0100 | [diff] [blame] | 1749 | if (list_empty(&epi->ffd.file->f_tfile_llink)) { |
Al Viro | 8238ee9 | 2020-09-02 11:30:48 -0400 | [diff] [blame] | 1750 | if (get_file_rcu(epi->ffd.file)) |
| 1751 | list_add(&epi->ffd.file->f_tfile_llink, |
| 1752 | &tfile_check_list); |
Marc Zyngier | 9bbd203 | 2020-08-19 17:12:17 +0100 | [diff] [blame] | 1753 | } |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1754 | } |
| 1755 | } |
| 1756 | mutex_unlock(&ep->mtx); |
| 1757 | |
| 1758 | return error; |
| 1759 | } |
| 1760 | |
| 1761 | /** |
| 1762 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) |
| 1763 | * another epoll file (represented by @ep) does not create |
| 1764 | * closed loops or too deep chains. |
| 1765 | * |
| 1766 | * @ep: Pointer to the epoll private data structure. |
| 1767 | * @file: Pointer to the epoll file to be checked. |
| 1768 | * |
| 1769 | * Returns: Returns zero if adding the epoll @file inside current epoll |
| 1770 | * structure @ep does not violate the constraints, or -1 otherwise. |
| 1771 | */ |
| 1772 | static int ep_loop_check(struct eventpoll *ep, struct file *file) |
| 1773 | { |
Al Viro | 0ece241 | 2020-09-10 08:30:05 -0400 | [diff] [blame] | 1774 | return ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1775 | ep_loop_check_proc, file, ep, current); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1776 | } |
| 1777 | |
| 1778 | static void clear_tfile_check_list(void) |
| 1779 | { |
| 1780 | struct file *file; |
| 1781 | |
| 1782 | /* first clear the tfile_check_list */ |
| 1783 | while (!list_empty(&tfile_check_list)) { |
| 1784 | file = list_first_entry(&tfile_check_list, struct file, |
| 1785 | f_tfile_llink); |
| 1786 | list_del_init(&file->f_tfile_llink); |
Marc Zyngier | 9bbd203 | 2020-08-19 17:12:17 +0100 | [diff] [blame] | 1787 | fput(file); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1788 | } |
| 1789 | INIT_LIST_HEAD(&tfile_check_list); |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1790 | } |
| 1791 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1792 | /* |
Andrew Morton | 523723b | 2008-08-12 15:09:01 -0700 | [diff] [blame] | 1793 | * Open an eventpoll file descriptor. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1794 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1795 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1796 | { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1797 | int error, fd; |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1798 | struct eventpoll *ep = NULL; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1799 | struct file *file; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1800 | |
Ulrich Drepper | e38b36f | 2008-07-23 21:29:42 -0700 | [diff] [blame] | 1801 | /* Check the EPOLL_* constant for consistency. */ |
| 1802 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); |
| 1803 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1804 | if (flags & ~EPOLL_CLOEXEC) |
| 1805 | return -EINVAL; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1806 | /* |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1807 | * Create the internal data structure ("struct eventpoll"). |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1808 | */ |
Ulrich Drepper | 9fe5ad9 | 2008-07-23 21:29:43 -0700 | [diff] [blame] | 1809 | error = ep_alloc(&ep); |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1810 | if (error < 0) |
| 1811 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1812 | /* |
| 1813 | * Creates all the items needed to setup an eventpoll file. That is, |
Al Viro | 2030a42 | 2008-02-23 06:46:49 -0500 | [diff] [blame] | 1814 | * a file structure and a free file descriptor. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1815 | */ |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1816 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
| 1817 | if (fd < 0) { |
| 1818 | error = fd; |
| 1819 | goto out_free_ep; |
| 1820 | } |
| 1821 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, |
Roland Dreier | 628ff7c | 2009-12-18 09:41:24 -0800 | [diff] [blame] | 1822 | O_RDWR | (flags & O_CLOEXEC)); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1823 | if (IS_ERR(file)) { |
| 1824 | error = PTR_ERR(file); |
| 1825 | goto out_free_fd; |
| 1826 | } |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1827 | ep->file = file; |
Al Viro | 9802274 | 2012-08-17 22:42:36 -0400 | [diff] [blame] | 1828 | fd_install(fd, file); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1829 | return fd; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1830 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1831 | out_free_fd: |
| 1832 | put_unused_fd(fd); |
| 1833 | out_free_ep: |
| 1834 | ep_free(ep); |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1835 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1836 | } |
| 1837 | |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1838 | SYSCALL_DEFINE1(epoll_create, int, size) |
Ulrich Drepper | a0998b5 | 2008-07-23 21:29:27 -0700 | [diff] [blame] | 1839 | { |
Davide Libenzi | bfe3891 | 2009-05-12 13:19:44 -0700 | [diff] [blame] | 1840 | if (size <= 0) |
Ulrich Drepper | 9fe5ad9 | 2008-07-23 21:29:43 -0700 | [diff] [blame] | 1841 | return -EINVAL; |
| 1842 | |
| 1843 | return sys_epoll_create1(0); |
Ulrich Drepper | a0998b5 | 2008-07-23 21:29:27 -0700 | [diff] [blame] | 1844 | } |
| 1845 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1846 | /* |
| 1847 | * The following function implements the controller interface for |
| 1848 | * the eventpoll file that enables the insertion/removal/change of |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 1849 | * file descriptors inside the interest set. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1850 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1851 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
| 1852 | struct epoll_event __user *, event) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1853 | { |
| 1854 | int error; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1855 | int full_check = 0; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 1856 | struct fd f, tf; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1857 | struct eventpoll *ep; |
| 1858 | struct epitem *epi; |
| 1859 | struct epoll_event epds; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1860 | struct eventpoll *tep = NULL; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1861 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1862 | error = -EFAULT; |
| 1863 | if (ep_op_has_event(op) && |
| 1864 | copy_from_user(&epds, event, sizeof(struct epoll_event))) |
| 1865 | goto error_return; |
| 1866 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1867 | error = -EBADF; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 1868 | f = fdget(epfd); |
| 1869 | if (!f.file) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1870 | goto error_return; |
| 1871 | |
| 1872 | /* Get the "struct file *" for the target file */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 1873 | tf = fdget(fd); |
| 1874 | if (!tf.file) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1875 | goto error_fput; |
| 1876 | |
| 1877 | /* The target file descriptor must support poll */ |
| 1878 | error = -EPERM; |
Al Viro | 72c2d53 | 2013-09-22 16:27:52 -0400 | [diff] [blame] | 1879 | if (!tf.file->f_op->poll) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1880 | goto error_tgt_fput; |
| 1881 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1882 | /* Check if EPOLLWAKEUP is allowed */ |
Nicolas Iooss | c680e41 | 2014-09-09 14:50:51 -0700 | [diff] [blame] | 1883 | if (ep_op_has_event(op)) |
| 1884 | ep_take_care_of_epollwakeup(&epds); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1885 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1886 | /* |
| 1887 | * We have to check that the file structure underneath the file descriptor |
| 1888 | * the user passed to us _is_ an eventpoll file. And also we do not permit |
| 1889 | * adding an epoll file descriptor inside itself. |
| 1890 | */ |
| 1891 | error = -EINVAL; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 1892 | if (f.file == tf.file || !is_file_epoll(f.file)) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1893 | goto error_tgt_fput; |
| 1894 | |
| 1895 | /* |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1896 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, |
| 1897 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. |
| 1898 | * Also, we do not currently supported nested exclusive wakeups. |
| 1899 | */ |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 1900 | if (epds.events & EPOLLEXCLUSIVE) { |
| 1901 | if (op == EPOLL_CTL_MOD) |
| 1902 | goto error_tgt_fput; |
| 1903 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || |
| 1904 | (epds.events & ~EPOLLEXCLUSIVE_OK_BITS))) |
| 1905 | goto error_tgt_fput; |
| 1906 | } |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1907 | |
| 1908 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1909 | * At this point it is safe to assume that the "private_data" contains |
| 1910 | * our own data structure. |
| 1911 | */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 1912 | ep = f.file->private_data; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1913 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1914 | /* |
| 1915 | * When we insert an epoll file descriptor, inside another epoll file |
| 1916 | * descriptor, there is the change of creating closed loops, which are |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1917 | * better be handled here, than in more critical paths. While we are |
| 1918 | * checking for loops we also determine the list of files reachable |
| 1919 | * and hang them on the tfile_check_list, so we can check that we |
| 1920 | * haven't created too many possible wakeup paths. |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1921 | * |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1922 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
| 1923 | * the epoll file descriptor is attaching directly to a wakeup source, |
| 1924 | * unless the epoll file descriptor is nested. The purpose of taking the |
| 1925 | * 'epmutex' on add is to prevent complex toplogies such as loops and |
| 1926 | * deep wakeup paths from forming in parallel through multiple |
| 1927 | * EPOLL_CTL_ADD operations. |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1928 | */ |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 1929 | mutex_lock_nested(&ep->mtx, 0); |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1930 | if (op == EPOLL_CTL_ADD) { |
| 1931 | if (!list_empty(&f.file->f_ep_links) || |
Al Viro | 187af78 | 2020-09-10 08:33:27 -0400 | [diff] [blame] | 1932 | ep->gen == loop_check_gen || |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1933 | is_file_epoll(tf.file)) { |
| 1934 | full_check = 1; |
| 1935 | mutex_unlock(&ep->mtx); |
| 1936 | mutex_lock(&epmutex); |
| 1937 | if (is_file_epoll(tf.file)) { |
| 1938 | error = -ELOOP; |
Al Viro | b3ce6ca | 2020-08-22 18:25:52 -0400 | [diff] [blame] | 1939 | if (ep_loop_check(ep, tf.file) != 0) |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1940 | goto error_tgt_fput; |
Marc Zyngier | 9bbd203 | 2020-08-19 17:12:17 +0100 | [diff] [blame] | 1941 | } else { |
| 1942 | get_file(tf.file); |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1943 | list_add(&tf.file->f_tfile_llink, |
| 1944 | &tfile_check_list); |
Marc Zyngier | 9bbd203 | 2020-08-19 17:12:17 +0100 | [diff] [blame] | 1945 | } |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1946 | mutex_lock_nested(&ep->mtx, 0); |
| 1947 | if (is_file_epoll(tf.file)) { |
| 1948 | tep = tf.file->private_data; |
| 1949 | mutex_lock_nested(&tep->mtx, 1); |
| 1950 | } |
| 1951 | } |
| 1952 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1953 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 1954 | /* |
| 1955 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" |
| 1956 | * above, we can be sure to be able to use the item looked up by |
| 1957 | * ep_find() till we release the mutex. |
| 1958 | */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 1959 | epi = ep_find(ep, tf.file, fd); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1960 | |
| 1961 | error = -EINVAL; |
| 1962 | switch (op) { |
| 1963 | case EPOLL_CTL_ADD: |
| 1964 | if (!epi) { |
| 1965 | epds.events |= POLLERR | POLLHUP; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1966 | error = ep_insert(ep, &epds, tf.file, fd, full_check); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1967 | } else |
| 1968 | error = -EEXIST; |
| 1969 | break; |
| 1970 | case EPOLL_CTL_DEL: |
| 1971 | if (epi) |
| 1972 | error = ep_remove(ep, epi); |
| 1973 | else |
| 1974 | error = -ENOENT; |
| 1975 | break; |
| 1976 | case EPOLL_CTL_MOD: |
| 1977 | if (epi) { |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 1978 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
| 1979 | epds.events |= POLLERR | POLLHUP; |
| 1980 | error = ep_modify(ep, epi, &epds); |
| 1981 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1982 | } else |
| 1983 | error = -ENOENT; |
| 1984 | break; |
| 1985 | } |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1986 | if (tep != NULL) |
| 1987 | mutex_unlock(&tep->mtx); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1988 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1989 | |
| 1990 | error_tgt_fput: |
Al Viro | b3ce6ca | 2020-08-22 18:25:52 -0400 | [diff] [blame] | 1991 | if (full_check) { |
| 1992 | clear_tfile_check_list(); |
Al Viro | 0ece241 | 2020-09-10 08:30:05 -0400 | [diff] [blame] | 1993 | loop_check_gen++; |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1994 | mutex_unlock(&epmutex); |
Al Viro | b3ce6ca | 2020-08-22 18:25:52 -0400 | [diff] [blame] | 1995 | } |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1996 | |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 1997 | fdput(tf); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1998 | error_fput: |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 1999 | fdput(f); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2000 | error_return: |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2001 | |
| 2002 | return error; |
| 2003 | } |
| 2004 | |
| 2005 | /* |
| 2006 | * Implement the event wait interface for the eventpoll file. It is the kernel |
| 2007 | * part of the user space epoll_wait(2). |
| 2008 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 2009 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
| 2010 | int, maxevents, int, timeout) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2011 | { |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2012 | int error; |
| 2013 | struct fd f; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2014 | struct eventpoll *ep; |
| 2015 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2016 | /* The maximum number of event must be greater than zero */ |
| 2017 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) |
| 2018 | return -EINVAL; |
| 2019 | |
| 2020 | /* Verify that the area passed by the user is writeable */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2021 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) |
| 2022 | return -EFAULT; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2023 | |
| 2024 | /* Get the "struct file *" for the eventpoll file */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2025 | f = fdget(epfd); |
| 2026 | if (!f.file) |
| 2027 | return -EBADF; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2028 | |
| 2029 | /* |
| 2030 | * We have to check that the file structure underneath the fd |
| 2031 | * the user passed to us _is_ an eventpoll file. |
| 2032 | */ |
| 2033 | error = -EINVAL; |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2034 | if (!is_file_epoll(f.file)) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2035 | goto error_fput; |
| 2036 | |
| 2037 | /* |
| 2038 | * At this point it is safe to assume that the "private_data" contains |
| 2039 | * our own data structure. |
| 2040 | */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2041 | ep = f.file->private_data; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2042 | |
| 2043 | /* Time to fish for events ... */ |
| 2044 | error = ep_poll(ep, events, maxevents, timeout); |
| 2045 | |
| 2046 | error_fput: |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2047 | fdput(f); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2048 | return error; |
| 2049 | } |
| 2050 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2051 | /* |
| 2052 | * Implement the event wait interface for the eventpoll file. It is the kernel |
| 2053 | * part of the user space epoll_pwait(2). |
| 2054 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 2055 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
| 2056 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, |
| 2057 | size_t, sigsetsize) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2058 | { |
| 2059 | int error; |
| 2060 | sigset_t ksigmask, sigsaved; |
| 2061 | |
| 2062 | /* |
| 2063 | * If the caller wants a certain signal mask to be set during the wait, |
| 2064 | * we apply it here. |
| 2065 | */ |
| 2066 | if (sigmask) { |
| 2067 | if (sigsetsize != sizeof(sigset_t)) |
| 2068 | return -EINVAL; |
| 2069 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) |
| 2070 | return -EFAULT; |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2071 | sigsaved = current->blocked; |
| 2072 | set_current_blocked(&ksigmask); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2073 | } |
| 2074 | |
| 2075 | error = sys_epoll_wait(epfd, events, maxevents, timeout); |
| 2076 | |
| 2077 | /* |
| 2078 | * If we changed the signal mask, we need to restore the original one. |
| 2079 | * In case we've got a signal while waiting, we do not restore the |
| 2080 | * signal mask yet, and we allow do_signal() to deliver the signal on |
| 2081 | * the way back to userspace, before the signal mask is restored. |
| 2082 | */ |
| 2083 | if (sigmask) { |
| 2084 | if (error == -EINTR) { |
| 2085 | memcpy(¤t->saved_sigmask, &sigsaved, |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 2086 | sizeof(sigsaved)); |
Roland McGrath | 4e4c22c | 2008-04-30 00:53:06 -0700 | [diff] [blame] | 2087 | set_restore_sigmask(); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2088 | } else |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2089 | set_current_blocked(&sigsaved); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2090 | } |
| 2091 | |
| 2092 | return error; |
| 2093 | } |
| 2094 | |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2095 | #ifdef CONFIG_COMPAT |
| 2096 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, |
| 2097 | struct epoll_event __user *, events, |
| 2098 | int, maxevents, int, timeout, |
| 2099 | const compat_sigset_t __user *, sigmask, |
| 2100 | compat_size_t, sigsetsize) |
| 2101 | { |
| 2102 | long err; |
| 2103 | compat_sigset_t csigmask; |
| 2104 | sigset_t ksigmask, sigsaved; |
| 2105 | |
| 2106 | /* |
| 2107 | * If the caller wants a certain signal mask to be set during the wait, |
| 2108 | * we apply it here. |
| 2109 | */ |
| 2110 | if (sigmask) { |
| 2111 | if (sigsetsize != sizeof(compat_sigset_t)) |
| 2112 | return -EINVAL; |
| 2113 | if (copy_from_user(&csigmask, sigmask, sizeof(csigmask))) |
| 2114 | return -EFAULT; |
| 2115 | sigset_from_compat(&ksigmask, &csigmask); |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2116 | sigsaved = current->blocked; |
| 2117 | set_current_blocked(&ksigmask); |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2118 | } |
| 2119 | |
| 2120 | err = sys_epoll_wait(epfd, events, maxevents, timeout); |
| 2121 | |
| 2122 | /* |
| 2123 | * If we changed the signal mask, we need to restore the original one. |
| 2124 | * In case we've got a signal while waiting, we do not restore the |
| 2125 | * signal mask yet, and we allow do_signal() to deliver the signal on |
| 2126 | * the way back to userspace, before the signal mask is restored. |
| 2127 | */ |
| 2128 | if (sigmask) { |
| 2129 | if (err == -EINTR) { |
| 2130 | memcpy(¤t->saved_sigmask, &sigsaved, |
| 2131 | sizeof(sigsaved)); |
| 2132 | set_restore_sigmask(); |
| 2133 | } else |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2134 | set_current_blocked(&sigsaved); |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2135 | } |
| 2136 | |
| 2137 | return err; |
| 2138 | } |
| 2139 | #endif |
| 2140 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2141 | static int __init eventpoll_init(void) |
| 2142 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 2143 | struct sysinfo si; |
| 2144 | |
| 2145 | si_meminfo(&si); |
Davide Libenzi | 9df04e1 | 2009-01-29 14:25:26 -0800 | [diff] [blame] | 2146 | /* |
| 2147 | * Allows top 4% of lomem to be allocated for epoll watches (per user). |
| 2148 | */ |
| 2149 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 2150 | EP_ITEM_COST; |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 2151 | BUG_ON(max_user_watches < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2153 | /* |
| 2154 | * Initialize the structure used to perform epoll file descriptor |
| 2155 | * inclusion loops checks. |
| 2156 | */ |
| 2157 | ep_nested_calls_init(&poll_loop_ncalls); |
| 2158 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2159 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 2160 | ep_nested_calls_init(&poll_safewake_ncalls); |
| 2161 | |
| 2162 | /* Initialize the structure used to perform file's f_op->poll() calls */ |
| 2163 | ep_nested_calls_init(&poll_readywalk_ncalls); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2164 | |
Eric Wong | 39732ca | 2013-04-30 15:27:38 -0700 | [diff] [blame] | 2165 | /* |
| 2166 | * We can have many thousands of epitems, so prevent this from |
| 2167 | * using an extra cache line on 64-bit (and smaller) CPUs |
| 2168 | */ |
| 2169 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); |
| 2170 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2171 | /* Allocates slab cache used to allocate "struct epitem" items */ |
| 2172 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 2173 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2174 | |
| 2175 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ |
| 2176 | pwq_cache = kmem_cache_create("eventpoll_pwq", |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 2177 | sizeof(struct eppoll_entry), 0, SLAB_PANIC, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2178 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2179 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2180 | } |
Davide Libenzi | cea6924 | 2007-05-10 22:23:22 -0700 | [diff] [blame] | 2181 | fs_initcall(eventpoll_init); |