Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 2 | * fs/eventpoll.c (Efficient event retrieval implementation) |
| 3 | * Copyright (C) 2001,...,2009 Davide Libenzi |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify |
| 6 | * it under the terms of the GNU General Public License as published by |
| 7 | * the Free Software Foundation; either version 2 of the License, or |
| 8 | * (at your option) any later version. |
| 9 | * |
| 10 | * Davide Libenzi <davidel@xmailserver.org> |
| 11 | * |
| 12 | */ |
| 13 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/init.h> |
| 15 | #include <linux/kernel.h> |
Ingo Molnar | 174cd4b | 2017-02-02 19:15:33 +0100 | [diff] [blame] | 16 | #include <linux/sched/signal.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #include <linux/fs.h> |
| 18 | #include <linux/file.h> |
| 19 | #include <linux/signal.h> |
| 20 | #include <linux/errno.h> |
| 21 | #include <linux/mm.h> |
| 22 | #include <linux/slab.h> |
| 23 | #include <linux/poll.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/string.h> |
| 25 | #include <linux/list.h> |
| 26 | #include <linux/hash.h> |
| 27 | #include <linux/spinlock.h> |
| 28 | #include <linux/syscalls.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/rbtree.h> |
| 30 | #include <linux/wait.h> |
| 31 | #include <linux/eventpoll.h> |
| 32 | #include <linux/mount.h> |
| 33 | #include <linux/bitops.h> |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 34 | #include <linux/mutex.h> |
Davide Libenzi | da66f7c | 2007-05-10 22:23:21 -0700 | [diff] [blame] | 35 | #include <linux/anon_inodes.h> |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 36 | #include <linux/device.h> |
Colin Cross | 0c39a6a | 2013-05-06 23:50:16 +0000 | [diff] [blame] | 37 | #include <linux/freezer.h> |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 38 | #include <linux/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> |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 46 | #include <net/busy_poll.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | /* |
| 49 | * LOCKING: |
| 50 | * There are three level of locking required by epoll : |
| 51 | * |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 52 | * 1) epmutex (mutex) |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 53 | * 2) ep->mtx (mutex) |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 54 | * 3) ep->wq.lock (spinlock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | * |
| 56 | * The acquire order is the one listed above, from 1 to 3. |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 57 | * We need a spinlock (ep->wq.lock) because we manipulate objects |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | * from inside the poll callback, that might be triggered from |
| 59 | * a wake_up() that in turn might be called from IRQ context. |
| 60 | * So we can't sleep inside the poll callback and hence we need |
| 61 | * a spinlock. During the event transfer loop (from kernel to |
| 62 | * user space) we could end up sleeping due a copy_to_user(), so |
| 63 | * 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] | 64 | * mutex (ep->mtx). It is acquired during the event transfer loop, |
| 65 | * during epoll_ctl(EPOLL_CTL_DEL) and during eventpoll_release_file(). |
| 66 | * Then we also need a global mutex to serialize eventpoll_release_file() |
| 67 | * and ep_free(). |
| 68 | * This mutex is acquired by ep_free() during the epoll file |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | * cleanup path and it is also acquired by eventpoll_release_file() |
| 70 | * 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] | 71 | * close()d without a previous call to epoll_ctl(EPOLL_CTL_DEL). |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 72 | * It is also acquired when inserting an epoll fd onto another epoll |
| 73 | * fd. We do this so that we walk the epoll tree and ensure that this |
| 74 | * insertion does not create a cycle of epoll file descriptors, which |
| 75 | * could lead to deadlock. We need a global mutex to prevent two |
| 76 | * simultaneous inserts (A into B and B into A) from racing and |
| 77 | * constructing a cycle without either insert observing that it is |
| 78 | * going to. |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 79 | * It is necessary to acquire multiple "ep->mtx"es at once in the |
| 80 | * case when one epoll fd is added to another. In this case, we |
| 81 | * always acquire the locks in the order of nesting (i.e. after |
| 82 | * epoll_ctl(e1, EPOLL_CTL_ADD, e2), e1->mtx will always be acquired |
| 83 | * before e2->mtx). Since we disallow cycles of epoll file |
| 84 | * descriptors, this ensures that the mutexes are well-ordered. In |
| 85 | * order to communicate this nesting to lockdep, when walking a tree |
| 86 | * of epoll file descriptors, we use the current recursion depth as |
| 87 | * the lockdep subkey. |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 88 | * It is possible to drop the "ep->mtx" and to use the global |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 89 | * mutex "epmutex" (together with "ep->wq.lock") to have it working, |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 90 | * but having "ep->mtx" will make the interface more scalable. |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 91 | * Events that require holding "epmutex" are very rare, while for |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 92 | * normal operations the epoll private "ep->mtx" will guarantee |
| 93 | * a better scalability. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | */ |
| 95 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | /* Epoll private bits inside the event mask */ |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 97 | #define EP_PRIVATE_BITS (EPOLLWAKEUP | EPOLLONESHOT | EPOLLET | EPOLLEXCLUSIVE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 99 | #define EPOLLINOUT_BITS (EPOLLIN | EPOLLOUT) |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 100 | |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 101 | #define EPOLLEXCLUSIVE_OK_BITS (EPOLLINOUT_BITS | EPOLLERR | EPOLLHUP | \ |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 102 | EPOLLWAKEUP | EPOLLET | EPOLLEXCLUSIVE) |
| 103 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 104 | /* Maximum number of nesting allowed inside epoll sets */ |
| 105 | #define EP_MAX_NESTS 4 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | |
Davide Libenzi | b611967 | 2006-10-11 01:21:44 -0700 | [diff] [blame] | 107 | #define EP_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) |
| 108 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 109 | #define EP_UNACTIVE_PTR ((void *) -1L) |
| 110 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 111 | #define EP_ITEM_COST (sizeof(struct epitem) + sizeof(struct eppoll_entry)) |
| 112 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | struct epoll_filefd { |
| 114 | struct file *file; |
| 115 | int fd; |
Eric Wong | 39732ca | 2013-04-30 15:27:38 -0700 | [diff] [blame] | 116 | } __packed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | |
| 118 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 119 | * Structure used to track possible nested calls, for too deep recursions |
| 120 | * and loop cycles. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 122 | struct nested_call_node { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | struct list_head llink; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 124 | void *cookie; |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 125 | void *ctx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | }; |
| 127 | |
| 128 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 129 | * This structure is used as collector for nested calls, to check for |
| 130 | * maximum recursion dept and loop cycles. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 131 | */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 132 | struct nested_calls { |
| 133 | struct list_head tasks_call_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | spinlock_t lock; |
| 135 | }; |
| 136 | |
| 137 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | * Each file descriptor added to the eventpoll interface will |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 139 | * have an entry of this type linked to the "rbr" RB tree. |
Eric Wong | 39732ca | 2013-04-30 15:27:38 -0700 | [diff] [blame] | 140 | * Avoid increasing the size of this struct, there can be many thousands |
| 141 | * 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] | 142 | */ |
| 143 | struct epitem { |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 144 | union { |
| 145 | /* RB tree node links this structure to the eventpoll RB tree */ |
| 146 | struct rb_node rbn; |
| 147 | /* Used to free the struct epitem */ |
| 148 | struct rcu_head rcu; |
| 149 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 150 | |
| 151 | /* List header used to link this structure to the eventpoll ready list */ |
| 152 | struct list_head rdllink; |
| 153 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 154 | /* |
| 155 | * Works together "struct eventpoll"->ovflist in keeping the |
| 156 | * single linked chain of items. |
| 157 | */ |
| 158 | struct epitem *next; |
| 159 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | /* The file descriptor information this item refers to */ |
| 161 | struct epoll_filefd ffd; |
| 162 | |
| 163 | /* Number of active wait queue attached to poll operations */ |
| 164 | int nwait; |
| 165 | |
| 166 | /* List containing poll wait queues */ |
| 167 | struct list_head pwqlist; |
| 168 | |
| 169 | /* The "container" of this item */ |
| 170 | struct eventpoll *ep; |
| 171 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | /* List header used to link this item to the "struct file" items list */ |
| 173 | struct list_head fllink; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 174 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 175 | /* wakeup_source used when EPOLLWAKEUP is set */ |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 176 | struct wakeup_source __rcu *ws; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 177 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 178 | /* The structure that describe the interested events and the source fd */ |
| 179 | struct epoll_event event; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 180 | }; |
| 181 | |
| 182 | /* |
| 183 | * This structure is stored inside the "private_data" member of the file |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 184 | * structure and represents the main data structure for the eventpoll |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 185 | * interface. |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 186 | * |
| 187 | * Access to it is protected by the lock inside wq. |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 188 | */ |
| 189 | struct eventpoll { |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 190 | /* |
| 191 | * This mutex is used to ensure that files are not removed |
| 192 | * while epoll is using them. This is held during the event |
| 193 | * collection loop, the file cleanup path, the epoll file exit |
| 194 | * code and the ctl operations. |
| 195 | */ |
| 196 | struct mutex mtx; |
| 197 | |
| 198 | /* Wait queue used by sys_epoll_wait() */ |
| 199 | wait_queue_head_t wq; |
| 200 | |
| 201 | /* Wait queue used by file->poll() */ |
| 202 | wait_queue_head_t poll_wait; |
| 203 | |
| 204 | /* List of ready file descriptors */ |
| 205 | struct list_head rdllist; |
| 206 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 207 | /* RB tree root used to store monitored fd structs */ |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 208 | struct rb_root_cached rbr; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 209 | |
| 210 | /* |
| 211 | * This is a single linked list that chains all the "struct epitem" that |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 212 | * happened while transferring ready events to userspace w/out |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 213 | * holding ->wq.lock. |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 214 | */ |
| 215 | struct epitem *ovflist; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 216 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 217 | /* wakeup_source used when ep_scan_ready_list is running */ |
| 218 | struct wakeup_source *ws; |
| 219 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 220 | /* The user that created the eventpoll descriptor */ |
| 221 | struct user_struct *user; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 222 | |
| 223 | struct file *file; |
| 224 | |
| 225 | /* used to optimize loop detection check */ |
| 226 | int visited; |
| 227 | struct list_head visited_list_link; |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 228 | |
| 229 | #ifdef CONFIG_NET_RX_BUSY_POLL |
| 230 | /* used to track busy poll napi_id */ |
| 231 | unsigned int napi_id; |
| 232 | #endif |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 233 | }; |
| 234 | |
| 235 | /* Wait structure used by the poll hooks */ |
| 236 | struct eppoll_entry { |
| 237 | /* List header used to link this structure to the "struct epitem" */ |
| 238 | struct list_head llink; |
| 239 | |
| 240 | /* The "base" pointer is set to the container "struct epitem" */ |
Tony Battersby | 4f0989d | 2009-03-31 15:24:16 -0700 | [diff] [blame] | 241 | struct epitem *base; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 242 | |
| 243 | /* |
| 244 | * Wait queue item that will be linked to the target file wait |
| 245 | * queue head. |
| 246 | */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 247 | wait_queue_entry_t wait; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 248 | |
| 249 | /* The wait queue head that linked the "wait" wait queue item */ |
| 250 | wait_queue_head_t *whead; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | }; |
| 252 | |
| 253 | /* Wrapper struct used by poll queueing */ |
| 254 | struct ep_pqueue { |
| 255 | poll_table pt; |
| 256 | struct epitem *epi; |
| 257 | }; |
| 258 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 259 | /* Used by the ep_send_events() function as callback private data */ |
| 260 | struct ep_send_events_data { |
| 261 | int maxevents; |
| 262 | struct epoll_event __user *events; |
Al Viro | d7ebbe4 | 2018-02-01 15:21:13 -0500 | [diff] [blame] | 263 | int res; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 264 | }; |
| 265 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | /* |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 267 | * Configuration options available inside /proc/sys/fs/epoll/ |
| 268 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 269 | /* Maximum number of epoll watched descriptors, per user */ |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 270 | static long max_user_watches __read_mostly; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 271 | |
| 272 | /* |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 273 | * This mutex is used to serialize ep_free() and eventpoll_release_file(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 274 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 275 | static DEFINE_MUTEX(epmutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 276 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 277 | /* Used to check for epoll file descriptor inclusion loops */ |
| 278 | static struct nested_calls poll_loop_ncalls; |
| 279 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | /* Slab cache used to allocate "struct epitem" */ |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 281 | static struct kmem_cache *epi_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | |
| 283 | /* Slab cache used to allocate "struct eppoll_entry" */ |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 284 | static struct kmem_cache *pwq_cache __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 286 | /* Visited nodes during ep_loop_check(), so we can unset them when we finish */ |
| 287 | static LIST_HEAD(visited_list); |
| 288 | |
| 289 | /* |
| 290 | * List of files with newly added links, where we may need to limit the number |
| 291 | * of emanating paths. Protected by the epmutex. |
| 292 | */ |
| 293 | static LIST_HEAD(tfile_check_list); |
| 294 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 295 | #ifdef CONFIG_SYSCTL |
| 296 | |
| 297 | #include <linux/sysctl.h> |
| 298 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 299 | static long zero; |
| 300 | static long long_max = LONG_MAX; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 301 | |
Joe Perches | 1f7e061 | 2014-06-06 14:38:05 -0700 | [diff] [blame] | 302 | struct ctl_table epoll_table[] = { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 303 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 304 | .procname = "max_user_watches", |
| 305 | .data = &max_user_watches, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 306 | .maxlen = sizeof(max_user_watches), |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 307 | .mode = 0644, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 308 | .proc_handler = proc_doulongvec_minmax, |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 309 | .extra1 = &zero, |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 310 | .extra2 = &long_max, |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 311 | }, |
Eric W. Biederman | ab09203 | 2009-11-05 14:25:10 -0800 | [diff] [blame] | 312 | { } |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 313 | }; |
| 314 | #endif /* CONFIG_SYSCTL */ |
| 315 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 316 | static const struct file_operations eventpoll_fops; |
| 317 | |
| 318 | static inline int is_file_epoll(struct file *f) |
| 319 | { |
| 320 | return f->f_op == &eventpoll_fops; |
| 321 | } |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 322 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 323 | /* Setup the structure that is used as key for the RB tree */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 324 | static inline void ep_set_ffd(struct epoll_filefd *ffd, |
| 325 | struct file *file, int fd) |
| 326 | { |
| 327 | ffd->file = file; |
| 328 | ffd->fd = fd; |
| 329 | } |
| 330 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 331 | /* Compare RB tree keys */ |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 332 | static inline int ep_cmp_ffd(struct epoll_filefd *p1, |
| 333 | struct epoll_filefd *p2) |
| 334 | { |
| 335 | return (p1->file > p2->file ? +1: |
| 336 | (p1->file < p2->file ? -1 : p1->fd - p2->fd)); |
| 337 | } |
| 338 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 339 | /* Tells us if the item is currently linked */ |
Davidlohr Bueso | 992991c | 2018-08-21 21:58:26 -0700 | [diff] [blame] | 340 | static inline int ep_is_linked(struct epitem *epi) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 341 | { |
Davidlohr Bueso | 992991c | 2018-08-21 21:58:26 -0700 | [diff] [blame] | 342 | return !list_empty(&epi->rdllink); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 343 | } |
| 344 | |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 345 | static inline struct eppoll_entry *ep_pwq_from_wait(wait_queue_entry_t *p) |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 346 | { |
| 347 | return container_of(p, struct eppoll_entry, wait); |
| 348 | } |
| 349 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 350 | /* Get the "struct epitem" from a wait queue pointer */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 351 | static inline struct epitem *ep_item_from_wait(wait_queue_entry_t *p) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 352 | { |
| 353 | return container_of(p, struct eppoll_entry, wait)->base; |
| 354 | } |
| 355 | |
| 356 | /* Get the "struct epitem" from an epoll queue wrapper */ |
Davide Libenzi | cdac75e | 2008-04-29 00:58:34 -0700 | [diff] [blame] | 357 | static inline struct epitem *ep_item_from_epqueue(poll_table *p) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 358 | { |
| 359 | return container_of(p, struct ep_pqueue, pt)->epi; |
| 360 | } |
| 361 | |
| 362 | /* 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] | 363 | static inline int ep_op_has_event(int op) |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 364 | { |
Andrew Morton | a80a6b8 | 2012-11-08 15:53:35 -0800 | [diff] [blame] | 365 | return op != EPOLL_CTL_DEL; |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 366 | } |
| 367 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | /* Initialize the poll safe wake up structure */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 369 | static void ep_nested_calls_init(struct nested_calls *ncalls) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 371 | INIT_LIST_HEAD(&ncalls->tasks_call_list); |
| 372 | spin_lock_init(&ncalls->lock); |
| 373 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 375 | /** |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 376 | * ep_events_available - Checks if ready events might be available. |
| 377 | * |
| 378 | * @ep: Pointer to the eventpoll context. |
| 379 | * |
| 380 | * Returns: Returns a value different than zero if ready events are available, |
| 381 | * or zero otherwise. |
| 382 | */ |
| 383 | static inline int ep_events_available(struct eventpoll *ep) |
| 384 | { |
| 385 | return !list_empty(&ep->rdllist) || ep->ovflist != EP_UNACTIVE_PTR; |
| 386 | } |
| 387 | |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 388 | #ifdef CONFIG_NET_RX_BUSY_POLL |
| 389 | static bool ep_busy_loop_end(void *p, unsigned long start_time) |
| 390 | { |
| 391 | struct eventpoll *ep = p; |
| 392 | |
| 393 | return ep_events_available(ep) || busy_loop_timeout(start_time); |
| 394 | } |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 395 | |
| 396 | /* |
| 397 | * Busy poll if globally on and supporting sockets found && no events, |
| 398 | * busy loop will return if need_resched or ep_events_available. |
| 399 | * |
| 400 | * we must do our busy polling with irqs enabled |
| 401 | */ |
| 402 | static void ep_busy_loop(struct eventpoll *ep, int nonblock) |
| 403 | { |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 404 | unsigned int napi_id = READ_ONCE(ep->napi_id); |
| 405 | |
| 406 | if ((napi_id >= MIN_NAPI_ID) && net_busy_loop_on()) |
| 407 | napi_busy_loop(napi_id, nonblock ? NULL : ep_busy_loop_end, ep); |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 408 | } |
| 409 | |
| 410 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) |
| 411 | { |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 412 | if (ep->napi_id) |
| 413 | ep->napi_id = 0; |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 414 | } |
| 415 | |
| 416 | /* |
| 417 | * Set epoll busy poll NAPI ID from sk. |
| 418 | */ |
| 419 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) |
| 420 | { |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 421 | struct eventpoll *ep; |
| 422 | unsigned int napi_id; |
| 423 | struct socket *sock; |
| 424 | struct sock *sk; |
| 425 | int err; |
| 426 | |
| 427 | if (!net_busy_loop_on()) |
| 428 | return; |
| 429 | |
| 430 | sock = sock_from_file(epi->ffd.file, &err); |
| 431 | if (!sock) |
| 432 | return; |
| 433 | |
| 434 | sk = sock->sk; |
| 435 | if (!sk) |
| 436 | return; |
| 437 | |
| 438 | napi_id = READ_ONCE(sk->sk_napi_id); |
| 439 | ep = epi->ep; |
| 440 | |
| 441 | /* Non-NAPI IDs can be rejected |
| 442 | * or |
| 443 | * Nothing to do if we already have this ID |
| 444 | */ |
| 445 | if (napi_id < MIN_NAPI_ID || napi_id == ep->napi_id) |
| 446 | return; |
| 447 | |
| 448 | /* record NAPI ID for use in next busy poll */ |
| 449 | ep->napi_id = napi_id; |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 450 | } |
| 451 | |
Davidlohr Bueso | 514056d | 2018-08-21 21:58:19 -0700 | [diff] [blame] | 452 | #else |
| 453 | |
| 454 | static inline void ep_busy_loop(struct eventpoll *ep, int nonblock) |
| 455 | { |
| 456 | } |
| 457 | |
| 458 | static inline void ep_reset_busy_poll_napi_id(struct eventpoll *ep) |
| 459 | { |
| 460 | } |
| 461 | |
| 462 | static inline void ep_set_busy_poll_napi_id(struct epitem *epi) |
| 463 | { |
| 464 | } |
| 465 | |
| 466 | #endif /* CONFIG_NET_RX_BUSY_POLL */ |
| 467 | |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 468 | /** |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 469 | * ep_call_nested - Perform a bound (possibly) nested call, by checking |
| 470 | * that the recursion limit is not exceeded, and that |
| 471 | * the same nested call (by the meaning of same cookie) is |
| 472 | * no re-entered. |
| 473 | * |
| 474 | * @ncalls: Pointer to the nested_calls structure to be used for this call. |
| 475 | * @max_nests: Maximum number of allowed nesting calls. |
| 476 | * @nproc: Nested call core function pointer. |
| 477 | * @priv: Opaque data to be passed to the @nproc callback. |
| 478 | * @cookie: Cookie to be used to identify this nested call. |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 479 | * @ctx: This instance context. |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 480 | * |
| 481 | * Returns: Returns the code returned by the @nproc callback, or -1 if |
| 482 | * the maximum recursion limit has been exceeded. |
| 483 | */ |
| 484 | static int ep_call_nested(struct nested_calls *ncalls, int max_nests, |
| 485 | int (*nproc)(void *, void *, int), void *priv, |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 486 | void *cookie, void *ctx) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 487 | { |
| 488 | int error, call_nests = 0; |
| 489 | unsigned long flags; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 490 | struct list_head *lsthead = &ncalls->tasks_call_list; |
| 491 | struct nested_call_node *tncur; |
| 492 | struct nested_call_node tnode; |
| 493 | |
| 494 | spin_lock_irqsave(&ncalls->lock, flags); |
| 495 | |
| 496 | /* |
| 497 | * Try to see if the current task is already inside this wakeup call. |
| 498 | * We use a list here, since the population inside this set is always |
| 499 | * very much limited. |
| 500 | */ |
| 501 | list_for_each_entry(tncur, lsthead, llink) { |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 502 | if (tncur->ctx == ctx && |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 503 | (tncur->cookie == cookie || ++call_nests > max_nests)) { |
| 504 | /* |
| 505 | * Ops ... loop detected or maximum nest level reached. |
| 506 | * We abort this wake by breaking the cycle itself. |
| 507 | */ |
Tony Battersby | abff55c | 2009-03-31 15:24:13 -0700 | [diff] [blame] | 508 | error = -1; |
| 509 | goto out_unlock; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 510 | } |
| 511 | } |
| 512 | |
| 513 | /* Add the current task and cookie to the list */ |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 514 | tnode.ctx = ctx; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 515 | tnode.cookie = cookie; |
| 516 | list_add(&tnode.llink, lsthead); |
| 517 | |
| 518 | spin_unlock_irqrestore(&ncalls->lock, flags); |
| 519 | |
| 520 | /* Call the nested function */ |
| 521 | error = (*nproc)(priv, cookie, call_nests); |
| 522 | |
| 523 | /* Remove the current task from the list */ |
| 524 | spin_lock_irqsave(&ncalls->lock, flags); |
| 525 | list_del(&tnode.llink); |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 526 | out_unlock: |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 527 | spin_unlock_irqrestore(&ncalls->lock, flags); |
| 528 | |
| 529 | return error; |
| 530 | } |
| 531 | |
Steven Rostedt | 02edc6f | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 532 | /* |
| 533 | * As described in commit 0ccf831cb lockdep: annotate epoll |
| 534 | * the use of wait queues used by epoll is done in a very controlled |
| 535 | * manner. Wake ups can nest inside each other, but are never done |
| 536 | * with the same locking. For example: |
| 537 | * |
| 538 | * dfd = socket(...); |
| 539 | * efd1 = epoll_create(); |
| 540 | * efd2 = epoll_create(); |
| 541 | * epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...); |
| 542 | * epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...); |
| 543 | * |
| 544 | * When a packet arrives to the device underneath "dfd", the net code will |
| 545 | * issue a wake_up() on its poll wake list. Epoll (efd1) has installed a |
| 546 | * callback wakeup entry on that queue, and the wake_up() performed by the |
| 547 | * "dfd" net code will end up in ep_poll_callback(). At this point epoll |
| 548 | * (efd1) notices that it may have some event ready, so it needs to wake up |
| 549 | * the waiters on its poll wait list (efd2). So it calls ep_poll_safewake() |
| 550 | * that ends up in another wake_up(), after having checked about the |
| 551 | * recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to |
| 552 | * avoid stack blasting. |
| 553 | * |
| 554 | * When CONFIG_DEBUG_LOCK_ALLOC is enabled, make sure lockdep can handle |
| 555 | * this special case of epoll. |
| 556 | */ |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 557 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 558 | |
Jason Baron | 57a173b | 2017-11-17 15:29:02 -0800 | [diff] [blame] | 559 | static struct nested_calls poll_safewake_ncalls; |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 560 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 561 | static int ep_poll_wakeup_proc(void *priv, void *cookie, int call_nests) |
| 562 | { |
Jason Baron | 57a173b | 2017-11-17 15:29:02 -0800 | [diff] [blame] | 563 | unsigned long flags; |
| 564 | wait_queue_head_t *wqueue = (wait_queue_head_t *)cookie; |
| 565 | |
| 566 | spin_lock_irqsave_nested(&wqueue->lock, flags, call_nests + 1); |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 567 | wake_up_locked_poll(wqueue, EPOLLIN); |
Jason Baron | 57a173b | 2017-11-17 15:29:02 -0800 | [diff] [blame] | 568 | spin_unlock_irqrestore(&wqueue->lock, flags); |
| 569 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 570 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 571 | } |
| 572 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 573 | static void ep_poll_safewake(wait_queue_head_t *wq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | { |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 575 | int this_cpu = get_cpu(); |
| 576 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 577 | ep_call_nested(&poll_safewake_ncalls, EP_MAX_NESTS, |
Davide Libenzi | 3fe4a97 | 2009-06-17 16:25:58 -0700 | [diff] [blame] | 578 | ep_poll_wakeup_proc, NULL, wq, (void *) (long) this_cpu); |
| 579 | |
| 580 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 581 | } |
| 582 | |
Jason Baron | 57a173b | 2017-11-17 15:29:02 -0800 | [diff] [blame] | 583 | #else |
| 584 | |
| 585 | static void ep_poll_safewake(wait_queue_head_t *wq) |
| 586 | { |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 587 | wake_up_poll(wq, EPOLLIN); |
Jason Baron | 57a173b | 2017-11-17 15:29:02 -0800 | [diff] [blame] | 588 | } |
| 589 | |
| 590 | #endif |
| 591 | |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 592 | static void ep_remove_wait_queue(struct eppoll_entry *pwq) |
| 593 | { |
| 594 | wait_queue_head_t *whead; |
| 595 | |
| 596 | rcu_read_lock(); |
Oleg Nesterov | 138e4ad | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 597 | /* |
| 598 | * If it is cleared by POLLFREE, it should be rcu-safe. |
| 599 | * If we read NULL we need a barrier paired with |
| 600 | * smp_store_release() in ep_poll_callback(), otherwise |
| 601 | * we rely on whead->lock. |
| 602 | */ |
| 603 | whead = smp_load_acquire(&pwq->whead); |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 604 | if (whead) |
| 605 | remove_wait_queue(whead, &pwq->wait); |
| 606 | rcu_read_unlock(); |
| 607 | } |
| 608 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 609 | /* |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 610 | * This function unregisters poll callbacks from the associated file |
| 611 | * descriptor. Must be called with "mtx" held (or "epmutex" if called from |
| 612 | * ep_free). |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 613 | */ |
| 614 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) |
| 615 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 616 | struct list_head *lsthead = &epi->pwqlist; |
| 617 | struct eppoll_entry *pwq; |
| 618 | |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 619 | while (!list_empty(lsthead)) { |
| 620 | pwq = list_first_entry(lsthead, struct eppoll_entry, llink); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 621 | |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 622 | list_del(&pwq->llink); |
Oleg Nesterov | 971316f | 2012-02-24 20:07:29 +0100 | [diff] [blame] | 623 | ep_remove_wait_queue(pwq); |
Tony Battersby | d1bc90d | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 624 | kmem_cache_free(pwq_cache, pwq); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 625 | } |
| 626 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 628 | /* call only when ep->mtx is held */ |
| 629 | static inline struct wakeup_source *ep_wakeup_source(struct epitem *epi) |
| 630 | { |
| 631 | return rcu_dereference_check(epi->ws, lockdep_is_held(&epi->ep->mtx)); |
| 632 | } |
| 633 | |
| 634 | /* call only when ep->mtx is held */ |
| 635 | static inline void ep_pm_stay_awake(struct epitem *epi) |
| 636 | { |
| 637 | struct wakeup_source *ws = ep_wakeup_source(epi); |
| 638 | |
| 639 | if (ws) |
| 640 | __pm_stay_awake(ws); |
| 641 | } |
| 642 | |
| 643 | static inline bool ep_has_wakeup_source(struct epitem *epi) |
| 644 | { |
| 645 | return rcu_access_pointer(epi->ws) ? true : false; |
| 646 | } |
| 647 | |
| 648 | /* call when ep->mtx cannot be held (ep_poll_callback) */ |
| 649 | static inline void ep_pm_stay_awake_rcu(struct epitem *epi) |
| 650 | { |
| 651 | struct wakeup_source *ws; |
| 652 | |
| 653 | rcu_read_lock(); |
| 654 | ws = rcu_dereference(epi->ws); |
| 655 | if (ws) |
| 656 | __pm_stay_awake(ws); |
| 657 | rcu_read_unlock(); |
| 658 | } |
| 659 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 660 | /** |
| 661 | * ep_scan_ready_list - Scans the ready list in a way that makes possible for |
| 662 | * the scan code, to call f_op->poll(). Also allows for |
| 663 | * O(NumReady) performance. |
| 664 | * |
| 665 | * @ep: Pointer to the epoll private data structure. |
| 666 | * @sproc: Pointer to the scan callback. |
| 667 | * @priv: Private opaque data passed to the @sproc callback. |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 668 | * @depth: The current depth of recursive f_op->poll calls. |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 669 | * @ep_locked: caller already holds ep->mtx |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 670 | * |
| 671 | * Returns: The same integer error code returned by the @sproc callback. |
| 672 | */ |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 673 | static __poll_t ep_scan_ready_list(struct eventpoll *ep, |
| 674 | __poll_t (*sproc)(struct eventpoll *, |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 675 | struct list_head *, void *), |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 676 | void *priv, int depth, bool ep_locked) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 677 | { |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 678 | __poll_t res; |
| 679 | int pwake = 0; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 680 | struct epitem *epi, *nepi; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 681 | LIST_HEAD(txlist); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 682 | |
Davidlohr Bueso | 92e6417 | 2018-08-21 21:56:45 -0700 | [diff] [blame] | 683 | lockdep_assert_irqs_enabled(); |
| 684 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 685 | /* |
| 686 | * We need to lock this because we could be hit by |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 687 | * eventpoll_release_file() and epoll_ctl(). |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 688 | */ |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 689 | |
| 690 | if (!ep_locked) |
| 691 | mutex_lock_nested(&ep->mtx, depth); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 692 | |
| 693 | /* |
| 694 | * Steal the ready list, and re-init the original one to the |
| 695 | * empty list. Also, set ep->ovflist to NULL so that events |
| 696 | * happening while looping w/out locks, are not lost. We cannot |
| 697 | * have the poll callback to queue directly on ep->rdllist, |
| 698 | * because we want the "sproc" callback to be able to do it |
| 699 | * in a lockless way. |
| 700 | */ |
Davidlohr Bueso | 002b343 | 2018-08-21 21:56:38 -0700 | [diff] [blame] | 701 | spin_lock_irq(&ep->wq.lock); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 702 | list_splice_init(&ep->rdllist, &txlist); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 703 | ep->ovflist = NULL; |
Davidlohr Bueso | 002b343 | 2018-08-21 21:56:38 -0700 | [diff] [blame] | 704 | spin_unlock_irq(&ep->wq.lock); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 705 | |
| 706 | /* |
| 707 | * Now call the callback function. |
| 708 | */ |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 709 | res = (*sproc)(ep, &txlist, priv); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 710 | |
Davidlohr Bueso | 002b343 | 2018-08-21 21:56:38 -0700 | [diff] [blame] | 711 | spin_lock_irq(&ep->wq.lock); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 712 | /* |
| 713 | * During the time we spent inside the "sproc" callback, some |
| 714 | * other events might have been queued by the poll callback. |
| 715 | * We re-insert them inside the main ready-list here. |
| 716 | */ |
| 717 | for (nepi = ep->ovflist; (epi = nepi) != NULL; |
| 718 | nepi = epi->next, epi->next = EP_UNACTIVE_PTR) { |
| 719 | /* |
| 720 | * We need to check if the item is already in the list. |
| 721 | * During the "sproc" callback execution time, items are |
| 722 | * queued into ->ovflist but the "txlist" might already |
| 723 | * contain them, and the list_splice() below takes care of them. |
| 724 | */ |
Davidlohr Bueso | 992991c | 2018-08-21 21:58:26 -0700 | [diff] [blame] | 725 | if (!ep_is_linked(epi)) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 726 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 727 | ep_pm_stay_awake(epi); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 728 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 729 | } |
| 730 | /* |
| 731 | * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after |
| 732 | * releasing the lock, events will be queued in the normal way inside |
| 733 | * ep->rdllist. |
| 734 | */ |
| 735 | ep->ovflist = EP_UNACTIVE_PTR; |
| 736 | |
| 737 | /* |
| 738 | * Quickly re-inject items left on "txlist". |
| 739 | */ |
| 740 | list_splice(&txlist, &ep->rdllist); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 741 | __pm_relax(ep->ws); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 742 | |
| 743 | if (!list_empty(&ep->rdllist)) { |
| 744 | /* |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 745 | * Wake up (if active) both the eventpoll wait list and |
| 746 | * the ->poll() wait list (delayed after we release the lock). |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 747 | */ |
| 748 | if (waitqueue_active(&ep->wq)) |
| 749 | wake_up_locked(&ep->wq); |
| 750 | if (waitqueue_active(&ep->poll_wait)) |
| 751 | pwake++; |
| 752 | } |
Davidlohr Bueso | 002b343 | 2018-08-21 21:56:38 -0700 | [diff] [blame] | 753 | spin_unlock_irq(&ep->wq.lock); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 754 | |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 755 | if (!ep_locked) |
| 756 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 757 | |
| 758 | /* We have to call this outside the lock */ |
| 759 | if (pwake) |
| 760 | ep_poll_safewake(&ep->poll_wait); |
| 761 | |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 762 | return res; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 763 | } |
| 764 | |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 765 | static void epi_rcu_free(struct rcu_head *head) |
| 766 | { |
| 767 | struct epitem *epi = container_of(head, struct epitem, rcu); |
| 768 | kmem_cache_free(epi_cache, epi); |
| 769 | } |
| 770 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 771 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 772 | * Removes a "struct epitem" from the eventpoll RB tree and deallocates |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 773 | * all the associated resources. Must be called with "mtx" held. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 774 | */ |
| 775 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) |
| 776 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 777 | struct file *file = epi->ffd.file; |
| 778 | |
Davidlohr Bueso | 92e6417 | 2018-08-21 21:56:45 -0700 | [diff] [blame] | 779 | lockdep_assert_irqs_enabled(); |
| 780 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 781 | /* |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 782 | * Removes poll wait queue hooks. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 783 | */ |
| 784 | ep_unregister_pollwait(ep, epi); |
| 785 | |
| 786 | /* Remove the current item from the list of epoll hooks */ |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 787 | spin_lock(&file->f_lock); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 788 | list_del_rcu(&epi->fllink); |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 789 | spin_unlock(&file->f_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 790 | |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 791 | rb_erase_cached(&epi->rbn, &ep->rbr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | |
Davidlohr Bueso | 304b18b | 2018-08-21 21:56:41 -0700 | [diff] [blame] | 793 | spin_lock_irq(&ep->wq.lock); |
Davidlohr Bueso | 992991c | 2018-08-21 21:58:26 -0700 | [diff] [blame] | 794 | if (ep_is_linked(epi)) |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 795 | list_del_init(&epi->rdllink); |
Davidlohr Bueso | 304b18b | 2018-08-21 21:56:41 -0700 | [diff] [blame] | 796 | spin_unlock_irq(&ep->wq.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 798 | wakeup_source_unregister(ep_wakeup_source(epi)); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 799 | /* |
| 800 | * At this point it is safe to free the eventpoll item. Use the union |
| 801 | * field epi->rcu, since we are trying to minimize the size of |
| 802 | * 'struct epitem'. The 'rbn' field is no longer in use. Protected by |
| 803 | * ep->mtx. The rcu read side, reverse_path_check_proc(), does not make |
| 804 | * use of the rbn field. |
| 805 | */ |
| 806 | call_rcu(&epi->rcu, epi_rcu_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 808 | atomic_long_dec(&ep->user->epoll_watches); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 809 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 810 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 811 | } |
| 812 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | static void ep_free(struct eventpoll *ep) |
| 814 | { |
| 815 | struct rb_node *rbp; |
| 816 | struct epitem *epi; |
| 817 | |
| 818 | /* We need to release all tasks waiting for these file */ |
| 819 | if (waitqueue_active(&ep->poll_wait)) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 820 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 821 | |
| 822 | /* |
| 823 | * We need to lock this because we could be hit by |
| 824 | * eventpoll_release_file() while we're freeing the "struct eventpoll". |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 825 | * 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] | 826 | * is on the way to be removed and no one has references to it |
| 827 | * anymore. The only hit might come from eventpoll_release_file() but |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 828 | * holding "epmutex" is sufficient here. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 829 | */ |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 830 | mutex_lock(&epmutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 831 | |
| 832 | /* |
| 833 | * Walks through the whole tree by unregistering poll callbacks. |
| 834 | */ |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 835 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 836 | epi = rb_entry(rbp, struct epitem, rbn); |
| 837 | |
| 838 | ep_unregister_pollwait(ep, epi); |
Eric Dumazet | 91cf5ab | 2013-09-11 14:24:06 -0700 | [diff] [blame] | 839 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 840 | } |
| 841 | |
| 842 | /* |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 843 | * Walks through the whole tree by freeing each "struct epitem". At this |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | * 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] | 845 | * holding "epmutex" we can be sure that no file cleanup code will hit |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 846 | * us during this operation. So we can avoid the lock on "ep->wq.lock". |
Eric Wong | ddf676c | 2013-04-30 15:27:40 -0700 | [diff] [blame] | 847 | * We do not need to lock ep->mtx, either, we only do it to prevent |
| 848 | * a lockdep warning. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 849 | */ |
Eric Wong | ddf676c | 2013-04-30 15:27:40 -0700 | [diff] [blame] | 850 | mutex_lock(&ep->mtx); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 851 | while ((rbp = rb_first_cached(&ep->rbr)) != NULL) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 852 | epi = rb_entry(rbp, struct epitem, rbn); |
| 853 | ep_remove(ep, epi); |
Eric Dumazet | 91cf5ab | 2013-09-11 14:24:06 -0700 | [diff] [blame] | 854 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 855 | } |
Eric Wong | ddf676c | 2013-04-30 15:27:40 -0700 | [diff] [blame] | 856 | mutex_unlock(&ep->mtx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 857 | |
Arjan van de Ven | 144efe3 | 2006-03-23 03:00:32 -0800 | [diff] [blame] | 858 | mutex_unlock(&epmutex); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 859 | mutex_destroy(&ep->mtx); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 860 | free_uid(ep->user); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 861 | wakeup_source_unregister(ep->ws); |
Davide Libenzi | f0ee9aa | 2007-05-15 01:40:57 -0700 | [diff] [blame] | 862 | kfree(ep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | } |
| 864 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 865 | static int ep_eventpoll_release(struct inode *inode, struct file *file) |
| 866 | { |
| 867 | struct eventpoll *ep = file->private_data; |
| 868 | |
Davide Libenzi | f0ee9aa | 2007-05-15 01:40:57 -0700 | [diff] [blame] | 869 | if (ep) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 870 | ep_free(ep); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 871 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 872 | return 0; |
| 873 | } |
| 874 | |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 875 | static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 876 | void *priv); |
| 877 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
| 878 | poll_table *pt); |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 879 | |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 880 | /* |
| 881 | * Differs from ep_eventpoll_poll() in that internal callers already have |
| 882 | * the ep->mtx so we need to start from depth=1, such that mutex_lock_nested() |
| 883 | * is correctly annotated. |
| 884 | */ |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 885 | static __poll_t ep_item_poll(const struct epitem *epi, poll_table *pt, |
Al Viro | bec1a502 | 2017-11-28 19:43:33 -0500 | [diff] [blame] | 886 | int depth) |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 887 | { |
| 888 | struct eventpoll *ep; |
| 889 | bool locked; |
| 890 | |
| 891 | pt->_key = epi->event.events; |
| 892 | if (!is_file_epoll(epi->ffd.file)) |
Christoph Hellwig | 9965ed17 | 2018-03-05 07:26:05 -0800 | [diff] [blame] | 893 | return vfs_poll(epi->ffd.file, pt) & epi->event.events; |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 894 | |
| 895 | ep = epi->ffd.file->private_data; |
| 896 | poll_wait(epi->ffd.file, &ep->poll_wait, pt); |
| 897 | locked = pt && (pt->_qproc == ep_ptable_queue_proc); |
| 898 | |
| 899 | return ep_scan_ready_list(epi->ffd.file->private_data, |
| 900 | ep_read_events_proc, &depth, depth, |
| 901 | locked) & epi->event.events; |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 902 | } |
| 903 | |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 904 | static __poll_t ep_read_events_proc(struct eventpoll *ep, struct list_head *head, |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 905 | void *priv) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 906 | { |
| 907 | struct epitem *epi, *tmp; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 908 | poll_table pt; |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 909 | int depth = *(int *)priv; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 910 | |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 911 | init_poll_funcptr(&pt, NULL); |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 912 | depth++; |
Eric Wong | 450d89e | 2013-04-30 15:27:42 -0700 | [diff] [blame] | 913 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 914 | list_for_each_entry_safe(epi, tmp, head, rdllink) { |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 915 | if (ep_item_poll(epi, &pt, depth)) { |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 916 | return EPOLLIN | EPOLLRDNORM; |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 917 | } else { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 918 | /* |
| 919 | * Item has been dropped into the ready list by the poll |
| 920 | * callback, but it's not actually ready, as far as |
| 921 | * caller requested events goes. We can remove it here. |
| 922 | */ |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 923 | __pm_relax(ep_wakeup_source(epi)); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 924 | list_del_init(&epi->rdllink); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 925 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 926 | } |
| 927 | |
| 928 | return 0; |
| 929 | } |
| 930 | |
Linus Torvalds | a11e1d4 | 2018-06-28 09:43:44 -0700 | [diff] [blame] | 931 | static __poll_t ep_eventpoll_poll(struct file *file, poll_table *wait) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 932 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 933 | struct eventpoll *ep = file->private_data; |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 934 | int depth = 0; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 935 | |
Linus Torvalds | a11e1d4 | 2018-06-28 09:43:44 -0700 | [diff] [blame] | 936 | /* Insert inside our poll wait queue */ |
| 937 | poll_wait(file, &ep->poll_wait, wait); |
| 938 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 939 | /* |
| 940 | * Proceed to find out if wanted events are really available inside |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 941 | * the ready list. |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 942 | */ |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 943 | return ep_scan_ready_list(ep, ep_read_events_proc, |
| 944 | &depth, depth, false); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 945 | } |
| 946 | |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 947 | #ifdef CONFIG_PROC_FS |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 948 | static void ep_show_fdinfo(struct seq_file *m, struct file *f) |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 949 | { |
| 950 | struct eventpoll *ep = f->private_data; |
| 951 | struct rb_node *rbp; |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 952 | |
| 953 | mutex_lock(&ep->mtx); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 954 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 955 | struct epitem *epi = rb_entry(rbp, struct epitem, rbn); |
Cyrill Gorcunov | 77493f0 | 2017-07-12 14:34:25 -0700 | [diff] [blame] | 956 | struct inode *inode = file_inode(epi->ffd.file); |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 957 | |
Cyrill Gorcunov | 77493f0 | 2017-07-12 14:34:25 -0700 | [diff] [blame] | 958 | seq_printf(m, "tfd: %8d events: %8x data: %16llx " |
| 959 | " pos:%lli ino:%lx sdev:%x\n", |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 960 | epi->ffd.fd, epi->event.events, |
Cyrill Gorcunov | 77493f0 | 2017-07-12 14:34:25 -0700 | [diff] [blame] | 961 | (long long)epi->event.data, |
| 962 | (long long)epi->ffd.file->f_pos, |
| 963 | inode->i_ino, inode->i_sb->s_dev); |
Joe Perches | a3816ab | 2014-09-29 16:08:25 -0700 | [diff] [blame] | 964 | if (seq_has_overflowed(m)) |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 965 | break; |
| 966 | } |
| 967 | mutex_unlock(&ep->mtx); |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 968 | } |
| 969 | #endif |
| 970 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 971 | /* File callbacks that implement the eventpoll file behaviour */ |
| 972 | static const struct file_operations eventpoll_fops = { |
Cyrill Gorcunov | 138d22b | 2012-12-17 16:05:02 -0800 | [diff] [blame] | 973 | #ifdef CONFIG_PROC_FS |
| 974 | .show_fdinfo = ep_show_fdinfo, |
| 975 | #endif |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 976 | .release = ep_eventpoll_release, |
Linus Torvalds | a11e1d4 | 2018-06-28 09:43:44 -0700 | [diff] [blame] | 977 | .poll = ep_eventpoll_poll, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 978 | .llseek = noop_llseek, |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 979 | }; |
| 980 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 981 | /* |
| 982 | * This is called from eventpoll_release() to unlink files from the eventpoll |
| 983 | * interface. We need to have this facility to cleanup correctly files that are |
| 984 | * closed without being removed from the eventpoll interface. |
| 985 | */ |
| 986 | void eventpoll_release_file(struct file *file) |
| 987 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 988 | struct eventpoll *ep; |
Konstantin Khlebnikov | ebe0618 | 2014-06-17 06:58:05 +0400 | [diff] [blame] | 989 | struct epitem *epi, *next; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 990 | |
| 991 | /* |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 992 | * 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] | 993 | * 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] | 994 | * 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] | 995 | * 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] | 996 | * point, the file counter already went to zero and fget() would fail. |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 997 | * 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] | 998 | * will correctly serialize the operation. We do need to acquire |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 999 | * "ep->mtx" after "epmutex" because ep_remove() requires it when called |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1000 | * from anywhere but ep_free(). |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 1001 | * |
| 1002 | * 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] | 1003 | */ |
| 1004 | mutex_lock(&epmutex); |
Konstantin Khlebnikov | ebe0618 | 2014-06-17 06:58:05 +0400 | [diff] [blame] | 1005 | list_for_each_entry_safe(epi, next, &file->f_ep_links, fllink) { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1006 | ep = epi->ep; |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 1007 | mutex_lock_nested(&ep->mtx, 0); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1008 | ep_remove(ep, epi); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1009 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1010 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1011 | mutex_unlock(&epmutex); |
| 1012 | } |
| 1013 | |
| 1014 | static int ep_alloc(struct eventpoll **pep) |
| 1015 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1016 | int error; |
| 1017 | struct user_struct *user; |
| 1018 | struct eventpoll *ep; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1019 | |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1020 | user = get_current_user(); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1021 | error = -ENOMEM; |
| 1022 | ep = kzalloc(sizeof(*ep), GFP_KERNEL); |
| 1023 | if (unlikely(!ep)) |
| 1024 | goto free_uid; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1025 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1026 | mutex_init(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1027 | init_waitqueue_head(&ep->wq); |
| 1028 | init_waitqueue_head(&ep->poll_wait); |
| 1029 | INIT_LIST_HEAD(&ep->rdllist); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1030 | ep->rbr = RB_ROOT_CACHED; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1031 | ep->ovflist = EP_UNACTIVE_PTR; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1032 | ep->user = user; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1033 | |
| 1034 | *pep = ep; |
| 1035 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1036 | return 0; |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1037 | |
| 1038 | free_uid: |
| 1039 | free_uid(user); |
| 1040 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1041 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1042 | |
| 1043 | /* |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1044 | * Search the file inside the eventpoll tree. The RB tree operations |
| 1045 | * are protected by the "mtx" mutex, and ep_find() must be called with |
| 1046 | * "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1047 | */ |
| 1048 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) |
| 1049 | { |
| 1050 | int kcmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1051 | struct rb_node *rbp; |
| 1052 | struct epitem *epi, *epir = NULL; |
| 1053 | struct epoll_filefd ffd; |
| 1054 | |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1055 | ep_set_ffd(&ffd, file, fd); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1056 | for (rbp = ep->rbr.rb_root.rb_node; rbp; ) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1057 | epi = rb_entry(rbp, struct epitem, rbn); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1058 | kcmp = ep_cmp_ffd(&ffd, &epi->ffd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | if (kcmp > 0) |
| 1060 | rbp = rbp->rb_right; |
| 1061 | else if (kcmp < 0) |
| 1062 | rbp = rbp->rb_left; |
| 1063 | else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1064 | epir = epi; |
| 1065 | break; |
| 1066 | } |
| 1067 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1068 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1069 | return epir; |
| 1070 | } |
| 1071 | |
Cyrill Gorcunov | 92ef6da | 2017-07-12 14:34:31 -0700 | [diff] [blame] | 1072 | #ifdef CONFIG_CHECKPOINT_RESTORE |
Cyrill Gorcunov | 0791e36 | 2017-07-12 14:34:28 -0700 | [diff] [blame] | 1073 | static struct epitem *ep_find_tfd(struct eventpoll *ep, int tfd, unsigned long toff) |
| 1074 | { |
| 1075 | struct rb_node *rbp; |
| 1076 | struct epitem *epi; |
| 1077 | |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1078 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
Cyrill Gorcunov | 0791e36 | 2017-07-12 14:34:28 -0700 | [diff] [blame] | 1079 | epi = rb_entry(rbp, struct epitem, rbn); |
| 1080 | if (epi->ffd.fd == tfd) { |
| 1081 | if (toff == 0) |
| 1082 | return epi; |
| 1083 | else |
| 1084 | toff--; |
| 1085 | } |
| 1086 | cond_resched(); |
| 1087 | } |
| 1088 | |
| 1089 | return NULL; |
| 1090 | } |
| 1091 | |
| 1092 | struct file *get_epoll_tfile_raw_ptr(struct file *file, int tfd, |
| 1093 | unsigned long toff) |
| 1094 | { |
| 1095 | struct file *file_raw; |
| 1096 | struct eventpoll *ep; |
| 1097 | struct epitem *epi; |
| 1098 | |
| 1099 | if (!is_file_epoll(file)) |
| 1100 | return ERR_PTR(-EINVAL); |
| 1101 | |
| 1102 | ep = file->private_data; |
| 1103 | |
| 1104 | mutex_lock(&ep->mtx); |
| 1105 | epi = ep_find_tfd(ep, tfd, toff); |
| 1106 | if (epi) |
| 1107 | file_raw = epi->ffd.file; |
| 1108 | else |
| 1109 | file_raw = ERR_PTR(-ENOENT); |
| 1110 | mutex_unlock(&ep->mtx); |
| 1111 | |
| 1112 | return file_raw; |
| 1113 | } |
Cyrill Gorcunov | 92ef6da | 2017-07-12 14:34:31 -0700 | [diff] [blame] | 1114 | #endif /* CONFIG_CHECKPOINT_RESTORE */ |
Cyrill Gorcunov | 0791e36 | 2017-07-12 14:34:28 -0700 | [diff] [blame] | 1115 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1116 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1117 | * This is the callback that is passed to the wait queue wakeup |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 1118 | * mechanism. It is called by the stored file descriptors when they |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1119 | * have events to report. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1121 | static int ep_poll_callback(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1122 | { |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1123 | int pwake = 0; |
| 1124 | unsigned long flags; |
| 1125 | struct epitem *epi = ep_item_from_wait(wait); |
| 1126 | struct eventpoll *ep = epi->ep; |
Al Viro | 3ad6f93 | 2017-07-03 20:14:56 -0400 | [diff] [blame] | 1127 | __poll_t pollflags = key_to_poll(key); |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1128 | int ewake = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 1130 | spin_lock_irqsave(&ep->wq.lock, flags); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1131 | |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 1132 | ep_set_busy_poll_napi_id(epi); |
| 1133 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1134 | /* |
| 1135 | * If the event mask does not contain any poll(2) event, we consider the |
| 1136 | * descriptor to be disabled. This condition is likely the effect of the |
| 1137 | * EPOLLONESHOT bit that disables the descriptor when an event is received, |
| 1138 | * until the next EPOLL_CTL_MOD will be issued. |
| 1139 | */ |
| 1140 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1141 | goto out_unlock; |
| 1142 | |
| 1143 | /* |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 1144 | * Check the events coming with the callback. At this stage, not |
| 1145 | * every device reports the events in the "key" parameter of the |
| 1146 | * callback. We need to be able to handle both cases here, hence the |
| 1147 | * test for "key" != NULL before the event match test. |
| 1148 | */ |
Al Viro | 3ad6f93 | 2017-07-03 20:14:56 -0400 | [diff] [blame] | 1149 | if (pollflags && !(pollflags & epi->event.events)) |
Davide Libenzi | 2dfa4ee | 2009-03-31 15:24:22 -0700 | [diff] [blame] | 1150 | goto out_unlock; |
| 1151 | |
| 1152 | /* |
Daniel Baluta | bf6a41d | 2011-01-30 23:42:29 +0200 | [diff] [blame] | 1153 | * If we are transferring events to userspace, we can hold no locks |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1154 | * (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] | 1155 | * semantics). All the events that happen during that period of time are |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1156 | * chained in ep->ovflist and requeued later on. |
| 1157 | */ |
Davidlohr Bueso | 9cb8f80 | 2019-01-03 15:27:09 -0800 | [diff] [blame] | 1158 | if (ep->ovflist != EP_UNACTIVE_PTR) { |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1159 | if (epi->next == EP_UNACTIVE_PTR) { |
| 1160 | epi->next = ep->ovflist; |
| 1161 | ep->ovflist = epi; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1162 | if (epi->ws) { |
| 1163 | /* |
| 1164 | * Activate ep->ws since epi->ws may get |
| 1165 | * deactivated at any time. |
| 1166 | */ |
| 1167 | __pm_stay_awake(ep->ws); |
| 1168 | } |
| 1169 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1170 | } |
| 1171 | goto out_unlock; |
| 1172 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1173 | |
| 1174 | /* If this file is already in the ready list we exit soon */ |
Davidlohr Bueso | 992991c | 2018-08-21 21:58:26 -0700 | [diff] [blame] | 1175 | if (!ep_is_linked(epi)) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1176 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1177 | ep_pm_stay_awake_rcu(epi); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1178 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1179 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1180 | /* |
| 1181 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() |
| 1182 | * wait list. |
| 1183 | */ |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1184 | if (waitqueue_active(&ep->wq)) { |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 1185 | if ((epi->event.events & EPOLLEXCLUSIVE) && |
Al Viro | 3ad6f93 | 2017-07-03 20:14:56 -0400 | [diff] [blame] | 1186 | !(pollflags & POLLFREE)) { |
| 1187 | switch (pollflags & EPOLLINOUT_BITS) { |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 1188 | case EPOLLIN: |
| 1189 | if (epi->event.events & EPOLLIN) |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 1190 | ewake = 1; |
| 1191 | break; |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 1192 | case EPOLLOUT: |
| 1193 | if (epi->event.events & EPOLLOUT) |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 1194 | ewake = 1; |
| 1195 | break; |
| 1196 | case 0: |
| 1197 | ewake = 1; |
| 1198 | break; |
| 1199 | } |
| 1200 | } |
Matthew Wilcox | 4a6e9e2c | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1201 | wake_up_locked(&ep->wq); |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1202 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1203 | if (waitqueue_active(&ep->poll_wait)) |
| 1204 | pwake++; |
| 1205 | |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1206 | out_unlock: |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 1207 | spin_unlock_irqrestore(&ep->wq.lock, flags); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1208 | |
| 1209 | /* We have to call this outside the lock */ |
| 1210 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1211 | ep_poll_safewake(&ep->poll_wait); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1212 | |
Oleg Nesterov | 138e4ad | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 1213 | if (!(epi->event.events & EPOLLEXCLUSIVE)) |
| 1214 | ewake = 1; |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1215 | |
Al Viro | 3ad6f93 | 2017-07-03 20:14:56 -0400 | [diff] [blame] | 1216 | if (pollflags & POLLFREE) { |
Oleg Nesterov | 138e4ad | 2017-09-01 18:55:33 +0200 | [diff] [blame] | 1217 | /* |
| 1218 | * If we race with ep_remove_wait_queue() it can miss |
| 1219 | * ->whead = NULL and do another remove_wait_queue() after |
| 1220 | * us, so we can't use __remove_wait_queue(). |
| 1221 | */ |
| 1222 | list_del_init(&wait->entry); |
| 1223 | /* |
| 1224 | * ->whead != NULL protects us from the race with ep_free() |
| 1225 | * or ep_remove(), ep_remove_wait_queue() takes whead->lock |
| 1226 | * held by the caller. Once we nullify it, nothing protects |
| 1227 | * ep/epi or even wait. |
| 1228 | */ |
| 1229 | smp_store_release(&ep_pwq_from_wait(wait)->whead, NULL); |
| 1230 | } |
| 1231 | |
| 1232 | return ewake; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1233 | } |
| 1234 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1235 | /* |
| 1236 | * This is the callback that is used to add our wait queue to the |
| 1237 | * target file wakeup lists. |
| 1238 | */ |
| 1239 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, |
| 1240 | poll_table *pt) |
| 1241 | { |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1242 | struct epitem *epi = ep_item_from_epqueue(pt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1243 | struct eppoll_entry *pwq; |
| 1244 | |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 1245 | if (epi->nwait >= 0 && (pwq = kmem_cache_alloc(pwq_cache, GFP_KERNEL))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1246 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); |
| 1247 | pwq->whead = whead; |
| 1248 | pwq->base = epi; |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 1249 | if (epi->event.events & EPOLLEXCLUSIVE) |
| 1250 | add_wait_queue_exclusive(whead, &pwq->wait); |
| 1251 | else |
| 1252 | add_wait_queue(whead, &pwq->wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1253 | list_add_tail(&pwq->llink, &epi->pwqlist); |
| 1254 | epi->nwait++; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1255 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1256 | /* We have to signal that an error occurred */ |
| 1257 | epi->nwait = -1; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1258 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1259 | } |
| 1260 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1261 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) |
| 1262 | { |
| 1263 | int kcmp; |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1264 | struct rb_node **p = &ep->rbr.rb_root.rb_node, *parent = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1265 | struct epitem *epic; |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1266 | bool leftmost = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | |
| 1268 | while (*p) { |
| 1269 | parent = *p; |
| 1270 | epic = rb_entry(parent, struct epitem, rbn); |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1271 | kcmp = ep_cmp_ffd(&epi->ffd, &epic->ffd); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1272 | if (kcmp > 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1273 | p = &parent->rb_right; |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1274 | leftmost = false; |
| 1275 | } else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1276 | p = &parent->rb_left; |
| 1277 | } |
| 1278 | rb_link_node(&epi->rbn, parent, p); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1279 | rb_insert_color_cached(&epi->rbn, &ep->rbr, leftmost); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1280 | } |
| 1281 | |
Andrew Morton | a80a6b8 | 2012-11-08 15:53:35 -0800 | [diff] [blame] | 1282 | |
| 1283 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1284 | #define PATH_ARR_SIZE 5 |
| 1285 | /* |
| 1286 | * These are the number paths of length 1 to 5, that we are allowing to emanate |
| 1287 | * from a single file of interest. For example, we allow 1000 paths of length |
| 1288 | * 1, to emanate from each file of interest. This essentially represents the |
| 1289 | * potential wakeup paths, which need to be limited in order to avoid massive |
| 1290 | * uncontrolled wakeup storms. The common use case should be a single ep which |
| 1291 | * is connected to n file sources. In this case each file source has 1 path |
| 1292 | * of length 1. Thus, the numbers below should be more than sufficient. These |
| 1293 | * path limits are enforced during an EPOLL_CTL_ADD operation, since a modify |
| 1294 | * and delete can't add additional paths. Protected by the epmutex. |
| 1295 | */ |
| 1296 | static const int path_limits[PATH_ARR_SIZE] = { 1000, 500, 100, 50, 10 }; |
| 1297 | static int path_count[PATH_ARR_SIZE]; |
| 1298 | |
| 1299 | static int path_count_inc(int nests) |
| 1300 | { |
Jason Baron | 93dc610 | 2012-03-16 16:34:03 -0400 | [diff] [blame] | 1301 | /* Allow an arbitrary number of depth 1 paths */ |
| 1302 | if (nests == 0) |
| 1303 | return 0; |
| 1304 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1305 | if (++path_count[nests] > path_limits[nests]) |
| 1306 | return -1; |
| 1307 | return 0; |
| 1308 | } |
| 1309 | |
| 1310 | static void path_count_init(void) |
| 1311 | { |
| 1312 | int i; |
| 1313 | |
| 1314 | for (i = 0; i < PATH_ARR_SIZE; i++) |
| 1315 | path_count[i] = 0; |
| 1316 | } |
| 1317 | |
| 1318 | static int reverse_path_check_proc(void *priv, void *cookie, int call_nests) |
| 1319 | { |
| 1320 | int error = 0; |
| 1321 | struct file *file = priv; |
| 1322 | struct file *child_file; |
| 1323 | struct epitem *epi; |
| 1324 | |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1325 | /* CTL_DEL can remove links here, but that can't increase our count */ |
| 1326 | rcu_read_lock(); |
| 1327 | list_for_each_entry_rcu(epi, &file->f_ep_links, fllink) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1328 | child_file = epi->ep->file; |
| 1329 | if (is_file_epoll(child_file)) { |
| 1330 | if (list_empty(&child_file->f_ep_links)) { |
| 1331 | if (path_count_inc(call_nests)) { |
| 1332 | error = -1; |
| 1333 | break; |
| 1334 | } |
| 1335 | } else { |
| 1336 | error = ep_call_nested(&poll_loop_ncalls, |
| 1337 | EP_MAX_NESTS, |
| 1338 | reverse_path_check_proc, |
| 1339 | child_file, child_file, |
| 1340 | current); |
| 1341 | } |
| 1342 | if (error != 0) |
| 1343 | break; |
| 1344 | } else { |
| 1345 | printk(KERN_ERR "reverse_path_check_proc: " |
| 1346 | "file is not an ep!\n"); |
| 1347 | } |
| 1348 | } |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1349 | rcu_read_unlock(); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1350 | return error; |
| 1351 | } |
| 1352 | |
| 1353 | /** |
| 1354 | * reverse_path_check - The tfile_check_list is list of file *, which have |
| 1355 | * links that are proposed to be newly added. We need to |
| 1356 | * make sure that those added links don't add too many |
| 1357 | * paths such that we will spend all our time waking up |
| 1358 | * eventpoll objects. |
| 1359 | * |
| 1360 | * Returns: Returns zero if the proposed links don't create too many paths, |
| 1361 | * -1 otherwise. |
| 1362 | */ |
| 1363 | static int reverse_path_check(void) |
| 1364 | { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1365 | int error = 0; |
| 1366 | struct file *current_file; |
| 1367 | |
| 1368 | /* let's call this for all tfiles */ |
| 1369 | list_for_each_entry(current_file, &tfile_check_list, f_tfile_llink) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1370 | path_count_init(); |
| 1371 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
| 1372 | reverse_path_check_proc, current_file, |
| 1373 | current_file, current); |
| 1374 | if (error) |
| 1375 | break; |
| 1376 | } |
| 1377 | return error; |
| 1378 | } |
| 1379 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1380 | static int ep_create_wakeup_source(struct epitem *epi) |
| 1381 | { |
| 1382 | const char *name; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1383 | struct wakeup_source *ws; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1384 | |
| 1385 | if (!epi->ep->ws) { |
| 1386 | epi->ep->ws = wakeup_source_register("eventpoll"); |
| 1387 | if (!epi->ep->ws) |
| 1388 | return -ENOMEM; |
| 1389 | } |
| 1390 | |
| 1391 | name = epi->ffd.file->f_path.dentry->d_name.name; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1392 | ws = wakeup_source_register(name); |
| 1393 | |
| 1394 | if (!ws) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1395 | return -ENOMEM; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1396 | rcu_assign_pointer(epi->ws, ws); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1397 | |
| 1398 | return 0; |
| 1399 | } |
| 1400 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1401 | /* rare code path, only used when EPOLL_CTL_MOD removes a wakeup source */ |
| 1402 | static noinline void ep_destroy_wakeup_source(struct epitem *epi) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1403 | { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1404 | struct wakeup_source *ws = ep_wakeup_source(epi); |
| 1405 | |
Eric Wong | d6d67e7 | 2013-04-30 15:27:43 -0700 | [diff] [blame] | 1406 | RCU_INIT_POINTER(epi->ws, NULL); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1407 | |
| 1408 | /* |
| 1409 | * wait for ep_pm_stay_awake_rcu to finish, synchronize_rcu is |
| 1410 | * used internally by wakeup_source_remove, too (called by |
| 1411 | * wakeup_source_unregister), so we cannot use call_rcu |
| 1412 | */ |
| 1413 | synchronize_rcu(); |
| 1414 | wakeup_source_unregister(ws); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1415 | } |
| 1416 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1417 | /* |
| 1418 | * Must be called with "mtx" held. |
| 1419 | */ |
Al Viro | bec1a502 | 2017-11-28 19:43:33 -0500 | [diff] [blame] | 1420 | static int ep_insert(struct eventpoll *ep, const struct epoll_event *event, |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1421 | struct file *tfile, int fd, int full_check) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1422 | { |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 1423 | int error, pwake = 0; |
| 1424 | __poll_t revents; |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1425 | long user_watches; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1426 | struct epitem *epi; |
| 1427 | struct ep_pqueue epq; |
| 1428 | |
Davidlohr Bueso | 92e6417 | 2018-08-21 21:56:45 -0700 | [diff] [blame] | 1429 | lockdep_assert_irqs_enabled(); |
| 1430 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1431 | user_watches = atomic_long_read(&ep->user->epoll_watches); |
| 1432 | if (unlikely(user_watches >= max_user_watches)) |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1433 | return -ENOSPC; |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 1434 | if (!(epi = kmem_cache_alloc(epi_cache, GFP_KERNEL))) |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1435 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1436 | |
| 1437 | /* Item initialization follow here ... */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1438 | INIT_LIST_HEAD(&epi->rdllink); |
| 1439 | INIT_LIST_HEAD(&epi->fllink); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1440 | INIT_LIST_HEAD(&epi->pwqlist); |
| 1441 | epi->ep = ep; |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1442 | ep_set_ffd(&epi->ffd, tfile, fd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | epi->event = *event; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1444 | epi->nwait = 0; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1445 | epi->next = EP_UNACTIVE_PTR; |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1446 | if (epi->event.events & EPOLLWAKEUP) { |
| 1447 | error = ep_create_wakeup_source(epi); |
| 1448 | if (error) |
| 1449 | goto error_create_wakeup_source; |
| 1450 | } else { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1451 | RCU_INIT_POINTER(epi->ws, NULL); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1452 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1453 | |
| 1454 | /* Initialize the poll table using the queue callback */ |
| 1455 | epq.epi = epi; |
| 1456 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); |
| 1457 | |
| 1458 | /* |
| 1459 | * Attach the item to the poll hooks and get current event bits. |
| 1460 | * We can safely use the file* here because its usage count has |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1461 | * been increased by the caller of this function. Note that after |
| 1462 | * this operation completes, the poll callback can start hitting |
| 1463 | * the new item. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1464 | */ |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 1465 | revents = ep_item_poll(epi, &epq.pt, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | |
| 1467 | /* |
| 1468 | * We have to check if something went wrong during the poll wait queue |
| 1469 | * install process. Namely an allocation for a wait queue failed due |
| 1470 | * high memory pressure. |
| 1471 | */ |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1472 | error = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | if (epi->nwait < 0) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1474 | goto error_unregister; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1475 | |
| 1476 | /* Add the current item to the list of active epoll hook for this file */ |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 1477 | spin_lock(&tfile->f_lock); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1478 | list_add_tail_rcu(&epi->fllink, &tfile->f_ep_links); |
Jonathan Corbet | 6849991 | 2009-02-06 13:52:43 -0700 | [diff] [blame] | 1479 | spin_unlock(&tfile->f_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1480 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1481 | /* |
| 1482 | * Add the current item to the RB tree. All RB tree operations are |
| 1483 | * protected by "mtx", and ep_insert() is called with "mtx" held. |
| 1484 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1485 | ep_rbtree_insert(ep, epi); |
| 1486 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1487 | /* now check if we've created too many backpaths */ |
| 1488 | error = -EINVAL; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 1489 | if (full_check && reverse_path_check()) |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1490 | goto error_remove_epi; |
| 1491 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1492 | /* We have to drop the new item inside our item list to keep track of it */ |
Davidlohr Bueso | 304b18b | 2018-08-21 21:56:41 -0700 | [diff] [blame] | 1493 | spin_lock_irq(&ep->wq.lock); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1494 | |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 1495 | /* record NAPI ID of new item if present */ |
| 1496 | ep_set_busy_poll_napi_id(epi); |
| 1497 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | /* If the file is already "ready" we drop it inside the ready list */ |
Davidlohr Bueso | 992991c | 2018-08-21 21:58:26 -0700 | [diff] [blame] | 1499 | if (revents && !ep_is_linked(epi)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1501 | ep_pm_stay_awake(epi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | |
| 1503 | /* Notify waiting tasks that events are available */ |
| 1504 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2c | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1505 | wake_up_locked(&ep->wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | if (waitqueue_active(&ep->poll_wait)) |
| 1507 | pwake++; |
| 1508 | } |
| 1509 | |
Davidlohr Bueso | 304b18b | 2018-08-21 21:56:41 -0700 | [diff] [blame] | 1510 | spin_unlock_irq(&ep->wq.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1511 | |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 1512 | atomic_long_inc(&ep->user->epoll_watches); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1513 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | /* We have to call this outside the lock */ |
| 1515 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1516 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1517 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1518 | return 0; |
| 1519 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1520 | error_remove_epi: |
| 1521 | spin_lock(&tfile->f_lock); |
Jason Baron | ae10b2b | 2013-11-12 15:10:16 -0800 | [diff] [blame] | 1522 | list_del_rcu(&epi->fllink); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1523 | spin_unlock(&tfile->f_lock); |
| 1524 | |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1525 | rb_erase_cached(&epi->rbn, &ep->rbr); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1526 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1527 | error_unregister: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1528 | ep_unregister_pollwait(ep, epi); |
| 1529 | |
| 1530 | /* |
| 1531 | * 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] | 1532 | * allocated wait queue. Note that we don't care about the ep->ovflist |
| 1533 | * list, since that is used/cleaned only inside a section bound by "mtx". |
| 1534 | * And ep_insert() is called with "mtx" held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1535 | */ |
Davidlohr Bueso | 304b18b | 2018-08-21 21:56:41 -0700 | [diff] [blame] | 1536 | spin_lock_irq(&ep->wq.lock); |
Davidlohr Bueso | 992991c | 2018-08-21 21:58:26 -0700 | [diff] [blame] | 1537 | if (ep_is_linked(epi)) |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 1538 | list_del_init(&epi->rdllink); |
Davidlohr Bueso | 304b18b | 2018-08-21 21:56:41 -0700 | [diff] [blame] | 1539 | spin_unlock_irq(&ep->wq.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1540 | |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1541 | wakeup_source_unregister(ep_wakeup_source(epi)); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1542 | |
| 1543 | error_create_wakeup_source: |
Pekka Enberg | b030a4d | 2005-06-23 00:10:03 -0700 | [diff] [blame] | 1544 | kmem_cache_free(epi_cache, epi); |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 1545 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1546 | return error; |
| 1547 | } |
| 1548 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1549 | /* |
| 1550 | * 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] | 1551 | * 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] | 1552 | */ |
Al Viro | bec1a502 | 2017-11-28 19:43:33 -0500 | [diff] [blame] | 1553 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, |
| 1554 | const struct epoll_event *event) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1555 | { |
| 1556 | int pwake = 0; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 1557 | poll_table pt; |
| 1558 | |
Davidlohr Bueso | 92e6417 | 2018-08-21 21:56:45 -0700 | [diff] [blame] | 1559 | lockdep_assert_irqs_enabled(); |
| 1560 | |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 1561 | init_poll_funcptr(&pt, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1562 | |
| 1563 | /* |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1564 | * Set the new event interest mask before calling f_op->poll(); |
| 1565 | * otherwise we might miss an event that happens between the |
| 1566 | * f_op->poll() call and the new event set registering. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1567 | */ |
Eric Wong | 128dd17 | 2013-01-01 21:20:27 +0000 | [diff] [blame] | 1568 | epi->event.events = event->events; /* need barrier below */ |
Tony Battersby | e057e15 | 2009-03-31 15:24:15 -0700 | [diff] [blame] | 1569 | epi->event.data = event->data; /* protected by mtx */ |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1570 | if (epi->event.events & EPOLLWAKEUP) { |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1571 | if (!ep_has_wakeup_source(epi)) |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1572 | ep_create_wakeup_source(epi); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1573 | } else if (ep_has_wakeup_source(epi)) { |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1574 | ep_destroy_wakeup_source(epi); |
| 1575 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1576 | |
| 1577 | /* |
Eric Wong | 128dd17 | 2013-01-01 21:20:27 +0000 | [diff] [blame] | 1578 | * The following barrier has two effects: |
| 1579 | * |
| 1580 | * 1) Flush epi changes above to other CPUs. This ensures |
| 1581 | * we do not miss events from ep_poll_callback if an |
| 1582 | * event occurs immediately after we call f_op->poll(). |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 1583 | * We need this because we did not take ep->wq.lock while |
Eric Wong | 128dd17 | 2013-01-01 21:20:27 +0000 | [diff] [blame] | 1584 | * changing epi above (but ep_poll_callback does take |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 1585 | * ep->wq.lock). |
Eric Wong | 128dd17 | 2013-01-01 21:20:27 +0000 | [diff] [blame] | 1586 | * |
| 1587 | * 2) We also need to ensure we do not miss _past_ events |
| 1588 | * when calling f_op->poll(). This barrier also |
| 1589 | * pairs with the barrier in wq_has_sleeper (see |
| 1590 | * comments for wq_has_sleeper). |
| 1591 | * |
| 1592 | * This barrier will now guarantee ep_poll_callback or f_op->poll |
| 1593 | * (or both) will notice the readiness of an item. |
| 1594 | */ |
| 1595 | smp_mb(); |
| 1596 | |
| 1597 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1598 | * Get current event bits. We can safely use the file* here because |
| 1599 | * its usage count has been increased by the caller of this function. |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1600 | * 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] | 1601 | * list, push it inside. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1602 | */ |
Al Viro | 6911273 | 2017-11-28 19:56:15 -0500 | [diff] [blame] | 1603 | if (ep_item_poll(epi, &pt, 1)) { |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 1604 | spin_lock_irq(&ep->wq.lock); |
Davidlohr Bueso | 992991c | 2018-08-21 21:58:26 -0700 | [diff] [blame] | 1605 | if (!ep_is_linked(epi)) { |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1606 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1607 | ep_pm_stay_awake(epi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1608 | |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1609 | /* Notify waiting tasks that events are available */ |
| 1610 | if (waitqueue_active(&ep->wq)) |
Matthew Wilcox | 4a6e9e2c | 2007-08-30 16:10:22 -0400 | [diff] [blame] | 1611 | wake_up_locked(&ep->wq); |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 1612 | if (waitqueue_active(&ep->poll_wait)) |
| 1613 | pwake++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | } |
Christoph Hellwig | ee8ef0a | 2018-08-21 21:56:26 -0700 | [diff] [blame] | 1615 | spin_unlock_irq(&ep->wq.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1616 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1617 | |
| 1618 | /* We have to call this outside the lock */ |
| 1619 | if (pwake) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1620 | ep_poll_safewake(&ep->poll_wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1621 | |
| 1622 | return 0; |
| 1623 | } |
| 1624 | |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 1625 | static __poll_t ep_send_events_proc(struct eventpoll *ep, struct list_head *head, |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1626 | void *priv) |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1627 | { |
| 1628 | struct ep_send_events_data *esed = priv; |
Al Viro | d85e2aa | 2018-02-01 15:24:58 -0500 | [diff] [blame] | 1629 | __poll_t revents; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1630 | struct epitem *epi; |
| 1631 | struct epoll_event __user *uevent; |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1632 | struct wakeup_source *ws; |
Hans Verkuil | 626cf23 | 2012-03-23 15:02:27 -0700 | [diff] [blame] | 1633 | poll_table pt; |
| 1634 | |
| 1635 | init_poll_funcptr(&pt, NULL); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1636 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1637 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1638 | * We can loop without lock because we are passed a task private list. |
| 1639 | * Items cannot vanish during the loop because ep_scan_ready_list() is |
| 1640 | * holding "mtx" during this call. |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1641 | */ |
Al Viro | d7ebbe4 | 2018-02-01 15:21:13 -0500 | [diff] [blame] | 1642 | for (esed->res = 0, uevent = esed->events; |
| 1643 | !list_empty(head) && esed->res < esed->maxevents;) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1644 | epi = list_first_entry(head, struct epitem, rdllink); |
| 1645 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 1646 | /* |
| 1647 | * Activate ep->ws before deactivating epi->ws to prevent |
| 1648 | * triggering auto-suspend here (in case we reactive epi->ws |
| 1649 | * below). |
| 1650 | * |
| 1651 | * This could be rearranged to delay the deactivation of epi->ws |
| 1652 | * instead, but then epi->ws would temporarily be out of sync |
| 1653 | * with ep_is_linked(). |
| 1654 | */ |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1655 | ws = ep_wakeup_source(epi); |
| 1656 | if (ws) { |
| 1657 | if (ws->active) |
| 1658 | __pm_stay_awake(ep->ws); |
| 1659 | __pm_relax(ws); |
| 1660 | } |
| 1661 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1662 | list_del_init(&epi->rdllink); |
| 1663 | |
Jason Baron | 37b5e52 | 2017-11-17 15:29:06 -0800 | [diff] [blame] | 1664 | revents = ep_item_poll(epi, &pt, 1); |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1665 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1666 | /* |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1667 | * If the event mask intersect the caller-requested one, |
| 1668 | * deliver the event to userspace. Again, ep_scan_ready_list() |
| 1669 | * is holding "mtx", so no operations coming from userspace |
| 1670 | * can change the item. |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1671 | */ |
| 1672 | if (revents) { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1673 | if (__put_user(revents, &uevent->events) || |
Tony Battersby | d030588 | 2009-03-31 15:24:14 -0700 | [diff] [blame] | 1674 | __put_user(epi->event.data, &uevent->data)) { |
| 1675 | list_add(&epi->rdllink, head); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1676 | ep_pm_stay_awake(epi); |
Al Viro | d7ebbe4 | 2018-02-01 15:21:13 -0500 | [diff] [blame] | 1677 | if (!esed->res) |
| 1678 | esed->res = -EFAULT; |
| 1679 | return 0; |
Tony Battersby | d030588 | 2009-03-31 15:24:14 -0700 | [diff] [blame] | 1680 | } |
Al Viro | d7ebbe4 | 2018-02-01 15:21:13 -0500 | [diff] [blame] | 1681 | esed->res++; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1682 | uevent++; |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1683 | if (epi->event.events & EPOLLONESHOT) |
| 1684 | epi->event.events &= EP_PRIVATE_BITS; |
| 1685 | else if (!(epi->event.events & EPOLLET)) { |
| 1686 | /* |
| 1687 | * If this file has been added with Level |
| 1688 | * Trigger mode, we need to insert back inside |
| 1689 | * the ready list, so that the next call to |
| 1690 | * epoll_wait() will check again the events |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1691 | * availability. At this point, no one can insert |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1692 | * into ep->rdllist besides us. The epoll_ctl() |
| 1693 | * callers are locked out by |
| 1694 | * ep_scan_ready_list() holding "mtx" and the |
| 1695 | * poll callback will queue them in ep->ovflist. |
| 1696 | */ |
| 1697 | list_add_tail(&epi->rdllink, &ep->rdllist); |
Eric Wong | eea1d58 | 2013-04-30 15:27:39 -0700 | [diff] [blame] | 1698 | ep_pm_stay_awake(epi); |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1699 | } |
| 1700 | } |
| 1701 | } |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1702 | |
Al Viro | d7ebbe4 | 2018-02-01 15:21:13 -0500 | [diff] [blame] | 1703 | return 0; |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1704 | } |
| 1705 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1706 | static int ep_send_events(struct eventpoll *ep, |
| 1707 | struct epoll_event __user *events, int maxevents) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1708 | { |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1709 | struct ep_send_events_data esed; |
Davide Libenzi | 6192bd5 | 2007-05-08 00:25:41 -0700 | [diff] [blame] | 1710 | |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 1711 | esed.maxevents = maxevents; |
| 1712 | esed.events = events; |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 1713 | |
Al Viro | d7ebbe4 | 2018-02-01 15:21:13 -0500 | [diff] [blame] | 1714 | ep_scan_ready_list(ep, ep_send_events_proc, &esed, 0, false); |
| 1715 | return esed.res; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1716 | } |
| 1717 | |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1718 | static inline struct timespec64 ep_set_mstimeout(long ms) |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1719 | { |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1720 | struct timespec64 now, ts = { |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1721 | .tv_sec = ms / MSEC_PER_SEC, |
| 1722 | .tv_nsec = NSEC_PER_MSEC * (ms % MSEC_PER_SEC), |
| 1723 | }; |
| 1724 | |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1725 | ktime_get_ts64(&now); |
| 1726 | return timespec64_add_safe(now, ts); |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1727 | } |
| 1728 | |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1729 | /** |
| 1730 | * ep_poll - Retrieves ready events, and delivers them to the caller supplied |
| 1731 | * event buffer. |
| 1732 | * |
| 1733 | * @ep: Pointer to the eventpoll context. |
| 1734 | * @events: Pointer to the userspace buffer where the ready events should be |
| 1735 | * stored. |
| 1736 | * @maxevents: Size (in terms of number of events) of the caller event buffer. |
| 1737 | * @timeout: Maximum timeout for the ready events fetch operation, in |
| 1738 | * milliseconds. If the @timeout is zero, the function will not block, |
| 1739 | * while if the @timeout is less than zero, the function will block |
| 1740 | * until at least one event has been retrieved (or an error |
| 1741 | * occurred). |
| 1742 | * |
| 1743 | * Returns: Returns the number of ready events which have been fetched, or an |
| 1744 | * error code, in case of error. |
| 1745 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1746 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, |
| 1747 | int maxevents, long timeout) |
| 1748 | { |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1749 | int res = 0, eavail, timed_out = 0; |
John Stultz | da8b44d | 2016-03-17 14:20:51 -0700 | [diff] [blame] | 1750 | u64 slack = 0; |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1751 | wait_queue_entry_t wait; |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1752 | ktime_t expires, *to = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | |
Davidlohr Bueso | 679abf3 | 2018-08-21 21:58:23 -0700 | [diff] [blame] | 1754 | lockdep_assert_irqs_enabled(); |
| 1755 | |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1756 | if (timeout > 0) { |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1757 | struct timespec64 end_time = ep_set_mstimeout(timeout); |
Eric Dumazet | 0781b90 | 2011-02-01 15:52:35 -0800 | [diff] [blame] | 1758 | |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1759 | slack = select_estimate_accuracy(&end_time); |
| 1760 | to = &expires; |
Deepa Dinamani | 766b9f9 | 2016-05-19 17:09:05 -0700 | [diff] [blame] | 1761 | *to = timespec64_to_ktime(end_time); |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1762 | } else if (timeout == 0) { |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1763 | /* |
| 1764 | * Avoid the unnecessary trip to the wait queue loop, if the |
| 1765 | * caller specified a non blocking operation. |
| 1766 | */ |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1767 | timed_out = 1; |
Davidlohr Bueso | 679abf3 | 2018-08-21 21:58:23 -0700 | [diff] [blame] | 1768 | spin_lock_irq(&ep->wq.lock); |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1769 | goto check_events; |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1770 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1771 | |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1772 | fetch_events: |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 1773 | |
| 1774 | if (!ep_events_available(ep)) |
| 1775 | ep_busy_loop(ep, timed_out); |
| 1776 | |
Davidlohr Bueso | 679abf3 | 2018-08-21 21:58:23 -0700 | [diff] [blame] | 1777 | spin_lock_irq(&ep->wq.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1778 | |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1779 | if (!ep_events_available(ep)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1780 | /* |
Sridhar Samudrala | bf3b9f6 | 2017-03-24 10:08:30 -0700 | [diff] [blame] | 1781 | * Busy poll timed out. Drop NAPI ID for now, we can add |
| 1782 | * it back in when we have moved a socket with a valid NAPI |
| 1783 | * ID onto the ready list. |
| 1784 | */ |
| 1785 | ep_reset_busy_poll_napi_id(ep); |
| 1786 | |
| 1787 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1788 | * We don't have any available event to return to the caller. |
| 1789 | * We need to sleep here, and we will be wake up by |
| 1790 | * ep_poll_callback() when events will become available. |
| 1791 | */ |
| 1792 | init_waitqueue_entry(&wait, current); |
Changli Gao | a93d2f1 | 2010-05-07 14:33:26 +0800 | [diff] [blame] | 1793 | __add_wait_queue_exclusive(&ep->wq, &wait); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1794 | |
| 1795 | for (;;) { |
| 1796 | /* |
| 1797 | * We don't want to sleep if the ep_poll_callback() sends us |
| 1798 | * a wakeup in between. That's why we set the task state |
| 1799 | * to TASK_INTERRUPTIBLE before doing the checks. |
| 1800 | */ |
| 1801 | set_current_state(TASK_INTERRUPTIBLE); |
David Rientjes | c257a34 | 2017-07-10 15:52:33 -0700 | [diff] [blame] | 1802 | /* |
| 1803 | * Always short-circuit for fatal signals to allow |
| 1804 | * threads to make a timely exit without the chance of |
| 1805 | * finding more events available and fetching |
| 1806 | * repeatedly. |
| 1807 | */ |
| 1808 | if (fatal_signal_pending(current)) { |
| 1809 | res = -EINTR; |
| 1810 | break; |
| 1811 | } |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1812 | if (ep_events_available(ep) || timed_out) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1813 | break; |
| 1814 | if (signal_pending(current)) { |
| 1815 | res = -EINTR; |
| 1816 | break; |
| 1817 | } |
| 1818 | |
Davidlohr Bueso | 679abf3 | 2018-08-21 21:58:23 -0700 | [diff] [blame] | 1819 | spin_unlock_irq(&ep->wq.lock); |
Colin Cross | 0c39a6a | 2013-05-06 23:50:16 +0000 | [diff] [blame] | 1820 | if (!freezable_schedule_hrtimeout_range(to, slack, |
| 1821 | HRTIMER_MODE_ABS)) |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1822 | timed_out = 1; |
| 1823 | |
Davidlohr Bueso | 679abf3 | 2018-08-21 21:58:23 -0700 | [diff] [blame] | 1824 | spin_lock_irq(&ep->wq.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1825 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1826 | |
Davidlohr Bueso | 4d5755b | 2015-02-13 14:39:08 -0800 | [diff] [blame] | 1827 | __remove_wait_queue(&ep->wq, &wait); |
| 1828 | __set_current_state(TASK_RUNNING); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1829 | } |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1830 | check_events: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1831 | /* Is it worth to try to dig for events ? */ |
Davide Libenzi | 3fb0e58 | 2011-03-22 16:34:46 -0700 | [diff] [blame] | 1832 | eavail = ep_events_available(ep); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1833 | |
Davidlohr Bueso | 679abf3 | 2018-08-21 21:58:23 -0700 | [diff] [blame] | 1834 | spin_unlock_irq(&ep->wq.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1835 | |
| 1836 | /* |
| 1837 | * Try to transfer events to user space. In case we get 0 events and |
| 1838 | * there's still timeout left over, we go trying again in search of |
| 1839 | * more luck. |
| 1840 | */ |
| 1841 | if (!res && eavail && |
Shawn Bohrer | 95aac7b | 2010-10-27 15:34:54 -0700 | [diff] [blame] | 1842 | !(res = ep_send_events(ep, events, maxevents)) && !timed_out) |
Shawn Bohrer | f4d93ad | 2011-03-22 16:34:47 -0700 | [diff] [blame] | 1843 | goto fetch_events; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1844 | |
| 1845 | return res; |
| 1846 | } |
| 1847 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1848 | /** |
| 1849 | * ep_loop_check_proc - Callback function to be passed to the @ep_call_nested() |
| 1850 | * API, to verify that adding an epoll file inside another |
| 1851 | * epoll structure, does not violate the constraints, in |
| 1852 | * terms of closed loops, or too deep chains (which can |
| 1853 | * result in excessive stack usage). |
| 1854 | * |
| 1855 | * @priv: Pointer to the epoll file to be currently checked. |
| 1856 | * @cookie: Original cookie for this call. This is the top-of-the-chain epoll |
| 1857 | * data structure pointer. |
| 1858 | * @call_nests: Current dept of the @ep_call_nested() call stack. |
| 1859 | * |
| 1860 | * Returns: Returns zero if adding the epoll @file inside current epoll |
| 1861 | * structure @ep does not violate the constraints, or -1 otherwise. |
| 1862 | */ |
| 1863 | static int ep_loop_check_proc(void *priv, void *cookie, int call_nests) |
| 1864 | { |
| 1865 | int error = 0; |
| 1866 | struct file *file = priv; |
| 1867 | struct eventpoll *ep = file->private_data; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1868 | struct eventpoll *ep_tovisit; |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1869 | struct rb_node *rbp; |
| 1870 | struct epitem *epi; |
| 1871 | |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 1872 | mutex_lock_nested(&ep->mtx, call_nests + 1); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1873 | ep->visited = 1; |
| 1874 | list_add(&ep->visited_list_link, &visited_list); |
Davidlohr Bueso | b2ac2ea | 2017-09-08 16:15:18 -0700 | [diff] [blame] | 1875 | for (rbp = rb_first_cached(&ep->rbr); rbp; rbp = rb_next(rbp)) { |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1876 | epi = rb_entry(rbp, struct epitem, rbn); |
| 1877 | if (unlikely(is_file_epoll(epi->ffd.file))) { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1878 | ep_tovisit = epi->ffd.file->private_data; |
| 1879 | if (ep_tovisit->visited) |
| 1880 | continue; |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1881 | error = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1882 | ep_loop_check_proc, epi->ffd.file, |
| 1883 | ep_tovisit, current); |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1884 | if (error != 0) |
| 1885 | break; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1886 | } else { |
| 1887 | /* |
| 1888 | * If we've reached a file that is not associated with |
| 1889 | * an ep, then we need to check if the newly added |
| 1890 | * links are going to add too many wakeup paths. We do |
| 1891 | * this by adding it to the tfile_check_list, if it's |
| 1892 | * not already there, and calling reverse_path_check() |
| 1893 | * during ep_insert(). |
| 1894 | */ |
| 1895 | if (list_empty(&epi->ffd.file->f_tfile_llink)) |
| 1896 | list_add(&epi->ffd.file->f_tfile_llink, |
| 1897 | &tfile_check_list); |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1898 | } |
| 1899 | } |
| 1900 | mutex_unlock(&ep->mtx); |
| 1901 | |
| 1902 | return error; |
| 1903 | } |
| 1904 | |
| 1905 | /** |
| 1906 | * ep_loop_check - Performs a check to verify that adding an epoll file (@file) |
| 1907 | * another epoll file (represented by @ep) does not create |
| 1908 | * closed loops or too deep chains. |
| 1909 | * |
| 1910 | * @ep: Pointer to the epoll private data structure. |
| 1911 | * @file: Pointer to the epoll file to be checked. |
| 1912 | * |
| 1913 | * Returns: Returns zero if adding the epoll @file inside current epoll |
| 1914 | * structure @ep does not violate the constraints, or -1 otherwise. |
| 1915 | */ |
| 1916 | static int ep_loop_check(struct eventpoll *ep, struct file *file) |
| 1917 | { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1918 | int ret; |
| 1919 | struct eventpoll *ep_cur, *ep_next; |
| 1920 | |
| 1921 | ret = ep_call_nested(&poll_loop_ncalls, EP_MAX_NESTS, |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1922 | ep_loop_check_proc, file, ep, current); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1923 | /* clear visited list */ |
| 1924 | list_for_each_entry_safe(ep_cur, ep_next, &visited_list, |
| 1925 | visited_list_link) { |
| 1926 | ep_cur->visited = 0; |
| 1927 | list_del(&ep_cur->visited_list_link); |
| 1928 | } |
| 1929 | return ret; |
| 1930 | } |
| 1931 | |
| 1932 | static void clear_tfile_check_list(void) |
| 1933 | { |
| 1934 | struct file *file; |
| 1935 | |
| 1936 | /* first clear the tfile_check_list */ |
| 1937 | while (!list_empty(&tfile_check_list)) { |
| 1938 | file = list_first_entry(&tfile_check_list, struct file, |
| 1939 | f_tfile_llink); |
| 1940 | list_del_init(&file->f_tfile_llink); |
| 1941 | } |
| 1942 | INIT_LIST_HEAD(&tfile_check_list); |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 1943 | } |
| 1944 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1945 | /* |
Andrew Morton | 523723b | 2008-08-12 15:09:01 -0700 | [diff] [blame] | 1946 | * Open an eventpoll file descriptor. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1947 | */ |
Dominik Brodowski | 791eb22 | 2018-03-11 11:34:30 +0100 | [diff] [blame] | 1948 | static int do_epoll_create(int flags) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1949 | { |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1950 | int error, fd; |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1951 | struct eventpoll *ep = NULL; |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1952 | struct file *file; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1953 | |
Ulrich Drepper | e38b36f | 2008-07-23 21:29:42 -0700 | [diff] [blame] | 1954 | /* Check the EPOLL_* constant for consistency. */ |
| 1955 | BUILD_BUG_ON(EPOLL_CLOEXEC != O_CLOEXEC); |
| 1956 | |
Davide Libenzi | 296e236 | 2009-03-31 15:24:11 -0700 | [diff] [blame] | 1957 | if (flags & ~EPOLL_CLOEXEC) |
| 1958 | return -EINVAL; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1959 | /* |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1960 | * Create the internal data structure ("struct eventpoll"). |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1961 | */ |
Ulrich Drepper | 9fe5ad9 | 2008-07-23 21:29:43 -0700 | [diff] [blame] | 1962 | error = ep_alloc(&ep); |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1963 | if (error < 0) |
| 1964 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1965 | /* |
| 1966 | * Creates all the items needed to setup an eventpoll file. That is, |
Al Viro | 2030a42 | 2008-02-23 06:46:49 -0500 | [diff] [blame] | 1967 | * a file structure and a free file descriptor. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1968 | */ |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1969 | fd = get_unused_fd_flags(O_RDWR | (flags & O_CLOEXEC)); |
| 1970 | if (fd < 0) { |
| 1971 | error = fd; |
| 1972 | goto out_free_ep; |
| 1973 | } |
| 1974 | file = anon_inode_getfile("[eventpoll]", &eventpoll_fops, ep, |
Roland Dreier | 628ff7c | 2009-12-18 09:41:24 -0800 | [diff] [blame] | 1975 | O_RDWR | (flags & O_CLOEXEC)); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1976 | if (IS_ERR(file)) { |
| 1977 | error = PTR_ERR(file); |
| 1978 | goto out_free_fd; |
| 1979 | } |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1980 | ep->file = file; |
Al Viro | 9802274 | 2012-08-17 22:42:36 -0400 | [diff] [blame] | 1981 | fd_install(fd, file); |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1982 | return fd; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1983 | |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 1984 | out_free_fd: |
| 1985 | put_unused_fd(fd); |
| 1986 | out_free_ep: |
| 1987 | ep_free(ep); |
Davide Libenzi | bb57c3e | 2009-03-31 15:24:12 -0700 | [diff] [blame] | 1988 | return error; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 1989 | } |
| 1990 | |
Dominik Brodowski | 791eb22 | 2018-03-11 11:34:30 +0100 | [diff] [blame] | 1991 | SYSCALL_DEFINE1(epoll_create1, int, flags) |
| 1992 | { |
| 1993 | return do_epoll_create(flags); |
| 1994 | } |
| 1995 | |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 1996 | SYSCALL_DEFINE1(epoll_create, int, size) |
Ulrich Drepper | a0998b5 | 2008-07-23 21:29:27 -0700 | [diff] [blame] | 1997 | { |
Davide Libenzi | bfe3891 | 2009-05-12 13:19:44 -0700 | [diff] [blame] | 1998 | if (size <= 0) |
Ulrich Drepper | 9fe5ad9 | 2008-07-23 21:29:43 -0700 | [diff] [blame] | 1999 | return -EINVAL; |
| 2000 | |
Dominik Brodowski | 791eb22 | 2018-03-11 11:34:30 +0100 | [diff] [blame] | 2001 | return do_epoll_create(0); |
Ulrich Drepper | a0998b5 | 2008-07-23 21:29:27 -0700 | [diff] [blame] | 2002 | } |
| 2003 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2004 | /* |
| 2005 | * The following function implements the controller interface for |
| 2006 | * the eventpoll file that enables the insertion/removal/change of |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 2007 | * file descriptors inside the interest set. |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2008 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 2009 | SYSCALL_DEFINE4(epoll_ctl, int, epfd, int, op, int, fd, |
| 2010 | struct epoll_event __user *, event) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2011 | { |
| 2012 | int error; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2013 | int full_check = 0; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2014 | struct fd f, tf; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2015 | struct eventpoll *ep; |
| 2016 | struct epitem *epi; |
| 2017 | struct epoll_event epds; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2018 | struct eventpoll *tep = NULL; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2019 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2020 | error = -EFAULT; |
| 2021 | if (ep_op_has_event(op) && |
| 2022 | copy_from_user(&epds, event, sizeof(struct epoll_event))) |
| 2023 | goto error_return; |
| 2024 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2025 | error = -EBADF; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2026 | f = fdget(epfd); |
| 2027 | if (!f.file) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2028 | goto error_return; |
| 2029 | |
| 2030 | /* Get the "struct file *" for the target file */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2031 | tf = fdget(fd); |
| 2032 | if (!tf.file) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2033 | goto error_fput; |
| 2034 | |
| 2035 | /* The target file descriptor must support poll */ |
| 2036 | error = -EPERM; |
Christoph Hellwig | 9965ed17 | 2018-03-05 07:26:05 -0800 | [diff] [blame] | 2037 | if (!file_can_poll(tf.file)) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2038 | goto error_tgt_fput; |
| 2039 | |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 2040 | /* Check if EPOLLWAKEUP is allowed */ |
Nicolas Iooss | c680e41 | 2014-09-09 14:50:51 -0700 | [diff] [blame] | 2041 | if (ep_op_has_event(op)) |
| 2042 | ep_take_care_of_epollwakeup(&epds); |
Arve Hjønnevåg | 4d7e30d | 2012-05-01 21:33:34 +0200 | [diff] [blame] | 2043 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2044 | /* |
| 2045 | * We have to check that the file structure underneath the file descriptor |
| 2046 | * the user passed to us _is_ an eventpoll file. And also we do not permit |
| 2047 | * adding an epoll file descriptor inside itself. |
| 2048 | */ |
| 2049 | error = -EINVAL; |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2050 | if (f.file == tf.file || !is_file_epoll(f.file)) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2051 | goto error_tgt_fput; |
| 2052 | |
| 2053 | /* |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 2054 | * epoll adds to the wakeup queue at EPOLL_CTL_ADD time only, |
| 2055 | * so EPOLLEXCLUSIVE is not allowed for a EPOLL_CTL_MOD operation. |
| 2056 | * Also, we do not currently supported nested exclusive wakeups. |
| 2057 | */ |
Cyrill Gorcunov | c857ab6 | 2017-02-27 14:27:40 -0800 | [diff] [blame] | 2058 | if (ep_op_has_event(op) && (epds.events & EPOLLEXCLUSIVE)) { |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 2059 | if (op == EPOLL_CTL_MOD) |
| 2060 | goto error_tgt_fput; |
| 2061 | if (op == EPOLL_CTL_ADD && (is_file_epoll(tf.file) || |
| 2062 | (epds.events & ~EPOLLEXCLUSIVE_OK_BITS))) |
| 2063 | goto error_tgt_fput; |
| 2064 | } |
Jason Baron | df0108c | 2016-01-20 14:59:24 -0800 | [diff] [blame] | 2065 | |
| 2066 | /* |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2067 | * At this point it is safe to assume that the "private_data" contains |
| 2068 | * our own data structure. |
| 2069 | */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2070 | ep = f.file->private_data; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2071 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2072 | /* |
| 2073 | * When we insert an epoll file descriptor, inside another epoll file |
| 2074 | * descriptor, there is the change of creating closed loops, which are |
Jason Baron | 28d82dc | 2012-01-12 17:17:43 -0800 | [diff] [blame] | 2075 | * better be handled here, than in more critical paths. While we are |
| 2076 | * checking for loops we also determine the list of files reachable |
| 2077 | * and hang them on the tfile_check_list, so we can check that we |
| 2078 | * haven't created too many possible wakeup paths. |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2079 | * |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2080 | * We do not need to take the global 'epumutex' on EPOLL_CTL_ADD when |
| 2081 | * the epoll file descriptor is attaching directly to a wakeup source, |
| 2082 | * unless the epoll file descriptor is nested. The purpose of taking the |
| 2083 | * 'epmutex' on add is to prevent complex toplogies such as loops and |
| 2084 | * deep wakeup paths from forming in parallel through multiple |
| 2085 | * EPOLL_CTL_ADD operations. |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2086 | */ |
Nelson Elhage | d8805e6 | 2011-10-31 17:13:14 -0700 | [diff] [blame] | 2087 | mutex_lock_nested(&ep->mtx, 0); |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2088 | if (op == EPOLL_CTL_ADD) { |
| 2089 | if (!list_empty(&f.file->f_ep_links) || |
| 2090 | is_file_epoll(tf.file)) { |
| 2091 | full_check = 1; |
| 2092 | mutex_unlock(&ep->mtx); |
| 2093 | mutex_lock(&epmutex); |
| 2094 | if (is_file_epoll(tf.file)) { |
| 2095 | error = -ELOOP; |
| 2096 | if (ep_loop_check(ep, tf.file) != 0) { |
| 2097 | clear_tfile_check_list(); |
| 2098 | goto error_tgt_fput; |
| 2099 | } |
| 2100 | } else |
| 2101 | list_add(&tf.file->f_tfile_llink, |
| 2102 | &tfile_check_list); |
| 2103 | mutex_lock_nested(&ep->mtx, 0); |
| 2104 | if (is_file_epoll(tf.file)) { |
| 2105 | tep = tf.file->private_data; |
| 2106 | mutex_lock_nested(&tep->mtx, 1); |
| 2107 | } |
| 2108 | } |
| 2109 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2110 | |
Davide Libenzi | 67647d0 | 2007-05-15 01:40:52 -0700 | [diff] [blame] | 2111 | /* |
| 2112 | * Try to lookup the file inside our RB tree, Since we grabbed "mtx" |
| 2113 | * above, we can be sure to be able to use the item looked up by |
| 2114 | * ep_find() till we release the mutex. |
| 2115 | */ |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2116 | epi = ep_find(ep, tf.file, fd); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2117 | |
| 2118 | error = -EINVAL; |
| 2119 | switch (op) { |
| 2120 | case EPOLL_CTL_ADD: |
| 2121 | if (!epi) { |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 2122 | epds.events |= EPOLLERR | EPOLLHUP; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2123 | error = ep_insert(ep, &epds, tf.file, fd, full_check); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2124 | } else |
| 2125 | error = -EEXIST; |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2126 | if (full_check) |
| 2127 | clear_tfile_check_list(); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2128 | break; |
| 2129 | case EPOLL_CTL_DEL: |
| 2130 | if (epi) |
| 2131 | error = ep_remove(ep, epi); |
| 2132 | else |
| 2133 | error = -ENOENT; |
| 2134 | break; |
| 2135 | case EPOLL_CTL_MOD: |
| 2136 | if (epi) { |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 2137 | if (!(epi->event.events & EPOLLEXCLUSIVE)) { |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 2138 | epds.events |= EPOLLERR | EPOLLHUP; |
Jason Baron | b6a515c | 2016-02-05 15:37:04 -0800 | [diff] [blame] | 2139 | error = ep_modify(ep, epi, &epds); |
| 2140 | } |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2141 | } else |
| 2142 | error = -ENOENT; |
| 2143 | break; |
| 2144 | } |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2145 | if (tep != NULL) |
| 2146 | mutex_unlock(&tep->mtx); |
Davide Libenzi | d47de16 | 2007-05-15 01:40:41 -0700 | [diff] [blame] | 2147 | mutex_unlock(&ep->mtx); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2148 | |
| 2149 | error_tgt_fput: |
Jason Baron | 67347fe | 2013-11-12 15:10:18 -0800 | [diff] [blame] | 2150 | if (full_check) |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2151 | mutex_unlock(&epmutex); |
| 2152 | |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2153 | fdput(tf); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2154 | error_fput: |
Al Viro | 7e3fb58 | 2013-08-30 12:47:21 -0400 | [diff] [blame] | 2155 | fdput(f); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2156 | error_return: |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2157 | |
| 2158 | return error; |
| 2159 | } |
| 2160 | |
| 2161 | /* |
| 2162 | * Implement the event wait interface for the eventpoll file. It is the kernel |
| 2163 | * part of the user space epoll_wait(2). |
| 2164 | */ |
Dominik Brodowski | 791eb22 | 2018-03-11 11:34:30 +0100 | [diff] [blame] | 2165 | static int do_epoll_wait(int epfd, struct epoll_event __user *events, |
| 2166 | int maxevents, int timeout) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2167 | { |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2168 | int error; |
| 2169 | struct fd f; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2170 | struct eventpoll *ep; |
| 2171 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2172 | /* The maximum number of event must be greater than zero */ |
| 2173 | if (maxevents <= 0 || maxevents > EP_MAX_EVENTS) |
| 2174 | return -EINVAL; |
| 2175 | |
| 2176 | /* Verify that the area passed by the user is writeable */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2177 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) |
| 2178 | return -EFAULT; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2179 | |
| 2180 | /* Get the "struct file *" for the eventpoll file */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2181 | f = fdget(epfd); |
| 2182 | if (!f.file) |
| 2183 | return -EBADF; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2184 | |
| 2185 | /* |
| 2186 | * We have to check that the file structure underneath the fd |
| 2187 | * the user passed to us _is_ an eventpoll file. |
| 2188 | */ |
| 2189 | error = -EINVAL; |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2190 | if (!is_file_epoll(f.file)) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2191 | goto error_fput; |
| 2192 | |
| 2193 | /* |
| 2194 | * At this point it is safe to assume that the "private_data" contains |
| 2195 | * our own data structure. |
| 2196 | */ |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2197 | ep = f.file->private_data; |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2198 | |
| 2199 | /* Time to fish for events ... */ |
| 2200 | error = ep_poll(ep, events, maxevents, timeout); |
| 2201 | |
| 2202 | error_fput: |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 2203 | fdput(f); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2204 | return error; |
| 2205 | } |
| 2206 | |
Dominik Brodowski | 791eb22 | 2018-03-11 11:34:30 +0100 | [diff] [blame] | 2207 | SYSCALL_DEFINE4(epoll_wait, int, epfd, struct epoll_event __user *, events, |
| 2208 | int, maxevents, int, timeout) |
| 2209 | { |
| 2210 | return do_epoll_wait(epfd, events, maxevents, timeout); |
| 2211 | } |
| 2212 | |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2213 | /* |
| 2214 | * Implement the event wait interface for the eventpoll file. It is the kernel |
| 2215 | * part of the user space epoll_pwait(2). |
| 2216 | */ |
Heiko Carstens | 5a8a82b | 2009-01-14 14:14:25 +0100 | [diff] [blame] | 2217 | SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct epoll_event __user *, events, |
| 2218 | int, maxevents, int, timeout, const sigset_t __user *, sigmask, |
| 2219 | size_t, sigsetsize) |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2220 | { |
| 2221 | int error; |
| 2222 | sigset_t ksigmask, sigsaved; |
| 2223 | |
| 2224 | /* |
| 2225 | * If the caller wants a certain signal mask to be set during the wait, |
| 2226 | * we apply it here. |
| 2227 | */ |
| 2228 | if (sigmask) { |
| 2229 | if (sigsetsize != sizeof(sigset_t)) |
| 2230 | return -EINVAL; |
| 2231 | if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) |
| 2232 | return -EFAULT; |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2233 | sigsaved = current->blocked; |
| 2234 | set_current_blocked(&ksigmask); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2235 | } |
| 2236 | |
Dominik Brodowski | 791eb22 | 2018-03-11 11:34:30 +0100 | [diff] [blame] | 2237 | error = do_epoll_wait(epfd, events, maxevents, timeout); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2238 | |
| 2239 | /* |
| 2240 | * If we changed the signal mask, we need to restore the original one. |
| 2241 | * In case we've got a signal while waiting, we do not restore the |
| 2242 | * signal mask yet, and we allow do_signal() to deliver the signal on |
| 2243 | * the way back to userspace, before the signal mask is restored. |
| 2244 | */ |
| 2245 | if (sigmask) { |
| 2246 | if (error == -EINTR) { |
| 2247 | memcpy(¤t->saved_sigmask, &sigsaved, |
Davide Libenzi | c7ea763 | 2007-05-15 01:40:47 -0700 | [diff] [blame] | 2248 | sizeof(sigsaved)); |
Roland McGrath | 4e4c22c | 2008-04-30 00:53:06 -0700 | [diff] [blame] | 2249 | set_restore_sigmask(); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2250 | } else |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2251 | set_current_blocked(&sigsaved); |
Davide Libenzi | 7699acd | 2007-05-10 22:23:23 -0700 | [diff] [blame] | 2252 | } |
| 2253 | |
| 2254 | return error; |
| 2255 | } |
| 2256 | |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2257 | #ifdef CONFIG_COMPAT |
| 2258 | COMPAT_SYSCALL_DEFINE6(epoll_pwait, int, epfd, |
| 2259 | struct epoll_event __user *, events, |
| 2260 | int, maxevents, int, timeout, |
| 2261 | const compat_sigset_t __user *, sigmask, |
| 2262 | compat_size_t, sigsetsize) |
| 2263 | { |
| 2264 | long err; |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2265 | sigset_t ksigmask, sigsaved; |
| 2266 | |
| 2267 | /* |
| 2268 | * If the caller wants a certain signal mask to be set during the wait, |
| 2269 | * we apply it here. |
| 2270 | */ |
| 2271 | if (sigmask) { |
| 2272 | if (sigsetsize != sizeof(compat_sigset_t)) |
| 2273 | return -EINVAL; |
Al Viro | 3968cf6 | 2017-09-03 21:45:17 -0400 | [diff] [blame] | 2274 | if (get_compat_sigset(&ksigmask, sigmask)) |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2275 | return -EFAULT; |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2276 | sigsaved = current->blocked; |
| 2277 | set_current_blocked(&ksigmask); |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2278 | } |
| 2279 | |
Dominik Brodowski | 791eb22 | 2018-03-11 11:34:30 +0100 | [diff] [blame] | 2280 | err = do_epoll_wait(epfd, events, maxevents, timeout); |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2281 | |
| 2282 | /* |
| 2283 | * If we changed the signal mask, we need to restore the original one. |
| 2284 | * In case we've got a signal while waiting, we do not restore the |
| 2285 | * signal mask yet, and we allow do_signal() to deliver the signal on |
| 2286 | * the way back to userspace, before the signal mask is restored. |
| 2287 | */ |
| 2288 | if (sigmask) { |
| 2289 | if (err == -EINTR) { |
| 2290 | memcpy(¤t->saved_sigmask, &sigsaved, |
| 2291 | sizeof(sigsaved)); |
| 2292 | set_restore_sigmask(); |
| 2293 | } else |
Oleg Nesterov | 77d5591 | 2013-07-03 15:08:14 -0700 | [diff] [blame] | 2294 | set_current_blocked(&sigsaved); |
Al Viro | 35280bd | 2013-02-24 14:52:17 -0500 | [diff] [blame] | 2295 | } |
| 2296 | |
| 2297 | return err; |
| 2298 | } |
| 2299 | #endif |
| 2300 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2301 | static int __init eventpoll_init(void) |
| 2302 | { |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 2303 | struct sysinfo si; |
| 2304 | |
| 2305 | si_meminfo(&si); |
Davide Libenzi | 9df04e1 | 2009-01-29 14:25:26 -0800 | [diff] [blame] | 2306 | /* |
| 2307 | * Allows top 4% of lomem to be allocated for epoll watches (per user). |
| 2308 | */ |
| 2309 | max_user_watches = (((si.totalram - si.totalhigh) / 25) << PAGE_SHIFT) / |
Davide Libenzi | 7ef9964 | 2008-12-01 13:13:55 -0800 | [diff] [blame] | 2310 | EP_ITEM_COST; |
Robin Holt | 52bd19f7 | 2011-01-12 17:00:01 -0800 | [diff] [blame] | 2311 | BUG_ON(max_user_watches < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | |
Davide Libenzi | 22bacca | 2011-02-25 14:44:12 -0800 | [diff] [blame] | 2313 | /* |
| 2314 | * Initialize the structure used to perform epoll file descriptor |
| 2315 | * inclusion loops checks. |
| 2316 | */ |
| 2317 | ep_nested_calls_init(&poll_loop_ncalls); |
| 2318 | |
Jason Baron | 57a173b | 2017-11-17 15:29:02 -0800 | [diff] [blame] | 2319 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2320 | /* Initialize the structure used to perform safe poll wait head wake ups */ |
Davide Libenzi | 5071f97 | 2009-03-31 15:24:10 -0700 | [diff] [blame] | 2321 | ep_nested_calls_init(&poll_safewake_ncalls); |
Jason Baron | 57a173b | 2017-11-17 15:29:02 -0800 | [diff] [blame] | 2322 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2323 | |
Eric Wong | 39732ca | 2013-04-30 15:27:38 -0700 | [diff] [blame] | 2324 | /* |
| 2325 | * We can have many thousands of epitems, so prevent this from |
| 2326 | * using an extra cache line on 64-bit (and smaller) CPUs |
| 2327 | */ |
| 2328 | BUILD_BUG_ON(sizeof(void *) <= 8 && sizeof(struct epitem) > 128); |
| 2329 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2330 | /* Allocates slab cache used to allocate "struct epitem" items */ |
| 2331 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), |
Shakeel Butt | 2ae928a | 2017-11-17 15:28:59 -0800 | [diff] [blame] | 2332 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2333 | |
| 2334 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ |
| 2335 | pwq_cache = kmem_cache_create("eventpoll_pwq", |
Shakeel Butt | 2ae928a | 2017-11-17 15:28:59 -0800 | [diff] [blame] | 2336 | sizeof(struct eppoll_entry), 0, SLAB_PANIC|SLAB_ACCOUNT, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2337 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2338 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2339 | } |
Davide Libenzi | cea6924 | 2007-05-10 22:23:22 -0700 | [diff] [blame] | 2340 | fs_initcall(eventpoll_init); |