blob: aea219d7d8d1afec462b546df8c3412c915eecf2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef __LINUX__AIO_H
2#define __LINUX__AIO_H
3
4#include <linux/list.h>
5#include <linux/workqueue.h>
6#include <linux/aio_abi.h>
Badari Pulavarty027445c2006-09-30 23:28:46 -07007#include <linux/uio.h>
Jens Axboeabf137d2008-12-09 08:11:22 +01008#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009
10#include <asm/atomic.h>
11
12#define AIO_MAXSEGS 4
13#define AIO_KIOGRP_NR_ATOMIC 8
14
15struct kioctx;
16
17/* Notes on cancelling a kiocb:
18 * If a kiocb is cancelled, aio_complete may return 0 to indicate
19 * that cancel has not yet disposed of the kiocb. All cancel
20 * operations *must* call aio_put_req to dispose of the kiocb
21 * to guard against races with the completion code.
22 */
23#define KIOCB_C_CANCELLED 0x01
24#define KIOCB_C_COMPLETE 0x02
25
26#define KIOCB_SYNC_KEY (~0U)
27
28/* ki_flags bits */
Zach Brown4faa5282005-10-17 16:43:33 -070029/*
30 * This may be used for cancel/retry serialization in the future, but
31 * for now it's unused and we probably don't want modules to even
32 * think they can use it.
33 */
34/* #define KIF_LOCKED 0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#define KIF_KICKED 1
36#define KIF_CANCELLED 2
37
38#define kiocbTryLock(iocb) test_and_set_bit(KIF_LOCKED, &(iocb)->ki_flags)
39#define kiocbTryKick(iocb) test_and_set_bit(KIF_KICKED, &(iocb)->ki_flags)
40
41#define kiocbSetLocked(iocb) set_bit(KIF_LOCKED, &(iocb)->ki_flags)
42#define kiocbSetKicked(iocb) set_bit(KIF_KICKED, &(iocb)->ki_flags)
43#define kiocbSetCancelled(iocb) set_bit(KIF_CANCELLED, &(iocb)->ki_flags)
44
45#define kiocbClearLocked(iocb) clear_bit(KIF_LOCKED, &(iocb)->ki_flags)
46#define kiocbClearKicked(iocb) clear_bit(KIF_KICKED, &(iocb)->ki_flags)
47#define kiocbClearCancelled(iocb) clear_bit(KIF_CANCELLED, &(iocb)->ki_flags)
48
49#define kiocbIsLocked(iocb) test_bit(KIF_LOCKED, &(iocb)->ki_flags)
50#define kiocbIsKicked(iocb) test_bit(KIF_KICKED, &(iocb)->ki_flags)
51#define kiocbIsCancelled(iocb) test_bit(KIF_CANCELLED, &(iocb)->ki_flags)
52
Zach Brown897f15f2005-09-30 11:58:55 -070053/* is there a better place to document function pointer methods? */
54/**
55 * ki_retry - iocb forward progress callback
56 * @kiocb: The kiocb struct to advance by performing an operation.
57 *
58 * This callback is called when the AIO core wants a given AIO operation
59 * to make forward progress. The kiocb argument describes the operation
60 * that is to be performed. As the operation proceeds, perhaps partially,
61 * ki_retry is expected to update the kiocb with progress made. Typically
62 * ki_retry is set in the AIO core and it itself calls file_operations
63 * helpers.
64 *
65 * ki_retry's return value determines when the AIO operation is completed
66 * and an event is generated in the AIO event ring. Except the special
67 * return values described below, the value that is returned from ki_retry
68 * is transferred directly into the completion ring as the operation's
69 * resulting status. Once this has happened ki_retry *MUST NOT* reference
70 * the kiocb pointer again.
71 *
72 * If ki_retry returns -EIOCBQUEUED it has made a promise that aio_complete()
73 * will be called on the kiocb pointer in the future. The AIO core will
74 * not ask the method again -- ki_retry must ensure forward progress.
75 * aio_complete() must be called once and only once in the future, multiple
76 * calls may result in undefined behaviour.
77 *
78 * If ki_retry returns -EIOCBRETRY it has made a promise that kick_iocb()
79 * will be called on the kiocb pointer in the future. This may happen
80 * through generic helpers that associate kiocb->ki_wait with a wait
81 * queue head that ki_retry uses via current->io_wait. It can also happen
82 * with custom tracking and manual calls to kick_iocb(), though that is
83 * discouraged. In either case, kick_iocb() must be called once and only
84 * once. ki_retry must ensure forward progress, the AIO core will wait
85 * indefinitely for kick_iocb() to be called.
86 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070087struct kiocb {
88 struct list_head ki_run_list;
David Brownell2ba2d002007-07-19 01:47:55 -070089 unsigned long ki_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 int ki_users;
91 unsigned ki_key; /* id of this request */
92
93 struct file *ki_filp;
94 struct kioctx *ki_ctx; /* may be NULL for sync ops */
95 int (*ki_cancel)(struct kiocb *, struct io_event *);
96 ssize_t (*ki_retry)(struct kiocb *);
97 void (*ki_dtor)(struct kiocb *);
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 union {
100 void __user *user;
101 struct task_struct *tsk;
102 } ki_obj;
Benjamin LaHaise59d91362006-01-08 01:04:34 -0800103
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 __u64 ki_user_data; /* user's data for completion */
Benjamin LaHaise59d91362006-01-08 01:04:34 -0800105 wait_queue_t ki_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 loff_t ki_pos;
Benjamin LaHaise59d91362006-01-08 01:04:34 -0800107
108 void *private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 /* State that we remember to be able to restart/retry */
110 unsigned short ki_opcode;
111 size_t ki_nbytes; /* copy of iocb->aio_nbytes */
112 char __user *ki_buf; /* remaining iocb->aio_buf */
113 size_t ki_left; /* remaining bytes */
Badari Pulavarty027445c2006-09-30 23:28:46 -0700114 struct iovec ki_inline_vec; /* inline vector */
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700115 struct iovec *ki_iovec;
116 unsigned long ki_nr_segs;
117 unsigned long ki_cur_seg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
Benjamin LaHaise59d91362006-01-08 01:04:34 -0800119 struct list_head ki_list; /* the aio core uses this
120 * for cancellation */
Davide Libenzi9c3060b2007-05-10 22:23:21 -0700121
122 /*
123 * If the aio_resfd field of the userspace iocb is not zero,
Davide Libenzi13389012009-06-30 11:41:11 -0700124 * this is the underlying eventfd context to deliver events to.
Davide Libenzi9c3060b2007-05-10 22:23:21 -0700125 */
Davide Libenzi13389012009-06-30 11:41:11 -0700126 struct eventfd_ctx *ki_eventfd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127};
128
129#define is_sync_kiocb(iocb) ((iocb)->ki_key == KIOCB_SYNC_KEY)
130#define init_sync_kiocb(x, filp) \
131 do { \
132 struct task_struct *tsk = current; \
133 (x)->ki_flags = 0; \
134 (x)->ki_users = 1; \
135 (x)->ki_key = KIOCB_SYNC_KEY; \
136 (x)->ki_filp = (filp); \
Zach Brown20dcae32005-11-13 16:07:33 -0800137 (x)->ki_ctx = NULL; \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 (x)->ki_cancel = NULL; \
Benjamin LaHaise59d91362006-01-08 01:04:34 -0800139 (x)->ki_retry = NULL; \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 (x)->ki_dtor = NULL; \
141 (x)->ki_obj.tsk = tsk; \
142 (x)->ki_user_data = 0; \
143 init_wait((&(x)->ki_wait)); \
144 } while (0)
145
146#define AIO_RING_MAGIC 0xa10a10a1
147#define AIO_RING_COMPAT_FEATURES 1
148#define AIO_RING_INCOMPAT_FEATURES 0
149struct aio_ring {
150 unsigned id; /* kernel internal index number */
151 unsigned nr; /* number of io_events */
152 unsigned head;
153 unsigned tail;
154
155 unsigned magic;
156 unsigned compat_features;
157 unsigned incompat_features;
158 unsigned header_length; /* size of aio_ring */
159
160
161 struct io_event io_events[0];
162}; /* 128 bytes + ring size */
163
164#define aio_ring_avail(info, ring) (((ring)->head + (info)->nr - 1 - (ring)->tail) % (info)->nr)
165
166#define AIO_RING_PAGES 8
167struct aio_ring_info {
168 unsigned long mmap_base;
169 unsigned long mmap_size;
170
171 struct page **ring_pages;
172 spinlock_t ring_lock;
173 long nr_pages;
174
175 unsigned nr, tail;
176
177 struct page *internal_pages[AIO_RING_PAGES];
178};
179
180struct kioctx {
181 atomic_t users;
182 int dead;
183 struct mm_struct *mm;
184
185 /* This needs improving */
186 unsigned long user_id;
Jens Axboeabf137d2008-12-09 08:11:22 +0100187 struct hlist_node list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 wait_queue_head_t wait;
190
191 spinlock_t ctx_lock;
192
193 int reqs_active;
194 struct list_head active_reqs; /* used for cancellation */
195 struct list_head run_list; /* used for kicked reqs */
196
Zach Brownd55b5fd2005-11-07 00:59:31 -0800197 /* sys_io_setup currently limits this to an unsigned int */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 unsigned max_reqs;
199
200 struct aio_ring_info ring_info;
201
David Howells52bad642006-11-22 14:54:01 +0000202 struct delayed_work wq;
Jens Axboeabf137d2008-12-09 08:11:22 +0100203
204 struct rcu_head rcu_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205};
206
207/* prototypes */
208extern unsigned aio_max_size;
209
Thomas Petazzoniebf3f092008-10-15 22:05:12 -0700210#ifdef CONFIG_AIO
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800211extern ssize_t wait_on_sync_kiocb(struct kiocb *iocb);
212extern int aio_put_req(struct kiocb *iocb);
213extern void kick_iocb(struct kiocb *iocb);
214extern int aio_complete(struct kiocb *iocb, long res, long res2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215struct mm_struct;
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800216extern void exit_aio(struct mm_struct *mm);
Thomas Petazzoniebf3f092008-10-15 22:05:12 -0700217#else
218static inline ssize_t wait_on_sync_kiocb(struct kiocb *iocb) { return 0; }
219static inline int aio_put_req(struct kiocb *iocb) { return 0; }
220static inline void kick_iocb(struct kiocb *iocb) { }
221static inline int aio_complete(struct kiocb *iocb, long res, long res2) { return 0; }
222struct mm_struct;
223static inline void exit_aio(struct mm_struct *mm) { }
224#endif /* CONFIG_AIO */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226#define io_wait_to_kiocb(wait) container_of(wait, struct kiocb, ki_wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228static inline struct kiocb *list_kiocb(struct list_head *h)
229{
230 return list_entry(h, struct kiocb, ki_list);
231}
232
233/* for sysctl: */
Zach Brownd55b5fd2005-11-07 00:59:31 -0800234extern unsigned long aio_nr;
235extern unsigned long aio_max_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
237#endif /* __LINUX__AIO_H */