1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * Copyright (c) 2016 Trond Myklebust 3 * Copyright (c) 2016 Trond Myklebust
4 * Copyright (c) 2019 Jeff Layton 4 * Copyright (c) 2019 Jeff Layton
5 * 5 *
6 * I/O and data path helper functionality. 6 * I/O and data path helper functionality.
7 * 7 *
8 * Heavily borrowed from equivalent code in fs 8 * Heavily borrowed from equivalent code in fs/nfs/io.c
9 */ 9 */
10 10
11 #include <linux/ceph/ceph_debug.h> 11 #include <linux/ceph/ceph_debug.h>
12 12
13 #include <linux/types.h> 13 #include <linux/types.h>
14 #include <linux/kernel.h> 14 #include <linux/kernel.h>
15 #include <linux/rwsem.h> 15 #include <linux/rwsem.h>
16 #include <linux/fs.h> 16 #include <linux/fs.h>
17 17
18 #include "super.h" 18 #include "super.h"
19 #include "io.h" 19 #include "io.h"
20 20
21 /* Call with exclusively locked inode->i_rwsem 21 /* Call with exclusively locked inode->i_rwsem */
22 static void ceph_block_o_direct(struct ceph_in 22 static void ceph_block_o_direct(struct ceph_inode_info *ci, struct inode *inode)
23 { 23 {
24 lockdep_assert_held_write(&inode->i_rw 24 lockdep_assert_held_write(&inode->i_rwsem);
25 25
26 if (READ_ONCE(ci->i_ceph_flags) & CEPH 26 if (READ_ONCE(ci->i_ceph_flags) & CEPH_I_ODIRECT) {
27 spin_lock(&ci->i_ceph_lock); 27 spin_lock(&ci->i_ceph_lock);
28 ci->i_ceph_flags &= ~CEPH_I_OD 28 ci->i_ceph_flags &= ~CEPH_I_ODIRECT;
29 spin_unlock(&ci->i_ceph_lock); 29 spin_unlock(&ci->i_ceph_lock);
30 inode_dio_wait(inode); 30 inode_dio_wait(inode);
31 } 31 }
32 } 32 }
33 33
34 /** 34 /**
35 * ceph_start_io_read - declare the file is be 35 * ceph_start_io_read - declare the file is being used for buffered reads
36 * @inode: file inode 36 * @inode: file inode
37 * 37 *
38 * Declare that a buffered read operation is a 38 * Declare that a buffered read operation is about to start, and ensure
39 * that we block all direct I/O. 39 * that we block all direct I/O.
40 * On exit, the function ensures that the CEPH 40 * On exit, the function ensures that the CEPH_I_ODIRECT flag is unset,
41 * and holds a shared lock on inode->i_rwsem t 41 * and holds a shared lock on inode->i_rwsem to ensure that the flag
42 * cannot be changed. 42 * cannot be changed.
43 * In practice, this means that buffered read 43 * In practice, this means that buffered read operations are allowed to
44 * execute in parallel, thanks to the shared l 44 * execute in parallel, thanks to the shared lock, whereas direct I/O
45 * operations need to wait to grab an exclusiv 45 * operations need to wait to grab an exclusive lock in order to set
46 * CEPH_I_ODIRECT. 46 * CEPH_I_ODIRECT.
47 * Note that buffered writes and truncates bot 47 * Note that buffered writes and truncates both take a write lock on
48 * inode->i_rwsem, meaning that those are seri 48 * inode->i_rwsem, meaning that those are serialised w.r.t. the reads.
49 */ 49 */
50 void 50 void
51 ceph_start_io_read(struct inode *inode) 51 ceph_start_io_read(struct inode *inode)
52 { 52 {
53 struct ceph_inode_info *ci = ceph_inod 53 struct ceph_inode_info *ci = ceph_inode(inode);
54 54
55 /* Be an optimist! */ 55 /* Be an optimist! */
56 down_read(&inode->i_rwsem); 56 down_read(&inode->i_rwsem);
57 if (!(READ_ONCE(ci->i_ceph_flags) & CE 57 if (!(READ_ONCE(ci->i_ceph_flags) & CEPH_I_ODIRECT))
58 return; 58 return;
59 up_read(&inode->i_rwsem); 59 up_read(&inode->i_rwsem);
60 /* Slow path.... */ 60 /* Slow path.... */
61 down_write(&inode->i_rwsem); 61 down_write(&inode->i_rwsem);
62 ceph_block_o_direct(ci, inode); 62 ceph_block_o_direct(ci, inode);
63 downgrade_write(&inode->i_rwsem); 63 downgrade_write(&inode->i_rwsem);
64 } 64 }
65 65
66 /** 66 /**
67 * ceph_end_io_read - declare that the buffere 67 * ceph_end_io_read - declare that the buffered read operation is done
68 * @inode: file inode 68 * @inode: file inode
69 * 69 *
70 * Declare that a buffered read operation is d 70 * Declare that a buffered read operation is done, and release the shared
71 * lock on inode->i_rwsem. 71 * lock on inode->i_rwsem.
72 */ 72 */
73 void 73 void
74 ceph_end_io_read(struct inode *inode) 74 ceph_end_io_read(struct inode *inode)
75 { 75 {
76 up_read(&inode->i_rwsem); 76 up_read(&inode->i_rwsem);
77 } 77 }
78 78
79 /** 79 /**
80 * ceph_start_io_write - declare the file is b 80 * ceph_start_io_write - declare the file is being used for buffered writes
81 * @inode: file inode 81 * @inode: file inode
82 * 82 *
83 * Declare that a buffered write operation is 83 * Declare that a buffered write operation is about to start, and ensure
84 * that we block all direct I/O. 84 * that we block all direct I/O.
85 */ 85 */
86 void 86 void
87 ceph_start_io_write(struct inode *inode) 87 ceph_start_io_write(struct inode *inode)
88 { 88 {
89 down_write(&inode->i_rwsem); 89 down_write(&inode->i_rwsem);
90 ceph_block_o_direct(ceph_inode(inode), 90 ceph_block_o_direct(ceph_inode(inode), inode);
91 } 91 }
92 92
93 /** 93 /**
94 * ceph_end_io_write - declare that the buffer 94 * ceph_end_io_write - declare that the buffered write operation is done
95 * @inode: file inode 95 * @inode: file inode
96 * 96 *
97 * Declare that a buffered write operation is 97 * Declare that a buffered write operation is done, and release the
98 * lock on inode->i_rwsem. 98 * lock on inode->i_rwsem.
99 */ 99 */
100 void 100 void
101 ceph_end_io_write(struct inode *inode) 101 ceph_end_io_write(struct inode *inode)
102 { 102 {
103 up_write(&inode->i_rwsem); 103 up_write(&inode->i_rwsem);
104 } 104 }
105 105
106 /* Call with exclusively locked inode->i_rwsem 106 /* Call with exclusively locked inode->i_rwsem */
107 static void ceph_block_buffered(struct ceph_in 107 static void ceph_block_buffered(struct ceph_inode_info *ci, struct inode *inode)
108 { 108 {
109 lockdep_assert_held_write(&inode->i_rw 109 lockdep_assert_held_write(&inode->i_rwsem);
110 110
111 if (!(READ_ONCE(ci->i_ceph_flags) & CE 111 if (!(READ_ONCE(ci->i_ceph_flags) & CEPH_I_ODIRECT)) {
112 spin_lock(&ci->i_ceph_lock); 112 spin_lock(&ci->i_ceph_lock);
113 ci->i_ceph_flags |= CEPH_I_ODI 113 ci->i_ceph_flags |= CEPH_I_ODIRECT;
114 spin_unlock(&ci->i_ceph_lock); 114 spin_unlock(&ci->i_ceph_lock);
115 /* FIXME: unmap_mapping_range? 115 /* FIXME: unmap_mapping_range? */
116 filemap_write_and_wait(inode-> 116 filemap_write_and_wait(inode->i_mapping);
117 } 117 }
118 } 118 }
119 119
120 /** 120 /**
121 * ceph_start_io_direct - declare the file is !! 121 * ceph_end_io_direct - declare the file is being used for direct i/o
122 * @inode: file inode 122 * @inode: file inode
123 * 123 *
124 * Declare that a direct I/O operation is abou 124 * Declare that a direct I/O operation is about to start, and ensure
125 * that we block all buffered I/O. 125 * that we block all buffered I/O.
126 * On exit, the function ensures that the CEPH 126 * On exit, the function ensures that the CEPH_I_ODIRECT flag is set,
127 * and holds a shared lock on inode->i_rwsem t 127 * and holds a shared lock on inode->i_rwsem to ensure that the flag
128 * cannot be changed. 128 * cannot be changed.
129 * In practice, this means that direct I/O ope 129 * In practice, this means that direct I/O operations are allowed to
130 * execute in parallel, thanks to the shared l 130 * execute in parallel, thanks to the shared lock, whereas buffered I/O
131 * operations need to wait to grab an exclusiv 131 * operations need to wait to grab an exclusive lock in order to clear
132 * CEPH_I_ODIRECT. 132 * CEPH_I_ODIRECT.
133 * Note that buffered writes and truncates bot 133 * Note that buffered writes and truncates both take a write lock on
134 * inode->i_rwsem, meaning that those are seri 134 * inode->i_rwsem, meaning that those are serialised w.r.t. O_DIRECT.
135 */ 135 */
136 void 136 void
137 ceph_start_io_direct(struct inode *inode) 137 ceph_start_io_direct(struct inode *inode)
138 { 138 {
139 struct ceph_inode_info *ci = ceph_inod 139 struct ceph_inode_info *ci = ceph_inode(inode);
140 140
141 /* Be an optimist! */ 141 /* Be an optimist! */
142 down_read(&inode->i_rwsem); 142 down_read(&inode->i_rwsem);
143 if (READ_ONCE(ci->i_ceph_flags) & CEPH 143 if (READ_ONCE(ci->i_ceph_flags) & CEPH_I_ODIRECT)
144 return; 144 return;
145 up_read(&inode->i_rwsem); 145 up_read(&inode->i_rwsem);
146 /* Slow path.... */ 146 /* Slow path.... */
147 down_write(&inode->i_rwsem); 147 down_write(&inode->i_rwsem);
148 ceph_block_buffered(ci, inode); 148 ceph_block_buffered(ci, inode);
149 downgrade_write(&inode->i_rwsem); 149 downgrade_write(&inode->i_rwsem);
150 } 150 }
151 151
152 /** 152 /**
153 * ceph_end_io_direct - declare that the direc 153 * ceph_end_io_direct - declare that the direct i/o operation is done
154 * @inode: file inode 154 * @inode: file inode
155 * 155 *
156 * Declare that a direct I/O operation is done 156 * Declare that a direct I/O operation is done, and release the shared
157 * lock on inode->i_rwsem. 157 * lock on inode->i_rwsem.
158 */ 158 */
159 void 159 void
160 ceph_end_io_direct(struct inode *inode) 160 ceph_end_io_direct(struct inode *inode)
161 { 161 {
162 up_read(&inode->i_rwsem); 162 up_read(&inode->i_rwsem);
163 } 163 }
164 164
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.