source: trunk/kernel/mm/mapper.h @ 313

Last change on this file since 313 was 313, checked in by alain, 5 years ago

RSeveral modifs in the page-fault handling.

File size: 10.8 KB
Line 
1/*
2 * mapper.h - Map memory, file or device in process virtual address space.
3 *
4 * Authors   Mohamed Lamine Karaoui (2015)
5 *           Alain Greiner (2016)
6 *
7 * Copyright (c)  UPMC Sorbonne Universites
8 *
9 * This file is part of ALMOS-MKH.
10 *
11 * ALMOS-MKH is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; version 2.0 of the License.
14 *
15 * ALMOS-MKH is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with ALMOS-MKH; if not, write to the Free Software Foundation,
22 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
25#ifndef _MAPPER_H_
26#define _MAPPER_H_
27
28#include <hal_types.h>
29#include <hal_atomic.h>
30#include <xlist.h>
31#include <grdxt.h>
32#include <rwlock.h>
33
34/****  Forward declarations ****/
35
36struct page_s;
37struct vfs_inode_s;
38
39/*******************************************************************************************
40 * The mapper implements the kernel cache for a given file or directory.
41 * There is one mapper per file/dir. It is implemented as a three levels radix tree,
42 * entirely stored in the same cluster as the inode representing the file/dir.
43 * - The fast retrieval key is the page index in the file.
44 *   The ix1_width, ix2_width, ix3_width sub-indexes are configuration parameters.
45 * - The leaves are pointers on physical page descriptors, dynamically allocated
46 *   in the local cluster.
47 * - In a given cluster, a mapper is a "private" structure: a thread accessing the mapper
48 *   must be running in the cluster containing it (can be a local thread or a RPC thread).
49 * - The mapper is protected by a blocking "rwlock", to support several simultaneous
50 *   readers, and only one writer. This lock implement a busy waiting policy.
51 * - The mapper_get_page() function that return a page descriptor pointer from a page
52 *   index in file is in charge of handling the miss on the mapper cache.
53 * - The vfs_mapper_move_page() function access the file system to handle a mapper miss,
54 *   or update a dirty page on device.
55 * - The vfs_mapper_load_all() functions is used to load all pages of a given file
56 *   or directory into the mapper.
57 * - the mapper_move_user() function is used to move data to or from an user buffer.
58 *   This user space buffer can be physically distributed in several clusters.
59 * - the mapper_move_kernel() function is used to move data to or from a remote kernel
60 *   buffer, that can be physically located in any cluster.
61 * - In the present implementation the cache size for a given file increases on demand,
62 *   and the  allocated memory is only released when the mapper/inode is destroyed.
63 ******************************************************************************************/
64
65
66/*******************************************************************************************
67 * This structure defines the mapper descriptor.
68 ******************************************************************************************/
69
70typedef struct mapper_s
71{
72        struct vfs_inode_s * inode;           /*! owner inode                                     */
73    uint32_t             type;        /*! file system type                                */
74        grdxt_t              radix;           /*! pages cache implemented as a radix tree         */
75        rwlock_t             lock;        /*! several readers / only one writer               */
76        uint32_t                 refcount;    /*! several vsegs can refer the same file           */
77        xlist_entry_t        vsegs_root;  /*! root of list of vsegs refering this mapper      */
78        xlist_entry_t        wait_root;   /*! root of list of threads waiting on mapper       */
79    list_entry_t         dirty_root;  /*! root of list of dirty pages                     */
80}
81mapper_t;
82
83/*******************************************************************************************
84 * This structure defines a "fragment". It is used to move data between the kernel mapper,
85 * and an user buffer, that can be split in several distributed physical pages located
86 * in different clusters. A fragment is a set of contiguous bytes in the file.
87 * - It can be stored in one single physical page in the user buffer.
88 * - It can spread two successive physical pages in the kernel mapper.
89 ******************************************************************************************/
90
91typedef struct fragment_s
92{
93    uint32_t    file_offset;         /*! offset of fragment in file (i.e. in mapper)      */
94    uint32_t    size;                /*! number of bytes in fragment                      */
95    cxy_t       buf_cxy;             /*! user buffer cluster identifier                   */
96    void      * buf_ptr;             /*! local pointer on first byte in user buffer       */
97}
98fragment_t;
99
100/*******************************************************************************************
101 * This function allocates physical memory for a mapper descriptor, and initializes it
102 * (refcount <= 0) / inode <= NULL).
103 * It must be executed by a thread running in the cluster containing the mapper.
104 *******************************************************************************************
105 * @ return pointer on created mapper if success / return NULL if no memory
106 ******************************************************************************************/
107mapper_t * mapper_create();
108
109/*******************************************************************************************
110 * This function releases all physical pages allocated for the mapper.
111 * It synchronizes all dirty pages (i.e. update the file on disk) if required.
112 * The mapper descriptor and the radix tree themselves are released.
113 * It must be executed by a thread running in the cluster containing the mapper.
114 *******************************************************************************************
115 * @ mapper      : target mapper.
116 * @ return 0 if success / return EIO if a dirty page cannot be updated on device.
117 ******************************************************************************************/
118error_t mapper_destroy( mapper_t * mapper );
119
120/*******************************************************************************************
121 * This function move data between a mapper and a - possibly distributed - user buffer.
122 * It must be called by a thread running in the cluster containing the mapper.
123 * It is called by the vfs_user_move() function to implement sys_read() and sys_write().
124 * If required, the data transfer is split in "fragments", where one fragment contains
125 * contiguous bytes in the same mapper page.
126 * It uses "hal_uspace" accesses to move a fragment to/from the user buffer.
127 * In case of write, the dirty bit is set for all pages written in the mapper.
128 * The offset in the file descriptor is not modified by this function.
129 *******************************************************************************************
130 * @ mapper       : local pointer on mapper.
131 * @ to_buffer    : mapper -> buffer if true / buffer -> mapper if false.
132 * @ is_user      : user space buffer if true / kernel local buffer if false.
133 * @ file_offset  : first byte to move in file.
134 * @ buffer       : user space pointer on user buffer.
135 * @ size         : number of bytes to move.
136 * returns O if success / returns EINVAL if error.
137 ******************************************************************************************/
138error_t mapper_move_user( mapper_t * mapper,
139                          bool_t     to_buffer,
140                          bool_t     is_user,
141                          uint32_t   file_offset,
142                          void     * buffer,
143                          uint32_t   size );
144
145/*******************************************************************************************
146 * This function move data between a mapper and a remote kernel buffer.
147 * It must be called by a thread running in the cluster containing the mapper.
148 * If required, the data transfer is split in "fragments", where one fragment contains
149 * contiguous bytes in the same mapper page.
150 * It uses a "remote_memcpy" to move a fragment to/from the kernel buffer.
151 * In case of write, the dirty bit is set for all pages written in the mapper.
152 * The offset in the file descriptor is not modified by this function.
153 *******************************************************************************************
154 * @ mapper       : local pointer on mapper.
155 * @ to_buffer    : mapper -> buffer if true / buffer -> mapper if false.
156 * @ file_offset  : first byte to move in file.
157 * @ buffer_xp    : extended pointer on kernel buffer.
158 * @ size         : number of bytes to move.
159 * returns O if success / returns EINVAL if error.
160 ******************************************************************************************/
161error_t mapper_move_kernel( mapper_t * mapper,
162                            bool_t     to_buffer,
163                            uint32_t   file_offset,
164                            xptr_t     buffer_xp,
165                            uint32_t   size );
166
167
168/*******************************************************************************************
169 * This function removes a physical page from the mapper, update the FS if the page
170 * is dirty, and releases the page to PPM. It is called by the mapper_destroy() function.
171 * It must be executed by a thread running in the cluster containing the mapper.
172 * It takes both the page lock and the mapper lock in WRITE_MODE to release the page.
173 *******************************************************************************************
174 * @ mapper     : local pointer on the mapper.
175 * @ page       : pointer on page to remove.
176 * @ return 0 if success / return EIO if a dirty page cannot be copied to FS.
177 ******************************************************************************************/
178error_t mapper_release_page( mapper_t      * mapper,
179                             struct page_s * page );
180
181/*******************************************************************************************
182 * This function searches a physical page descriptor from its index in mapper.
183 * It must be executed by a thread running in the cluster containing the mapper.
184 * In case of miss, it takes the mapper lock in WRITE_MODE, load the missing
185 * page from device to the mapper, and release the mapper lock.
186 *******************************************************************************************
187 * @ mapper     : local pointer on the mapper.
188 * @ index      : page index in file
189 * @ returns pointer on page descriptor if success / return NULL if error.
190 ******************************************************************************************/
191struct page_s * mapper_get_page( mapper_t * mapper,
192                                 uint32_t   index );
193
194 
195
196#endif /* _MAPPER_H_ */
Note: See TracBrowser for help on using the repository browser.