blob: f05fa826f89ec4664099a05d92a4ae6abbea6b62 [file] [log] [blame]
Boaz Harrosh8ff660a2011-08-06 19:26:31 -07001/*
2 * Copyright (C) 2011
3 * Boaz Harrosh <bharrosh@panasas.com>
4 *
5 * Public Declarations of the ORE API
6 *
7 * This file is part of the ORE (Object Raid Engine) library.
8 *
9 * ORE is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as published
11 * by the Free Software Foundation. (GPL v2)
12 *
13 * ORE is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with the ORE; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22#ifndef __ORE_H__
23#define __ORE_H__
24
25#include <scsi/osd_initiator.h>
26#include <scsi/osd_attributes.h>
27#include <scsi/osd_sec.h>
28#include <linux/pnfs_osd_xdr.h>
29
30struct ore_comp {
31 struct osd_obj_id obj;
32 u8 cred[OSD_CAP_LEN];
33};
34
35struct ore_layout {
36 /* Our way of looking at the data_map */
Boaz Harrosh8d2d83a2011-08-10 14:15:02 -070037 enum pnfs_osd_raid_algorithm4
38 raid_algorithm;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -070039 unsigned stripe_unit;
40 unsigned mirrors_p1;
41
42 unsigned group_width;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +020043 unsigned parity;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -070044 u64 group_depth;
45 unsigned group_count;
Boaz Harrosh5a51c0c2011-09-28 13:18:45 +030046
47 /* Cached often needed calculations filled in by
48 * ore_verify_layout
49 */
50 unsigned long max_io_length; /* Max length that should be passed to
51 * ore_get_rw_state
52 */
Boaz Harrosh8ff660a2011-08-06 19:26:31 -070053};
54
Boaz Harroshd866d872011-09-28 14:43:09 +030055struct ore_dev {
56 struct osd_dev *od;
57};
58
Boaz Harrosh8ff660a2011-08-06 19:26:31 -070059struct ore_components {
Boaz Harrosh3bd98562011-09-28 12:04:23 +030060 unsigned first_dev; /* First logical device no */
Boaz Harrosh8ff660a2011-08-06 19:26:31 -070061 unsigned numdevs; /* Num of devices in array */
62 /* If @single_comp == EC_SINGLE_COMP, @comps points to a single
63 * component. else there are @numdevs components
64 */
65 enum EC_COMP_USAGE {
66 EC_SINGLE_COMP = 0, EC_MULTPLE_COMPS = 0xffffffff
67 } single_comp;
68 struct ore_comp *comps;
Boaz Harroshd866d872011-09-28 14:43:09 +030069
70 /* Array of pointers to ore_dev-* . User will usually have these pointed
71 * too a bigger struct which contain an "ore_dev ored" member and use
72 * container_of(oc->ods[i], struct foo_dev, ored) to access the bigger
73 * structure.
74 */
75 struct ore_dev **ods;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -070076};
77
Boaz Harroshd866d872011-09-28 14:43:09 +030078/* ore_comp_dev Recievies a logical device index */
79static inline struct osd_dev *ore_comp_dev(
80 const struct ore_components *oc, unsigned i)
81{
Boaz Harrosh3bd98562011-09-28 12:04:23 +030082 BUG_ON((i < oc->first_dev) || (oc->first_dev + oc->numdevs <= i));
83 return oc->ods[i - oc->first_dev]->od;
Boaz Harroshd866d872011-09-28 14:43:09 +030084}
85
86static inline void ore_comp_set_dev(
87 struct ore_components *oc, unsigned i, struct osd_dev *od)
88{
Boaz Harrosh3bd98562011-09-28 12:04:23 +030089 oc->ods[i - oc->first_dev]->od = od;
Boaz Harroshd866d872011-09-28 14:43:09 +030090}
91
Boaz Harrosheb507bc2011-08-10 14:17:28 -070092struct ore_striping_info {
Boaz Harrosha1fec1d2011-10-12 18:42:22 +020093 u64 offset;
Boaz Harrosheb507bc2011-08-10 14:17:28 -070094 u64 obj_offset;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +020095 u64 length;
96 u64 first_stripe_start; /* only used in raid writes */
Boaz Harrosheb507bc2011-08-10 14:17:28 -070097 u64 M; /* for truncate */
Boaz Harrosha1fec1d2011-10-12 18:42:22 +020098 unsigned bytes_in_stripe;
Boaz Harrosheb507bc2011-08-10 14:17:28 -070099 unsigned dev;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200100 unsigned par_dev;
Boaz Harrosheb507bc2011-08-10 14:17:28 -0700101 unsigned unit_off;
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200102 unsigned cur_pg;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200103 unsigned cur_comp;
Boaz Harrosheb507bc2011-08-10 14:17:28 -0700104};
105
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700106struct ore_io_state;
107typedef void (*ore_io_done_fn)(struct ore_io_state *ios, void *private);
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200108struct _ore_r4w_op {
109 /* @Priv given here is passed ios->private */
110 struct page * (*get_page)(void *priv, u64 page_index, bool *uptodate);
111 void (*put_page)(void *priv, struct page *page);
112};
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700113
114struct ore_io_state {
115 struct kref kref;
Boaz Harrosh98260752011-10-02 15:32:50 +0200116 struct ore_striping_info si;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700117
118 void *private;
119 ore_io_done_fn done;
120
121 struct ore_layout *layout;
Boaz Harrosh5bf696d2011-09-28 11:39:59 +0300122 struct ore_components *oc;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700123
124 /* Global read/write IO*/
125 loff_t offset;
126 unsigned long length;
127 void *kern_buff;
128
129 struct page **pages;
130 unsigned nr_pages;
131 unsigned pgbase;
132 unsigned pages_consumed;
133
134 /* Attributes */
135 unsigned in_attr_len;
136 struct osd_attr *in_attr;
137 unsigned out_attr_len;
138 struct osd_attr *out_attr;
139
140 bool reading;
141
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200142 /* House keeping of Parity pages */
143 bool extra_part_alloc;
144 struct page **parity_pages;
145 unsigned max_par_pages;
146 unsigned cur_par_page;
147 unsigned sgs_per_dev;
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200148 struct __stripe_pages_2d *sp2d;
149 struct ore_io_state *ios_read_4_write;
150 const struct _ore_r4w_op *r4w;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200151
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700152 /* Variable array of size numdevs */
153 unsigned numdevs;
154 struct ore_per_dev_state {
155 struct osd_request *or;
156 struct bio *bio;
157 loff_t offset;
158 unsigned length;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200159 unsigned last_sgs_total;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700160 unsigned dev;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200161 struct osd_sg_entry *sglist;
162 unsigned cur_sg;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700163 } per_dev[];
164};
165
166static inline unsigned ore_io_state_size(unsigned numdevs)
167{
168 return sizeof(struct ore_io_state) +
169 sizeof(struct ore_per_dev_state) * numdevs;
170}
171
172/* ore.c */
Boaz Harrosh5a51c0c2011-09-28 13:18:45 +0300173int ore_verify_layout(unsigned total_comps, struct ore_layout *layout);
Boaz Harrosh611d7a52011-10-04 14:20:17 +0200174void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200175 u64 length, struct ore_striping_info *si);
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700176int ore_get_rw_state(struct ore_layout *layout, struct ore_components *comps,
177 bool is_reading, u64 offset, u64 length,
178 struct ore_io_state **ios);
179int ore_get_io_state(struct ore_layout *layout, struct ore_components *comps,
180 struct ore_io_state **ios);
181void ore_put_io_state(struct ore_io_state *ios);
182
Boaz Harrosh4b46c9f2011-09-28 13:25:50 +0300183typedef void (*ore_on_dev_error)(struct ore_io_state *ios, struct ore_dev *od,
184 unsigned dev_index, enum osd_err_priority oep,
185 u64 dev_offset, u64 dev_len);
186int ore_check_io(struct ore_io_state *ios, ore_on_dev_error rep);
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700187
188int ore_create(struct ore_io_state *ios);
189int ore_remove(struct ore_io_state *ios);
190int ore_write(struct ore_io_state *ios);
191int ore_read(struct ore_io_state *ios);
192int ore_truncate(struct ore_layout *layout, struct ore_components *comps,
193 u64 size);
194
195int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr);
196
197extern const struct osd_attr g_attr_logical_length;
198
199#endif