Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2005, 2006 |
Boaz Harrosh | 27d2e14 | 2009-06-14 17:23:09 +0300 | [diff] [blame] | 3 | * Avishay Traeger (avishay@gmail.com) |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 4 | * Copyright (C) 2008, 2009 |
| 5 | * Boaz Harrosh <bharrosh@panasas.com> |
| 6 | * |
| 7 | * Copyrights for code taken from ext2: |
| 8 | * Copyright (C) 1992, 1993, 1994, 1995 |
| 9 | * Remy Card (card@masi.ibp.fr) |
| 10 | * Laboratoire MASI - Institut Blaise Pascal |
| 11 | * Universite Pierre et Marie Curie (Paris VI) |
| 12 | * from |
| 13 | * linux/fs/minix/inode.c |
| 14 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 15 | * |
| 16 | * This file is part of exofs. |
| 17 | * |
| 18 | * exofs is free software; you can redistribute it and/or modify |
| 19 | * it under the terms of the GNU General Public License as published by |
| 20 | * the Free Software Foundation. Since it is based on ext2, and the only |
| 21 | * valid version of GPL for the Linux kernel is version 2, the only valid |
| 22 | * version of GPL for exofs is version 2. |
| 23 | * |
| 24 | * exofs is distributed in the hope that it will be useful, |
| 25 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 26 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 27 | * GNU General Public License for more details. |
| 28 | * |
| 29 | * You should have received a copy of the GNU General Public License |
| 30 | * along with exofs; if not, write to the Free Software |
| 31 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 32 | */ |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 33 | #ifndef __EXOFS_H__ |
| 34 | #define __EXOFS_H__ |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 35 | |
| 36 | #include <linux/fs.h> |
| 37 | #include <linux/time.h> |
Jens Axboe | b3d0ab7 | 2010-04-22 12:26:04 +0200 | [diff] [blame] | 38 | #include <linux/backing-dev.h> |
Boaz Harrosh | 26ae93c | 2010-02-02 15:56:53 +0200 | [diff] [blame] | 39 | #include <linux/pnfs_osd_xdr.h> |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 40 | #include "common.h" |
| 41 | |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 42 | #define EXOFS_ERR(fmt, a...) printk(KERN_ERR "exofs: " fmt, ##a) |
| 43 | |
| 44 | #ifdef CONFIG_EXOFS_DEBUG |
| 45 | #define EXOFS_DBGMSG(fmt, a...) \ |
| 46 | printk(KERN_NOTICE "exofs @%s:%d: " fmt, __func__, __LINE__, ##a) |
| 47 | #else |
| 48 | #define EXOFS_DBGMSG(fmt, a...) \ |
| 49 | do { if (0) printk(fmt, ##a); } while (0) |
| 50 | #endif |
| 51 | |
| 52 | /* u64 has problems with printk this will cast it to unsigned long long */ |
| 53 | #define _LLU(x) (unsigned long long)(x) |
| 54 | |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 55 | struct exofs_comp { |
| 56 | struct osd_obj_id obj; |
| 57 | u8 cred[OSD_CAP_LEN]; |
| 58 | }; |
Boaz Harrosh | 45d3abc | 2010-01-28 11:46:16 +0200 | [diff] [blame] | 59 | |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 60 | struct exofs_layout { |
Boaz Harrosh | 5d952b8 | 2010-02-01 13:35:51 +0200 | [diff] [blame] | 61 | /* Our way of looking at the data_map */ |
| 62 | unsigned stripe_unit; |
| 63 | unsigned mirrors_p1; |
| 64 | |
| 65 | unsigned group_width; |
Boaz Harrosh | 50a76fd | 2010-02-11 13:01:39 +0200 | [diff] [blame] | 66 | u64 group_depth; |
| 67 | unsigned group_count; |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 68 | }; |
Boaz Harrosh | 5d952b8 | 2010-02-01 13:35:51 +0200 | [diff] [blame] | 69 | |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 70 | struct exofs_components { |
| 71 | unsigned numdevs; /* Num of devices in array */ |
| 72 | /* If @single_comp == EC_SINGLE_COMP, @comps points to a single |
| 73 | * component. else there are @numdevs components |
| 74 | */ |
| 75 | enum EC_COMP_USAGE { |
| 76 | EC_SINGLE_COMP = 0, EC_MULTPLE_COMPS = 0xffffffff |
| 77 | } single_comp; |
| 78 | struct exofs_comp *comps; |
| 79 | struct osd_dev **ods; /* osd_dev array */ |
Boaz Harrosh | 45d3abc | 2010-01-28 11:46:16 +0200 | [diff] [blame] | 80 | }; |
| 81 | |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 82 | /* |
| 83 | * our extension to the in-memory superblock |
| 84 | */ |
| 85 | struct exofs_sb_info { |
Boaz Harrosh | 6d4073e | 2011-07-27 17:51:53 -0700 | [diff] [blame] | 86 | struct backing_dev_info bdi; /* register our bdi with VFS */ |
Boaz Harrosh | 1cea312 | 2011-02-03 17:53:25 +0200 | [diff] [blame] | 87 | struct exofs_sb_stats s_ess; /* Written often, pre-allocate*/ |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 88 | int s_timeout; /* timeout for OSD operations */ |
| 89 | uint64_t s_nextid; /* highest object ID used */ |
| 90 | uint32_t s_numfiles; /* number of files on fs */ |
| 91 | spinlock_t s_next_gen_lock; /* spinlock for gen # update */ |
| 92 | u32 s_next_generation; /* next gen # to use */ |
| 93 | atomic_t s_curr_pending; /* number of pending commands */ |
Boaz Harrosh | 04dc1e8 | 2009-11-16 16:03:05 +0200 | [diff] [blame] | 94 | |
Boaz Harrosh | 45d3abc | 2010-01-28 11:46:16 +0200 | [diff] [blame] | 95 | struct pnfs_osd_data_map data_map; /* Default raid to use |
| 96 | * FIXME: Needed ? |
| 97 | */ |
Boaz Harrosh | 6d4073e | 2011-07-27 17:51:53 -0700 | [diff] [blame] | 98 | struct exofs_layout layout; /* Default files layout */ |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 99 | struct exofs_comp one_comp; /* id & cred of partition id=0*/ |
| 100 | struct exofs_components comps; /* comps for the partition */ |
Boaz Harrosh | 45d3abc | 2010-01-28 11:46:16 +0200 | [diff] [blame] | 101 | struct osd_dev *_min_one_dev[1]; /* Place holder for one dev */ |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 102 | }; |
| 103 | |
| 104 | /* |
| 105 | * our extension to the in-memory inode |
| 106 | */ |
| 107 | struct exofs_i_info { |
Boaz Harrosh | 518f167 | 2010-01-21 20:00:02 +0200 | [diff] [blame] | 108 | struct inode vfs_inode; /* normal in-memory inode */ |
| 109 | wait_queue_head_t i_wq; /* wait queue for inode */ |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 110 | unsigned long i_flags; /* various atomic flags */ |
| 111 | uint32_t i_data[EXOFS_IDATA];/*short symlink names and device #s*/ |
| 112 | uint32_t i_dir_start_lookup; /* which page to start lookup */ |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 113 | uint64_t i_commit_size; /* the object's written length */ |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 114 | struct exofs_comp one_comp; /* same component for all devices */ |
| 115 | struct exofs_components comps; /* inode view of the device table */ |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 116 | }; |
| 117 | |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 118 | static inline osd_id exofs_oi_objno(struct exofs_i_info *oi) |
| 119 | { |
| 120 | return oi->vfs_inode.i_ino + EXOFS_OBJ_OFF; |
| 121 | } |
| 122 | |
| 123 | struct exofs_io_state; |
Boaz Harrosh | e1042ba | 2010-11-16 20:09:58 +0200 | [diff] [blame] | 124 | typedef void (*exofs_io_done_fn)(struct exofs_io_state *ios, void *private); |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 125 | |
| 126 | struct exofs_io_state { |
| 127 | struct kref kref; |
| 128 | |
| 129 | void *private; |
| 130 | exofs_io_done_fn done; |
| 131 | |
Boaz Harrosh | 45d3abc | 2010-01-28 11:46:16 +0200 | [diff] [blame] | 132 | struct exofs_layout *layout; |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 133 | struct exofs_components *comps; |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 134 | |
| 135 | /* Global read/write IO*/ |
| 136 | loff_t offset; |
| 137 | unsigned long length; |
| 138 | void *kern_buff; |
Boaz Harrosh | 86093aa | 2010-01-28 18:24:06 +0200 | [diff] [blame] | 139 | |
| 140 | struct page **pages; |
| 141 | unsigned nr_pages; |
| 142 | unsigned pgbase; |
Boaz Harrosh | 50a76fd | 2010-02-11 13:01:39 +0200 | [diff] [blame] | 143 | unsigned pages_consumed; |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 144 | |
| 145 | /* Attributes */ |
| 146 | unsigned in_attr_len; |
| 147 | struct osd_attr *in_attr; |
| 148 | unsigned out_attr_len; |
| 149 | struct osd_attr *out_attr; |
| 150 | |
Boaz Harrosh | e1042ba | 2010-11-16 20:09:58 +0200 | [diff] [blame] | 151 | bool reading; |
| 152 | |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 153 | /* Variable array of size numdevs */ |
| 154 | unsigned numdevs; |
| 155 | struct exofs_per_dev_state { |
| 156 | struct osd_request *or; |
| 157 | struct bio *bio; |
Boaz Harrosh | 5d952b8 | 2010-02-01 13:35:51 +0200 | [diff] [blame] | 158 | loff_t offset; |
| 159 | unsigned length; |
| 160 | unsigned dev; |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 161 | } per_dev[]; |
| 162 | }; |
| 163 | |
| 164 | static inline unsigned exofs_io_state_size(unsigned numdevs) |
| 165 | { |
| 166 | return sizeof(struct exofs_io_state) + |
| 167 | sizeof(struct exofs_per_dev_state) * numdevs; |
| 168 | } |
| 169 | |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 170 | /* |
| 171 | * our inode flags |
| 172 | */ |
| 173 | #define OBJ_2BCREATED 0 /* object will be created soon*/ |
| 174 | #define OBJ_CREATED 1 /* object has been created on the osd*/ |
| 175 | |
| 176 | static inline int obj_2bcreated(struct exofs_i_info *oi) |
| 177 | { |
| 178 | return test_bit(OBJ_2BCREATED, &oi->i_flags); |
| 179 | } |
| 180 | |
| 181 | static inline void set_obj_2bcreated(struct exofs_i_info *oi) |
| 182 | { |
| 183 | set_bit(OBJ_2BCREATED, &oi->i_flags); |
| 184 | } |
| 185 | |
| 186 | static inline int obj_created(struct exofs_i_info *oi) |
| 187 | { |
| 188 | return test_bit(OBJ_CREATED, &oi->i_flags); |
| 189 | } |
| 190 | |
| 191 | static inline void set_obj_created(struct exofs_i_info *oi) |
| 192 | { |
| 193 | set_bit(OBJ_CREATED, &oi->i_flags); |
| 194 | } |
| 195 | |
| 196 | int __exofs_wait_obj_created(struct exofs_i_info *oi); |
| 197 | static inline int wait_obj_created(struct exofs_i_info *oi) |
| 198 | { |
| 199 | if (likely(obj_created(oi))) |
| 200 | return 0; |
| 201 | |
| 202 | return __exofs_wait_obj_created(oi); |
| 203 | } |
| 204 | |
| 205 | /* |
| 206 | * get to our inode from the vfs inode |
| 207 | */ |
| 208 | static inline struct exofs_i_info *exofs_i(struct inode *inode) |
| 209 | { |
| 210 | return container_of(inode, struct exofs_i_info, vfs_inode); |
| 211 | } |
| 212 | |
Boaz Harrosh | e6af00f | 2008-10-28 15:38:12 +0200 | [diff] [blame] | 213 | /* |
| 214 | * Maximum count of links to a file |
| 215 | */ |
| 216 | #define EXOFS_LINK_MAX 32000 |
| 217 | |
Boaz Harrosh | e806271 | 2008-10-27 18:37:02 +0200 | [diff] [blame] | 218 | /************************* |
| 219 | * function declarations * |
| 220 | *************************/ |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 221 | |
| 222 | /* ios.c */ |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 223 | int exofs_get_rw_state(struct exofs_layout *layout, |
| 224 | struct exofs_components *comps, |
| 225 | bool is_reading, u64 offset, u64 length, |
| 226 | struct exofs_io_state **ios); |
| 227 | int exofs_get_io_state(struct exofs_layout *layout, |
| 228 | struct exofs_components *comps, |
| 229 | struct exofs_io_state **ios); |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 230 | void exofs_put_io_state(struct exofs_io_state *ios); |
| 231 | |
| 232 | int exofs_check_io(struct exofs_io_state *ios, u64 *resid); |
| 233 | |
| 234 | int exofs_sbi_create(struct exofs_io_state *ios); |
| 235 | int exofs_sbi_remove(struct exofs_io_state *ios); |
| 236 | int exofs_sbi_write(struct exofs_io_state *ios); |
| 237 | int exofs_sbi_read(struct exofs_io_state *ios); |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 238 | int exofs_truncate(struct exofs_layout *layout, struct exofs_components *comps, |
| 239 | u64 size); |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 240 | |
| 241 | int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr); |
Boaz Harrosh | 85e44df | 2011-05-16 15:26:47 +0300 | [diff] [blame] | 242 | extern const struct osd_attr g_attr_logical_length; |
Boaz Harrosh | 06886a5 | 2009-11-08 14:54:08 +0200 | [diff] [blame] | 243 | |
Boaz Harrosh | e806271 | 2008-10-27 18:37:02 +0200 | [diff] [blame] | 244 | /* inode.c */ |
bharrosh@panasas.com | 66cd6ca | 2010-10-07 14:28:18 -0400 | [diff] [blame] | 245 | unsigned exofs_max_io_pages(struct exofs_layout *layout, |
| 246 | unsigned expected_pages); |
Boaz Harrosh | e806271 | 2008-10-27 18:37:02 +0200 | [diff] [blame] | 247 | int exofs_setattr(struct dentry *, struct iattr *); |
Boaz Harrosh | beaec07 | 2008-10-27 19:31:34 +0200 | [diff] [blame] | 248 | int exofs_write_begin(struct file *file, struct address_space *mapping, |
| 249 | loff_t pos, unsigned len, unsigned flags, |
| 250 | struct page **pagep, void **fsdata); |
Boaz Harrosh | e6af00f | 2008-10-28 15:38:12 +0200 | [diff] [blame] | 251 | extern struct inode *exofs_iget(struct super_block *, unsigned long); |
| 252 | struct inode *exofs_new_inode(struct inode *, int); |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 253 | extern int exofs_write_inode(struct inode *, struct writeback_control *wbc); |
Al Viro | 4ec70c9 | 2010-06-07 11:42:26 -0400 | [diff] [blame] | 254 | extern void exofs_evict_inode(struct inode *); |
Boaz Harrosh | e6af00f | 2008-10-28 15:38:12 +0200 | [diff] [blame] | 255 | |
| 256 | /* dir.c: */ |
| 257 | int exofs_add_link(struct dentry *, struct inode *); |
| 258 | ino_t exofs_inode_by_name(struct inode *, struct dentry *); |
| 259 | int exofs_delete_entry(struct exofs_dir_entry *, struct page *); |
| 260 | int exofs_make_empty(struct inode *, struct inode *); |
| 261 | struct exofs_dir_entry *exofs_find_entry(struct inode *, struct dentry *, |
| 262 | struct page **); |
| 263 | int exofs_empty_dir(struct inode *); |
| 264 | struct exofs_dir_entry *exofs_dotdot(struct inode *, struct page **); |
Boaz Harrosh | 8cf74b3 | 2009-03-22 12:47:26 +0200 | [diff] [blame] | 265 | ino_t exofs_parent_ino(struct dentry *child); |
Boaz Harrosh | e6af00f | 2008-10-28 15:38:12 +0200 | [diff] [blame] | 266 | int exofs_set_link(struct inode *, struct exofs_dir_entry *, struct page *, |
| 267 | struct inode *); |
Boaz Harrosh | e806271 | 2008-10-27 18:37:02 +0200 | [diff] [blame] | 268 | |
Boaz Harrosh | baaf94c | 2009-06-14 16:52:10 +0300 | [diff] [blame] | 269 | /* super.c */ |
Boaz Harrosh | 85e44df | 2011-05-16 15:26:47 +0300 | [diff] [blame] | 270 | void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], |
| 271 | const struct osd_obj_id *obj); |
Boaz Harrosh | 1cea312 | 2011-02-03 17:53:25 +0200 | [diff] [blame] | 272 | int exofs_sbi_write_stats(struct exofs_sb_info *sbi); |
Boaz Harrosh | baaf94c | 2009-06-14 16:52:10 +0300 | [diff] [blame] | 273 | |
Boaz Harrosh | e806271 | 2008-10-27 18:37:02 +0200 | [diff] [blame] | 274 | /********************* |
| 275 | * operation vectors * |
| 276 | *********************/ |
Boaz Harrosh | e6af00f | 2008-10-28 15:38:12 +0200 | [diff] [blame] | 277 | /* dir.c: */ |
| 278 | extern const struct file_operations exofs_dir_operations; |
| 279 | |
Boaz Harrosh | e806271 | 2008-10-27 18:37:02 +0200 | [diff] [blame] | 280 | /* file.c */ |
| 281 | extern const struct inode_operations exofs_file_inode_operations; |
| 282 | extern const struct file_operations exofs_file_operations; |
| 283 | |
Boaz Harrosh | beaec07 | 2008-10-27 19:31:34 +0200 | [diff] [blame] | 284 | /* inode.c */ |
| 285 | extern const struct address_space_operations exofs_aops; |
| 286 | |
Boaz Harrosh | e6af00f | 2008-10-28 15:38:12 +0200 | [diff] [blame] | 287 | /* namei.c */ |
| 288 | extern const struct inode_operations exofs_dir_inode_operations; |
| 289 | extern const struct inode_operations exofs_special_inode_operations; |
| 290 | |
Boaz Harrosh | 982980d | 2008-10-27 19:04:34 +0200 | [diff] [blame] | 291 | /* symlink.c */ |
| 292 | extern const struct inode_operations exofs_symlink_inode_operations; |
| 293 | extern const struct inode_operations exofs_fast_symlink_inode_operations; |
| 294 | |
Boaz Harrosh | 9e9db45 | 2011-08-05 15:06:04 -0700 | [diff] [blame^] | 295 | /* exofs_init_comps will initialize an exofs_components device array |
| 296 | * pointing to a single exofs_comp struct, and a round-robin view |
| 297 | * of the device table. |
| 298 | * The first device of each inode is the [inode->ino % num_devices] |
| 299 | * and the rest of the devices sequentially following where the |
| 300 | * first device is after the last device. |
| 301 | * It is assumed that the global device array at @sbi is twice |
| 302 | * bigger and that the device table repeats twice. |
| 303 | * See: exofs_read_lookup_dev_table() |
| 304 | */ |
| 305 | static inline void exofs_init_comps(struct exofs_components *comps, |
| 306 | struct exofs_comp *one_comp, |
| 307 | struct exofs_sb_info *sbi, osd_id oid) |
| 308 | { |
| 309 | unsigned dev_mod = (unsigned)oid, first_dev; |
| 310 | |
| 311 | one_comp->obj.partition = sbi->one_comp.obj.partition; |
| 312 | one_comp->obj.id = oid; |
| 313 | exofs_make_credential(one_comp->cred, &one_comp->obj); |
| 314 | |
| 315 | comps->numdevs = sbi->comps.numdevs; |
| 316 | comps->single_comp = EC_SINGLE_COMP; |
| 317 | comps->comps = one_comp; |
| 318 | |
| 319 | /* Round robin device view of the table */ |
| 320 | first_dev = (dev_mod * sbi->layout.mirrors_p1) % sbi->comps.numdevs; |
| 321 | comps->ods = sbi->comps.ods + first_dev; |
| 322 | } |
| 323 | |
Boaz Harrosh | b14f8ab | 2008-10-27 18:27:55 +0200 | [diff] [blame] | 324 | #endif |