fs/pohmelfs/pohmelfs.h

   1 /*
   2  * Copyright (C) 2011+ Evgeniy Polyakov <zbr@ioremap.net>
   3  */
   4
   5 #ifndef __POHMELFS_H
   6 #define __POHMELFS_H
   7
   8 #include <linux/backing-dev.h>
   9 #include <linux/crypto.h>
  10 #include <linux/fs.h>
  11 #include <linux/kref.h>
  12 #include <linux/list.h>
  13 #include <linux/mutex.h>
  14 #include <linux/net.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/printk.h>
  18 #include <linux/slab.h>
  19 #include <linux/time.h>
  20 #include <linux/wait.h>
  21 #include <linux/workqueue.h>
  22
  23 #include <crypto/sha.h>
  24
  25 #define dnet_bswap16(x)         cpu_to_le16(x)
  26 #define dnet_bswap32(x)         cpu_to_le32(x)
  27 #define dnet_bswap64(x)         cpu_to_le64(x)
  28
  29 /* theese are needed for packet.h below to compile */
  30 #define DNET_ID_SIZE    SHA512_DIGEST_SIZE
  31 #define DNET_CSUM_SIZE  SHA512_DIGEST_SIZE
  32
  33 #define POHMELFS_INODE_COLUMN           3
  34
  35 /*
  36  * is not used in kernel, but we want to share the same header
  37  * with userspace, so I put it here for compiler to shut up
  38  */
  39 int gettimeofday(struct timeval *, struct timezone *);
  40
  41 #include "packet.h"
  42
  43 static inline struct timespec pohmelfs_date(struct dnet_time *tm)
  44 {
  45         struct timespec ts;
  46
  47         ts.tv_sec = tm->tsec;
  48         ts.tv_nsec = tm->tnsec;
  49
  50         return ts;
  51 }
  52
  53 struct pohmelfs_cmd {
  54         struct dnet_cmd         cmd;
  55         struct dnet_attr        attr;
  56         union {
  57                 struct dnet_io_attr     io;
  58         } p;
  59 };
  60
  61 /*
  62  * Compare two IDs.
  63  * Returns  1 when id1 > id2
  64  *         -1 when id1 < id2
  65  *          0 when id1 = id2
  66  */
  67 static inline int dnet_id_cmp_str(const unsigned char *id1, const unsigned char *id2)
  68 {
  69         unsigned int i = 0;
  70
  71         for (i*=sizeof(unsigned long); i<DNET_ID_SIZE; ++i) {
  72                 if (id1[i] < id2[i])
  73                         return -1;
  74                 if (id1[i] > id2[i])
  75                         return 1;
  76         }
  77
  78         return 0;
  79 }
  80
  81 struct pohmelfs_state;
  82 struct pohmelfs_sb;
  83 struct pohmelfs_trans;
  84
  85 struct pohmelfs_trans_cb {
  86         int                             (* init)(struct pohmelfs_trans *t);
  87         int                             (* complete)(struct pohmelfs_trans *t, struct pohmelfs_state *recv);
  88         int                             (* recv_reply)(struct pohmelfs_trans *t, struct pohmelfs_state *recv);
  89         void                            (* destroy)(struct pohmelfs_trans *t);
  90 };
  91
  92 struct pohmelfs_trans {
  93         struct list_head        trans_entry;
  94
  95         struct kref             refcnt;
  96
  97         unsigned long           trans;
  98
  99         struct inode            *inode;
 100
 101         struct pohmelfs_state   *st;
 102
 103         struct pohmelfs_cmd     cmd;
 104
 105         u64                     header_size, data_size;
 106
 107         void                    *data;
 108
 109         unsigned long long      recv_offset;
 110         void                    *recv_data;
 111
 112         struct pohmelfs_write_ctl       *wctl;
 113         void                    *priv;
 114
 115         struct pohmelfs_trans_cb        cb;
 116 };
 117
 118 struct pohmelfs_trans *pohmelfs_trans_alloc(struct inode *inode);
 119 struct pohmelfs_trans *pohmelfs_trans_alloc_io_buf(struct inode *inode, int group, int command,
 120                 void *data, u64 offset, u64 size, int aflags, int ioflags, int type);
 121 void pohmelfs_trans_put(struct pohmelfs_trans *t);
 122
 123 int pohmelfs_trans_insert(struct pohmelfs_trans *t);
 124 struct pohmelfs_trans *pohmelfs_trans_lookup(struct pohmelfs_state *st, struct dnet_cmd *cmd);
 125
 126 struct pohmelfs_state {
 127         struct pohmelfs_sb      *psb;
 128         struct list_head        state_entry;
 129
 130         struct sockaddr_storage sa;
 131         int                     addrlen;
 132         struct socket           *sock;
 133
 134         int                     group_id;
 135
 136         struct mutex            trans_lock;
 137         struct list_head        trans_list;
 138         struct list_head        sent_trans_list;
 139
 140         struct kref             refcnt;
 141
 142         int                     routes;
 143
 144         /* Waiting/polling machinery */
 145         wait_queue_t            wait;
 146         wait_queue_head_t       *whead;
 147
 148         struct work_struct      send_work;
 149         struct work_struct      recv_work;
 150
 151         /* is set when dnet_cmd is being read, otherwise attached data */
 152         int                     cmd_read;
 153         /* currently read command reply */
 154         struct dnet_cmd         cmd;
 155
 156         uint64_t                bsize;          /* Block size */
 157         uint64_t                frsize;         /* Fragment size */
 158         uint64_t                blocks;         /* Filesystem size in frsize units */
 159         uint64_t                bfree;          /* # free blocks */
 160         uint64_t                bavail;         /* # free blocks for non-root */
 161 };
 162
 163 struct pohmelfs_state *pohmelfs_state_create(struct pohmelfs_sb *psb, struct sockaddr_storage *sa, int addrlen,
 164                 int ask_route, int group_id);
 165 struct pohmelfs_state *pohmelfs_state_lookup(struct pohmelfs_sb *psb, struct dnet_raw_id *id, int group);
 166 int pohmelfs_grab_states(struct pohmelfs_sb *psb, struct pohmelfs_state ***stp);
 167
 168 static inline void pohmelfs_state_get(struct pohmelfs_state *st)
 169 {
 170         kref_get(&st->refcnt);
 171 }
 172
 173 void pohmelfs_state_put(struct pohmelfs_state *st);
 174 void pohmelfs_state_kill(struct pohmelfs_state *st);
 175
 176 struct pohmelfs_state *pohmelfs_addr_exist(struct pohmelfs_sb *psb, struct sockaddr_storage *sa, int addrlen);
 177
 178 void pohmelfs_state_schedule(struct pohmelfs_state *st);
 179
 180 __attribute__ ((format (printf, 2, 3))) void pohmelfs_print_addr(struct sockaddr_storage *addr, const char *fmt, ...);
 181
 182 #define POHMELFS_INODE_INFO_REMOVED             (1<<0)
 183
 184 struct pohmelfs_inode_info {
 185         struct dnet_raw_id      id;
 186
 187         unsigned int            mode;
 188         unsigned int            nlink;
 189         unsigned int            uid;
 190         unsigned int            gid;
 191         unsigned int            blocksize;
 192         unsigned int            namelen;
 193         __u64                   ino;
 194         __u64                   blocks;
 195         __u64                   rdev;
 196         __u64                   size;
 197         __u64                   version;
 198
 199         __u64                   flags;
 200
 201         struct dnet_time        ctime;
 202         struct dnet_time        mtime;
 203         struct dnet_time        atime;
 204 } __attribute__ ((packed));
 205
 206 void pohmelfs_fill_inode_info(struct inode *inode, struct pohmelfs_inode_info *info);
 207 void pohmelfs_fill_inode(struct inode *inode, struct pohmelfs_inode_info *info);
 208 void pohmelfs_convert_inode_info(struct pohmelfs_inode_info *info);
 209
 210 struct pohmelfs_inode {
 211         struct inode            vfs_inode;
 212         struct dnet_raw_id      id;
 213         struct dnet_raw_id      parent_id;
 214
 215         struct rb_node          node;
 216
 217         struct mutex            lock;
 218
 219         int                     *groups;
 220         int                     group_num;
 221
 222         time_t                  update;
 223         int                     local;
 224 };
 225
 226 int pohmelfs_send_dentry(struct pohmelfs_inode *pi, struct dnet_raw_id *id, const char *sname, int len, int sync);
 227 struct pohmelfs_inode *pohmelfs_sb_inode_lookup(struct pohmelfs_sb *psb, struct dnet_raw_id *id);
 228
 229 struct pohmelfs_reconnect {
 230         struct list_head        reconnect_entry;
 231         struct sockaddr_storage sa;
 232         int                     addrlen;
 233         int                     group_id;
 234 };
 235
 236 int pohmelfs_state_add_reconnect(struct pohmelfs_state *st);
 237
 238 struct pohmelfs_path {
 239         struct mutex            lock;
 240         char                    *data;
 241 };
 242
 243 int pohmelfs_http_compat_id(struct pohmelfs_inode *pi);
 244
 245 struct pohmelfs_sb {
 246         struct super_block      *sb;
 247         struct backing_dev_info bdi;
 248
 249         struct pohmelfs_inode   *root;
 250
 251         spinlock_t              inode_lock;
 252         struct rb_root          inode_root;
 253
 254         int                     http_compat;
 255         struct pohmelfs_path    *path;
 256
 257         int                     bdi_num;
 258
 259         struct rb_root          route_root;
 260         struct list_head        state_list;
 261         spinlock_t              state_lock;
 262
 263         long                    read_wait_timeout;
 264         long                    write_wait_timeout;
 265
 266         long                    sync_timeout;
 267         struct delayed_work     sync_work;
 268
 269         char                    *fsid;
 270         int                     fsid_len;
 271
 272         int                     no_read_csum;
 273
 274         int                     need_exit;
 275
 276         atomic_long_t           ino;
 277         atomic_long_t           trans;
 278
 279         struct crypto_hash      *hash;
 280
 281         struct workqueue_struct *wq;
 282
 283         int                     *groups;
 284         int                     group_num;
 285
 286         /*
 287          * number of copies to be successfully written to mark write as successful
 288          * if not set, half of groups plus one must be successfully written, i.e. plain write quorum
 289          */
 290         int                     successful_write_count;
 291
 292         struct mutex            reconnect_lock;
 293         struct list_head        reconnect_list;
 294         struct list_head        kill_state_list;
 295         struct delayed_work     reconnect_work;
 296         long                    reconnect_timeout;
 297
 298         int                     keepalive_cnt, keepalive_interval, keepalive_idle;
 299
 300         int                     readdir_allocation;
 301
 302         int                     sync_on_close;
 303 };
 304
 305 static inline struct pohmelfs_sb *pohmelfs_sb(struct super_block *sb)
 306 {
 307         return (struct pohmelfs_sb *)sb->s_fs_info;
 308 }
 309
 310 static inline struct pohmelfs_inode *pohmelfs_inode(struct inode *inode)
 311 {
 312         return container_of(inode, struct pohmelfs_inode, vfs_inode);
 313 }
 314
 315 struct pohmelfs_wait {
 316         wait_queue_head_t       wq;
 317         struct pohmelfs_inode   *pi;
 318         void                    *ret;
 319         atomic_long_t           count;
 320         int                     condition;
 321         struct kref             refcnt;
 322 };
 323
 324 int pohmelfs_wait_init(struct pohmelfs_wait *wait, struct pohmelfs_inode *pi);
 325 struct pohmelfs_wait *pohmelfs_wait_alloc(struct pohmelfs_inode *pi);
 326 void pohmelfs_wait_put(struct pohmelfs_wait *wait);
 327 static inline void pohmelfs_wait_get(struct pohmelfs_wait *wait)
 328 {
 329         kref_get(&wait->refcnt);
 330 }
 331
 332 struct pohmelfs_inode_info_binary_package {
 333         struct pohmelfs_inode_info      info;
 334
 335         struct pohmelfs_wait            wait;
 336 };
 337
 338 struct pohmelfs_write_ctl {
 339         struct pagevec                  pvec;
 340         struct pohmelfs_inode_info      *info;
 341
 342         struct kref                     refcnt;
 343         atomic_t                        good_writes;
 344 };
 345
 346 struct pohmelfs_dentry_disk {
 347         struct dnet_raw_id              id;
 348         uint64_t                        ino;
 349         int                             pad0;
 350         short                           pad1;
 351         unsigned char                   type;
 352         unsigned char                   len;
 353         char                            name[0];
 354 };
 355
 356 struct pohmelfs_dentry {
 357         struct dnet_raw_id              parent_id;
 358         struct pohmelfs_dentry_disk     disk;
 359 };
 360
 361 extern struct kmem_cache *pohmelfs_inode_cache;
 362 extern struct kmem_cache *pohmelfs_trans_cache;
 363 extern struct kmem_cache *pohmelfs_inode_info_cache;
 364 extern struct kmem_cache *pohmelfs_route_cache;
 365 extern struct kmem_cache *pohmelfs_wait_cache;
 366 extern struct kmem_cache *pohmelfs_io_cache;
 367 extern struct kmem_cache *pohmelfs_inode_info_binary_package_cache;
 368 extern struct kmem_cache *pohmelfs_write_cache;
 369 extern struct kmem_cache *pohmelfs_dentry_cache;
 370
 371 struct inode *pohmelfs_alloc_inode(struct super_block *sb);
 372 void pohmelfs_destroy_inode(struct inode *);
 373
 374 struct pohmelfs_inode *pohmelfs_existing_inode(struct pohmelfs_sb *psb, struct pohmelfs_inode_info *info);
 375 struct pohmelfs_inode *pohmelfs_new_inode(struct pohmelfs_sb *psb, int mode);
 376 int pohmelfs_hash(struct pohmelfs_sb *psb, const void *data, const size_t size, struct dnet_raw_id *id);
 377
 378 char *pohmelfs_dump_id(const unsigned char *id);
 379 char *pohmelfs_dump_id_len_raw(const unsigned char *id, unsigned int len, char *dst);
 380
 381 int pohmelfs_write_command(struct pohmelfs_inode *pi, struct pohmelfs_write_ctl *ctl, loff_t offset, size_t len);
 382 void pohmelfs_write_ctl_release(struct kref *kref);
 383 int pohmelfs_metadata_inode(struct pohmelfs_inode *pi, int sync);
 384
 385 extern const struct file_operations pohmelfs_dir_fops;
 386 extern const struct inode_operations pohmelfs_dir_inode_operations;
 387
 388 extern const struct file_operations pohmelfs_file_ops;
 389 extern const struct inode_operations pohmelfs_file_inode_operations;
 390
 391 extern const struct inode_operations pohmelfs_symlink_inode_operations;
 392 extern const struct inode_operations pohmelfs_special_inode_operations;
 393
 394 extern void *pohmelfs_scratch_buf;
 395 extern int pohmelfs_scratch_buf_size;
 396
 397 /*
 398  * if this flag is set, pohmelfs_inode_info->data is owned by the caller,
 399  * so sending path may use it on its own and free (using kfree) when it's done
 400  *
 401  * This logic does not work for shared buffers or
 402  * when multiple transactions will be sent for single pohmelfs_inode_info
 403  */
 404 #define POHMELFS_IO_OWN                 (1<<0)
 405
 406 struct pohmelfs_io {
 407         struct pohmelfs_inode           *pi;
 408
 409         struct dnet_raw_id              *id;
 410
 411         int                             cmd;
 412         int                             type;
 413
 414         u64                             offset, size;
 415         u64                             start, num;
 416
 417         u32                             cflags;
 418         u32                             aflags;
 419         u32                             ioflags;
 420
 421         int                             group_id;
 422
 423         u32                             alloc_flags;
 424         void                            *data;
 425
 426         struct pohmelfs_write_ctl       *wctl;
 427         void                            *priv;
 428
 429         struct pohmelfs_trans_cb        cb;
 430 };
 431
 432 int pohmelfs_send_io_group(struct pohmelfs_io *pio, int group_id);
 433 int pohmelfs_send_io(struct pohmelfs_io *pio);
 434 int pohmelfs_send_buf_single(struct pohmelfs_io *pio, struct pohmelfs_state *st);
 435 int pohmelfs_send_buf(struct pohmelfs_io *pio);
 436
 437 int pohmelfs_data_recv(struct pohmelfs_state *st, void *buf, u64 size, unsigned int flags);
 438 int pohmelfs_recv(struct pohmelfs_trans *t, struct pohmelfs_state *recv, void *data, int size);
 439
 440 struct pohmelfs_route {
 441         struct rb_node                  node;
 442         int                             group_id;
 443         struct dnet_raw_id              id;
 444         struct pohmelfs_state           *st;
 445 };
 446
 447 int pohmelfs_route_request(struct pohmelfs_state *st);
 448 void pohmelfs_route_remove_all(struct pohmelfs_state *st);
 449
 450 struct pohmelfs_script_req {
 451         char                    *obj_name;
 452         int                     obj_len;
 453
 454         char                    *script_name;
 455         int                     script_namelen;
 456
 457         void                    *binary;
 458         int                     binary_size;
 459
 460         int                     group_id;
 461
 462         unsigned int            cflags;
 463         int                     sync;
 464
 465         struct dnet_raw_id      *id;
 466
 467         int                     (* complete)(struct pohmelfs_trans *t, struct pohmelfs_state *recv);
 468         void                    *ret;
 469         int                     ret_cond;
 470 };
 471
 472 int pohmelfs_send_script_request(struct pohmelfs_inode *parent, struct pohmelfs_script_req *req);
 473
 474 int pohmelfs_stat(struct pohmelfs_sb *psb, int sync);
 475
 476 static inline int pohmelfs_need_resync(struct pohmelfs_inode *pi)
 477 {
 478         struct pohmelfs_sb *psb = pohmelfs_sb(pi->vfs_inode.i_sb);
 479         return get_seconds() > pi->update + psb->sync_timeout;
 480 }
 481
 482 #endif /* __POHMELFS_H */