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