LXR linux/fs/ceph/mds

   1#ifndef _FS_CEPH_MDS_CLIENT_H
   2#define _FS_CEPH_MDS_CLIENT_H
   3
   4#include <linux/completion.h>
   5#include <linux/kref.h>
   6#include <linux/list.h>
   7#include <linux/mutex.h>
   8#include <linux/rbtree.h>
   9#include <linux/spinlock.h>
  10
  11#include <linux/ceph/types.h>
  12#include <linux/ceph/messenger.h>
  13#include <linux/ceph/mdsmap.h>
  14#include <linux/ceph/auth.h>
  15
  16/*
  17 * Some lock dependencies:
  18 *
  19 * session->s_mutex
  20 *         mdsc->mutex
  21 *
  22 *         mdsc->snap_rwsem
  23 *
  24 *         ci->i_ceph_lock
  25 *                 mdsc->snap_flush_lock
  26 *                 mdsc->cap_delay_lock
  27 *
  28 */
  29
  30struct ceph_fs_client;
  31struct ceph_cap;
  32
  33/*
  34 * parsed info about a single inode.  pointers are into the encoded
  35 * on-wire structures within the mds reply message payload.
  36 */
  37struct ceph_mds_reply_info_in {
  38        struct ceph_mds_reply_inode *in;
  39        struct ceph_dir_layout dir_layout;
  40        u32 symlink_len;
  41        char *symlink;
  42        u32 xattr_len;
  43        char *xattr_data;
  44};
  45
  46/*
  47 * parsed info about an mds reply, including information about
  48 * either: 1) the target inode and/or its parent directory and dentry,
  49 * and directory contents (for readdir results), or
  50 * 2) the file range lock info (for fcntl F_GETLK results).
  51 */
  52struct ceph_mds_reply_info_parsed {
  53        struct ceph_mds_reply_head    *head;
  54
  55        /* trace */
  56        struct ceph_mds_reply_info_in diri, targeti;
  57        struct ceph_mds_reply_dirfrag *dirfrag;
  58        char                          *dname;
  59        u32                           dname_len;
  60        struct ceph_mds_reply_lease   *dlease;
  61
  62        /* extra */
  63        union {
  64                /* for fcntl F_GETLK results */
  65                struct ceph_filelock *filelock_reply;
  66
  67                /* for readdir results */
  68                struct {
  69                        struct ceph_mds_reply_dirfrag *dir_dir;
  70                        size_t                        dir_buf_size;
  71                        int                           dir_nr;
  72                        char                          **dir_dname;
  73                        u32                           *dir_dname_len;
  74                        struct ceph_mds_reply_lease   **dir_dlease;
  75                        struct ceph_mds_reply_info_in *dir_in;
  76                        u8                            dir_complete, dir_end;
  77                };
  78
  79                /* for create results */
  80                struct {
  81                        bool has_create_ino;
  82                        u64 ino;
  83                };
  84        };
  85
  86        /* encoded blob describing snapshot contexts for certain
  87           operations (e.g., open) */
  88        void *snapblob;
  89        int snapblob_len;
  90};
  91
  92
  93/*
  94 * cap releases are batched and sent to the MDS en masse.
  95 */
  96#define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE -                       \
  97                                sizeof(struct ceph_mds_cap_release)) /  \
  98                               sizeof(struct ceph_mds_cap_item))
  99
 100
 101/*
 102 * state associated with each MDS<->client session
 103 */
 104enum {
 105        CEPH_MDS_SESSION_NEW = 1,
 106        CEPH_MDS_SESSION_OPENING = 2,
 107        CEPH_MDS_SESSION_OPEN = 3,
 108        CEPH_MDS_SESSION_HUNG = 4,
 109        CEPH_MDS_SESSION_CLOSING = 5,
 110        CEPH_MDS_SESSION_RESTARTING = 6,
 111        CEPH_MDS_SESSION_RECONNECTING = 7,
 112};
 113
 114struct ceph_mds_session {
 115        struct ceph_mds_client *s_mdsc;
 116        int               s_mds;
 117        int               s_state;
 118        unsigned long     s_ttl;      /* time until mds kills us */
 119        u64               s_seq;      /* incoming msg seq # */
 120        struct mutex      s_mutex;    /* serialize session messages */
 121
 122        struct ceph_connection s_con;
 123
 124        struct ceph_auth_handshake s_auth;
 125
 126        /* protected by s_gen_ttl_lock */
 127        spinlock_t        s_gen_ttl_lock;
 128        u32               s_cap_gen;  /* inc each time we get mds stale msg */
 129        unsigned long     s_cap_ttl;  /* when session caps expire */
 130
 131        /* protected by s_cap_lock */
 132        spinlock_t        s_cap_lock;
 133        struct list_head  s_caps;     /* all caps issued by this session */
 134        int               s_nr_caps, s_trim_caps;
 135        int               s_num_cap_releases;
 136        int               s_cap_reconnect;
 137        struct list_head  s_cap_releases; /* waiting cap_release messages */
 138        struct list_head  s_cap_releases_done; /* ready to send */
 139        struct ceph_cap  *s_cap_iterator;
 140
 141        /* protected by mutex */
 142        struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
 143        struct list_head  s_cap_snaps_flushing;
 144        unsigned long     s_renew_requested; /* last time we sent a renew req */
 145        u64               s_renew_seq;
 146
 147        atomic_t          s_ref;
 148        struct list_head  s_waiting;  /* waiting requests */
 149        struct list_head  s_unsafe;   /* unsafe requests */
 150};
 151
 152/*
 153 * modes of choosing which MDS to send a request to
 154 */
 155enum {
 156        USE_ANY_MDS,
 157        USE_RANDOM_MDS,
 158        USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
 159};
 160
 161struct ceph_mds_request;
 162struct ceph_mds_client;
 163
 164/*
 165 * request completion callback
 166 */
 167typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
 168                                             struct ceph_mds_request *req);
 169
 170/*
 171 * an in-flight mds request
 172 */
 173struct ceph_mds_request {
 174        u64 r_tid;                   /* transaction id */
 175        struct rb_node r_node;
 176        struct ceph_mds_client *r_mdsc;
 177
 178        int r_op;                    /* mds op code */
 179
 180        /* operation on what? */
 181        struct inode *r_inode;              /* arg1 */
 182        struct dentry *r_dentry;            /* arg1 */
 183        struct dentry *r_old_dentry;        /* arg2: rename from or link from */
 184        struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
 185        char *r_path1, *r_path2;
 186        struct ceph_vino r_ino1, r_ino2;
 187
 188        struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
 189        struct inode *r_target_inode;       /* resulting inode */
 190
 191        struct mutex r_fill_mutex;
 192
 193        union ceph_mds_request_args r_args;
 194        int r_fmode;        /* file mode, if expecting cap */
 195        kuid_t r_uid;
 196        kgid_t r_gid;
 197        struct timespec r_stamp;
 198
 199        /* for choosing which mds to send this request to */
 200        int r_direct_mode;
 201        u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
 202        bool r_direct_is_hash;  /* true if r_direct_hash is valid */
 203
 204        /* data payload is used for xattr ops */
 205        struct page **r_pages;
 206        int r_num_pages;
 207        int r_data_len;
 208
 209        /* what caps shall we drop? */
 210        int r_inode_drop, r_inode_unless;
 211        int r_dentry_drop, r_dentry_unless;
 212        int r_old_dentry_drop, r_old_dentry_unless;
 213        struct inode *r_old_inode;
 214        int r_old_inode_drop, r_old_inode_unless;
 215
 216        struct ceph_msg  *r_request;  /* original request */
 217        int r_request_release_offset;
 218        struct ceph_msg  *r_reply;
 219        struct ceph_mds_reply_info_parsed r_reply_info;
 220        int r_err;
 221        bool r_aborted;
 222
 223        unsigned long r_timeout;  /* optional.  jiffies */
 224        unsigned long r_started;  /* start time to measure timeout against */
 225        unsigned long r_request_started; /* start time for mds request only,
 226                                            used to measure lease durations */
 227
 228        /* link unsafe requests to parent directory, for fsync */
 229        struct inode    *r_unsafe_dir;
 230        struct list_head r_unsafe_dir_item;
 231
 232        struct ceph_mds_session *r_session;
 233
 234        int               r_attempts;   /* resend attempts */
 235        int               r_num_fwd;    /* number of forward attempts */
 236        int               r_resend_mds; /* mds to resend to next, if any*/
 237        u32               r_sent_on_mseq; /* cap mseq request was sent at*/
 238
 239        struct kref       r_kref;
 240        struct list_head  r_wait;
 241        struct completion r_completion;
 242        struct completion r_safe_completion;
 243        ceph_mds_request_callback_t r_callback;
 244        struct list_head  r_unsafe_item;  /* per-session unsafe list item */
 245        bool              r_got_unsafe, r_got_safe, r_got_result;
 246
 247        bool              r_did_prepopulate;
 248        u32               r_readdir_offset;
 249
 250        struct ceph_cap_reservation r_caps_reservation;
 251        int r_num_caps;
 252};
 253
 254/*
 255 * mds client state
 256 */
 257struct ceph_mds_client {
 258        struct ceph_fs_client  *fsc;
 259        struct mutex            mutex;         /* all nested structures */
 260
 261        struct ceph_mdsmap      *mdsmap;
 262        struct completion       safe_umount_waiters;
 263        wait_queue_head_t       session_close_wq;
 264        struct list_head        waiting_for_map;
 265
 266        struct ceph_mds_session **sessions;    /* NULL for mds if no session */
 267        int                     max_sessions;  /* len of s_mds_sessions */
 268        int                     stopping;      /* true if shutting down */
 269
 270        /*
 271         * snap_rwsem will cover cap linkage into snaprealms, and
 272         * realm snap contexts.  (later, we can do per-realm snap
 273         * contexts locks..)  the empty list contains realms with no
 274         * references (implying they contain no inodes with caps) that
 275         * should be destroyed.
 276         */
 277        struct rw_semaphore     snap_rwsem;
 278        struct rb_root          snap_realms;
 279        struct list_head        snap_empty;
 280        spinlock_t              snap_empty_lock;  /* protect snap_empty */
 281
 282        u64                    last_tid;      /* most recent mds request */
 283        struct rb_root         request_tree;  /* pending mds requests */
 284        struct delayed_work    delayed_work;  /* delayed work */
 285        unsigned long    last_renew_caps;  /* last time we renewed our caps */
 286        struct list_head cap_delay_list;   /* caps with delayed release */
 287        spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
 288        struct list_head snap_flush_list;  /* cap_snaps ready to flush */
 289        spinlock_t       snap_flush_lock;
 290
 291        u64               cap_flush_seq;
 292        struct list_head  cap_dirty;        /* inodes with dirty caps */
 293        struct list_head  cap_dirty_migrating; /* ...that are migration... */
 294        int               num_cap_flushing; /* # caps we are flushing */
 295        spinlock_t        cap_dirty_lock;   /* protects above items */
 296        wait_queue_head_t cap_flushing_wq;
 297
 298        /*
 299         * Cap reservations
 300         *
 301         * Maintain a global pool of preallocated struct ceph_caps, referenced
 302         * by struct ceph_caps_reservations.  This ensures that we preallocate
 303         * memory needed to successfully process an MDS response.  (If an MDS
 304         * sends us cap information and we fail to process it, we will have
 305         * problems due to the client and MDS being out of sync.)
 306         *
 307         * Reservations are 'owned' by a ceph_cap_reservation context.
 308         */
 309        spinlock_t      caps_list_lock;
 310        struct          list_head caps_list; /* unused (reserved or
 311                                                unreserved) */
 312        int             caps_total_count;    /* total caps allocated */
 313        int             caps_use_count;      /* in use */
 314        int             caps_reserve_count;  /* unused, reserved */
 315        int             caps_avail_count;    /* unused, unreserved */
 316        int             caps_min_count;      /* keep at least this many
 317                                                (unreserved) */
 318        spinlock_t        dentry_lru_lock;
 319        struct list_head  dentry_lru;
 320        int               num_dentry;
 321};
 322
 323extern const char *ceph_mds_op_name(int op);
 324
 325extern struct ceph_mds_session *
 326__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
 327
 328static inline struct ceph_mds_session *
 329ceph_get_mds_session(struct ceph_mds_session *s)
 330{
 331        atomic_inc(&s->s_ref);
 332        return s;
 333}
 334
 335extern void ceph_put_mds_session(struct ceph_mds_session *s);
 336
 337extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
 338                             struct ceph_msg *msg, int mds);
 339
 340extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
 341extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
 342extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
 343
 344extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
 345
 346extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
 347                                    struct inode *inode,
 348                                    struct dentry *dn);
 349
 350extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
 351extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
 352                                           struct inode *dir);
 353extern struct ceph_mds_request *
 354ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
 355extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
 356                                     struct ceph_mds_request *req);
 357extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
 358                                struct inode *dir,
 359                                struct ceph_mds_request *req);
 360static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
 361{
 362        kref_get(&req->r_kref);
 363}
 364extern void ceph_mdsc_release_request(struct kref *kref);
 365static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
 366{
 367        kref_put(&req->r_kref, ceph_mdsc_release_request);
 368}
 369
 370extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
 371                                 struct ceph_mds_session *session);
 372extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
 373                                   struct ceph_mds_session *session);
 374
 375extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
 376
 377extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
 378                                  int stop_on_nosnap);
 379
 380extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
 381extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
 382                                     struct inode *inode,
 383                                     struct dentry *dentry, char action,
 384                                     u32 seq);
 385
 386extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
 387                                 struct ceph_msg *msg);
 388
 389extern struct ceph_mds_session *
 390ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
 391extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 392                                          struct ceph_mds_session *session);
 393
 394#endif
 395