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        u64 inline_version;
  45        u32 inline_len;
  46        char *inline_data;
  47        u32 pool_ns_len;
  48};
  49
  50/*
  51 * parsed info about an mds reply, including information about
  52 * either: 1) the target inode and/or its parent directory and dentry,
  53 * and directory contents (for readdir results), or
  54 * 2) the file range lock info (for fcntl F_GETLK results).
  55 */
  56struct ceph_mds_reply_info_parsed {
  57        struct ceph_mds_reply_head    *head;
  58
  59        /* trace */
  60        struct ceph_mds_reply_info_in diri, targeti;
  61        struct ceph_mds_reply_dirfrag *dirfrag;
  62        char                          *dname;
  63        u32                           dname_len;
  64        struct ceph_mds_reply_lease   *dlease;
  65
  66        /* extra */
  67        union {
  68                /* for fcntl F_GETLK results */
  69                struct ceph_filelock *filelock_reply;
  70
  71                /* for readdir results */
  72                struct {
  73                        struct ceph_mds_reply_dirfrag *dir_dir;
  74                        size_t                        dir_buf_size;
  75                        int                           dir_nr;
  76                        char                          **dir_dname;
  77                        u32                           *dir_dname_len;
  78                        struct ceph_mds_reply_lease   **dir_dlease;
  79                        struct ceph_mds_reply_info_in *dir_in;
  80                        u8                            dir_complete, dir_end;
  81                };
  82
  83                /* for create results */
  84                struct {
  85                        bool has_create_ino;
  86                        u64 ino;
  87                };
  88        };
  89
  90        /* encoded blob describing snapshot contexts for certain
  91           operations (e.g., open) */
  92        void *snapblob;
  93        int snapblob_len;
  94};
  95
  96
  97/*
  98 * cap releases are batched and sent to the MDS en masse.
  99 */
 100#define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE -                     \
 101                                sizeof(struct ceph_mds_cap_release)) /  \
 102                               sizeof(struct ceph_mds_cap_item))
 103
 104
 105/*
 106 * state associated with each MDS<->client session
 107 */
 108enum {
 109        CEPH_MDS_SESSION_NEW = 1,
 110        CEPH_MDS_SESSION_OPENING = 2,
 111        CEPH_MDS_SESSION_OPEN = 3,
 112        CEPH_MDS_SESSION_HUNG = 4,
 113        CEPH_MDS_SESSION_CLOSING = 5,
 114        CEPH_MDS_SESSION_RESTARTING = 6,
 115        CEPH_MDS_SESSION_RECONNECTING = 7,
 116};
 117
 118struct ceph_mds_session {
 119        struct ceph_mds_client *s_mdsc;
 120        int               s_mds;
 121        int               s_state;
 122        unsigned long     s_ttl;      /* time until mds kills us */
 123        u64               s_seq;      /* incoming msg seq # */
 124        struct mutex      s_mutex;    /* serialize session messages */
 125
 126        struct ceph_connection s_con;
 127
 128        struct ceph_auth_handshake s_auth;
 129
 130        /* protected by s_gen_ttl_lock */
 131        spinlock_t        s_gen_ttl_lock;
 132        u32               s_cap_gen;  /* inc each time we get mds stale msg */
 133        unsigned long     s_cap_ttl;  /* when session caps expire */
 134
 135        /* protected by s_cap_lock */
 136        spinlock_t        s_cap_lock;
 137        struct list_head  s_caps;     /* all caps issued by this session */
 138        int               s_nr_caps, s_trim_caps;
 139        int               s_num_cap_releases;
 140        int               s_cap_reconnect;
 141        int               s_readonly;
 142        struct list_head  s_cap_releases; /* waiting cap_release messages */
 143        struct ceph_cap  *s_cap_iterator;
 144
 145        /* protected by mutex */
 146        struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
 147        struct list_head  s_cap_snaps_flushing;
 148        unsigned long     s_renew_requested; /* last time we sent a renew req */
 149        u64               s_renew_seq;
 150
 151        atomic_t          s_ref;
 152        struct list_head  s_waiting;  /* waiting requests */
 153        struct list_head  s_unsafe;   /* unsafe requests */
 154};
 155
 156/*
 157 * modes of choosing which MDS to send a request to
 158 */
 159enum {
 160        USE_ANY_MDS,
 161        USE_RANDOM_MDS,
 162        USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
 163};
 164
 165struct ceph_mds_request;
 166struct ceph_mds_client;
 167
 168/*
 169 * request completion callback
 170 */
 171typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
 172                                             struct ceph_mds_request *req);
 173/*
 174 * wait for request completion callback
 175 */
 176typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
 177                                                 struct ceph_mds_request *req);
 178
 179/*
 180 * an in-flight mds request
 181 */
 182struct ceph_mds_request {
 183        u64 r_tid;                   /* transaction id */
 184        struct rb_node r_node;
 185        struct ceph_mds_client *r_mdsc;
 186
 187        int r_op;                    /* mds op code */
 188
 189        /* operation on what? */
 190        struct inode *r_inode;              /* arg1 */
 191        struct dentry *r_dentry;            /* arg1 */
 192        struct dentry *r_old_dentry;        /* arg2: rename from or link from */
 193        struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
 194        char *r_path1, *r_path2;
 195        struct ceph_vino r_ino1, r_ino2;
 196
 197        struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
 198        struct inode *r_target_inode;       /* resulting inode */
 199
 200        struct mutex r_fill_mutex;
 201
 202        union ceph_mds_request_args r_args;
 203        int r_fmode;        /* file mode, if expecting cap */
 204        kuid_t r_uid;
 205        kgid_t r_gid;
 206        struct timespec r_stamp;
 207
 208        /* for choosing which mds to send this request to */
 209        int r_direct_mode;
 210        u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
 211        bool r_direct_is_hash;  /* true if r_direct_hash is valid */
 212
 213        /* data payload is used for xattr ops */
 214        struct ceph_pagelist *r_pagelist;
 215
 216        /* what caps shall we drop? */
 217        int r_inode_drop, r_inode_unless;
 218        int r_dentry_drop, r_dentry_unless;
 219        int r_old_dentry_drop, r_old_dentry_unless;
 220        struct inode *r_old_inode;
 221        int r_old_inode_drop, r_old_inode_unless;
 222
 223        struct ceph_msg  *r_request;  /* original request */
 224        int r_request_release_offset;
 225        struct ceph_msg  *r_reply;
 226        struct ceph_mds_reply_info_parsed r_reply_info;
 227        struct page *r_locked_page;
 228        int r_err;
 229        bool r_aborted;
 230
 231        unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
 232        unsigned long r_started;  /* start time to measure timeout against */
 233        unsigned long r_request_started; /* start time for mds request only,
 234                                            used to measure lease durations */
 235
 236        /* link unsafe requests to parent directory, for fsync */
 237        struct inode    *r_unsafe_dir;
 238        struct list_head r_unsafe_dir_item;
 239
 240        /* unsafe requests that modify the target inode */
 241        struct list_head r_unsafe_target_item;
 242
 243        struct ceph_mds_session *r_session;
 244
 245        int               r_attempts;   /* resend attempts */
 246        int               r_num_fwd;    /* number of forward attempts */
 247        int               r_resend_mds; /* mds to resend to next, if any*/
 248        u32               r_sent_on_mseq; /* cap mseq request was sent at*/
 249
 250        struct kref       r_kref;
 251        struct list_head  r_wait;
 252        struct completion r_completion;
 253        struct completion r_safe_completion;
 254        ceph_mds_request_callback_t r_callback;
 255        ceph_mds_request_wait_callback_t r_wait_for_completion;
 256        struct list_head  r_unsafe_item;  /* per-session unsafe list item */
 257        bool              r_got_unsafe, r_got_safe, r_got_result;
 258
 259        bool              r_did_prepopulate;
 260        long long         r_dir_release_cnt;
 261        long long         r_dir_ordered_cnt;
 262        int               r_readdir_cache_idx;
 263        u32               r_readdir_offset;
 264
 265        struct ceph_cap_reservation r_caps_reservation;
 266        int r_num_caps;
 267};
 268
 269struct ceph_pool_perm {
 270        struct rb_node node;
 271        u32 pool;
 272        int perm;
 273};
 274
 275/*
 276 * mds client state
 277 */
 278struct ceph_mds_client {
 279        struct ceph_fs_client  *fsc;
 280        struct mutex            mutex;         /* all nested structures */
 281
 282        struct ceph_mdsmap      *mdsmap;
 283        struct completion       safe_umount_waiters;
 284        wait_queue_head_t       session_close_wq;
 285        struct list_head        waiting_for_map;
 286
 287        struct ceph_mds_session **sessions;    /* NULL for mds if no session */
 288        atomic_t                num_sessions;
 289        int                     max_sessions;  /* len of s_mds_sessions */
 290        int                     stopping;      /* true if shutting down */
 291
 292        /*
 293         * snap_rwsem will cover cap linkage into snaprealms, and
 294         * realm snap contexts.  (later, we can do per-realm snap
 295         * contexts locks..)  the empty list contains realms with no
 296         * references (implying they contain no inodes with caps) that
 297         * should be destroyed.
 298         */
 299        u64                     last_snap_seq;
 300        struct rw_semaphore     snap_rwsem;
 301        struct rb_root          snap_realms;
 302        struct list_head        snap_empty;
 303        spinlock_t              snap_empty_lock;  /* protect snap_empty */
 304
 305        u64                    last_tid;      /* most recent mds request */
 306        u64                    oldest_tid;    /* oldest incomplete mds request,
 307                                                 excluding setfilelock requests */
 308        struct rb_root         request_tree;  /* pending mds requests */
 309        struct delayed_work    delayed_work;  /* delayed work */
 310        unsigned long    last_renew_caps;  /* last time we renewed our caps */
 311        struct list_head cap_delay_list;   /* caps with delayed release */
 312        spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
 313        struct list_head snap_flush_list;  /* cap_snaps ready to flush */
 314        spinlock_t       snap_flush_lock;
 315
 316        u64               last_cap_flush_tid;
 317        struct rb_root    cap_flush_tree;
 318        struct list_head  cap_dirty;        /* inodes with dirty caps */
 319        struct list_head  cap_dirty_migrating; /* ...that are migration... */
 320        int               num_cap_flushing; /* # caps we are flushing */
 321        spinlock_t        cap_dirty_lock;   /* protects above items */
 322        wait_queue_head_t cap_flushing_wq;
 323
 324        /*
 325         * Cap reservations
 326         *
 327         * Maintain a global pool of preallocated struct ceph_caps, referenced
 328         * by struct ceph_caps_reservations.  This ensures that we preallocate
 329         * memory needed to successfully process an MDS response.  (If an MDS
 330         * sends us cap information and we fail to process it, we will have
 331         * problems due to the client and MDS being out of sync.)
 332         *
 333         * Reservations are 'owned' by a ceph_cap_reservation context.
 334         */
 335        spinlock_t      caps_list_lock;
 336        struct          list_head caps_list; /* unused (reserved or
 337                                                unreserved) */
 338        int             caps_total_count;    /* total caps allocated */
 339        int             caps_use_count;      /* in use */
 340        int             caps_reserve_count;  /* unused, reserved */
 341        int             caps_avail_count;    /* unused, unreserved */
 342        int             caps_min_count;      /* keep at least this many
 343                                                (unreserved) */
 344        spinlock_t        dentry_lru_lock;
 345        struct list_head  dentry_lru;
 346        int               num_dentry;
 347
 348        struct rw_semaphore     pool_perm_rwsem;
 349        struct rb_root          pool_perm_tree;
 350};
 351
 352extern const char *ceph_mds_op_name(int op);
 353
 354extern struct ceph_mds_session *
 355__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
 356
 357static inline struct ceph_mds_session *
 358ceph_get_mds_session(struct ceph_mds_session *s)
 359{
 360        atomic_inc(&s->s_ref);
 361        return s;
 362}
 363
 364extern const char *ceph_session_state_name(int s);
 365
 366extern void ceph_put_mds_session(struct ceph_mds_session *s);
 367
 368extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
 369                             struct ceph_msg *msg, int mds);
 370
 371extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
 372extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
 373extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
 374extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
 375
 376extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
 377
 378extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
 379                                    struct inode *inode,
 380                                    struct dentry *dn);
 381
 382extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
 383extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
 384                                           struct inode *dir);
 385extern struct ceph_mds_request *
 386ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
 387extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
 388                                     struct ceph_mds_request *req);
 389extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
 390                                struct inode *dir,
 391                                struct ceph_mds_request *req);
 392static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
 393{
 394        kref_get(&req->r_kref);
 395}
 396extern void ceph_mdsc_release_request(struct kref *kref);
 397static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
 398{
 399        kref_put(&req->r_kref, ceph_mdsc_release_request);
 400}
 401
 402extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
 403                                   struct ceph_mds_session *session);
 404
 405extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
 406
 407extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
 408                                  int stop_on_nosnap);
 409
 410extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
 411extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
 412                                     struct inode *inode,
 413                                     struct dentry *dentry, char action,
 414                                     u32 seq);
 415
 416extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
 417                                 struct ceph_msg *msg);
 418
 419extern struct ceph_mds_session *
 420ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
 421extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 422                                          struct ceph_mds_session *session);
 423
 424#endif
 425