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