fs/ceph/mds_client.h

   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  *         inode->i_lock
  24  *                 mdsc->snap_flush_lock
  25  *                 mdsc->cap_delay_lock
  26  *
  27  */
  28
  29 struct ceph_fs_client;
  30 struct 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  */
  36 struct 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  */
  51 struct 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  */
  96 enum {
  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
 106 struct 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_cap_lock */
 121         spinlock_t        s_cap_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         struct list_head  s_caps;     /* all caps issued by this session */
 125         int               s_nr_caps, s_trim_caps;
 126         int               s_num_cap_releases;
 127         struct list_head  s_cap_releases; /* waiting cap_release messages */
 128         struct list_head  s_cap_releases_done; /* ready to send */
 129         struct ceph_cap  *s_cap_iterator;
 130
 131         /* protected by mutex */
 132         struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
 133         struct list_head  s_cap_snaps_flushing;
 134         unsigned long     s_renew_requested; /* last time we sent a renew req */
 135         u64               s_renew_seq;
 136
 137         atomic_t          s_ref;
 138         struct list_head  s_waiting;  /* waiting requests */
 139         struct list_head  s_unsafe;   /* unsafe requests */
 140 };
 141
 142 /*
 143  * modes of choosing which MDS to send a request to
 144  */
 145 enum {
 146         USE_ANY_MDS,
 147         USE_RANDOM_MDS,
 148         USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
 149 };
 150
 151 struct ceph_mds_request;
 152 struct ceph_mds_client;
 153
 154 /*
 155  * request completion callback
 156  */
 157 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
 158                                              struct ceph_mds_request *req);
 159
 160 /*
 161  * an in-flight mds request
 162  */
 163 struct ceph_mds_request {
 164         u64 r_tid;                   /* transaction id */
 165         struct rb_node r_node;
 166         struct ceph_mds_client *r_mdsc;
 167
 168         int r_op;                    /* mds op code */
 169
 170         /* operation on what? */
 171         struct inode *r_inode;              /* arg1 */
 172         struct dentry *r_dentry;            /* arg1 */
 173         struct dentry *r_old_dentry;        /* arg2: rename from or link from */
 174         char *r_path1, *r_path2;
 175         struct ceph_vino r_ino1, r_ino2;
 176
 177         struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
 178         struct inode *r_target_inode;       /* resulting inode */
 179
 180         struct mutex r_fill_mutex;
 181
 182         union ceph_mds_request_args r_args;
 183         int r_fmode;        /* file mode, if expecting cap */
 184         uid_t r_uid;
 185         gid_t r_gid;
 186
 187         /* for choosing which mds to send this request to */
 188         int r_direct_mode;
 189         u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
 190         bool r_direct_is_hash;  /* true if r_direct_hash is valid */
 191
 192         /* data payload is used for xattr ops */
 193         struct page **r_pages;
 194         int r_num_pages;
 195         int r_data_len;
 196
 197         /* what caps shall we drop? */
 198         int r_inode_drop, r_inode_unless;
 199         int r_dentry_drop, r_dentry_unless;
 200         int r_old_dentry_drop, r_old_dentry_unless;
 201         struct inode *r_old_inode;
 202         int r_old_inode_drop, r_old_inode_unless;
 203
 204         struct ceph_msg  *r_request;  /* original request */
 205         int r_request_release_offset;
 206         struct ceph_msg  *r_reply;
 207         struct ceph_mds_reply_info_parsed r_reply_info;
 208         int r_err;
 209         bool r_aborted;
 210
 211         unsigned long r_timeout;  /* optional.  jiffies */
 212         unsigned long r_started;  /* start time to measure timeout against */
 213         unsigned long r_request_started; /* start time for mds request only,
 214                                             used to measure lease durations */
 215
 216         /* link unsafe requests to parent directory, for fsync */
 217         struct inode    *r_unsafe_dir;
 218         struct list_head r_unsafe_dir_item;
 219
 220         struct ceph_mds_session *r_session;
 221
 222         int               r_attempts;   /* resend attempts */
 223         int               r_num_fwd;    /* number of forward attempts */
 224         int               r_resend_mds; /* mds to resend to next, if any*/
 225         u32               r_sent_on_mseq; /* cap mseq request was sent at*/
 226
 227         struct kref       r_kref;
 228         struct list_head  r_wait;
 229         struct completion r_completion;
 230         struct completion r_safe_completion;
 231         ceph_mds_request_callback_t r_callback;
 232         struct list_head  r_unsafe_item;  /* per-session unsafe list item */
 233         bool              r_got_unsafe, r_got_safe, r_got_result;
 234
 235         bool              r_did_prepopulate;
 236         u32               r_readdir_offset;
 237
 238         struct ceph_cap_reservation r_caps_reservation;
 239         int r_num_caps;
 240 };
 241
 242 /*
 243  * mds client state
 244  */
 245 struct ceph_mds_client {
 246         struct ceph_fs_client  *fsc;
 247         struct mutex            mutex;         /* all nested structures */
 248
 249         struct ceph_mdsmap      *mdsmap;
 250         struct completion       safe_umount_waiters;
 251         wait_queue_head_t       session_close_wq;
 252         struct list_head        waiting_for_map;
 253
 254         struct ceph_mds_session **sessions;    /* NULL for mds if no session */
 255         int                     max_sessions;  /* len of s_mds_sessions */
 256         int                     stopping;      /* true if shutting down */
 257
 258         /*
 259          * snap_rwsem will cover cap linkage into snaprealms, and
 260          * realm snap contexts.  (later, we can do per-realm snap
 261          * contexts locks..)  the empty list contains realms with no
 262          * references (implying they contain no inodes with caps) that
 263          * should be destroyed.
 264          */
 265         struct rw_semaphore     snap_rwsem;
 266         struct rb_root          snap_realms;
 267         struct list_head        snap_empty;
 268         spinlock_t              snap_empty_lock;  /* protect snap_empty */
 269
 270         u64                    last_tid;      /* most recent mds request */
 271         struct rb_root         request_tree;  /* pending mds requests */
 272         struct delayed_work    delayed_work;  /* delayed work */
 273         unsigned long    last_renew_caps;  /* last time we renewed our caps */
 274         struct list_head cap_delay_list;   /* caps with delayed release */
 275         spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
 276         struct list_head snap_flush_list;  /* cap_snaps ready to flush */
 277         spinlock_t       snap_flush_lock;
 278
 279         u64               cap_flush_seq;
 280         struct list_head  cap_dirty;        /* inodes with dirty caps */
 281         struct list_head  cap_dirty_migrating; /* ...that are migration... */
 282         int               num_cap_flushing; /* # caps we are flushing */
 283         spinlock_t        cap_dirty_lock;   /* protects above items */
 284         wait_queue_head_t cap_flushing_wq;
 285
 286         /*
 287          * Cap reservations
 288          *
 289          * Maintain a global pool of preallocated struct ceph_caps, referenced
 290          * by struct ceph_caps_reservations.  This ensures that we preallocate
 291          * memory needed to successfully process an MDS response.  (If an MDS
 292          * sends us cap information and we fail to process it, we will have
 293          * problems due to the client and MDS being out of sync.)
 294          *
 295          * Reservations are 'owned' by a ceph_cap_reservation context.
 296          */
 297         spinlock_t      caps_list_lock;
 298         struct          list_head caps_list; /* unused (reserved or
 299                                                 unreserved) */
 300         int             caps_total_count;    /* total caps allocated */
 301         int             caps_use_count;      /* in use */
 302         int             caps_reserve_count;  /* unused, reserved */
 303         int             caps_avail_count;    /* unused, unreserved */
 304         int             caps_min_count;      /* keep at least this many
 305                                                 (unreserved) */
 306         spinlock_t        dentry_lru_lock;
 307         struct list_head  dentry_lru;
 308         int               num_dentry;
 309 };
 310
 311 extern const char *ceph_mds_op_name(int op);
 312
 313 extern struct ceph_mds_session *
 314 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
 315
 316 static inline struct ceph_mds_session *
 317 ceph_get_mds_session(struct ceph_mds_session *s)
 318 {
 319         atomic_inc(&s->s_ref);
 320         return s;
 321 }
 322
 323 extern void ceph_put_mds_session(struct ceph_mds_session *s);
 324
 325 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
 326                              struct ceph_msg *msg, int mds);
 327
 328 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
 329 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
 330 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
 331
 332 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
 333
 334 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
 335                                     struct inode *inode,
 336                                     struct dentry *dn, int mask);
 337
 338 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
 339
 340 extern struct ceph_mds_request *
 341 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
 342 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
 343                                      struct ceph_mds_request *req);
 344 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
 345                                 struct inode *dir,
 346                                 struct ceph_mds_request *req);
 347 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
 348 {
 349         kref_get(&req->r_kref);
 350 }
 351 extern void ceph_mdsc_release_request(struct kref *kref);
 352 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
 353 {
 354         kref_put(&req->r_kref, ceph_mdsc_release_request);
 355 }
 356
 357 extern int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
 358                                  struct ceph_mds_session *session);
 359 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
 360                                    struct ceph_mds_session *session);
 361
 362 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
 363
 364 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
 365                                   int stop_on_nosnap);
 366
 367 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
 368 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
 369                                      struct inode *inode,
 370                                      struct dentry *dentry, char action,
 371                                      u32 seq);
 372
 373 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
 374                                  struct ceph_msg *msg);
 375
 376 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 377                                           struct ceph_mds_session *session);
 378
 379 #endif