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