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 #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
  30 struct ceph_fs_client;
  31 struct 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  */
  37 struct 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 struct ceph_mds_reply_dir_entry {
  51         char                          *name;
  52         u32                           name_len;
  53         struct ceph_mds_reply_lease   *lease;
  54         struct ceph_mds_reply_info_in inode;
  55 };
  56
  57 /*
  58  * parsed info about an mds reply, including information about
  59  * either: 1) the target inode and/or its parent directory and dentry,
  60  * and directory contents (for readdir results), or
  61  * 2) the file range lock info (for fcntl F_GETLK results).
  62  */
  63 struct ceph_mds_reply_info_parsed {
  64         struct ceph_mds_reply_head    *head;
  65
  66         /* trace */
  67         struct ceph_mds_reply_info_in diri, targeti;
  68         struct ceph_mds_reply_dirfrag *dirfrag;
  69         char                          *dname;
  70         u32                           dname_len;
  71         struct ceph_mds_reply_lease   *dlease;
  72
  73         /* extra */
  74         union {
  75                 /* for fcntl F_GETLK results */
  76                 struct ceph_filelock *filelock_reply;
  77
  78                 /* for readdir results */
  79                 struct {
  80                         struct ceph_mds_reply_dirfrag *dir_dir;
  81                         size_t                        dir_buf_size;
  82                         int                           dir_nr;
  83                         u8                            dir_complete, dir_end;
  84                         struct ceph_mds_reply_dir_entry  *dir_entries;
  85                 };
  86
  87                 /* for create results */
  88                 struct {
  89                         bool has_create_ino;
  90                         u64 ino;
  91                 };
  92         };
  93
  94         /* encoded blob describing snapshot contexts for certain
  95            operations (e.g., open) */
  96         void *snapblob;
  97         int snapblob_len;
  98 };
  99
 100
 101 /*
 102  * cap releases are batched and sent to the MDS en masse.
 103  */
 104 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE -                     \
 105                                 sizeof(struct ceph_mds_cap_release)) /  \
 106                                sizeof(struct ceph_mds_cap_item))
 107
 108
 109 /*
 110  * state associated with each MDS<->client session
 111  */
 112 enum {
 113         CEPH_MDS_SESSION_NEW = 1,
 114         CEPH_MDS_SESSION_OPENING = 2,
 115         CEPH_MDS_SESSION_OPEN = 3,
 116         CEPH_MDS_SESSION_HUNG = 4,
 117         CEPH_MDS_SESSION_CLOSING = 5,
 118         CEPH_MDS_SESSION_RESTARTING = 6,
 119         CEPH_MDS_SESSION_RECONNECTING = 7,
 120 };
 121
 122 struct ceph_mds_session {
 123         struct ceph_mds_client *s_mdsc;
 124         int               s_mds;
 125         int               s_state;
 126         unsigned long     s_ttl;      /* time until mds kills us */
 127         u64               s_seq;      /* incoming msg seq # */
 128         struct mutex      s_mutex;    /* serialize session messages */
 129
 130         struct ceph_connection s_con;
 131
 132         struct ceph_auth_handshake s_auth;
 133
 134         /* protected by s_gen_ttl_lock */
 135         spinlock_t        s_gen_ttl_lock;
 136         u32               s_cap_gen;  /* inc each time we get mds stale msg */
 137         unsigned long     s_cap_ttl;  /* when session caps expire */
 138
 139         /* protected by s_cap_lock */
 140         spinlock_t        s_cap_lock;
 141         struct list_head  s_caps;     /* all caps issued by this session */
 142         int               s_nr_caps, s_trim_caps;
 143         int               s_num_cap_releases;
 144         int               s_cap_reconnect;
 145         int               s_readonly;
 146         struct list_head  s_cap_releases; /* waiting cap_release messages */
 147         struct ceph_cap  *s_cap_iterator;
 148
 149         /* protected by mutex */
 150         struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
 151         struct list_head  s_cap_snaps_flushing;
 152         unsigned long     s_renew_requested; /* last time we sent a renew req */
 153         u64               s_renew_seq;
 154
 155         atomic_t          s_ref;
 156         struct list_head  s_waiting;  /* waiting requests */
 157         struct list_head  s_unsafe;   /* unsafe requests */
 158 };
 159
 160 /*
 161  * modes of choosing which MDS to send a request to
 162  */
 163 enum {
 164         USE_ANY_MDS,
 165         USE_RANDOM_MDS,
 166         USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
 167 };
 168
 169 struct ceph_mds_request;
 170 struct ceph_mds_client;
 171
 172 /*
 173  * request completion callback
 174  */
 175 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
 176                                              struct ceph_mds_request *req);
 177 /*
 178  * wait for request completion callback
 179  */
 180 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
 181                                                  struct ceph_mds_request *req);
 182
 183 /*
 184  * an in-flight mds request
 185  */
 186 struct ceph_mds_request {
 187         u64 r_tid;                   /* transaction id */
 188         struct rb_node r_node;
 189         struct ceph_mds_client *r_mdsc;
 190
 191         int r_op;                    /* mds op code */
 192
 193         /* operation on what? */
 194         struct inode *r_inode;              /* arg1 */
 195         struct dentry *r_dentry;            /* arg1 */
 196         struct dentry *r_old_dentry;        /* arg2: rename from or link from */
 197         struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
 198         char *r_path1, *r_path2;
 199         struct ceph_vino r_ino1, r_ino2;
 200
 201         struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
 202         struct inode *r_target_inode;       /* resulting inode */
 203
 204         struct mutex r_fill_mutex;
 205
 206         union ceph_mds_request_args r_args;
 207         int r_fmode;        /* file mode, if expecting cap */
 208         kuid_t r_uid;
 209         kgid_t r_gid;
 210         struct timespec r_stamp;
 211
 212         /* for choosing which mds to send this request to */
 213         int r_direct_mode;
 214         u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
 215         bool r_direct_is_hash;  /* true if r_direct_hash is valid */
 216
 217         /* data payload is used for xattr ops */
 218         struct ceph_pagelist *r_pagelist;
 219
 220         /* what caps shall we drop? */
 221         int r_inode_drop, r_inode_unless;
 222         int r_dentry_drop, r_dentry_unless;
 223         int r_old_dentry_drop, r_old_dentry_unless;
 224         struct inode *r_old_inode;
 225         int r_old_inode_drop, r_old_inode_unless;
 226
 227         struct ceph_msg  *r_request;  /* original request */
 228         int r_request_release_offset;
 229         struct ceph_msg  *r_reply;
 230         struct ceph_mds_reply_info_parsed r_reply_info;
 231         struct page *r_locked_page;
 232         int r_err;
 233         bool r_aborted;
 234
 235         unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
 236         unsigned long r_started;  /* start time to measure timeout against */
 237         unsigned long r_request_started; /* start time for mds request only,
 238                                             used to measure lease durations */
 239
 240         /* link unsafe requests to parent directory, for fsync */
 241         struct inode    *r_unsafe_dir;
 242         struct list_head r_unsafe_dir_item;
 243
 244         /* unsafe requests that modify the target inode */
 245         struct list_head r_unsafe_target_item;
 246
 247         struct ceph_mds_session *r_session;
 248
 249         int               r_attempts;   /* resend attempts */
 250         int               r_num_fwd;    /* number of forward attempts */
 251         int               r_resend_mds; /* mds to resend to next, if any*/
 252         u32               r_sent_on_mseq; /* cap mseq request was sent at*/
 253
 254         struct kref       r_kref;
 255         struct list_head  r_wait;
 256         struct completion r_completion;
 257         struct completion r_safe_completion;
 258         ceph_mds_request_callback_t r_callback;
 259         ceph_mds_request_wait_callback_t r_wait_for_completion;
 260         struct list_head  r_unsafe_item;  /* per-session unsafe list item */
 261         bool              r_got_unsafe, r_got_safe, r_got_result;
 262
 263         bool              r_did_prepopulate;
 264         long long         r_dir_release_cnt;
 265         long long         r_dir_ordered_cnt;
 266         int               r_readdir_cache_idx;
 267         u32               r_readdir_offset;
 268
 269         struct ceph_cap_reservation r_caps_reservation;
 270         int r_num_caps;
 271 };
 272
 273 struct ceph_pool_perm {
 274         struct rb_node node;
 275         u32 pool;
 276         int perm;
 277 };
 278
 279 /*
 280  * mds client state
 281  */
 282 struct ceph_mds_client {
 283         struct ceph_fs_client  *fsc;
 284         struct mutex            mutex;         /* all nested structures */
 285
 286         struct ceph_mdsmap      *mdsmap;
 287         struct completion       safe_umount_waiters;
 288         wait_queue_head_t       session_close_wq;
 289         struct list_head        waiting_for_map;
 290
 291         struct ceph_mds_session **sessions;    /* NULL for mds if no session */
 292         atomic_t                num_sessions;
 293         int                     max_sessions;  /* len of s_mds_sessions */
 294         int                     stopping;      /* true if shutting down */
 295
 296         /*
 297          * snap_rwsem will cover cap linkage into snaprealms, and
 298          * realm snap contexts.  (later, we can do per-realm snap
 299          * contexts locks..)  the empty list contains realms with no
 300          * references (implying they contain no inodes with caps) that
 301          * should be destroyed.
 302          */
 303         u64                     last_snap_seq;
 304         struct rw_semaphore     snap_rwsem;
 305         struct rb_root          snap_realms;
 306         struct list_head        snap_empty;
 307         spinlock_t              snap_empty_lock;  /* protect snap_empty */
 308
 309         u64                    last_tid;      /* most recent mds request */
 310         u64                    oldest_tid;    /* oldest incomplete mds request,
 311                                                  excluding setfilelock requests */
 312         struct rb_root         request_tree;  /* pending mds requests */
 313         struct delayed_work    delayed_work;  /* delayed work */
 314         unsigned long    last_renew_caps;  /* last time we renewed our caps */
 315         struct list_head cap_delay_list;   /* caps with delayed release */
 316         spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
 317         struct list_head snap_flush_list;  /* cap_snaps ready to flush */
 318         spinlock_t       snap_flush_lock;
 319
 320         u64               last_cap_flush_tid;
 321         struct rb_root    cap_flush_tree;
 322         struct list_head  cap_dirty;        /* inodes with dirty caps */
 323         struct list_head  cap_dirty_migrating; /* ...that are migration... */
 324         int               num_cap_flushing; /* # caps we are flushing */
 325         spinlock_t        cap_dirty_lock;   /* protects above items */
 326         wait_queue_head_t cap_flushing_wq;
 327
 328         /*
 329          * Cap reservations
 330          *
 331          * Maintain a global pool of preallocated struct ceph_caps, referenced
 332          * by struct ceph_caps_reservations.  This ensures that we preallocate
 333          * memory needed to successfully process an MDS response.  (If an MDS
 334          * sends us cap information and we fail to process it, we will have
 335          * problems due to the client and MDS being out of sync.)
 336          *
 337          * Reservations are 'owned' by a ceph_cap_reservation context.
 338          */
 339         spinlock_t      caps_list_lock;
 340         struct          list_head caps_list; /* unused (reserved or
 341                                                 unreserved) */
 342         int             caps_total_count;    /* total caps allocated */
 343         int             caps_use_count;      /* in use */
 344         int             caps_reserve_count;  /* unused, reserved */
 345         int             caps_avail_count;    /* unused, unreserved */
 346         int             caps_min_count;      /* keep at least this many
 347                                                 (unreserved) */
 348         spinlock_t        dentry_lru_lock;
 349         struct list_head  dentry_lru;
 350         int               num_dentry;
 351
 352         struct rw_semaphore     pool_perm_rwsem;
 353         struct rb_root          pool_perm_tree;
 354 };
 355
 356 extern const char *ceph_mds_op_name(int op);
 357
 358 extern struct ceph_mds_session *
 359 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
 360
 361 static inline struct ceph_mds_session *
 362 ceph_get_mds_session(struct ceph_mds_session *s)
 363 {
 364         atomic_inc(&s->s_ref);
 365         return s;
 366 }
 367
 368 extern const char *ceph_session_state_name(int s);
 369
 370 extern void ceph_put_mds_session(struct ceph_mds_session *s);
 371
 372 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
 373                              struct ceph_msg *msg, int mds);
 374
 375 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
 376 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
 377 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
 378 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
 379
 380 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
 381
 382 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
 383                                     struct inode *inode,
 384                                     struct dentry *dn);
 385
 386 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
 387 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
 388                                            struct inode *dir);
 389 extern struct ceph_mds_request *
 390 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
 391 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
 392                                      struct ceph_mds_request *req);
 393 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
 394                                 struct inode *dir,
 395                                 struct ceph_mds_request *req);
 396 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
 397 {
 398         kref_get(&req->r_kref);
 399 }
 400 extern void ceph_mdsc_release_request(struct kref *kref);
 401 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
 402 {
 403         kref_put(&req->r_kref, ceph_mdsc_release_request);
 404 }
 405
 406 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
 407                                    struct ceph_mds_session *session);
 408
 409 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
 410
 411 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
 412                                   int stop_on_nosnap);
 413
 414 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
 415 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
 416                                      struct inode *inode,
 417                                      struct dentry *dentry, char action,
 418                                      u32 seq);
 419
 420 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
 421                                  struct ceph_msg *msg);
 422
 423 extern struct ceph_mds_session *
 424 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
 425 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 426                                           struct ceph_mds_session *session);
 427
 428 #endif