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         char *pool_ns_data;
  49 };
  50
  51 struct ceph_mds_reply_dir_entry {
  52         char                          *name;
  53         u32                           name_len;
  54         struct ceph_mds_reply_lease   *lease;
  55         struct ceph_mds_reply_info_in inode;
  56         loff_t                        offset;
  57 };
  58
  59 /*
  60  * parsed info about an mds reply, including information about
  61  * either: 1) the target inode and/or its parent directory and dentry,
  62  * and directory contents (for readdir results), or
  63  * 2) the file range lock info (for fcntl F_GETLK results).
  64  */
  65 struct ceph_mds_reply_info_parsed {
  66         struct ceph_mds_reply_head    *head;
  67
  68         /* trace */
  69         struct ceph_mds_reply_info_in diri, targeti;
  70         struct ceph_mds_reply_dirfrag *dirfrag;
  71         char                          *dname;
  72         u32                           dname_len;
  73         struct ceph_mds_reply_lease   *dlease;
  74
  75         /* extra */
  76         union {
  77                 /* for fcntl F_GETLK results */
  78                 struct ceph_filelock *filelock_reply;
  79
  80                 /* for readdir results */
  81                 struct {
  82                         struct ceph_mds_reply_dirfrag *dir_dir;
  83                         size_t                        dir_buf_size;
  84                         int                           dir_nr;
  85                         bool                          dir_complete;
  86                         bool                          dir_end;
  87                         bool                          hash_order;
  88                         struct ceph_mds_reply_dir_entry  *dir_entries;
  89                 };
  90
  91                 /* for create results */
  92                 struct {
  93                         bool has_create_ino;
  94                         u64 ino;
  95                 };
  96         };
  97
  98         /* encoded blob describing snapshot contexts for certain
  99            operations (e.g., open) */
 100         void *snapblob;
 101         int snapblob_len;
 102 };
 103
 104
 105 /*
 106  * cap releases are batched and sent to the MDS en masse.
 107  */
 108 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE -                     \
 109                                 sizeof(struct ceph_mds_cap_release)) /  \
 110                                sizeof(struct ceph_mds_cap_item))
 111
 112
 113 /*
 114  * state associated with each MDS<->client session
 115  */
 116 enum {
 117         CEPH_MDS_SESSION_NEW = 1,
 118         CEPH_MDS_SESSION_OPENING = 2,
 119         CEPH_MDS_SESSION_OPEN = 3,
 120         CEPH_MDS_SESSION_HUNG = 4,
 121         CEPH_MDS_SESSION_CLOSING = 5,
 122         CEPH_MDS_SESSION_RESTARTING = 6,
 123         CEPH_MDS_SESSION_RECONNECTING = 7,
 124         CEPH_MDS_SESSION_REJECTED = 8,
 125 };
 126
 127 struct ceph_mds_session {
 128         struct ceph_mds_client *s_mdsc;
 129         int               s_mds;
 130         int               s_state;
 131         unsigned long     s_ttl;      /* time until mds kills us */
 132         u64               s_seq;      /* incoming msg seq # */
 133         struct mutex      s_mutex;    /* serialize session messages */
 134
 135         struct ceph_connection s_con;
 136
 137         struct ceph_auth_handshake s_auth;
 138
 139         /* protected by s_gen_ttl_lock */
 140         spinlock_t        s_gen_ttl_lock;
 141         u32               s_cap_gen;  /* inc each time we get mds stale msg */
 142         unsigned long     s_cap_ttl;  /* when session caps expire */
 143
 144         /* protected by s_cap_lock */
 145         spinlock_t        s_cap_lock;
 146         struct list_head  s_caps;     /* all caps issued by this session */
 147         int               s_nr_caps, s_trim_caps;
 148         int               s_num_cap_releases;
 149         int               s_cap_reconnect;
 150         int               s_readonly;
 151         struct list_head  s_cap_releases; /* waiting cap_release messages */
 152         struct ceph_cap  *s_cap_iterator;
 153
 154         /* protected by mutex */
 155         struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
 156         unsigned long     s_renew_requested; /* last time we sent a renew req */
 157         u64               s_renew_seq;
 158
 159         atomic_t          s_ref;
 160         struct list_head  s_waiting;  /* waiting requests */
 161         struct list_head  s_unsafe;   /* unsafe requests */
 162 };
 163
 164 /*
 165  * modes of choosing which MDS to send a request to
 166  */
 167 enum {
 168         USE_ANY_MDS,
 169         USE_RANDOM_MDS,
 170         USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
 171 };
 172
 173 struct ceph_mds_request;
 174 struct ceph_mds_client;
 175
 176 /*
 177  * request completion callback
 178  */
 179 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
 180                                              struct ceph_mds_request *req);
 181 /*
 182  * wait for request completion callback
 183  */
 184 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
 185                                                  struct ceph_mds_request *req);
 186
 187 /*
 188  * an in-flight mds request
 189  */
 190 struct ceph_mds_request {
 191         u64 r_tid;                   /* transaction id */
 192         struct rb_node r_node;
 193         struct ceph_mds_client *r_mdsc;
 194
 195         int r_op;                    /* mds op code */
 196
 197         /* operation on what? */
 198         struct inode *r_inode;              /* arg1 */
 199         struct dentry *r_dentry;            /* arg1 */
 200         struct dentry *r_old_dentry;        /* arg2: rename from or link from */
 201         struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
 202         char *r_path1, *r_path2;
 203         struct ceph_vino r_ino1, r_ino2;
 204
 205         struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
 206         struct inode *r_target_inode;       /* resulting inode */
 207
 208         struct mutex r_fill_mutex;
 209
 210         union ceph_mds_request_args r_args;
 211         int r_fmode;        /* file mode, if expecting cap */
 212         kuid_t r_uid;
 213         kgid_t r_gid;
 214         struct timespec r_stamp;
 215
 216         /* for choosing which mds to send this request to */
 217         int r_direct_mode;
 218         u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
 219         bool r_direct_is_hash;  /* true if r_direct_hash is valid */
 220
 221         /* data payload is used for xattr ops */
 222         struct ceph_pagelist *r_pagelist;
 223
 224         /* what caps shall we drop? */
 225         int r_inode_drop, r_inode_unless;
 226         int r_dentry_drop, r_dentry_unless;
 227         int r_old_dentry_drop, r_old_dentry_unless;
 228         struct inode *r_old_inode;
 229         int r_old_inode_drop, r_old_inode_unless;
 230
 231         struct ceph_msg  *r_request;  /* original request */
 232         int r_request_release_offset;
 233         struct ceph_msg  *r_reply;
 234         struct ceph_mds_reply_info_parsed r_reply_info;
 235         struct page *r_locked_page;
 236         int r_err;
 237         bool r_aborted;
 238
 239         unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
 240         unsigned long r_started;  /* start time to measure timeout against */
 241         unsigned long r_request_started; /* start time for mds request only,
 242                                             used to measure lease durations */
 243
 244         /* link unsafe requests to parent directory, for fsync */
 245         struct inode    *r_unsafe_dir;
 246         struct list_head r_unsafe_dir_item;
 247
 248         /* unsafe requests that modify the target inode */
 249         struct list_head r_unsafe_target_item;
 250
 251         struct ceph_mds_session *r_session;
 252
 253         int               r_attempts;   /* resend attempts */
 254         int               r_num_fwd;    /* number of forward attempts */
 255         int               r_resend_mds; /* mds to resend to next, if any*/
 256         u32               r_sent_on_mseq; /* cap mseq request was sent at*/
 257
 258         struct kref       r_kref;
 259         struct list_head  r_wait;
 260         struct completion r_completion;
 261         struct completion r_safe_completion;
 262         ceph_mds_request_callback_t r_callback;
 263         ceph_mds_request_wait_callback_t r_wait_for_completion;
 264         struct list_head  r_unsafe_item;  /* per-session unsafe list item */
 265         bool              r_got_unsafe, r_got_safe, r_got_result;
 266
 267         bool              r_did_prepopulate;
 268         long long         r_dir_release_cnt;
 269         long long         r_dir_ordered_cnt;
 270         int               r_readdir_cache_idx;
 271         u32               r_readdir_offset;
 272
 273         struct ceph_cap_reservation r_caps_reservation;
 274         int r_num_caps;
 275 };
 276
 277 struct ceph_pool_perm {
 278         struct rb_node node;
 279         int perm;
 280         s64 pool;
 281         size_t pool_ns_len;
 282         char pool_ns[];
 283 };
 284
 285 /*
 286  * mds client state
 287  */
 288 struct ceph_mds_client {
 289         struct ceph_fs_client  *fsc;
 290         struct mutex            mutex;         /* all nested structures */
 291
 292         struct ceph_mdsmap      *mdsmap;
 293         struct completion       safe_umount_waiters;
 294         wait_queue_head_t       session_close_wq;
 295         struct list_head        waiting_for_map;
 296         int                     mdsmap_err;
 297
 298         struct ceph_mds_session **sessions;    /* NULL for mds if no session */
 299         atomic_t                num_sessions;
 300         int                     max_sessions;  /* len of s_mds_sessions */
 301         int                     stopping;      /* true if shutting down */
 302
 303         /*
 304          * snap_rwsem will cover cap linkage into snaprealms, and
 305          * realm snap contexts.  (later, we can do per-realm snap
 306          * contexts locks..)  the empty list contains realms with no
 307          * references (implying they contain no inodes with caps) that
 308          * should be destroyed.
 309          */
 310         u64                     last_snap_seq;
 311         struct rw_semaphore     snap_rwsem;
 312         struct rb_root          snap_realms;
 313         struct list_head        snap_empty;
 314         spinlock_t              snap_empty_lock;  /* protect snap_empty */
 315
 316         u64                    last_tid;      /* most recent mds request */
 317         u64                    oldest_tid;    /* oldest incomplete mds request,
 318                                                  excluding setfilelock requests */
 319         struct rb_root         request_tree;  /* pending mds requests */
 320         struct delayed_work    delayed_work;  /* delayed work */
 321         unsigned long    last_renew_caps;  /* last time we renewed our caps */
 322         struct list_head cap_delay_list;   /* caps with delayed release */
 323         spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
 324         struct list_head snap_flush_list;  /* cap_snaps ready to flush */
 325         spinlock_t       snap_flush_lock;
 326
 327         u64               last_cap_flush_tid;
 328         struct list_head  cap_flush_list;
 329         struct list_head  cap_dirty;        /* inodes with dirty caps */
 330         struct list_head  cap_dirty_migrating; /* ...that are migration... */
 331         int               num_cap_flushing; /* # caps we are flushing */
 332         spinlock_t        cap_dirty_lock;   /* protects above items */
 333         wait_queue_head_t cap_flushing_wq;
 334
 335         /*
 336          * Cap reservations
 337          *
 338          * Maintain a global pool of preallocated struct ceph_caps, referenced
 339          * by struct ceph_caps_reservations.  This ensures that we preallocate
 340          * memory needed to successfully process an MDS response.  (If an MDS
 341          * sends us cap information and we fail to process it, we will have
 342          * problems due to the client and MDS being out of sync.)
 343          *
 344          * Reservations are 'owned' by a ceph_cap_reservation context.
 345          */
 346         spinlock_t      caps_list_lock;
 347         struct          list_head caps_list; /* unused (reserved or
 348                                                 unreserved) */
 349         int             caps_total_count;    /* total caps allocated */
 350         int             caps_use_count;      /* in use */
 351         int             caps_reserve_count;  /* unused, reserved */
 352         int             caps_avail_count;    /* unused, unreserved */
 353         int             caps_min_count;      /* keep at least this many
 354                                                 (unreserved) */
 355         spinlock_t        dentry_lru_lock;
 356         struct list_head  dentry_lru;
 357         int               num_dentry;
 358
 359         struct rw_semaphore     pool_perm_rwsem;
 360         struct rb_root          pool_perm_tree;
 361 };
 362
 363 extern const char *ceph_mds_op_name(int op);
 364
 365 extern struct ceph_mds_session *
 366 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
 367
 368 static inline struct ceph_mds_session *
 369 ceph_get_mds_session(struct ceph_mds_session *s)
 370 {
 371         atomic_inc(&s->s_ref);
 372         return s;
 373 }
 374
 375 extern const char *ceph_session_state_name(int s);
 376
 377 extern void ceph_put_mds_session(struct ceph_mds_session *s);
 378
 379 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
 380                              struct ceph_msg *msg, int mds);
 381
 382 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
 383 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
 384 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
 385 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
 386
 387 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
 388
 389 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
 390 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
 391                                            struct inode *dir);
 392 extern struct ceph_mds_request *
 393 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
 394 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
 395                                      struct ceph_mds_request *req);
 396 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
 397                                 struct inode *dir,
 398                                 struct ceph_mds_request *req);
 399 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
 400 {
 401         kref_get(&req->r_kref);
 402 }
 403 extern void ceph_mdsc_release_request(struct kref *kref);
 404 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
 405 {
 406         kref_put(&req->r_kref, ceph_mdsc_release_request);
 407 }
 408
 409 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
 410                                    struct ceph_mds_session *session);
 411
 412 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
 413
 414 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
 415                                   int stop_on_nosnap);
 416
 417 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
 418 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
 419                                      struct inode *inode,
 420                                      struct dentry *dentry, char action,
 421                                      u32 seq);
 422
 423 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
 424                                     struct ceph_msg *msg);
 425 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
 426                                    struct ceph_msg *msg);
 427
 428 extern struct ceph_mds_session *
 429 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
 430 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 431                                           struct ceph_mds_session *session);
 432
 433 #endif