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 };
  48
  49 /*
  50  * parsed info about an mds reply, including information about
  51  * either: 1) the target inode and/or its parent directory and dentry,
  52  * and directory contents (for readdir results), or
  53  * 2) the file range lock info (for fcntl F_GETLK results).
  54  */
  55 struct ceph_mds_reply_info_parsed {
  56         struct ceph_mds_reply_head    *head;
  57
  58         /* trace */
  59         struct ceph_mds_reply_info_in diri, targeti;
  60         struct ceph_mds_reply_dirfrag *dirfrag;
  61         char                          *dname;
  62         u32                           dname_len;
  63         struct ceph_mds_reply_lease   *dlease;
  64
  65         /* extra */
  66         union {
  67                 /* for fcntl F_GETLK results */
  68                 struct ceph_filelock *filelock_reply;
  69
  70                 /* for readdir results */
  71                 struct {
  72                         struct ceph_mds_reply_dirfrag *dir_dir;
  73                         size_t                        dir_buf_size;
  74                         int                           dir_nr;
  75                         char                          **dir_dname;
  76                         u32                           *dir_dname_len;
  77                         struct ceph_mds_reply_lease   **dir_dlease;
  78                         struct ceph_mds_reply_info_in *dir_in;
  79                         u8                            dir_complete, dir_end;
  80                 };
  81
  82                 /* for create results */
  83                 struct {
  84                         bool has_create_ino;
  85                         u64 ino;
  86                 };
  87         };
  88
  89         /* encoded blob describing snapshot contexts for certain
  90            operations (e.g., open) */
  91         void *snapblob;
  92         int snapblob_len;
  93 };
  94
  95
  96 /*
  97  * cap releases are batched and sent to the MDS en masse.
  98  */
  99 #define CEPH_CAPS_PER_RELEASE ((PAGE_CACHE_SIZE -                       \
 100                                 sizeof(struct ceph_mds_cap_release)) /  \
 101                                sizeof(struct ceph_mds_cap_item))
 102
 103
 104 /*
 105  * state associated with each MDS<->client session
 106  */
 107 enum {
 108         CEPH_MDS_SESSION_NEW = 1,
 109         CEPH_MDS_SESSION_OPENING = 2,
 110         CEPH_MDS_SESSION_OPEN = 3,
 111         CEPH_MDS_SESSION_HUNG = 4,
 112         CEPH_MDS_SESSION_CLOSING = 5,
 113         CEPH_MDS_SESSION_RESTARTING = 6,
 114         CEPH_MDS_SESSION_RECONNECTING = 7,
 115 };
 116
 117 struct ceph_mds_session {
 118         struct ceph_mds_client *s_mdsc;
 119         int               s_mds;
 120         int               s_state;
 121         unsigned long     s_ttl;      /* time until mds kills us */
 122         u64               s_seq;      /* incoming msg seq # */
 123         struct mutex      s_mutex;    /* serialize session messages */
 124
 125         struct ceph_connection s_con;
 126
 127         struct ceph_auth_handshake s_auth;
 128
 129         /* protected by s_gen_ttl_lock */
 130         spinlock_t        s_gen_ttl_lock;
 131         u32               s_cap_gen;  /* inc each time we get mds stale msg */
 132         unsigned long     s_cap_ttl;  /* when session caps expire */
 133
 134         /* protected by s_cap_lock */
 135         spinlock_t        s_cap_lock;
 136         struct list_head  s_caps;     /* all caps issued by this session */
 137         int               s_nr_caps, s_trim_caps;
 138         int               s_num_cap_releases;
 139         int               s_cap_reconnect;
 140         int               s_readonly;
 141         struct list_head  s_cap_releases; /* waiting cap_release messages */
 142         struct ceph_cap  *s_cap_iterator;
 143
 144         /* protected by mutex */
 145         struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
 146         struct list_head  s_cap_snaps_flushing;
 147         unsigned long     s_renew_requested; /* last time we sent a renew req */
 148         u64               s_renew_seq;
 149
 150         atomic_t          s_ref;
 151         struct list_head  s_waiting;  /* waiting requests */
 152         struct list_head  s_unsafe;   /* unsafe requests */
 153 };
 154
 155 /*
 156  * modes of choosing which MDS to send a request to
 157  */
 158 enum {
 159         USE_ANY_MDS,
 160         USE_RANDOM_MDS,
 161         USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
 162 };
 163
 164 struct ceph_mds_request;
 165 struct ceph_mds_client;
 166
 167 /*
 168  * request completion callback
 169  */
 170 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
 171                                              struct ceph_mds_request *req);
 172 /*
 173  * wait for request completion callback
 174  */
 175 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
 176                                                  struct ceph_mds_request *req);
 177
 178 /*
 179  * an in-flight mds request
 180  */
 181 struct ceph_mds_request {
 182         u64 r_tid;                   /* transaction id */
 183         struct rb_node r_node;
 184         struct ceph_mds_client *r_mdsc;
 185
 186         int r_op;                    /* mds op code */
 187
 188         /* operation on what? */
 189         struct inode *r_inode;              /* arg1 */
 190         struct dentry *r_dentry;            /* arg1 */
 191         struct dentry *r_old_dentry;        /* arg2: rename from or link from */
 192         struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
 193         char *r_path1, *r_path2;
 194         struct ceph_vino r_ino1, r_ino2;
 195
 196         struct inode *r_locked_dir; /* dir (if any) i_mutex locked by vfs */
 197         struct inode *r_target_inode;       /* resulting inode */
 198
 199         struct mutex r_fill_mutex;
 200
 201         union ceph_mds_request_args r_args;
 202         int r_fmode;        /* file mode, if expecting cap */
 203         kuid_t r_uid;
 204         kgid_t r_gid;
 205         struct timespec r_stamp;
 206
 207         /* for choosing which mds to send this request to */
 208         int r_direct_mode;
 209         u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
 210         bool r_direct_is_hash;  /* true if r_direct_hash is valid */
 211
 212         /* data payload is used for xattr ops */
 213         struct ceph_pagelist *r_pagelist;
 214
 215         /* what caps shall we drop? */
 216         int r_inode_drop, r_inode_unless;
 217         int r_dentry_drop, r_dentry_unless;
 218         int r_old_dentry_drop, r_old_dentry_unless;
 219         struct inode *r_old_inode;
 220         int r_old_inode_drop, r_old_inode_unless;
 221
 222         struct ceph_msg  *r_request;  /* original request */
 223         int r_request_release_offset;
 224         struct ceph_msg  *r_reply;
 225         struct ceph_mds_reply_info_parsed r_reply_info;
 226         struct page *r_locked_page;
 227         int r_err;
 228         bool r_aborted;
 229
 230         unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
 231         unsigned long r_started;  /* start time to measure timeout against */
 232         unsigned long r_request_started; /* start time for mds request only,
 233                                             used to measure lease durations */
 234
 235         /* link unsafe requests to parent directory, for fsync */
 236         struct inode    *r_unsafe_dir;
 237         struct list_head r_unsafe_dir_item;
 238
 239         struct ceph_mds_session *r_session;
 240
 241         int               r_attempts;   /* resend attempts */
 242         int               r_num_fwd;    /* number of forward attempts */
 243         int               r_resend_mds; /* mds to resend to next, if any*/
 244         u32               r_sent_on_mseq; /* cap mseq request was sent at*/
 245
 246         struct kref       r_kref;
 247         struct list_head  r_wait;
 248         struct completion r_completion;
 249         struct completion r_safe_completion;
 250         ceph_mds_request_callback_t r_callback;
 251         ceph_mds_request_wait_callback_t r_wait_for_completion;
 252         struct list_head  r_unsafe_item;  /* per-session unsafe list item */
 253         bool              r_got_unsafe, r_got_safe, r_got_result;
 254
 255         bool              r_did_prepopulate;
 256         long long         r_dir_release_cnt;
 257         long long         r_dir_ordered_cnt;
 258         int               r_readdir_cache_idx;
 259         u32               r_readdir_offset;
 260
 261         struct ceph_cap_reservation r_caps_reservation;
 262         int r_num_caps;
 263 };
 264
 265 struct ceph_pool_perm {
 266         struct rb_node node;
 267         u32 pool;
 268         int perm;
 269 };
 270
 271 /*
 272  * mds client state
 273  */
 274 struct ceph_mds_client {
 275         struct ceph_fs_client  *fsc;
 276         struct mutex            mutex;         /* all nested structures */
 277
 278         struct ceph_mdsmap      *mdsmap;
 279         struct completion       safe_umount_waiters;
 280         wait_queue_head_t       session_close_wq;
 281         struct list_head        waiting_for_map;
 282
 283         struct ceph_mds_session **sessions;    /* NULL for mds if no session */
 284         atomic_t                num_sessions;
 285         int                     max_sessions;  /* len of s_mds_sessions */
 286         int                     stopping;      /* true if shutting down */
 287
 288         /*
 289          * snap_rwsem will cover cap linkage into snaprealms, and
 290          * realm snap contexts.  (later, we can do per-realm snap
 291          * contexts locks..)  the empty list contains realms with no
 292          * references (implying they contain no inodes with caps) that
 293          * should be destroyed.
 294          */
 295         u64                     last_snap_seq;
 296         struct rw_semaphore     snap_rwsem;
 297         struct rb_root          snap_realms;
 298         struct list_head        snap_empty;
 299         spinlock_t              snap_empty_lock;  /* protect snap_empty */
 300
 301         u64                    last_tid;      /* most recent mds request */
 302         u64                    oldest_tid;    /* oldest incomplete mds request,
 303                                                  excluding setfilelock requests */
 304         struct rb_root         request_tree;  /* pending mds requests */
 305         struct delayed_work    delayed_work;  /* delayed work */
 306         unsigned long    last_renew_caps;  /* last time we renewed our caps */
 307         struct list_head cap_delay_list;   /* caps with delayed release */
 308         spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
 309         struct list_head snap_flush_list;  /* cap_snaps ready to flush */
 310         spinlock_t       snap_flush_lock;
 311
 312         u64               last_cap_flush_tid;
 313         struct rb_root    cap_flush_tree;
 314         struct list_head  cap_dirty;        /* inodes with dirty caps */
 315         struct list_head  cap_dirty_migrating; /* ...that are migration... */
 316         int               num_cap_flushing; /* # caps we are flushing */
 317         spinlock_t        cap_dirty_lock;   /* protects above items */
 318         wait_queue_head_t cap_flushing_wq;
 319
 320         /*
 321          * Cap reservations
 322          *
 323          * Maintain a global pool of preallocated struct ceph_caps, referenced
 324          * by struct ceph_caps_reservations.  This ensures that we preallocate
 325          * memory needed to successfully process an MDS response.  (If an MDS
 326          * sends us cap information and we fail to process it, we will have
 327          * problems due to the client and MDS being out of sync.)
 328          *
 329          * Reservations are 'owned' by a ceph_cap_reservation context.
 330          */
 331         spinlock_t      caps_list_lock;
 332         struct          list_head caps_list; /* unused (reserved or
 333                                                 unreserved) */
 334         int             caps_total_count;    /* total caps allocated */
 335         int             caps_use_count;      /* in use */
 336         int             caps_reserve_count;  /* unused, reserved */
 337         int             caps_avail_count;    /* unused, unreserved */
 338         int             caps_min_count;      /* keep at least this many
 339                                                 (unreserved) */
 340         spinlock_t        dentry_lru_lock;
 341         struct list_head  dentry_lru;
 342         int               num_dentry;
 343
 344         struct rw_semaphore     pool_perm_rwsem;
 345         struct rb_root          pool_perm_tree;
 346 };
 347
 348 extern const char *ceph_mds_op_name(int op);
 349
 350 extern struct ceph_mds_session *
 351 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
 352
 353 static inline struct ceph_mds_session *
 354 ceph_get_mds_session(struct ceph_mds_session *s)
 355 {
 356         atomic_inc(&s->s_ref);
 357         return s;
 358 }
 359
 360 extern const char *ceph_session_state_name(int s);
 361
 362 extern void ceph_put_mds_session(struct ceph_mds_session *s);
 363
 364 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
 365                              struct ceph_msg *msg, int mds);
 366
 367 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
 368 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
 369 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
 370
 371 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
 372
 373 extern void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc,
 374                                     struct inode *inode,
 375                                     struct dentry *dn);
 376
 377 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
 378 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
 379                                            struct inode *dir);
 380 extern struct ceph_mds_request *
 381 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
 382 extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
 383                                      struct ceph_mds_request *req);
 384 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
 385                                 struct inode *dir,
 386                                 struct ceph_mds_request *req);
 387 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
 388 {
 389         kref_get(&req->r_kref);
 390 }
 391 extern void ceph_mdsc_release_request(struct kref *kref);
 392 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
 393 {
 394         kref_put(&req->r_kref, ceph_mdsc_release_request);
 395 }
 396
 397 extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
 398                                    struct ceph_mds_session *session);
 399
 400 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
 401
 402 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
 403                                   int stop_on_nosnap);
 404
 405 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
 406 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
 407                                      struct inode *inode,
 408                                      struct dentry *dentry, char action,
 409                                      u32 seq);
 410
 411 extern void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc,
 412                                  struct ceph_msg *msg);
 413
 414 extern struct ceph_mds_session *
 415 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
 416 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 417                                           struct ceph_mds_session *session);
 418
 419 #endif