Back to home page

OSCL-LXR

 
 

    


0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 #ifndef _FS_CEPH_MDS_CLIENT_H
0003 #define _FS_CEPH_MDS_CLIENT_H
0004 
0005 #include <linux/completion.h>
0006 #include <linux/kref.h>
0007 #include <linux/list.h>
0008 #include <linux/mutex.h>
0009 #include <linux/rbtree.h>
0010 #include <linux/spinlock.h>
0011 #include <linux/refcount.h>
0012 #include <linux/utsname.h>
0013 #include <linux/ktime.h>
0014 
0015 #include <linux/ceph/types.h>
0016 #include <linux/ceph/messenger.h>
0017 #include <linux/ceph/mdsmap.h>
0018 #include <linux/ceph/auth.h>
0019 
0020 #include "metric.h"
0021 #include "super.h"
0022 
0023 /* The first 8 bits are reserved for old ceph releases */
0024 enum ceph_feature_type {
0025     CEPHFS_FEATURE_MIMIC = 8,
0026     CEPHFS_FEATURE_REPLY_ENCODING,
0027     CEPHFS_FEATURE_RECLAIM_CLIENT,
0028     CEPHFS_FEATURE_LAZY_CAP_WANTED,
0029     CEPHFS_FEATURE_MULTI_RECONNECT,
0030     CEPHFS_FEATURE_DELEG_INO,
0031     CEPHFS_FEATURE_METRIC_COLLECT,
0032     CEPHFS_FEATURE_ALTERNATE_NAME,
0033     CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
0034 
0035     CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
0036 };
0037 
0038 #define CEPHFS_FEATURES_CLIENT_SUPPORTED {  \
0039     0, 1, 2, 3, 4, 5, 6, 7,         \
0040     CEPHFS_FEATURE_MIMIC,           \
0041     CEPHFS_FEATURE_REPLY_ENCODING,      \
0042     CEPHFS_FEATURE_LAZY_CAP_WANTED,     \
0043     CEPHFS_FEATURE_MULTI_RECONNECT,     \
0044     CEPHFS_FEATURE_DELEG_INO,       \
0045     CEPHFS_FEATURE_METRIC_COLLECT,      \
0046     CEPHFS_FEATURE_NOTIFY_SESSION_STATE,    \
0047 }
0048 
0049 /*
0050  * Some lock dependencies:
0051  *
0052  * session->s_mutex
0053  *         mdsc->mutex
0054  *
0055  *         mdsc->snap_rwsem
0056  *
0057  *         ci->i_ceph_lock
0058  *                 mdsc->snap_flush_lock
0059  *                 mdsc->cap_delay_lock
0060  *
0061  */
0062 
0063 struct ceph_fs_client;
0064 struct ceph_cap;
0065 
0066 /*
0067  * parsed info about a single inode.  pointers are into the encoded
0068  * on-wire structures within the mds reply message payload.
0069  */
0070 struct ceph_mds_reply_info_in {
0071     struct ceph_mds_reply_inode *in;
0072     struct ceph_dir_layout dir_layout;
0073     u32 symlink_len;
0074     char *symlink;
0075     u32 xattr_len;
0076     char *xattr_data;
0077     u64 inline_version;
0078     u32 inline_len;
0079     char *inline_data;
0080     u32 pool_ns_len;
0081     char *pool_ns_data;
0082     u64 max_bytes;
0083     u64 max_files;
0084     s32 dir_pin;
0085     struct ceph_timespec btime;
0086     struct ceph_timespec snap_btime;
0087     u64 rsnaps;
0088     u64 change_attr;
0089 };
0090 
0091 struct ceph_mds_reply_dir_entry {
0092     char                          *name;
0093     u32                           name_len;
0094     struct ceph_mds_reply_lease   *lease;
0095     struct ceph_mds_reply_info_in inode;
0096     loff_t                offset;
0097 };
0098 
0099 struct ceph_mds_reply_xattr {
0100     char *xattr_value;
0101     size_t xattr_value_len;
0102 };
0103 
0104 /*
0105  * parsed info about an mds reply, including information about
0106  * either: 1) the target inode and/or its parent directory and dentry,
0107  * and directory contents (for readdir results), or
0108  * 2) the file range lock info (for fcntl F_GETLK results).
0109  */
0110 struct ceph_mds_reply_info_parsed {
0111     struct ceph_mds_reply_head    *head;
0112 
0113     /* trace */
0114     struct ceph_mds_reply_info_in diri, targeti;
0115     struct ceph_mds_reply_dirfrag *dirfrag;
0116     char                          *dname;
0117     u32                           dname_len;
0118     struct ceph_mds_reply_lease   *dlease;
0119     struct ceph_mds_reply_xattr   xattr_info;
0120 
0121     /* extra */
0122     union {
0123         /* for fcntl F_GETLK results */
0124         struct ceph_filelock *filelock_reply;
0125 
0126         /* for readdir results */
0127         struct {
0128             struct ceph_mds_reply_dirfrag *dir_dir;
0129             size_t                dir_buf_size;
0130             int                           dir_nr;
0131             bool                  dir_end;
0132             bool                  dir_complete;
0133             bool                  hash_order;
0134             bool                  offset_hash;
0135             struct ceph_mds_reply_dir_entry  *dir_entries;
0136         };
0137 
0138         /* for create results */
0139         struct {
0140             bool has_create_ino;
0141             u64 ino;
0142         };
0143     };
0144 
0145     /* encoded blob describing snapshot contexts for certain
0146        operations (e.g., open) */
0147     void *snapblob;
0148     int snapblob_len;
0149 };
0150 
0151 
0152 /*
0153  * cap releases are batched and sent to the MDS en masse.
0154  *
0155  * Account for per-message overhead of mds_cap_release header
0156  * and __le32 for osd epoch barrier trailing field.
0157  */
0158 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) -       \
0159                 sizeof(struct ceph_mds_cap_release)) /  \
0160                     sizeof(struct ceph_mds_cap_item))
0161 
0162 
0163 /*
0164  * state associated with each MDS<->client session
0165  */
0166 enum {
0167     CEPH_MDS_SESSION_NEW = 1,
0168     CEPH_MDS_SESSION_OPENING = 2,
0169     CEPH_MDS_SESSION_OPEN = 3,
0170     CEPH_MDS_SESSION_HUNG = 4,
0171     CEPH_MDS_SESSION_RESTARTING = 5,
0172     CEPH_MDS_SESSION_RECONNECTING = 6,
0173     CEPH_MDS_SESSION_CLOSING = 7,
0174     CEPH_MDS_SESSION_CLOSED = 8,
0175     CEPH_MDS_SESSION_REJECTED = 9,
0176 };
0177 
0178 struct ceph_mds_session {
0179     struct ceph_mds_client *s_mdsc;
0180     int               s_mds;
0181     int               s_state;
0182     unsigned long     s_ttl;      /* time until mds kills us */
0183     unsigned long     s_features;
0184     u64               s_seq;      /* incoming msg seq # */
0185     struct mutex      s_mutex;    /* serialize session messages */
0186 
0187     struct ceph_connection s_con;
0188 
0189     struct ceph_auth_handshake s_auth;
0190 
0191     atomic_t          s_cap_gen;  /* inc each time we get mds stale msg */
0192     unsigned long     s_cap_ttl;  /* when session caps expire. protected by s_mutex */
0193 
0194     /* protected by s_cap_lock */
0195     spinlock_t        s_cap_lock;
0196     refcount_t        s_ref;
0197     struct list_head  s_caps;     /* all caps issued by this session */
0198     struct ceph_cap  *s_cap_iterator;
0199     int               s_nr_caps;
0200     int               s_num_cap_releases;
0201     int       s_cap_reconnect;
0202     int       s_readonly;
0203     struct list_head  s_cap_releases; /* waiting cap_release messages */
0204     struct work_struct s_cap_release_work;
0205 
0206     /* See ceph_inode_info->i_dirty_item. */
0207     struct list_head  s_cap_dirty;        /* inodes w/ dirty caps */
0208 
0209     /* See ceph_inode_info->i_flushing_item. */
0210     struct list_head  s_cap_flushing;     /* inodes w/ flushing caps */
0211 
0212     unsigned long     s_renew_requested; /* last time we sent a renew req */
0213     u64               s_renew_seq;
0214 
0215     struct list_head  s_waiting;  /* waiting requests */
0216     struct list_head  s_unsafe;   /* unsafe requests */
0217     struct xarray     s_delegated_inos;
0218 };
0219 
0220 /*
0221  * modes of choosing which MDS to send a request to
0222  */
0223 enum {
0224     USE_ANY_MDS,
0225     USE_RANDOM_MDS,
0226     USE_AUTH_MDS,   /* prefer authoritative mds for this metadata item */
0227 };
0228 
0229 struct ceph_mds_request;
0230 struct ceph_mds_client;
0231 
0232 /*
0233  * request completion callback
0234  */
0235 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
0236                          struct ceph_mds_request *req);
0237 /*
0238  * wait for request completion callback
0239  */
0240 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
0241                          struct ceph_mds_request *req);
0242 
0243 /*
0244  * an in-flight mds request
0245  */
0246 struct ceph_mds_request {
0247     u64 r_tid;                   /* transaction id */
0248     struct rb_node r_node;
0249     struct ceph_mds_client *r_mdsc;
0250 
0251     struct kref       r_kref;
0252     int r_op;                    /* mds op code */
0253 
0254     /* operation on what? */
0255     struct inode *r_inode;              /* arg1 */
0256     struct dentry *r_dentry;            /* arg1 */
0257     struct dentry *r_old_dentry;        /* arg2: rename from or link from */
0258     struct inode *r_old_dentry_dir;     /* arg2: old dentry's parent dir */
0259     char *r_path1, *r_path2;
0260     struct ceph_vino r_ino1, r_ino2;
0261 
0262     struct inode *r_parent;         /* parent dir inode */
0263     struct inode *r_target_inode;       /* resulting inode */
0264 
0265 #define CEPH_MDS_R_DIRECT_IS_HASH   (1) /* r_direct_hash is valid */
0266 #define CEPH_MDS_R_ABORTED      (2) /* call was aborted */
0267 #define CEPH_MDS_R_GOT_UNSAFE       (3) /* got an unsafe reply */
0268 #define CEPH_MDS_R_GOT_SAFE     (4) /* got a safe reply */
0269 #define CEPH_MDS_R_GOT_RESULT       (5) /* got a result */
0270 #define CEPH_MDS_R_DID_PREPOPULATE  (6) /* prepopulated readdir */
0271 #define CEPH_MDS_R_PARENT_LOCKED    (7) /* is r_parent->i_rwsem wlocked? */
0272 #define CEPH_MDS_R_ASYNC        (8) /* async request */
0273     unsigned long   r_req_flags;
0274 
0275     struct mutex r_fill_mutex;
0276 
0277     union ceph_mds_request_args r_args;
0278     int r_fmode;        /* file mode, if expecting cap */
0279     int r_request_release_offset;
0280     const struct cred *r_cred;
0281     struct timespec64 r_stamp;
0282 
0283     /* for choosing which mds to send this request to */
0284     int r_direct_mode;
0285     u32 r_direct_hash;      /* choose dir frag based on this dentry hash */
0286 
0287     /* data payload is used for xattr ops */
0288     struct ceph_pagelist *r_pagelist;
0289 
0290     /* what caps shall we drop? */
0291     int r_inode_drop, r_inode_unless;
0292     int r_dentry_drop, r_dentry_unless;
0293     int r_old_dentry_drop, r_old_dentry_unless;
0294     struct inode *r_old_inode;
0295     int r_old_inode_drop, r_old_inode_unless;
0296 
0297     struct ceph_msg  *r_request;  /* original request */
0298     struct ceph_msg  *r_reply;
0299     struct ceph_mds_reply_info_parsed r_reply_info;
0300     int r_err;
0301     u32               r_readdir_offset;
0302 
0303     struct page *r_locked_page;
0304     int r_dir_caps;
0305     int r_num_caps;
0306 
0307     unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
0308     unsigned long r_started;  /* start time to measure timeout against */
0309     unsigned long r_start_latency;  /* start time to measure latency */
0310     unsigned long r_end_latency;    /* finish time to measure latency */
0311     unsigned long r_request_started; /* start time for mds request only,
0312                         used to measure lease durations */
0313 
0314     /* link unsafe requests to parent directory, for fsync */
0315     struct inode    *r_unsafe_dir;
0316     struct list_head r_unsafe_dir_item;
0317 
0318     /* unsafe requests that modify the target inode */
0319     struct list_head r_unsafe_target_item;
0320 
0321     struct ceph_mds_session *r_session;
0322 
0323     int               r_attempts;   /* resend attempts */
0324     int               r_num_fwd;    /* number of forward attempts */
0325     int               r_resend_mds; /* mds to resend to next, if any*/
0326     u32               r_sent_on_mseq; /* cap mseq request was sent at*/
0327     u64       r_deleg_ino;
0328 
0329     struct list_head  r_wait;
0330     struct completion r_completion;
0331     struct completion r_safe_completion;
0332     ceph_mds_request_callback_t r_callback;
0333     struct list_head  r_unsafe_item;  /* per-session unsafe list item */
0334 
0335     long long     r_dir_release_cnt;
0336     long long     r_dir_ordered_cnt;
0337     int       r_readdir_cache_idx;
0338 
0339     struct ceph_cap_reservation r_caps_reservation;
0340 };
0341 
0342 struct ceph_pool_perm {
0343     struct rb_node node;
0344     int perm;
0345     s64 pool;
0346     size_t pool_ns_len;
0347     char pool_ns[];
0348 };
0349 
0350 struct ceph_snapid_map {
0351     struct rb_node node;
0352     struct list_head lru;
0353     atomic_t ref;
0354     u64 snap;
0355     dev_t dev;
0356     unsigned long last_used;
0357 };
0358 
0359 /*
0360  * node for list of quotarealm inodes that are not visible from the filesystem
0361  * mountpoint, but required to handle, e.g. quotas.
0362  */
0363 struct ceph_quotarealm_inode {
0364     struct rb_node node;
0365     u64 ino;
0366     unsigned long timeout; /* last time a lookup failed for this inode */
0367     struct mutex mutex;
0368     struct inode *inode;
0369 };
0370 
0371 struct cap_wait {
0372     struct list_head    list;
0373     u64         ino;
0374     pid_t           tgid;
0375     int         need;
0376     int         want;
0377 };
0378 
0379 /*
0380  * mds client state
0381  */
0382 struct ceph_mds_client {
0383     struct ceph_fs_client  *fsc;
0384     struct mutex            mutex;         /* all nested structures */
0385 
0386     struct ceph_mdsmap      *mdsmap;
0387     struct completion       safe_umount_waiters;
0388     wait_queue_head_t       session_close_wq;
0389     struct list_head        waiting_for_map;
0390     int             mdsmap_err;
0391 
0392     struct ceph_mds_session **sessions;    /* NULL for mds if no session */
0393     atomic_t        num_sessions;
0394     int                     max_sessions;  /* len of sessions array */
0395     int                     stopping;      /* true if shutting down */
0396 
0397     atomic64_t      quotarealms_count; /* # realms with quota */
0398     /*
0399      * We keep a list of inodes we don't see in the mountpoint but that we
0400      * need to track quota realms.
0401      */
0402     struct rb_root      quotarealms_inodes;
0403     struct mutex        quotarealms_inodes_mutex;
0404 
0405     /*
0406      * snap_rwsem will cover cap linkage into snaprealms, and
0407      * realm snap contexts.  (later, we can do per-realm snap
0408      * contexts locks..)  the empty list contains realms with no
0409      * references (implying they contain no inodes with caps) that
0410      * should be destroyed.
0411      */
0412     u64         last_snap_seq;
0413     struct rw_semaphore     snap_rwsem;
0414     struct rb_root          snap_realms;
0415     struct list_head        snap_empty;
0416     int         num_snap_realms;
0417     spinlock_t              snap_empty_lock;  /* protect snap_empty */
0418 
0419     u64                    last_tid;      /* most recent mds request */
0420     u64                    oldest_tid;    /* oldest incomplete mds request,
0421                          excluding setfilelock requests */
0422     struct rb_root         request_tree;  /* pending mds requests */
0423     struct delayed_work    delayed_work;  /* delayed work */
0424     unsigned long    last_renew_caps;  /* last time we renewed our caps */
0425     struct list_head cap_delay_list;   /* caps with delayed release */
0426     spinlock_t       cap_delay_lock;   /* protects cap_delay_list */
0427     struct list_head snap_flush_list;  /* cap_snaps ready to flush */
0428     spinlock_t       snap_flush_lock;
0429 
0430     u64               last_cap_flush_tid;
0431     struct list_head  cap_flush_list;
0432     struct list_head  cap_dirty_migrating; /* ...that are migration... */
0433     int               num_cap_flushing; /* # caps we are flushing */
0434     spinlock_t        cap_dirty_lock;   /* protects above items */
0435     wait_queue_head_t cap_flushing_wq;
0436 
0437     struct work_struct cap_reclaim_work;
0438     atomic_t       cap_reclaim_pending;
0439 
0440     /*
0441      * Cap reservations
0442      *
0443      * Maintain a global pool of preallocated struct ceph_caps, referenced
0444      * by struct ceph_caps_reservations.  This ensures that we preallocate
0445      * memory needed to successfully process an MDS response.  (If an MDS
0446      * sends us cap information and we fail to process it, we will have
0447      * problems due to the client and MDS being out of sync.)
0448      *
0449      * Reservations are 'owned' by a ceph_cap_reservation context.
0450      */
0451     spinlock_t  caps_list_lock;
0452     struct      list_head caps_list; /* unused (reserved or
0453                         unreserved) */
0454     struct      list_head cap_wait_list;
0455     int     caps_total_count;    /* total caps allocated */
0456     int     caps_use_count;      /* in use */
0457     int     caps_use_max;        /* max used caps */
0458     int     caps_reserve_count;  /* unused, reserved */
0459     int     caps_avail_count;    /* unused, unreserved */
0460     int     caps_min_count;      /* keep at least this many
0461                         (unreserved) */
0462     spinlock_t    dentry_list_lock;
0463     struct list_head  dentry_leases;     /* fifo list */
0464     struct list_head  dentry_dir_leases; /* lru list */
0465 
0466     struct ceph_client_metric metric;
0467 
0468     spinlock_t      snapid_map_lock;
0469     struct rb_root      snapid_map_tree;
0470     struct list_head    snapid_map_lru;
0471 
0472     struct rw_semaphore     pool_perm_rwsem;
0473     struct rb_root      pool_perm_tree;
0474 
0475     char nodename[__NEW_UTS_LEN + 1];
0476 };
0477 
0478 extern const char *ceph_mds_op_name(int op);
0479 
0480 extern bool check_session_state(struct ceph_mds_session *s);
0481 void inc_session_sequence(struct ceph_mds_session *s);
0482 
0483 extern struct ceph_mds_session *
0484 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
0485 
0486 extern const char *ceph_session_state_name(int s);
0487 
0488 extern struct ceph_mds_session *
0489 ceph_get_mds_session(struct ceph_mds_session *s);
0490 extern void ceph_put_mds_session(struct ceph_mds_session *s);
0491 
0492 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
0493                  struct ceph_msg *msg, int mds);
0494 
0495 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
0496 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
0497 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
0498 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
0499 
0500 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
0501 
0502 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
0503 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
0504                        struct inode *dir);
0505 extern struct ceph_mds_request *
0506 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
0507 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
0508                     struct inode *dir,
0509                     struct ceph_mds_request *req);
0510 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
0511             struct ceph_mds_request *req,
0512             ceph_mds_request_wait_callback_t wait_func);
0513 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
0514                 struct inode *dir,
0515                 struct ceph_mds_request *req);
0516 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
0517 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
0518 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
0519 {
0520     kref_get(&req->r_kref);
0521 }
0522 extern void ceph_mdsc_release_request(struct kref *kref);
0523 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
0524 {
0525     kref_put(&req->r_kref, ceph_mdsc_release_request);
0526 }
0527 
0528 extern void send_flush_mdlog(struct ceph_mds_session *s);
0529 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
0530                        void (*cb)(struct ceph_mds_session *),
0531                        bool check_state);
0532 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
0533 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
0534                     struct ceph_cap *cap);
0535 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
0536                     struct ceph_mds_session *session);
0537 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
0538 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
0539 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
0540                      int (*cb)(struct inode *,
0541                            struct ceph_cap *, void *),
0542                      void *arg);
0543 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
0544 
0545 static inline void ceph_mdsc_free_path(char *path, int len)
0546 {
0547     if (!IS_ERR_OR_NULL(path))
0548         __putname(path - (PATH_MAX - 1 - len));
0549 }
0550 
0551 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
0552                   int stop_on_nosnap);
0553 
0554 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
0555 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
0556                      struct dentry *dentry, char action,
0557                      u32 seq);
0558 
0559 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
0560                     struct ceph_msg *msg);
0561 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
0562                    struct ceph_msg *msg);
0563 
0564 extern struct ceph_mds_session *
0565 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
0566 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
0567                       struct ceph_mds_session *session);
0568 
0569 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
0570               struct ceph_mds_session *session,
0571               int max_caps);
0572 
0573 static inline int ceph_wait_on_async_create(struct inode *inode)
0574 {
0575     struct ceph_inode_info *ci = ceph_inode(inode);
0576 
0577     return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
0578                TASK_KILLABLE);
0579 }
0580 
0581 extern int ceph_wait_on_conflict_unlink(struct dentry *dentry);
0582 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
0583 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
0584 #endif