1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _FS_CEPH_MDS_CLIENT_H
3 #define _FS_CEPH_MDS_CLIENT_H
4
5 #include <linux/completion.h>
6 #include <linux/kref.h>
7 #include <linux/list.h>
8 #include <linux/mutex.h>
9 #include <linux/rbtree.h>
10 #include <linux/spinlock.h>
11 #include <linux/refcount.h>
12 #include <linux/utsname.h>
13 #include <linux/ktime.h>
14
15 #include <linux/ceph/types.h>
16 #include <linux/ceph/messenger.h>
17 #include <linux/ceph/mdsmap.h>
18 #include <linux/ceph/auth.h>
19
20 #include "metric.h"
21 #include "super.h"
22
23 /* The first 8 bits are reserved for old ceph releases */
24 enum ceph_feature_type {
25 CEPHFS_FEATURE_MIMIC = 8,
26 CEPHFS_FEATURE_REPLY_ENCODING,
27 CEPHFS_FEATURE_RECLAIM_CLIENT,
28 CEPHFS_FEATURE_LAZY_CAP_WANTED,
29 CEPHFS_FEATURE_MULTI_RECONNECT,
30 CEPHFS_FEATURE_DELEG_INO,
31 CEPHFS_FEATURE_METRIC_COLLECT,
32
33 CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_METRIC_COLLECT,
34 };
35
36 #define CEPHFS_FEATURES_CLIENT_SUPPORTED { \
37 0, 1, 2, 3, 4, 5, 6, 7, \
38 CEPHFS_FEATURE_MIMIC, \
39 CEPHFS_FEATURE_REPLY_ENCODING, \
40 CEPHFS_FEATURE_LAZY_CAP_WANTED, \
41 CEPHFS_FEATURE_MULTI_RECONNECT, \
42 CEPHFS_FEATURE_DELEG_INO, \
43 CEPHFS_FEATURE_METRIC_COLLECT, \
44 }
45 #define CEPHFS_FEATURES_CLIENT_REQUIRED {}
46
47 /*
48 * Some lock dependencies:
49 *
50 * session->s_mutex
51 * mdsc->mutex
52 *
53 * mdsc->snap_rwsem
54 *
55 * ci->i_ceph_lock
56 * mdsc->snap_flush_lock
57 * mdsc->cap_delay_lock
58 *
59 */
60
61 struct ceph_fs_client;
62 struct ceph_cap;
63
64 /*
65 * parsed info about a single inode. pointers are into the encoded
66 * on-wire structures within the mds reply message payload.
67 */
68 struct ceph_mds_reply_info_in {
69 struct ceph_mds_reply_inode *in;
70 struct ceph_dir_layout dir_layout;
71 u32 symlink_len;
72 char *symlink;
73 u32 xattr_len;
74 char *xattr_data;
75 u64 inline_version;
76 u32 inline_len;
77 char *inline_data;
78 u32 pool_ns_len;
79 char *pool_ns_data;
80 u64 max_bytes;
81 u64 max_files;
82 s32 dir_pin;
83 struct ceph_timespec btime;
84 struct ceph_timespec snap_btime;
85 u64 change_attr;
86 };
87
88 struct ceph_mds_reply_dir_entry {
89 char *name;
90 u32 name_len;
91 struct ceph_mds_reply_lease *lease;
92 struct ceph_mds_reply_info_in inode;
93 loff_t offset;
94 };
95
96 /*
97 * parsed info about an mds reply, including information about
98 * either: 1) the target inode and/or its parent directory and dentry,
99 * and directory contents (for readdir results), or
100 * 2) the file range lock info (for fcntl F_GETLK results).
101 */
102 struct ceph_mds_reply_info_parsed {
103 struct ceph_mds_reply_head *head;
104
105 /* trace */
106 struct ceph_mds_reply_info_in diri, targeti;
107 struct ceph_mds_reply_dirfrag *dirfrag;
108 char *dname;
109 u32 dname_len;
110 struct ceph_mds_reply_lease *dlease;
111
112 /* extra */
113 union {
114 /* for fcntl F_GETLK results */
115 struct ceph_filelock *filelock_reply;
116
117 /* for readdir results */
118 struct {
119 struct ceph_mds_reply_dirfrag *dir_dir;
120 size_t dir_buf_size;
121 int dir_nr;
122 bool dir_end;
123 bool dir_complete;
124 bool hash_order;
125 bool offset_hash;
126 struct ceph_mds_reply_dir_entry *dir_entries;
127 };
128
129 /* for create results */
130 struct {
131 bool has_create_ino;
132 u64 ino;
133 };
134 };
135
136 /* encoded blob describing snapshot contexts for certain
137 operations (e.g., open) */
138 void *snapblob;
139 int snapblob_len;
140 };
141
142
143 /*
144 * cap releases are batched and sent to the MDS en masse.
145 *
146 * Account for per-message overhead of mds_cap_release header
147 * and __le32 for osd epoch barrier trailing field.
148 */
149 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
150 sizeof(struct ceph_mds_cap_release)) / \
151 sizeof(struct ceph_mds_cap_item))
152
153
154 /*
155 * state associated with each MDS<->client session
156 */
157 enum {
158 CEPH_MDS_SESSION_NEW = 1,
159 CEPH_MDS_SESSION_OPENING = 2,
160 CEPH_MDS_SESSION_OPEN = 3,
161 CEPH_MDS_SESSION_HUNG = 4,
162 CEPH_MDS_SESSION_RESTARTING = 5,
163 CEPH_MDS_SESSION_RECONNECTING = 6,
164 CEPH_MDS_SESSION_CLOSING = 7,
165 CEPH_MDS_SESSION_CLOSED = 8,
166 CEPH_MDS_SESSION_REJECTED = 9,
167 };
168
169 struct ceph_mds_session {
170 struct ceph_mds_client *s_mdsc;
171 int s_mds;
172 int s_state;
173 unsigned long s_ttl; /* time until mds kills us */
174 unsigned long s_features;
175 u64 s_seq; /* incoming msg seq # */
176 struct mutex s_mutex; /* serialize session messages */
177
178 struct ceph_connection s_con;
179
180 struct ceph_auth_handshake s_auth;
181
182 /* protected by s_gen_ttl_lock */
183 spinlock_t s_gen_ttl_lock;
184 u32 s_cap_gen; /* inc each time we get mds stale msg */
185 unsigned long s_cap_ttl; /* when session caps expire */
186
187 /* protected by s_cap_lock */
188 spinlock_t s_cap_lock;
189 refcount_t s_ref;
190 struct list_head s_caps; /* all caps issued by this session */
191 struct ceph_cap *s_cap_iterator;
192 int s_nr_caps;
193 int s_num_cap_releases;
194 int s_cap_reconnect;
195 int s_readonly;
196 struct list_head s_cap_releases; /* waiting cap_release messages */
197 struct work_struct s_cap_release_work;
198
199 /* See ceph_inode_info->i_dirty_item. */
200 struct list_head s_cap_dirty; /* inodes w/ dirty caps */
201
202 /* See ceph_inode_info->i_flushing_item. */
203 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
204
205 unsigned long s_renew_requested; /* last time we sent a renew req */
206 u64 s_renew_seq;
207
208 struct list_head s_waiting; /* waiting requests */
209 struct list_head s_unsafe; /* unsafe requests */
210 struct xarray s_delegated_inos;
211 };
212
213 /*
214 * modes of choosing which MDS to send a request to
215 */
216 enum {
217 USE_ANY_MDS,
218 USE_RANDOM_MDS,
219 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
220 };
221
222 struct ceph_mds_request;
223 struct ceph_mds_client;
224
225 /*
226 * request completion callback
227 */
228 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
229 struct ceph_mds_request *req);
230 /*
231 * wait for request completion callback
232 */
233 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
234 struct ceph_mds_request *req);
235
236 /*
237 * an in-flight mds request
238 */
239 struct ceph_mds_request {
240 u64 r_tid; /* transaction id */
241 struct rb_node r_node;
242 struct ceph_mds_client *r_mdsc;
243
244 struct kref r_kref;
245 int r_op; /* mds op code */
246
247 /* operation on what? */
248 struct inode *r_inode; /* arg1 */
249 struct dentry *r_dentry; /* arg1 */
250 struct dentry *r_old_dentry; /* arg2: rename from or link from */
251 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
252 char *r_path1, *r_path2;
253 struct ceph_vino r_ino1, r_ino2;
254
255 struct inode *r_parent; /* parent dir inode */
256 struct inode *r_target_inode; /* resulting inode */
257
258 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
259 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
260 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
261 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
262 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
263 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
264 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
265 #define CEPH_MDS_R_ASYNC (8) /* async request */
266 unsigned long r_req_flags;
267
268 struct mutex r_fill_mutex;
269
270 union ceph_mds_request_args r_args;
271 int r_fmode; /* file mode, if expecting cap */
272 kuid_t r_uid;
273 kgid_t r_gid;
274 int r_request_release_offset;
275 struct timespec64 r_stamp;
276
277 /* for choosing which mds to send this request to */
278 int r_direct_mode;
279 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
280
281 /* data payload is used for xattr ops */
282 struct ceph_pagelist *r_pagelist;
283
284 /* what caps shall we drop? */
285 int r_inode_drop, r_inode_unless;
286 int r_dentry_drop, r_dentry_unless;
287 int r_old_dentry_drop, r_old_dentry_unless;
288 struct inode *r_old_inode;
289 int r_old_inode_drop, r_old_inode_unless;
290
291 struct ceph_msg *r_request; /* original request */
292 struct ceph_msg *r_reply;
293 struct ceph_mds_reply_info_parsed r_reply_info;
294 int r_err;
295
296
297 struct page *r_locked_page;
298 int r_dir_caps;
299 int r_num_caps;
300 u32 r_readdir_offset;
301
302 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
303 unsigned long r_started; /* start time to measure timeout against */
304 unsigned long r_start_latency; /* start time to measure latency */
305 unsigned long r_end_latency; /* finish time to measure latency */
306 unsigned long r_request_started; /* start time for mds request only,
307 used to measure lease durations */
308
309 /* link unsafe requests to parent directory, for fsync */
310 struct inode *r_unsafe_dir;
311 struct list_head r_unsafe_dir_item;
312
313 /* unsafe requests that modify the target inode */
314 struct list_head r_unsafe_target_item;
315
316 struct ceph_mds_session *r_session;
317
318 int r_attempts; /* resend attempts */
319 int r_num_fwd; /* number of forward attempts */
320 int r_resend_mds; /* mds to resend to next, if any*/
321 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
322 u64 r_deleg_ino;
323
324 struct list_head r_wait;
325 struct completion r_completion;
326 struct completion r_safe_completion;
327 ceph_mds_request_callback_t r_callback;
328 ceph_mds_request_wait_callback_t r_wait_for_completion;
329 struct list_head r_unsafe_item; /* per-session unsafe list item */
330
331 long long r_dir_release_cnt;
332 long long r_dir_ordered_cnt;
333 int r_readdir_cache_idx;
334
335 struct ceph_cap_reservation r_caps_reservation;
336 };
337
338 struct ceph_pool_perm {
339 struct rb_node node;
340 int perm;
341 s64 pool;
342 size_t pool_ns_len;
343 char pool_ns[];
344 };
345
346 struct ceph_snapid_map {
347 struct rb_node node;
348 struct list_head lru;
349 atomic_t ref;
350 u64 snap;
351 dev_t dev;
352 unsigned long last_used;
353 };
354
355 /*
356 * node for list of quotarealm inodes that are not visible from the filesystem
357 * mountpoint, but required to handle, e.g. quotas.
358 */
359 struct ceph_quotarealm_inode {
360 struct rb_node node;
361 u64 ino;
362 unsigned long timeout; /* last time a lookup failed for this inode */
363 struct mutex mutex;
364 struct inode *inode;
365 };
366
367 struct cap_wait {
368 struct list_head list;
369 u64 ino;
370 pid_t tgid;
371 int need;
372 int want;
373 };
374
375 enum {
376 CEPH_MDSC_STOPPING_BEGIN = 1,
377 CEPH_MDSC_STOPPING_FLUSHED = 2,
378 };
379
380 /*
381 * mds client state
382 */
383 struct ceph_mds_client {
384 struct ceph_fs_client *fsc;
385 struct mutex mutex; /* all nested structures */
386
387 struct ceph_mdsmap *mdsmap;
388 struct completion safe_umount_waiters;
389 wait_queue_head_t session_close_wq;
390 struct list_head waiting_for_map;
391 int mdsmap_err;
392
393 struct ceph_mds_session **sessions; /* NULL for mds if no session */
394 atomic_t num_sessions;
395 int max_sessions; /* len of sessions array */
396 int stopping; /* true if shutting down */
397
398 atomic64_t quotarealms_count; /* # realms with quota */
399 /*
400 * We keep a list of inodes we don't see in the mountpoint but that we
401 * need to track quota realms.
402 */
403 struct rb_root quotarealms_inodes;
404 struct mutex quotarealms_inodes_mutex;
405
406 /*
407 * snap_rwsem will cover cap linkage into snaprealms, and
408 * realm snap contexts. (later, we can do per-realm snap
409 * contexts locks..) the empty list contains realms with no
410 * references (implying they contain no inodes with caps) that
411 * should be destroyed.
412 */
413 u64 last_snap_seq;
414 struct rw_semaphore snap_rwsem;
415 struct rb_root snap_realms;
416 struct list_head snap_empty;
417 int num_snap_realms;
418 spinlock_t snap_empty_lock; /* protect snap_empty */
419
420 u64 last_tid; /* most recent mds request */
421 u64 oldest_tid; /* oldest incomplete mds request,
422 excluding setfilelock requests */
423 struct rb_root request_tree; /* pending mds requests */
424 struct delayed_work delayed_work; /* delayed work */
425 unsigned long last_renew_caps; /* last time we renewed our caps */
426 struct list_head cap_delay_list; /* caps with delayed release */
427 spinlock_t cap_delay_lock; /* protects cap_delay_list */
428 struct list_head snap_flush_list; /* cap_snaps ready to flush */
429 spinlock_t snap_flush_lock;
430
431 u64 last_cap_flush_tid;
432 struct list_head cap_flush_list;
433 struct list_head cap_dirty_migrating; /* ...that are migration... */
434 int num_cap_flushing; /* # caps we are flushing */
435 spinlock_t cap_dirty_lock; /* protects above items */
436 wait_queue_head_t cap_flushing_wq;
437
438 struct work_struct cap_reclaim_work;
439 atomic_t cap_reclaim_pending;
440
441 /*
442 * Cap reservations
443 *
444 * Maintain a global pool of preallocated struct ceph_caps, referenced
445 * by struct ceph_caps_reservations. This ensures that we preallocate
446 * memory needed to successfully process an MDS response. (If an MDS
447 * sends us cap information and we fail to process it, we will have
448 * problems due to the client and MDS being out of sync.)
449 *
450 * Reservations are 'owned' by a ceph_cap_reservation context.
451 */
452 spinlock_t caps_list_lock;
453 struct list_head caps_list; /* unused (reserved or
454 unreserved) */
455 struct list_head cap_wait_list;
456 int caps_total_count; /* total caps allocated */
457 int caps_use_count; /* in use */
458 int caps_use_max; /* max used caps */
459 int caps_reserve_count; /* unused, reserved */
460 int caps_avail_count; /* unused, unreserved */
461 int caps_min_count; /* keep at least this many
462 (unreserved) */
463 spinlock_t dentry_list_lock;
464 struct list_head dentry_leases; /* fifo list */
465 struct list_head dentry_dir_leases; /* lru list */
466
467 struct ceph_client_metric metric;
468
469 spinlock_t snapid_map_lock;
470 struct rb_root snapid_map_tree;
471 struct list_head snapid_map_lru;
472
473 struct rw_semaphore pool_perm_rwsem;
474 struct rb_root pool_perm_tree;
475
476 char nodename[__NEW_UTS_LEN + 1];
477 };
478
479 extern const char *ceph_mds_op_name(int op);
480
481 extern bool check_session_state(struct ceph_mds_session *s);
482 void inc_session_sequence(struct ceph_mds_session *s);
483
484 extern struct ceph_mds_session *
485 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
486
487 extern const char *ceph_session_state_name(int s);
488
489 extern struct ceph_mds_session *
490 ceph_get_mds_session(struct ceph_mds_session *s);
491 extern void ceph_put_mds_session(struct ceph_mds_session *s);
492
493 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
494 struct ceph_msg *msg, int mds);
495
496 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
497 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
498 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
499 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
500
501 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
502
503 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
504 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
505 struct inode *dir);
506 extern struct ceph_mds_request *
507 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
508 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
509 struct inode *dir,
510 struct ceph_mds_request *req);
511 extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
512 struct inode *dir,
513 struct ceph_mds_request *req);
514 extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
515 extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
ceph_mdsc_get_request(struct ceph_mds_request * req)516 static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
517 {
518 kref_get(&req->r_kref);
519 }
520 extern void ceph_mdsc_release_request(struct kref *kref);
ceph_mdsc_put_request(struct ceph_mds_request * req)521 static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
522 {
523 kref_put(&req->r_kref, ceph_mdsc_release_request);
524 }
525
526 extern void send_flush_mdlog(struct ceph_mds_session *s);
527 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
528 void (*cb)(struct ceph_mds_session *),
529 bool check_state);
530 extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
531 extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
532 struct ceph_cap *cap);
533 extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
534 struct ceph_mds_session *session);
535 extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
536 extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
537 extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
538 int (*cb)(struct inode *,
539 struct ceph_cap *, void *),
540 void *arg);
541 extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
542
ceph_mdsc_free_path(char * path,int len)543 static inline void ceph_mdsc_free_path(char *path, int len)
544 {
545 if (!IS_ERR_OR_NULL(path))
546 __putname(path - (PATH_MAX - 1 - len));
547 }
548
549 extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
550 int stop_on_nosnap);
551
552 extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
553 extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
554 struct dentry *dentry, char action,
555 u32 seq);
556
557 extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
558 struct ceph_msg *msg);
559 extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
560 struct ceph_msg *msg);
561
562 extern struct ceph_mds_session *
563 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
564 extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
565 struct ceph_mds_session *session);
566
567 extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
568 struct ceph_mds_session *session,
569 int max_caps);
570
ceph_wait_on_async_create(struct inode * inode)571 static inline int ceph_wait_on_async_create(struct inode *inode)
572 {
573 struct ceph_inode_info *ci = ceph_inode(inode);
574
575 return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
576 TASK_INTERRUPTIBLE);
577 }
578
579 extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
580 extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
581 #endif
582