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 rsnaps;
86 u64 change_attr;
87 };
88
89 struct ceph_mds_reply_dir_entry {
90 char *name;
91 u32 name_len;
92 struct ceph_mds_reply_lease *lease;
93 struct ceph_mds_reply_info_in inode;
94 loff_t offset;
95 };
96
97 struct ceph_mds_reply_xattr {
98 char *xattr_value;
99 size_t xattr_value_len;
100 };
101
102 /*
103 * parsed info about an mds reply, including information about
104 * either: 1) the target inode and/or its parent directory and dentry,
105 * and directory contents (for readdir results), or
106 * 2) the file range lock info (for fcntl F_GETLK results).
107 */
108 struct ceph_mds_reply_info_parsed {
109 struct ceph_mds_reply_head *head;
110
111 /* trace */
112 struct ceph_mds_reply_info_in diri, targeti;
113 struct ceph_mds_reply_dirfrag *dirfrag;
114 char *dname;
115 u32 dname_len;
116 struct ceph_mds_reply_lease *dlease;
117 struct ceph_mds_reply_xattr xattr_info;
118
119 /* extra */
120 union {
121 /* for fcntl F_GETLK results */
122 struct ceph_filelock *filelock_reply;
123
124 /* for readdir results */
125 struct {
126 struct ceph_mds_reply_dirfrag *dir_dir;
127 size_t dir_buf_size;
128 int dir_nr;
129 bool dir_end;
130 bool dir_complete;
131 bool hash_order;
132 bool offset_hash;
133 struct ceph_mds_reply_dir_entry *dir_entries;
134 };
135
136 /* for create results */
137 struct {
138 bool has_create_ino;
139 u64 ino;
140 };
141 };
142
143 /* encoded blob describing snapshot contexts for certain
144 operations (e.g., open) */
145 void *snapblob;
146 int snapblob_len;
147 };
148
149
150 /*
151 * cap releases are batched and sent to the MDS en masse.
152 *
153 * Account for per-message overhead of mds_cap_release header
154 * and __le32 for osd epoch barrier trailing field.
155 */
156 #define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
157 sizeof(struct ceph_mds_cap_release)) / \
158 sizeof(struct ceph_mds_cap_item))
159
160
161 /*
162 * state associated with each MDS<->client session
163 */
164 enum {
165 CEPH_MDS_SESSION_NEW = 1,
166 CEPH_MDS_SESSION_OPENING = 2,
167 CEPH_MDS_SESSION_OPEN = 3,
168 CEPH_MDS_SESSION_HUNG = 4,
169 CEPH_MDS_SESSION_RESTARTING = 5,
170 CEPH_MDS_SESSION_RECONNECTING = 6,
171 CEPH_MDS_SESSION_CLOSING = 7,
172 CEPH_MDS_SESSION_CLOSED = 8,
173 CEPH_MDS_SESSION_REJECTED = 9,
174 };
175
176 struct ceph_mds_session {
177 struct ceph_mds_client *s_mdsc;
178 int s_mds;
179 int s_state;
180 unsigned long s_ttl; /* time until mds kills us */
181 unsigned long s_features;
182 u64 s_seq; /* incoming msg seq # */
183 struct mutex s_mutex; /* serialize session messages */
184
185 struct ceph_connection s_con;
186
187 struct ceph_auth_handshake s_auth;
188
189 atomic_t s_cap_gen; /* inc each time we get mds stale msg */
190 unsigned long s_cap_ttl; /* when session caps expire. protected by s_mutex */
191
192 /* protected by s_cap_lock */
193 spinlock_t s_cap_lock;
194 refcount_t s_ref;
195 struct list_head s_caps; /* all caps issued by this session */
196 struct ceph_cap *s_cap_iterator;
197 int s_nr_caps;
198 int s_num_cap_releases;
199 int s_cap_reconnect;
200 int s_readonly;
201 struct list_head s_cap_releases; /* waiting cap_release messages */
202 struct work_struct s_cap_release_work;
203
204 /* See ceph_inode_info->i_dirty_item. */
205 struct list_head s_cap_dirty; /* inodes w/ dirty caps */
206
207 /* See ceph_inode_info->i_flushing_item. */
208 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
209
210 unsigned long s_renew_requested; /* last time we sent a renew req */
211 u64 s_renew_seq;
212
213 struct list_head s_waiting; /* waiting requests */
214 struct list_head s_unsafe; /* unsafe requests */
215 struct xarray s_delegated_inos;
216 };
217
218 /*
219 * modes of choosing which MDS to send a request to
220 */
221 enum {
222 USE_ANY_MDS,
223 USE_RANDOM_MDS,
224 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
225 };
226
227 struct ceph_mds_request;
228 struct ceph_mds_client;
229
230 /*
231 * request completion callback
232 */
233 typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
234 struct ceph_mds_request *req);
235 /*
236 * wait for request completion callback
237 */
238 typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
239 struct ceph_mds_request *req);
240
241 /*
242 * an in-flight mds request
243 */
244 struct ceph_mds_request {
245 u64 r_tid; /* transaction id */
246 struct rb_node r_node;
247 struct ceph_mds_client *r_mdsc;
248
249 struct kref r_kref;
250 int r_op; /* mds op code */
251
252 /* operation on what? */
253 struct inode *r_inode; /* arg1 */
254 struct dentry *r_dentry; /* arg1 */
255 struct dentry *r_old_dentry; /* arg2: rename from or link from */
256 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
257 char *r_path1, *r_path2;
258 struct ceph_vino r_ino1, r_ino2;
259
260 struct inode *r_parent; /* parent dir inode */
261 struct inode *r_target_inode; /* resulting inode */
262
263 #define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
264 #define CEPH_MDS_R_ABORTED (2) /* call was aborted */
265 #define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
266 #define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
267 #define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
268 #define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
269 #define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
270 #define CEPH_MDS_R_ASYNC (8) /* async request */
271 unsigned long r_req_flags;
272
273 struct mutex r_fill_mutex;
274
275 union ceph_mds_request_args r_args;
276 int r_fmode; /* file mode, if expecting cap */
277 int r_request_release_offset;
278 const struct cred *r_cred;
279 struct timespec64 r_stamp;
280
281 /* for choosing which mds to send this request to */
282 int r_direct_mode;
283 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
284
285 /* data payload is used for xattr ops */
286 struct ceph_pagelist *r_pagelist;
287
288 /* what caps shall we drop? */
289 int r_inode_drop, r_inode_unless;
290 int r_dentry_drop, r_dentry_unless;
291 int r_old_dentry_drop, r_old_dentry_unless;
292 struct inode *r_old_inode;
293 int r_old_inode_drop, r_old_inode_unless;
294
295 struct ceph_msg *r_request; /* original request */
296 struct ceph_msg *r_reply;
297 struct ceph_mds_reply_info_parsed r_reply_info;
298 int r_err;
299 u32 r_readdir_offset;
300
301 struct page *r_locked_page;
302 int r_dir_caps;
303 int r_num_caps;
304
305 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
306 unsigned long r_started; /* start time to measure timeout against */
307 unsigned long r_start_latency; /* start time to measure latency */
308 unsigned long r_end_latency; /* finish time to measure latency */
309 unsigned long r_request_started; /* start time for mds request only,
310 used to measure lease durations */
311
312 /* link unsafe requests to parent directory, for fsync */
313 struct inode *r_unsafe_dir;
314 struct list_head r_unsafe_dir_item;
315
316 /* unsafe requests that modify the target inode */
317 struct list_head r_unsafe_target_item;
318
319 struct ceph_mds_session *r_session;
320
321 int r_attempts; /* resend attempts */
322 int r_num_fwd; /* number of forward attempts */
323 int r_resend_mds; /* mds to resend to next, if any*/
324 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
325 u64 r_deleg_ino;
326
327 struct list_head r_wait;
328 struct completion r_completion;
329 struct completion r_safe_completion;
330 ceph_mds_request_callback_t r_callback;
331 struct list_head r_unsafe_item; /* per-session unsafe list item */
332
333 long long r_dir_release_cnt;
334 long long r_dir_ordered_cnt;
335 int r_readdir_cache_idx;
336
337 struct ceph_cap_reservation r_caps_reservation;
338 };
339
340 struct ceph_pool_perm {
341 struct rb_node node;
342 int perm;
343 s64 pool;
344 size_t pool_ns_len;
345 char pool_ns[];
346 };
347
348 struct ceph_snapid_map {
349 struct rb_node node;
350 struct list_head lru;
351 atomic_t ref;
352 u64 snap;
353 dev_t dev;
354 unsigned long last_used;
355 };
356
357 /*
358 * node for list of quotarealm inodes that are not visible from the filesystem
359 * mountpoint, but required to handle, e.g. quotas.
360 */
361 struct ceph_quotarealm_inode {
362 struct rb_node node;
363 u64 ino;
364 unsigned long timeout; /* last time a lookup failed for this inode */
365 struct mutex mutex;
366 struct inode *inode;
367 };
368
369 struct cap_wait {
370 struct list_head list;
371 u64 ino;
372 pid_t tgid;
373 int need;
374 int want;
375 };
376
377 /*
378 * mds client state
379 */
380 struct ceph_mds_client {
381 struct ceph_fs_client *fsc;
382 struct mutex mutex; /* all nested structures */
383
384 struct ceph_mdsmap *mdsmap;
385 struct completion safe_umount_waiters;
386 wait_queue_head_t session_close_wq;
387 struct list_head waiting_for_map;
388 int mdsmap_err;
389
390 struct ceph_mds_session **sessions; /* NULL for mds if no session */
391 atomic_t num_sessions;
392 int max_sessions; /* len of sessions array */
393 int stopping; /* true if shutting down */
394
395 atomic64_t quotarealms_count; /* # realms with quota */
396 /*
397 * We keep a list of inodes we don't see in the mountpoint but that we
398 * need to track quota realms.
399 */
400 struct rb_root quotarealms_inodes;
401 struct mutex quotarealms_inodes_mutex;
402
403 /*
404 * snap_rwsem will cover cap linkage into snaprealms, and
405 * realm snap contexts. (later, we can do per-realm snap
406 * contexts locks..) the empty list contains realms with no
407 * references (implying they contain no inodes with caps) that
408 * should be destroyed.
409 */
410 u64 last_snap_seq;
411 struct rw_semaphore snap_rwsem;
412 struct rb_root snap_realms;
413 struct list_head snap_empty;
414 int num_snap_realms;
415 spinlock_t snap_empty_lock; /* protect snap_empty */
416
417 u64 last_tid; /* most recent mds request */
418 u64 oldest_tid; /* oldest incomplete mds request,
419 excluding setfilelock requests */
420 struct rb_root request_tree; /* pending mds requests */
421 struct delayed_work delayed_work; /* delayed work */
422 unsigned long last_renew_caps; /* last time we renewed our caps */
423 struct list_head cap_delay_list; /* caps with delayed release */
424 spinlock_t cap_delay_lock; /* protects cap_delay_list */
425 struct list_head snap_flush_list; /* cap_snaps ready to flush */
426 spinlock_t snap_flush_lock;
427
428 u64 last_cap_flush_tid;
429 struct list_head cap_flush_list;
430 struct list_head cap_dirty_migrating; /* ...that are migration... */
431 int num_cap_flushing; /* # caps we are flushing */
432 spinlock_t cap_dirty_lock; /* protects above items */
433 wait_queue_head_t cap_flushing_wq;
434
435 struct work_struct cap_reclaim_work;
436 atomic_t cap_reclaim_pending;
437
438 /*
439 * Cap reservations
440 *
441 * Maintain a global pool of preallocated struct ceph_caps, referenced
442 * by struct ceph_caps_reservations. This ensures that we preallocate
443 * memory needed to successfully process an MDS response. (If an MDS
444 * sends us cap information and we fail to process it, we will have
445 * problems due to the client and MDS being out of sync.)
446 *
447 * Reservations are 'owned' by a ceph_cap_reservation context.
448 */
449 spinlock_t caps_list_lock;
450 struct list_head caps_list; /* unused (reserved or
451 unreserved) */
452 struct list_head cap_wait_list;
453 int caps_total_count; /* total caps allocated */
454 int caps_use_count; /* in use */
455 int caps_use_max; /* max used caps */
456 int caps_reserve_count; /* unused, reserved */
457 int caps_avail_count; /* unused, unreserved */
458 int caps_min_count; /* keep at least this many
459 (unreserved) */
460 spinlock_t dentry_list_lock;
461 struct list_head dentry_leases; /* fifo list */
462 struct list_head dentry_dir_leases; /* lru list */
463
464 struct ceph_client_metric metric;
465
466 spinlock_t snapid_map_lock;
467 struct rb_root snapid_map_tree;
468 struct list_head snapid_map_lru;
469
470 struct rw_semaphore pool_perm_rwsem;
471 struct rb_root pool_perm_tree;
472
473 char nodename[__NEW_UTS_LEN + 1];
474 };
475
476 extern const char *ceph_mds_op_name(int op);
477
478 extern bool check_session_state(struct ceph_mds_session *s);
479 void inc_session_sequence(struct ceph_mds_session *s);
480
481 extern struct ceph_mds_session *
482 __ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
483
484 extern const char *ceph_session_state_name(int s);
485
486 extern struct ceph_mds_session *
487 ceph_get_mds_session(struct ceph_mds_session *s);
488 extern void ceph_put_mds_session(struct ceph_mds_session *s);
489
490 extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
491 struct ceph_msg *msg, int mds);
492
493 extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
494 extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
495 extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
496 extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
497
498 extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
499
500 extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
501 extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
502 struct inode *dir);
503 extern struct ceph_mds_request *
504 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
505 extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
506 struct inode *dir,
507 struct ceph_mds_request *req);
508 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
509 struct ceph_mds_request *req,
510 ceph_mds_request_wait_callback_t wait_func);
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_KILLABLE);
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