Loading...
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
30struct ceph_fs_client;
31struct 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 */
37struct 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
51struct 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 */
65struct 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 */
116enum {
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
127struct 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 */
167enum {
168 USE_ANY_MDS,
169 USE_RANDOM_MDS,
170 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
171};
172
173struct ceph_mds_request;
174struct ceph_mds_client;
175
176/*
177 * request completion callback
178 */
179typedef 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 */
184typedef 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 */
190struct 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
277struct 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 */
288struct 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
363extern const char *ceph_mds_op_name(int op);
364
365extern struct ceph_mds_session *
366__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
367
368static inline struct ceph_mds_session *
369ceph_get_mds_session(struct ceph_mds_session *s)
370{
371 atomic_inc(&s->s_ref);
372 return s;
373}
374
375extern const char *ceph_session_state_name(int s);
376
377extern void ceph_put_mds_session(struct ceph_mds_session *s);
378
379extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
380 struct ceph_msg *msg, int mds);
381
382extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
383extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
384extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
385extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
386
387extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
388
389extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
390extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
391 struct inode *dir);
392extern struct ceph_mds_request *
393ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
394extern void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
395 struct ceph_mds_request *req);
396extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
397 struct inode *dir,
398 struct ceph_mds_request *req);
399static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
400{
401 kref_get(&req->r_kref);
402}
403extern void ceph_mdsc_release_request(struct kref *kref);
404static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
405{
406 kref_put(&req->r_kref, ceph_mdsc_release_request);
407}
408
409extern void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
410 struct ceph_mds_session *session);
411
412extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
413
414extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
415 int stop_on_nosnap);
416
417extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
418extern 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
423extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
424 struct ceph_msg *msg);
425extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
426 struct ceph_msg *msg);
427
428extern struct ceph_mds_session *
429ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
430extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
431 struct ceph_mds_session *session);
432
433#endif
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 */
24enum 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 CEPHFS_FEATURE_ALTERNATE_NAME,
33 CEPHFS_FEATURE_NOTIFY_SESSION_STATE,
34 CEPHFS_FEATURE_OP_GETVXATTR,
35
36 CEPHFS_FEATURE_MAX = CEPHFS_FEATURE_OP_GETVXATTR,
37};
38
39#define CEPHFS_FEATURES_CLIENT_SUPPORTED { \
40 0, 1, 2, 3, 4, 5, 6, 7, \
41 CEPHFS_FEATURE_MIMIC, \
42 CEPHFS_FEATURE_REPLY_ENCODING, \
43 CEPHFS_FEATURE_LAZY_CAP_WANTED, \
44 CEPHFS_FEATURE_MULTI_RECONNECT, \
45 CEPHFS_FEATURE_DELEG_INO, \
46 CEPHFS_FEATURE_METRIC_COLLECT, \
47 CEPHFS_FEATURE_NOTIFY_SESSION_STATE, \
48 CEPHFS_FEATURE_OP_GETVXATTR, \
49}
50
51/*
52 * Some lock dependencies:
53 *
54 * session->s_mutex
55 * mdsc->mutex
56 *
57 * mdsc->snap_rwsem
58 *
59 * ci->i_ceph_lock
60 * mdsc->snap_flush_lock
61 * mdsc->cap_delay_lock
62 *
63 */
64
65struct ceph_fs_client;
66struct ceph_cap;
67
68/*
69 * parsed info about a single inode. pointers are into the encoded
70 * on-wire structures within the mds reply message payload.
71 */
72struct ceph_mds_reply_info_in {
73 struct ceph_mds_reply_inode *in;
74 struct ceph_dir_layout dir_layout;
75 u32 symlink_len;
76 char *symlink;
77 u32 xattr_len;
78 char *xattr_data;
79 u64 inline_version;
80 u32 inline_len;
81 char *inline_data;
82 u32 pool_ns_len;
83 char *pool_ns_data;
84 u64 max_bytes;
85 u64 max_files;
86 s32 dir_pin;
87 struct ceph_timespec btime;
88 struct ceph_timespec snap_btime;
89 u64 rsnaps;
90 u64 change_attr;
91};
92
93struct ceph_mds_reply_dir_entry {
94 char *name;
95 u32 name_len;
96 struct ceph_mds_reply_lease *lease;
97 struct ceph_mds_reply_info_in inode;
98 loff_t offset;
99};
100
101struct ceph_mds_reply_xattr {
102 char *xattr_value;
103 size_t xattr_value_len;
104};
105
106/*
107 * parsed info about an mds reply, including information about
108 * either: 1) the target inode and/or its parent directory and dentry,
109 * and directory contents (for readdir results), or
110 * 2) the file range lock info (for fcntl F_GETLK results).
111 */
112struct ceph_mds_reply_info_parsed {
113 struct ceph_mds_reply_head *head;
114
115 /* trace */
116 struct ceph_mds_reply_info_in diri, targeti;
117 struct ceph_mds_reply_dirfrag *dirfrag;
118 char *dname;
119 u32 dname_len;
120 struct ceph_mds_reply_lease *dlease;
121 struct ceph_mds_reply_xattr xattr_info;
122
123 /* extra */
124 union {
125 /* for fcntl F_GETLK results */
126 struct ceph_filelock *filelock_reply;
127
128 /* for readdir results */
129 struct {
130 struct ceph_mds_reply_dirfrag *dir_dir;
131 size_t dir_buf_size;
132 int dir_nr;
133 bool dir_end;
134 bool dir_complete;
135 bool hash_order;
136 bool offset_hash;
137 struct ceph_mds_reply_dir_entry *dir_entries;
138 };
139
140 /* for create results */
141 struct {
142 bool has_create_ino;
143 u64 ino;
144 };
145 };
146
147 /* encoded blob describing snapshot contexts for certain
148 operations (e.g., open) */
149 void *snapblob;
150 int snapblob_len;
151};
152
153
154/*
155 * cap releases are batched and sent to the MDS en masse.
156 *
157 * Account for per-message overhead of mds_cap_release header
158 * and __le32 for osd epoch barrier trailing field.
159 */
160#define CEPH_CAPS_PER_RELEASE ((PAGE_SIZE - sizeof(u32) - \
161 sizeof(struct ceph_mds_cap_release)) / \
162 sizeof(struct ceph_mds_cap_item))
163
164
165/*
166 * state associated with each MDS<->client session
167 */
168enum {
169 CEPH_MDS_SESSION_NEW = 1,
170 CEPH_MDS_SESSION_OPENING = 2,
171 CEPH_MDS_SESSION_OPEN = 3,
172 CEPH_MDS_SESSION_HUNG = 4,
173 CEPH_MDS_SESSION_RESTARTING = 5,
174 CEPH_MDS_SESSION_RECONNECTING = 6,
175 CEPH_MDS_SESSION_CLOSING = 7,
176 CEPH_MDS_SESSION_CLOSED = 8,
177 CEPH_MDS_SESSION_REJECTED = 9,
178};
179
180struct ceph_mds_session {
181 struct ceph_mds_client *s_mdsc;
182 int s_mds;
183 int s_state;
184 unsigned long s_ttl; /* time until mds kills us */
185 unsigned long s_features;
186 u64 s_seq; /* incoming msg seq # */
187 struct mutex s_mutex; /* serialize session messages */
188
189 struct ceph_connection s_con;
190
191 struct ceph_auth_handshake s_auth;
192
193 atomic_t s_cap_gen; /* inc each time we get mds stale msg */
194 unsigned long s_cap_ttl; /* when session caps expire. protected by s_mutex */
195
196 /* protected by s_cap_lock */
197 spinlock_t s_cap_lock;
198 refcount_t s_ref;
199 struct list_head s_caps; /* all caps issued by this session */
200 struct ceph_cap *s_cap_iterator;
201 int s_nr_caps;
202 int s_num_cap_releases;
203 int s_cap_reconnect;
204 int s_readonly;
205 struct list_head s_cap_releases; /* waiting cap_release messages */
206 struct work_struct s_cap_release_work;
207
208 /* See ceph_inode_info->i_dirty_item. */
209 struct list_head s_cap_dirty; /* inodes w/ dirty caps */
210
211 /* See ceph_inode_info->i_flushing_item. */
212 struct list_head s_cap_flushing; /* inodes w/ flushing caps */
213
214 unsigned long s_renew_requested; /* last time we sent a renew req */
215 u64 s_renew_seq;
216
217 struct list_head s_waiting; /* waiting requests */
218 struct list_head s_unsafe; /* unsafe requests */
219 struct xarray s_delegated_inos;
220};
221
222/*
223 * modes of choosing which MDS to send a request to
224 */
225enum {
226 USE_ANY_MDS,
227 USE_RANDOM_MDS,
228 USE_AUTH_MDS, /* prefer authoritative mds for this metadata item */
229};
230
231struct ceph_mds_request;
232struct ceph_mds_client;
233
234/*
235 * request completion callback
236 */
237typedef void (*ceph_mds_request_callback_t) (struct ceph_mds_client *mdsc,
238 struct ceph_mds_request *req);
239/*
240 * wait for request completion callback
241 */
242typedef int (*ceph_mds_request_wait_callback_t) (struct ceph_mds_client *mdsc,
243 struct ceph_mds_request *req);
244
245/*
246 * an in-flight mds request
247 */
248struct ceph_mds_request {
249 u64 r_tid; /* transaction id */
250 struct rb_node r_node;
251 struct ceph_mds_client *r_mdsc;
252
253 struct kref r_kref;
254 int r_op; /* mds op code */
255
256 /* operation on what? */
257 struct inode *r_inode; /* arg1 */
258 struct dentry *r_dentry; /* arg1 */
259 struct dentry *r_old_dentry; /* arg2: rename from or link from */
260 struct inode *r_old_dentry_dir; /* arg2: old dentry's parent dir */
261 char *r_path1, *r_path2;
262 struct ceph_vino r_ino1, r_ino2;
263
264 struct inode *r_parent; /* parent dir inode */
265 struct inode *r_target_inode; /* resulting inode */
266
267#define CEPH_MDS_R_DIRECT_IS_HASH (1) /* r_direct_hash is valid */
268#define CEPH_MDS_R_ABORTED (2) /* call was aborted */
269#define CEPH_MDS_R_GOT_UNSAFE (3) /* got an unsafe reply */
270#define CEPH_MDS_R_GOT_SAFE (4) /* got a safe reply */
271#define CEPH_MDS_R_GOT_RESULT (5) /* got a result */
272#define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
273#define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
274#define CEPH_MDS_R_ASYNC (8) /* async request */
275 unsigned long r_req_flags;
276
277 struct mutex r_fill_mutex;
278
279 union ceph_mds_request_args r_args;
280 int r_fmode; /* file mode, if expecting cap */
281 int r_request_release_offset;
282 const struct cred *r_cred;
283 struct timespec64 r_stamp;
284
285 /* for choosing which mds to send this request to */
286 int r_direct_mode;
287 u32 r_direct_hash; /* choose dir frag based on this dentry hash */
288
289 /* data payload is used for xattr ops */
290 struct ceph_pagelist *r_pagelist;
291
292 /* what caps shall we drop? */
293 int r_inode_drop, r_inode_unless;
294 int r_dentry_drop, r_dentry_unless;
295 int r_old_dentry_drop, r_old_dentry_unless;
296 struct inode *r_old_inode;
297 int r_old_inode_drop, r_old_inode_unless;
298
299 struct ceph_msg *r_request; /* original request */
300 struct ceph_msg *r_reply;
301 struct ceph_mds_reply_info_parsed r_reply_info;
302 int r_err;
303 u32 r_readdir_offset;
304
305 struct page *r_locked_page;
306 int r_dir_caps;
307 int r_num_caps;
308
309 unsigned long r_timeout; /* optional. jiffies, 0 is "wait forever" */
310 unsigned long r_started; /* start time to measure timeout against */
311 unsigned long r_start_latency; /* start time to measure latency */
312 unsigned long r_end_latency; /* finish time to measure latency */
313 unsigned long r_request_started; /* start time for mds request only,
314 used to measure lease durations */
315
316 /* link unsafe requests to parent directory, for fsync */
317 struct inode *r_unsafe_dir;
318 struct list_head r_unsafe_dir_item;
319
320 /* unsafe requests that modify the target inode */
321 struct list_head r_unsafe_target_item;
322
323 struct ceph_mds_session *r_session;
324
325 int r_attempts; /* resend attempts */
326 int r_num_fwd; /* number of forward attempts */
327 int r_resend_mds; /* mds to resend to next, if any*/
328 u32 r_sent_on_mseq; /* cap mseq request was sent at*/
329 u64 r_deleg_ino;
330
331 struct list_head r_wait;
332 struct completion r_completion;
333 struct completion r_safe_completion;
334 ceph_mds_request_callback_t r_callback;
335 struct list_head r_unsafe_item; /* per-session unsafe list item */
336
337 long long r_dir_release_cnt;
338 long long r_dir_ordered_cnt;
339 int r_readdir_cache_idx;
340
341 int r_feature_needed;
342
343 struct ceph_cap_reservation r_caps_reservation;
344};
345
346struct ceph_pool_perm {
347 struct rb_node node;
348 int perm;
349 s64 pool;
350 size_t pool_ns_len;
351 char pool_ns[];
352};
353
354struct ceph_snapid_map {
355 struct rb_node node;
356 struct list_head lru;
357 atomic_t ref;
358 u64 snap;
359 dev_t dev;
360 unsigned long last_used;
361};
362
363/*
364 * node for list of quotarealm inodes that are not visible from the filesystem
365 * mountpoint, but required to handle, e.g. quotas.
366 */
367struct ceph_quotarealm_inode {
368 struct rb_node node;
369 u64 ino;
370 unsigned long timeout; /* last time a lookup failed for this inode */
371 struct mutex mutex;
372 struct inode *inode;
373};
374
375struct cap_wait {
376 struct list_head list;
377 u64 ino;
378 pid_t tgid;
379 int need;
380 int want;
381};
382
383/*
384 * mds client state
385 */
386struct ceph_mds_client {
387 struct ceph_fs_client *fsc;
388 struct mutex mutex; /* all nested structures */
389
390 struct ceph_mdsmap *mdsmap;
391 struct completion safe_umount_waiters;
392 wait_queue_head_t session_close_wq;
393 struct list_head waiting_for_map;
394 int mdsmap_err;
395
396 struct ceph_mds_session **sessions; /* NULL for mds if no session */
397 atomic_t num_sessions;
398 int max_sessions; /* len of sessions array */
399 int stopping; /* true if shutting down */
400
401 atomic64_t quotarealms_count; /* # realms with quota */
402 /*
403 * We keep a list of inodes we don't see in the mountpoint but that we
404 * need to track quota realms.
405 */
406 struct rb_root quotarealms_inodes;
407 struct mutex quotarealms_inodes_mutex;
408
409 /*
410 * snap_rwsem will cover cap linkage into snaprealms, and
411 * realm snap contexts. (later, we can do per-realm snap
412 * contexts locks..) the empty list contains realms with no
413 * references (implying they contain no inodes with caps) that
414 * should be destroyed.
415 */
416 u64 last_snap_seq;
417 struct rw_semaphore snap_rwsem;
418 struct rb_root snap_realms;
419 struct list_head snap_empty;
420 int num_snap_realms;
421 spinlock_t snap_empty_lock; /* protect snap_empty */
422
423 u64 last_tid; /* most recent mds request */
424 u64 oldest_tid; /* oldest incomplete mds request,
425 excluding setfilelock requests */
426 struct rb_root request_tree; /* pending mds requests */
427 struct delayed_work delayed_work; /* delayed work */
428 unsigned long last_renew_caps; /* last time we renewed our caps */
429 struct list_head cap_delay_list; /* caps with delayed release */
430 spinlock_t cap_delay_lock; /* protects cap_delay_list */
431 struct list_head snap_flush_list; /* cap_snaps ready to flush */
432 spinlock_t snap_flush_lock;
433
434 u64 last_cap_flush_tid;
435 struct list_head cap_flush_list;
436 struct list_head cap_dirty_migrating; /* ...that are migration... */
437 int num_cap_flushing; /* # caps we are flushing */
438 spinlock_t cap_dirty_lock; /* protects above items */
439 wait_queue_head_t cap_flushing_wq;
440
441 struct work_struct cap_reclaim_work;
442 atomic_t cap_reclaim_pending;
443
444 /*
445 * Cap reservations
446 *
447 * Maintain a global pool of preallocated struct ceph_caps, referenced
448 * by struct ceph_caps_reservations. This ensures that we preallocate
449 * memory needed to successfully process an MDS response. (If an MDS
450 * sends us cap information and we fail to process it, we will have
451 * problems due to the client and MDS being out of sync.)
452 *
453 * Reservations are 'owned' by a ceph_cap_reservation context.
454 */
455 spinlock_t caps_list_lock;
456 struct list_head caps_list; /* unused (reserved or
457 unreserved) */
458 struct list_head cap_wait_list;
459 int caps_total_count; /* total caps allocated */
460 int caps_use_count; /* in use */
461 int caps_use_max; /* max used caps */
462 int caps_reserve_count; /* unused, reserved */
463 int caps_avail_count; /* unused, unreserved */
464 int caps_min_count; /* keep at least this many
465 (unreserved) */
466 spinlock_t dentry_list_lock;
467 struct list_head dentry_leases; /* fifo list */
468 struct list_head dentry_dir_leases; /* lru list */
469
470 struct ceph_client_metric metric;
471
472 spinlock_t snapid_map_lock;
473 struct rb_root snapid_map_tree;
474 struct list_head snapid_map_lru;
475
476 struct rw_semaphore pool_perm_rwsem;
477 struct rb_root pool_perm_tree;
478
479 char nodename[__NEW_UTS_LEN + 1];
480};
481
482extern const char *ceph_mds_op_name(int op);
483
484extern bool check_session_state(struct ceph_mds_session *s);
485void inc_session_sequence(struct ceph_mds_session *s);
486
487extern struct ceph_mds_session *
488__ceph_lookup_mds_session(struct ceph_mds_client *, int mds);
489
490extern const char *ceph_session_state_name(int s);
491
492extern struct ceph_mds_session *
493ceph_get_mds_session(struct ceph_mds_session *s);
494extern void ceph_put_mds_session(struct ceph_mds_session *s);
495
496extern int ceph_send_msg_mds(struct ceph_mds_client *mdsc,
497 struct ceph_msg *msg, int mds);
498
499extern int ceph_mdsc_init(struct ceph_fs_client *fsc);
500extern void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc);
501extern void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc);
502extern void ceph_mdsc_destroy(struct ceph_fs_client *fsc);
503
504extern void ceph_mdsc_sync(struct ceph_mds_client *mdsc);
505
506extern void ceph_invalidate_dir_request(struct ceph_mds_request *req);
507extern int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
508 struct inode *dir);
509extern struct ceph_mds_request *
510ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode);
511extern int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
512 struct inode *dir,
513 struct ceph_mds_request *req);
514int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
515 struct ceph_mds_request *req,
516 ceph_mds_request_wait_callback_t wait_func);
517extern int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
518 struct inode *dir,
519 struct ceph_mds_request *req);
520extern void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req);
521extern void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req);
522static inline void ceph_mdsc_get_request(struct ceph_mds_request *req)
523{
524 kref_get(&req->r_kref);
525}
526extern void ceph_mdsc_release_request(struct kref *kref);
527static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
528{
529 kref_put(&req->r_kref, ceph_mdsc_release_request);
530}
531
532extern void send_flush_mdlog(struct ceph_mds_session *s);
533extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
534 void (*cb)(struct ceph_mds_session *),
535 bool check_state);
536extern struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq);
537extern void __ceph_queue_cap_release(struct ceph_mds_session *session,
538 struct ceph_cap *cap);
539extern void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
540 struct ceph_mds_session *session);
541extern void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc);
542extern void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr);
543extern int ceph_iterate_session_caps(struct ceph_mds_session *session,
544 int (*cb)(struct inode *,
545 struct ceph_cap *, void *),
546 void *arg);
547extern void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc);
548
549static inline void ceph_mdsc_free_path(char *path, int len)
550{
551 if (!IS_ERR_OR_NULL(path))
552 __putname(path - (PATH_MAX - 1 - len));
553}
554
555extern char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
556 int stop_on_nosnap);
557
558extern void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry);
559extern void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
560 struct dentry *dentry, char action,
561 u32 seq);
562
563extern void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc,
564 struct ceph_msg *msg);
565extern void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc,
566 struct ceph_msg *msg);
567
568extern struct ceph_mds_session *
569ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target);
570extern void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
571 struct ceph_mds_session *session);
572
573extern int ceph_trim_caps(struct ceph_mds_client *mdsc,
574 struct ceph_mds_session *session,
575 int max_caps);
576
577static inline int ceph_wait_on_async_create(struct inode *inode)
578{
579 struct ceph_inode_info *ci = ceph_inode(inode);
580
581 return wait_on_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT,
582 TASK_KILLABLE);
583}
584
585extern int ceph_wait_on_conflict_unlink(struct dentry *dentry);
586extern u64 ceph_get_deleg_ino(struct ceph_mds_session *session);
587extern int ceph_restore_deleg_ino(struct ceph_mds_session *session, u64 ino);
588#endif