Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
2f2dc053 | 2 | |
496e5955 | 3 | #include <linux/fs.h> |
2f2dc053 | 4 | #include <linux/wait.h> |
5a0e3ad6 | 5 | #include <linux/slab.h> |
54008399 | 6 | #include <linux/gfp.h> |
2f2dc053 | 7 | #include <linux/sched.h> |
3d14c5d2 YS |
8 | #include <linux/debugfs.h> |
9 | #include <linux/seq_file.h> | |
dbd0c8bf | 10 | #include <linux/utsname.h> |
3e0708b9 | 11 | #include <linux/ratelimit.h> |
2f2dc053 | 12 | |
2f2dc053 | 13 | #include "super.h" |
3d14c5d2 YS |
14 | #include "mds_client.h" |
15 | ||
1fe60e51 | 16 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
17 | #include <linux/ceph/messenger.h> |
18 | #include <linux/ceph/decode.h> | |
19 | #include <linux/ceph/pagelist.h> | |
20 | #include <linux/ceph/auth.h> | |
21 | #include <linux/ceph/debugfs.h> | |
2f2dc053 SW |
22 | |
23 | /* | |
24 | * A cluster of MDS (metadata server) daemons is responsible for | |
25 | * managing the file system namespace (the directory hierarchy and | |
26 | * inodes) and for coordinating shared access to storage. Metadata is | |
27 | * partitioning hierarchically across a number of servers, and that | |
28 | * partition varies over time as the cluster adjusts the distribution | |
29 | * in order to balance load. | |
30 | * | |
31 | * The MDS client is primarily responsible to managing synchronous | |
32 | * metadata requests for operations like open, unlink, and so forth. | |
33 | * If there is a MDS failure, we find out about it when we (possibly | |
34 | * request and) receive a new MDS map, and can resubmit affected | |
35 | * requests. | |
36 | * | |
37 | * For the most part, though, we take advantage of a lossless | |
38 | * communications channel to the MDS, and do not need to worry about | |
39 | * timing out or resubmitting requests. | |
40 | * | |
41 | * We maintain a stateful "session" with each MDS we interact with. | |
42 | * Within each session, we sent periodic heartbeat messages to ensure | |
43 | * any capabilities or leases we have been issues remain valid. If | |
44 | * the session times out and goes stale, our leases and capabilities | |
45 | * are no longer valid. | |
46 | */ | |
47 | ||
20cb34ae | 48 | struct ceph_reconnect_state { |
44c99757 | 49 | int nr_caps; |
20cb34ae SW |
50 | struct ceph_pagelist *pagelist; |
51 | bool flock; | |
52 | }; | |
53 | ||
2f2dc053 SW |
54 | static void __wake_requests(struct ceph_mds_client *mdsc, |
55 | struct list_head *head); | |
56 | ||
9e32789f | 57 | static const struct ceph_connection_operations mds_con_ops; |
2f2dc053 SW |
58 | |
59 | ||
60 | /* | |
61 | * mds reply parsing | |
62 | */ | |
63 | ||
64 | /* | |
65 | * parse individual inode info | |
66 | */ | |
67 | static int parse_reply_info_in(void **p, void *end, | |
14303d20 | 68 | struct ceph_mds_reply_info_in *info, |
12b4629a | 69 | u64 features) |
2f2dc053 SW |
70 | { |
71 | int err = -EIO; | |
72 | ||
73 | info->in = *p; | |
74 | *p += sizeof(struct ceph_mds_reply_inode) + | |
75 | sizeof(*info->in->fragtree.splits) * | |
76 | le32_to_cpu(info->in->fragtree.nsplits); | |
77 | ||
78 | ceph_decode_32_safe(p, end, info->symlink_len, bad); | |
79 | ceph_decode_need(p, end, info->symlink_len, bad); | |
80 | info->symlink = *p; | |
81 | *p += info->symlink_len; | |
82 | ||
14303d20 SW |
83 | if (features & CEPH_FEATURE_DIRLAYOUTHASH) |
84 | ceph_decode_copy_safe(p, end, &info->dir_layout, | |
85 | sizeof(info->dir_layout), bad); | |
86 | else | |
87 | memset(&info->dir_layout, 0, sizeof(info->dir_layout)); | |
88 | ||
2f2dc053 SW |
89 | ceph_decode_32_safe(p, end, info->xattr_len, bad); |
90 | ceph_decode_need(p, end, info->xattr_len, bad); | |
91 | info->xattr_data = *p; | |
92 | *p += info->xattr_len; | |
fb01d1f8 YZ |
93 | |
94 | if (features & CEPH_FEATURE_MDS_INLINE_DATA) { | |
95 | ceph_decode_64_safe(p, end, info->inline_version, bad); | |
96 | ceph_decode_32_safe(p, end, info->inline_len, bad); | |
97 | ceph_decode_need(p, end, info->inline_len, bad); | |
98 | info->inline_data = *p; | |
99 | *p += info->inline_len; | |
100 | } else | |
101 | info->inline_version = CEPH_INLINE_NONE; | |
102 | ||
5ea5c5e0 YZ |
103 | if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) { |
104 | ceph_decode_32_safe(p, end, info->pool_ns_len, bad); | |
105 | ceph_decode_need(p, end, info->pool_ns_len, bad); | |
106 | *p += info->pool_ns_len; | |
107 | } else { | |
108 | info->pool_ns_len = 0; | |
109 | } | |
110 | ||
2f2dc053 SW |
111 | return 0; |
112 | bad: | |
113 | return err; | |
114 | } | |
115 | ||
116 | /* | |
117 | * parse a normal reply, which may contain a (dir+)dentry and/or a | |
118 | * target inode. | |
119 | */ | |
120 | static int parse_reply_info_trace(void **p, void *end, | |
14303d20 | 121 | struct ceph_mds_reply_info_parsed *info, |
12b4629a | 122 | u64 features) |
2f2dc053 SW |
123 | { |
124 | int err; | |
125 | ||
126 | if (info->head->is_dentry) { | |
14303d20 | 127 | err = parse_reply_info_in(p, end, &info->diri, features); |
2f2dc053 SW |
128 | if (err < 0) |
129 | goto out_bad; | |
130 | ||
131 | if (unlikely(*p + sizeof(*info->dirfrag) > end)) | |
132 | goto bad; | |
133 | info->dirfrag = *p; | |
134 | *p += sizeof(*info->dirfrag) + | |
135 | sizeof(u32)*le32_to_cpu(info->dirfrag->ndist); | |
136 | if (unlikely(*p > end)) | |
137 | goto bad; | |
138 | ||
139 | ceph_decode_32_safe(p, end, info->dname_len, bad); | |
140 | ceph_decode_need(p, end, info->dname_len, bad); | |
141 | info->dname = *p; | |
142 | *p += info->dname_len; | |
143 | info->dlease = *p; | |
144 | *p += sizeof(*info->dlease); | |
145 | } | |
146 | ||
147 | if (info->head->is_target) { | |
14303d20 | 148 | err = parse_reply_info_in(p, end, &info->targeti, features); |
2f2dc053 SW |
149 | if (err < 0) |
150 | goto out_bad; | |
151 | } | |
152 | ||
153 | if (unlikely(*p != end)) | |
154 | goto bad; | |
155 | return 0; | |
156 | ||
157 | bad: | |
158 | err = -EIO; | |
159 | out_bad: | |
160 | pr_err("problem parsing mds trace %d\n", err); | |
161 | return err; | |
162 | } | |
163 | ||
164 | /* | |
165 | * parse readdir results | |
166 | */ | |
167 | static int parse_reply_info_dir(void **p, void *end, | |
14303d20 | 168 | struct ceph_mds_reply_info_parsed *info, |
12b4629a | 169 | u64 features) |
2f2dc053 SW |
170 | { |
171 | u32 num, i = 0; | |
172 | int err; | |
173 | ||
174 | info->dir_dir = *p; | |
175 | if (*p + sizeof(*info->dir_dir) > end) | |
176 | goto bad; | |
177 | *p += sizeof(*info->dir_dir) + | |
178 | sizeof(u32)*le32_to_cpu(info->dir_dir->ndist); | |
179 | if (*p > end) | |
180 | goto bad; | |
181 | ||
182 | ceph_decode_need(p, end, sizeof(num) + 2, bad); | |
c89136ea | 183 | num = ceph_decode_32(p); |
956d39d6 YZ |
184 | { |
185 | u16 flags = ceph_decode_16(p); | |
186 | info->dir_end = !!(flags & CEPH_READDIR_FRAG_END); | |
187 | info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE); | |
f3c4ebe6 | 188 | info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER); |
956d39d6 | 189 | } |
2f2dc053 SW |
190 | if (num == 0) |
191 | goto done; | |
192 | ||
2a5beea3 YZ |
193 | BUG_ON(!info->dir_entries); |
194 | if ((unsigned long)(info->dir_entries + num) > | |
195 | (unsigned long)info->dir_entries + info->dir_buf_size) { | |
54008399 YZ |
196 | pr_err("dir contents are larger than expected\n"); |
197 | WARN_ON(1); | |
198 | goto bad; | |
199 | } | |
2f2dc053 | 200 | |
54008399 | 201 | info->dir_nr = num; |
2f2dc053 | 202 | while (num) { |
2a5beea3 | 203 | struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i; |
2f2dc053 SW |
204 | /* dentry */ |
205 | ceph_decode_need(p, end, sizeof(u32)*2, bad); | |
2a5beea3 YZ |
206 | rde->name_len = ceph_decode_32(p); |
207 | ceph_decode_need(p, end, rde->name_len, bad); | |
208 | rde->name = *p; | |
209 | *p += rde->name_len; | |
210 | dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name); | |
211 | rde->lease = *p; | |
2f2dc053 SW |
212 | *p += sizeof(struct ceph_mds_reply_lease); |
213 | ||
214 | /* inode */ | |
2a5beea3 | 215 | err = parse_reply_info_in(p, end, &rde->inode, features); |
2f2dc053 SW |
216 | if (err < 0) |
217 | goto out_bad; | |
8974eebd YZ |
218 | /* ceph_readdir_prepopulate() will update it */ |
219 | rde->offset = 0; | |
2f2dc053 SW |
220 | i++; |
221 | num--; | |
222 | } | |
223 | ||
224 | done: | |
225 | if (*p != end) | |
226 | goto bad; | |
227 | return 0; | |
228 | ||
229 | bad: | |
230 | err = -EIO; | |
231 | out_bad: | |
232 | pr_err("problem parsing dir contents %d\n", err); | |
233 | return err; | |
234 | } | |
235 | ||
25933abd HS |
236 | /* |
237 | * parse fcntl F_GETLK results | |
238 | */ | |
239 | static int parse_reply_info_filelock(void **p, void *end, | |
14303d20 | 240 | struct ceph_mds_reply_info_parsed *info, |
12b4629a | 241 | u64 features) |
25933abd HS |
242 | { |
243 | if (*p + sizeof(*info->filelock_reply) > end) | |
244 | goto bad; | |
245 | ||
246 | info->filelock_reply = *p; | |
247 | *p += sizeof(*info->filelock_reply); | |
248 | ||
249 | if (unlikely(*p != end)) | |
250 | goto bad; | |
251 | return 0; | |
252 | ||
253 | bad: | |
254 | return -EIO; | |
255 | } | |
256 | ||
6e8575fa SL |
257 | /* |
258 | * parse create results | |
259 | */ | |
260 | static int parse_reply_info_create(void **p, void *end, | |
261 | struct ceph_mds_reply_info_parsed *info, | |
12b4629a | 262 | u64 features) |
6e8575fa SL |
263 | { |
264 | if (features & CEPH_FEATURE_REPLY_CREATE_INODE) { | |
265 | if (*p == end) { | |
266 | info->has_create_ino = false; | |
267 | } else { | |
268 | info->has_create_ino = true; | |
269 | info->ino = ceph_decode_64(p); | |
270 | } | |
271 | } | |
272 | ||
273 | if (unlikely(*p != end)) | |
274 | goto bad; | |
275 | return 0; | |
276 | ||
277 | bad: | |
278 | return -EIO; | |
279 | } | |
280 | ||
25933abd HS |
281 | /* |
282 | * parse extra results | |
283 | */ | |
284 | static int parse_reply_info_extra(void **p, void *end, | |
14303d20 | 285 | struct ceph_mds_reply_info_parsed *info, |
12b4629a | 286 | u64 features) |
25933abd HS |
287 | { |
288 | if (info->head->op == CEPH_MDS_OP_GETFILELOCK) | |
14303d20 | 289 | return parse_reply_info_filelock(p, end, info, features); |
8a034497 YZ |
290 | else if (info->head->op == CEPH_MDS_OP_READDIR || |
291 | info->head->op == CEPH_MDS_OP_LSSNAP) | |
14303d20 | 292 | return parse_reply_info_dir(p, end, info, features); |
6e8575fa SL |
293 | else if (info->head->op == CEPH_MDS_OP_CREATE) |
294 | return parse_reply_info_create(p, end, info, features); | |
295 | else | |
296 | return -EIO; | |
25933abd HS |
297 | } |
298 | ||
2f2dc053 SW |
299 | /* |
300 | * parse entire mds reply | |
301 | */ | |
302 | static int parse_reply_info(struct ceph_msg *msg, | |
14303d20 | 303 | struct ceph_mds_reply_info_parsed *info, |
12b4629a | 304 | u64 features) |
2f2dc053 SW |
305 | { |
306 | void *p, *end; | |
307 | u32 len; | |
308 | int err; | |
309 | ||
310 | info->head = msg->front.iov_base; | |
311 | p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head); | |
312 | end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head); | |
313 | ||
314 | /* trace */ | |
315 | ceph_decode_32_safe(&p, end, len, bad); | |
316 | if (len > 0) { | |
32852a81 | 317 | ceph_decode_need(&p, end, len, bad); |
14303d20 | 318 | err = parse_reply_info_trace(&p, p+len, info, features); |
2f2dc053 SW |
319 | if (err < 0) |
320 | goto out_bad; | |
321 | } | |
322 | ||
25933abd | 323 | /* extra */ |
2f2dc053 SW |
324 | ceph_decode_32_safe(&p, end, len, bad); |
325 | if (len > 0) { | |
32852a81 | 326 | ceph_decode_need(&p, end, len, bad); |
14303d20 | 327 | err = parse_reply_info_extra(&p, p+len, info, features); |
2f2dc053 SW |
328 | if (err < 0) |
329 | goto out_bad; | |
330 | } | |
331 | ||
332 | /* snap blob */ | |
333 | ceph_decode_32_safe(&p, end, len, bad); | |
334 | info->snapblob_len = len; | |
335 | info->snapblob = p; | |
336 | p += len; | |
337 | ||
338 | if (p != end) | |
339 | goto bad; | |
340 | return 0; | |
341 | ||
342 | bad: | |
343 | err = -EIO; | |
344 | out_bad: | |
345 | pr_err("mds parse_reply err %d\n", err); | |
346 | return err; | |
347 | } | |
348 | ||
349 | static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info) | |
350 | { | |
2a5beea3 | 351 | if (!info->dir_entries) |
54008399 | 352 | return; |
2a5beea3 | 353 | free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size)); |
2f2dc053 SW |
354 | } |
355 | ||
356 | ||
357 | /* | |
358 | * sessions | |
359 | */ | |
a687ecaf | 360 | const char *ceph_session_state_name(int s) |
2f2dc053 SW |
361 | { |
362 | switch (s) { | |
363 | case CEPH_MDS_SESSION_NEW: return "new"; | |
364 | case CEPH_MDS_SESSION_OPENING: return "opening"; | |
365 | case CEPH_MDS_SESSION_OPEN: return "open"; | |
366 | case CEPH_MDS_SESSION_HUNG: return "hung"; | |
367 | case CEPH_MDS_SESSION_CLOSING: return "closing"; | |
44ca18f2 | 368 | case CEPH_MDS_SESSION_RESTARTING: return "restarting"; |
2f2dc053 SW |
369 | case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting"; |
370 | default: return "???"; | |
371 | } | |
372 | } | |
373 | ||
374 | static struct ceph_mds_session *get_session(struct ceph_mds_session *s) | |
375 | { | |
376 | if (atomic_inc_not_zero(&s->s_ref)) { | |
377 | dout("mdsc get_session %p %d -> %d\n", s, | |
378 | atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref)); | |
379 | return s; | |
380 | } else { | |
381 | dout("mdsc get_session %p 0 -- FAIL", s); | |
382 | return NULL; | |
383 | } | |
384 | } | |
385 | ||
386 | void ceph_put_mds_session(struct ceph_mds_session *s) | |
387 | { | |
388 | dout("mdsc put_session %p %d -> %d\n", s, | |
389 | atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1); | |
4e7a5dcd | 390 | if (atomic_dec_and_test(&s->s_ref)) { |
6c4a1915 | 391 | if (s->s_auth.authorizer) |
6c1ea260 | 392 | ceph_auth_destroy_authorizer(s->s_auth.authorizer); |
2f2dc053 | 393 | kfree(s); |
4e7a5dcd | 394 | } |
2f2dc053 SW |
395 | } |
396 | ||
397 | /* | |
398 | * called under mdsc->mutex | |
399 | */ | |
400 | struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc, | |
401 | int mds) | |
402 | { | |
403 | struct ceph_mds_session *session; | |
404 | ||
405 | if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL) | |
406 | return NULL; | |
407 | session = mdsc->sessions[mds]; | |
408 | dout("lookup_mds_session %p %d\n", session, | |
409 | atomic_read(&session->s_ref)); | |
410 | get_session(session); | |
411 | return session; | |
412 | } | |
413 | ||
414 | static bool __have_session(struct ceph_mds_client *mdsc, int mds) | |
415 | { | |
416 | if (mds >= mdsc->max_sessions) | |
417 | return false; | |
418 | return mdsc->sessions[mds]; | |
419 | } | |
420 | ||
2600d2dd SW |
421 | static int __verify_registered_session(struct ceph_mds_client *mdsc, |
422 | struct ceph_mds_session *s) | |
423 | { | |
424 | if (s->s_mds >= mdsc->max_sessions || | |
425 | mdsc->sessions[s->s_mds] != s) | |
426 | return -ENOENT; | |
427 | return 0; | |
428 | } | |
429 | ||
2f2dc053 SW |
430 | /* |
431 | * create+register a new session for given mds. | |
432 | * called under mdsc->mutex. | |
433 | */ | |
434 | static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc, | |
435 | int mds) | |
436 | { | |
437 | struct ceph_mds_session *s; | |
438 | ||
c338c07c NY |
439 | if (mds >= mdsc->mdsmap->m_max_mds) |
440 | return ERR_PTR(-EINVAL); | |
441 | ||
2f2dc053 | 442 | s = kzalloc(sizeof(*s), GFP_NOFS); |
4736b009 DC |
443 | if (!s) |
444 | return ERR_PTR(-ENOMEM); | |
2f2dc053 SW |
445 | s->s_mdsc = mdsc; |
446 | s->s_mds = mds; | |
447 | s->s_state = CEPH_MDS_SESSION_NEW; | |
448 | s->s_ttl = 0; | |
449 | s->s_seq = 0; | |
450 | mutex_init(&s->s_mutex); | |
451 | ||
b7a9e5dd | 452 | ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr); |
2f2dc053 | 453 | |
d8fb02ab | 454 | spin_lock_init(&s->s_gen_ttl_lock); |
2f2dc053 | 455 | s->s_cap_gen = 0; |
1ce208a6 | 456 | s->s_cap_ttl = jiffies - 1; |
d8fb02ab AE |
457 | |
458 | spin_lock_init(&s->s_cap_lock); | |
2f2dc053 SW |
459 | s->s_renew_requested = 0; |
460 | s->s_renew_seq = 0; | |
461 | INIT_LIST_HEAD(&s->s_caps); | |
462 | s->s_nr_caps = 0; | |
5dacf091 | 463 | s->s_trim_caps = 0; |
2f2dc053 SW |
464 | atomic_set(&s->s_ref, 1); |
465 | INIT_LIST_HEAD(&s->s_waiting); | |
466 | INIT_LIST_HEAD(&s->s_unsafe); | |
467 | s->s_num_cap_releases = 0; | |
99a9c273 | 468 | s->s_cap_reconnect = 0; |
7c1332b8 | 469 | s->s_cap_iterator = NULL; |
2f2dc053 | 470 | INIT_LIST_HEAD(&s->s_cap_releases); |
2f2dc053 SW |
471 | INIT_LIST_HEAD(&s->s_cap_flushing); |
472 | INIT_LIST_HEAD(&s->s_cap_snaps_flushing); | |
473 | ||
474 | dout("register_session mds%d\n", mds); | |
475 | if (mds >= mdsc->max_sessions) { | |
476 | int newmax = 1 << get_count_order(mds+1); | |
477 | struct ceph_mds_session **sa; | |
478 | ||
479 | dout("register_session realloc to %d\n", newmax); | |
480 | sa = kcalloc(newmax, sizeof(void *), GFP_NOFS); | |
481 | if (sa == NULL) | |
42ce56e5 | 482 | goto fail_realloc; |
2f2dc053 SW |
483 | if (mdsc->sessions) { |
484 | memcpy(sa, mdsc->sessions, | |
485 | mdsc->max_sessions * sizeof(void *)); | |
486 | kfree(mdsc->sessions); | |
487 | } | |
488 | mdsc->sessions = sa; | |
489 | mdsc->max_sessions = newmax; | |
490 | } | |
491 | mdsc->sessions[mds] = s; | |
86d8f67b | 492 | atomic_inc(&mdsc->num_sessions); |
2f2dc053 | 493 | atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */ |
42ce56e5 | 494 | |
b7a9e5dd SW |
495 | ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds, |
496 | ceph_mdsmap_get_addr(mdsc->mdsmap, mds)); | |
42ce56e5 | 497 | |
2f2dc053 | 498 | return s; |
42ce56e5 SW |
499 | |
500 | fail_realloc: | |
501 | kfree(s); | |
502 | return ERR_PTR(-ENOMEM); | |
2f2dc053 SW |
503 | } |
504 | ||
505 | /* | |
506 | * called under mdsc->mutex | |
507 | */ | |
2600d2dd | 508 | static void __unregister_session(struct ceph_mds_client *mdsc, |
42ce56e5 | 509 | struct ceph_mds_session *s) |
2f2dc053 | 510 | { |
2600d2dd SW |
511 | dout("__unregister_session mds%d %p\n", s->s_mds, s); |
512 | BUG_ON(mdsc->sessions[s->s_mds] != s); | |
42ce56e5 SW |
513 | mdsc->sessions[s->s_mds] = NULL; |
514 | ceph_con_close(&s->s_con); | |
515 | ceph_put_mds_session(s); | |
86d8f67b | 516 | atomic_dec(&mdsc->num_sessions); |
2f2dc053 SW |
517 | } |
518 | ||
519 | /* | |
520 | * drop session refs in request. | |
521 | * | |
522 | * should be last request ref, or hold mdsc->mutex | |
523 | */ | |
524 | static void put_request_session(struct ceph_mds_request *req) | |
525 | { | |
526 | if (req->r_session) { | |
527 | ceph_put_mds_session(req->r_session); | |
528 | req->r_session = NULL; | |
529 | } | |
530 | } | |
531 | ||
153c8e6b | 532 | void ceph_mdsc_release_request(struct kref *kref) |
2f2dc053 | 533 | { |
153c8e6b SW |
534 | struct ceph_mds_request *req = container_of(kref, |
535 | struct ceph_mds_request, | |
536 | r_kref); | |
54008399 | 537 | destroy_reply_info(&req->r_reply_info); |
153c8e6b SW |
538 | if (req->r_request) |
539 | ceph_msg_put(req->r_request); | |
54008399 | 540 | if (req->r_reply) |
153c8e6b | 541 | ceph_msg_put(req->r_reply); |
153c8e6b | 542 | if (req->r_inode) { |
41b02e1f | 543 | ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN); |
153c8e6b SW |
544 | iput(req->r_inode); |
545 | } | |
546 | if (req->r_locked_dir) | |
41b02e1f | 547 | ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN); |
e96a650a | 548 | iput(req->r_target_inode); |
153c8e6b SW |
549 | if (req->r_dentry) |
550 | dput(req->r_dentry); | |
844d87c3 SW |
551 | if (req->r_old_dentry) |
552 | dput(req->r_old_dentry); | |
553 | if (req->r_old_dentry_dir) { | |
41b02e1f SW |
554 | /* |
555 | * track (and drop pins for) r_old_dentry_dir | |
556 | * separately, since r_old_dentry's d_parent may have | |
557 | * changed between the dir mutex being dropped and | |
558 | * this request being freed. | |
559 | */ | |
560 | ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir), | |
561 | CEPH_CAP_PIN); | |
41b02e1f | 562 | iput(req->r_old_dentry_dir); |
2f2dc053 | 563 | } |
153c8e6b SW |
564 | kfree(req->r_path1); |
565 | kfree(req->r_path2); | |
25e6bae3 YZ |
566 | if (req->r_pagelist) |
567 | ceph_pagelist_release(req->r_pagelist); | |
153c8e6b | 568 | put_request_session(req); |
37151668 | 569 | ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation); |
153c8e6b | 570 | kfree(req); |
2f2dc053 SW |
571 | } |
572 | ||
fcd00b68 ID |
573 | DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node) |
574 | ||
2f2dc053 SW |
575 | /* |
576 | * lookup session, bump ref if found. | |
577 | * | |
578 | * called under mdsc->mutex. | |
579 | */ | |
fcd00b68 ID |
580 | static struct ceph_mds_request * |
581 | lookup_get_request(struct ceph_mds_client *mdsc, u64 tid) | |
2f2dc053 SW |
582 | { |
583 | struct ceph_mds_request *req; | |
44ca18f2 | 584 | |
fcd00b68 ID |
585 | req = lookup_request(&mdsc->request_tree, tid); |
586 | if (req) | |
587 | ceph_mdsc_get_request(req); | |
44ca18f2 | 588 | |
fcd00b68 | 589 | return req; |
2f2dc053 SW |
590 | } |
591 | ||
592 | /* | |
593 | * Register an in-flight request, and assign a tid. Link to directory | |
594 | * are modifying (if any). | |
595 | * | |
596 | * Called under mdsc->mutex. | |
597 | */ | |
598 | static void __register_request(struct ceph_mds_client *mdsc, | |
599 | struct ceph_mds_request *req, | |
600 | struct inode *dir) | |
601 | { | |
602 | req->r_tid = ++mdsc->last_tid; | |
603 | if (req->r_num_caps) | |
37151668 YS |
604 | ceph_reserve_caps(mdsc, &req->r_caps_reservation, |
605 | req->r_num_caps); | |
2f2dc053 SW |
606 | dout("__register_request %p tid %lld\n", req, req->r_tid); |
607 | ceph_mdsc_get_request(req); | |
fcd00b68 | 608 | insert_request(&mdsc->request_tree, req); |
2f2dc053 | 609 | |
cb4276cc SW |
610 | req->r_uid = current_fsuid(); |
611 | req->r_gid = current_fsgid(); | |
612 | ||
e8a7b8b1 YZ |
613 | if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK) |
614 | mdsc->oldest_tid = req->r_tid; | |
615 | ||
2f2dc053 | 616 | if (dir) { |
3b663780 | 617 | ihold(dir); |
2f2dc053 | 618 | req->r_unsafe_dir = dir; |
2f2dc053 SW |
619 | } |
620 | } | |
621 | ||
622 | static void __unregister_request(struct ceph_mds_client *mdsc, | |
623 | struct ceph_mds_request *req) | |
624 | { | |
625 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); | |
e8a7b8b1 YZ |
626 | |
627 | if (req->r_tid == mdsc->oldest_tid) { | |
628 | struct rb_node *p = rb_next(&req->r_node); | |
629 | mdsc->oldest_tid = 0; | |
630 | while (p) { | |
631 | struct ceph_mds_request *next_req = | |
632 | rb_entry(p, struct ceph_mds_request, r_node); | |
633 | if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) { | |
634 | mdsc->oldest_tid = next_req->r_tid; | |
635 | break; | |
636 | } | |
637 | p = rb_next(p); | |
638 | } | |
639 | } | |
640 | ||
fcd00b68 | 641 | erase_request(&mdsc->request_tree, req); |
2f2dc053 | 642 | |
4c06ace8 | 643 | if (req->r_unsafe_dir && req->r_got_unsafe) { |
2f2dc053 | 644 | struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir); |
2f2dc053 SW |
645 | spin_lock(&ci->i_unsafe_lock); |
646 | list_del_init(&req->r_unsafe_dir_item); | |
647 | spin_unlock(&ci->i_unsafe_lock); | |
4c06ace8 | 648 | } |
68cd5b4b YZ |
649 | if (req->r_target_inode && req->r_got_unsafe) { |
650 | struct ceph_inode_info *ci = ceph_inode(req->r_target_inode); | |
651 | spin_lock(&ci->i_unsafe_lock); | |
652 | list_del_init(&req->r_unsafe_target_item); | |
653 | spin_unlock(&ci->i_unsafe_lock); | |
654 | } | |
3b663780 | 655 | |
4c06ace8 | 656 | if (req->r_unsafe_dir) { |
3b663780 SW |
657 | iput(req->r_unsafe_dir); |
658 | req->r_unsafe_dir = NULL; | |
2f2dc053 | 659 | } |
94aa8ae1 | 660 | |
fc55d2c9 YZ |
661 | complete_all(&req->r_safe_completion); |
662 | ||
94aa8ae1 | 663 | ceph_mdsc_put_request(req); |
2f2dc053 SW |
664 | } |
665 | ||
666 | /* | |
667 | * Choose mds to send request to next. If there is a hint set in the | |
668 | * request (e.g., due to a prior forward hint from the mds), use that. | |
669 | * Otherwise, consult frag tree and/or caps to identify the | |
670 | * appropriate mds. If all else fails, choose randomly. | |
671 | * | |
672 | * Called under mdsc->mutex. | |
673 | */ | |
7fd7d101 | 674 | static struct dentry *get_nonsnap_parent(struct dentry *dentry) |
eb6bb1c5 | 675 | { |
d79698da SW |
676 | /* |
677 | * we don't need to worry about protecting the d_parent access | |
678 | * here because we never renaming inside the snapped namespace | |
679 | * except to resplice to another snapdir, and either the old or new | |
680 | * result is a valid result. | |
681 | */ | |
2b0143b5 | 682 | while (!IS_ROOT(dentry) && ceph_snap(d_inode(dentry)) != CEPH_NOSNAP) |
eb6bb1c5 SW |
683 | dentry = dentry->d_parent; |
684 | return dentry; | |
685 | } | |
686 | ||
2f2dc053 SW |
687 | static int __choose_mds(struct ceph_mds_client *mdsc, |
688 | struct ceph_mds_request *req) | |
689 | { | |
690 | struct inode *inode; | |
691 | struct ceph_inode_info *ci; | |
692 | struct ceph_cap *cap; | |
693 | int mode = req->r_direct_mode; | |
694 | int mds = -1; | |
695 | u32 hash = req->r_direct_hash; | |
696 | bool is_hash = req->r_direct_is_hash; | |
697 | ||
698 | /* | |
699 | * is there a specific mds we should try? ignore hint if we have | |
700 | * no session and the mds is not up (active or recovering). | |
701 | */ | |
702 | if (req->r_resend_mds >= 0 && | |
703 | (__have_session(mdsc, req->r_resend_mds) || | |
704 | ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) { | |
705 | dout("choose_mds using resend_mds mds%d\n", | |
706 | req->r_resend_mds); | |
707 | return req->r_resend_mds; | |
708 | } | |
709 | ||
710 | if (mode == USE_RANDOM_MDS) | |
711 | goto random; | |
712 | ||
713 | inode = NULL; | |
714 | if (req->r_inode) { | |
715 | inode = req->r_inode; | |
716 | } else if (req->r_dentry) { | |
d79698da SW |
717 | /* ignore race with rename; old or new d_parent is okay */ |
718 | struct dentry *parent = req->r_dentry->d_parent; | |
2b0143b5 | 719 | struct inode *dir = d_inode(parent); |
eb6bb1c5 | 720 | |
3d14c5d2 | 721 | if (dir->i_sb != mdsc->fsc->sb) { |
eb6bb1c5 | 722 | /* not this fs! */ |
2b0143b5 | 723 | inode = d_inode(req->r_dentry); |
eb6bb1c5 SW |
724 | } else if (ceph_snap(dir) != CEPH_NOSNAP) { |
725 | /* direct snapped/virtual snapdir requests | |
726 | * based on parent dir inode */ | |
d79698da | 727 | struct dentry *dn = get_nonsnap_parent(parent); |
2b0143b5 | 728 | inode = d_inode(dn); |
eb6bb1c5 | 729 | dout("__choose_mds using nonsnap parent %p\n", inode); |
ca18bede | 730 | } else { |
eb6bb1c5 | 731 | /* dentry target */ |
2b0143b5 | 732 | inode = d_inode(req->r_dentry); |
ca18bede YZ |
733 | if (!inode || mode == USE_AUTH_MDS) { |
734 | /* dir + name */ | |
735 | inode = dir; | |
736 | hash = ceph_dentry_hash(dir, req->r_dentry); | |
737 | is_hash = true; | |
738 | } | |
2f2dc053 SW |
739 | } |
740 | } | |
eb6bb1c5 | 741 | |
2f2dc053 SW |
742 | dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash, |
743 | (int)hash, mode); | |
744 | if (!inode) | |
745 | goto random; | |
746 | ci = ceph_inode(inode); | |
747 | ||
748 | if (is_hash && S_ISDIR(inode->i_mode)) { | |
749 | struct ceph_inode_frag frag; | |
750 | int found; | |
751 | ||
752 | ceph_choose_frag(ci, hash, &frag, &found); | |
753 | if (found) { | |
754 | if (mode == USE_ANY_MDS && frag.ndist > 0) { | |
755 | u8 r; | |
756 | ||
757 | /* choose a random replica */ | |
758 | get_random_bytes(&r, 1); | |
759 | r %= frag.ndist; | |
760 | mds = frag.dist[r]; | |
761 | dout("choose_mds %p %llx.%llx " | |
762 | "frag %u mds%d (%d/%d)\n", | |
763 | inode, ceph_vinop(inode), | |
d66bbd44 | 764 | frag.frag, mds, |
2f2dc053 | 765 | (int)r, frag.ndist); |
d66bbd44 SW |
766 | if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >= |
767 | CEPH_MDS_STATE_ACTIVE) | |
768 | return mds; | |
2f2dc053 SW |
769 | } |
770 | ||
771 | /* since this file/dir wasn't known to be | |
772 | * replicated, then we want to look for the | |
773 | * authoritative mds. */ | |
774 | mode = USE_AUTH_MDS; | |
775 | if (frag.mds >= 0) { | |
776 | /* choose auth mds */ | |
777 | mds = frag.mds; | |
778 | dout("choose_mds %p %llx.%llx " | |
779 | "frag %u mds%d (auth)\n", | |
780 | inode, ceph_vinop(inode), frag.frag, mds); | |
d66bbd44 SW |
781 | if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >= |
782 | CEPH_MDS_STATE_ACTIVE) | |
783 | return mds; | |
2f2dc053 SW |
784 | } |
785 | } | |
786 | } | |
787 | ||
be655596 | 788 | spin_lock(&ci->i_ceph_lock); |
2f2dc053 SW |
789 | cap = NULL; |
790 | if (mode == USE_AUTH_MDS) | |
791 | cap = ci->i_auth_cap; | |
792 | if (!cap && !RB_EMPTY_ROOT(&ci->i_caps)) | |
793 | cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node); | |
794 | if (!cap) { | |
be655596 | 795 | spin_unlock(&ci->i_ceph_lock); |
2f2dc053 SW |
796 | goto random; |
797 | } | |
798 | mds = cap->session->s_mds; | |
799 | dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n", | |
800 | inode, ceph_vinop(inode), mds, | |
801 | cap == ci->i_auth_cap ? "auth " : "", cap); | |
be655596 | 802 | spin_unlock(&ci->i_ceph_lock); |
2f2dc053 SW |
803 | return mds; |
804 | ||
805 | random: | |
806 | mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap); | |
807 | dout("choose_mds chose random mds%d\n", mds); | |
808 | return mds; | |
809 | } | |
810 | ||
811 | ||
812 | /* | |
813 | * session messages | |
814 | */ | |
815 | static struct ceph_msg *create_session_msg(u32 op, u64 seq) | |
816 | { | |
817 | struct ceph_msg *msg; | |
818 | struct ceph_mds_session_head *h; | |
819 | ||
b61c2763 SW |
820 | msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS, |
821 | false); | |
a79832f2 | 822 | if (!msg) { |
2f2dc053 | 823 | pr_err("create_session_msg ENOMEM creating msg\n"); |
a79832f2 | 824 | return NULL; |
2f2dc053 SW |
825 | } |
826 | h = msg->front.iov_base; | |
827 | h->op = cpu_to_le32(op); | |
828 | h->seq = cpu_to_le64(seq); | |
dbd0c8bf JS |
829 | |
830 | return msg; | |
831 | } | |
832 | ||
833 | /* | |
834 | * session message, specialization for CEPH_SESSION_REQUEST_OPEN | |
835 | * to include additional client metadata fields. | |
836 | */ | |
837 | static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq) | |
838 | { | |
839 | struct ceph_msg *msg; | |
840 | struct ceph_mds_session_head *h; | |
841 | int i = -1; | |
842 | int metadata_bytes = 0; | |
843 | int metadata_key_count = 0; | |
844 | struct ceph_options *opt = mdsc->fsc->client->options; | |
3f384954 | 845 | struct ceph_mount_options *fsopt = mdsc->fsc->mount_options; |
dbd0c8bf JS |
846 | void *p; |
847 | ||
a6a5ce4f | 848 | const char* metadata[][2] = { |
dbd0c8bf | 849 | {"hostname", utsname()->nodename}, |
a6a5ce4f | 850 | {"kernel_version", utsname()->release}, |
3f384954 YZ |
851 | {"entity_id", opt->name ? : ""}, |
852 | {"root", fsopt->server_path ? : "/"}, | |
dbd0c8bf JS |
853 | {NULL, NULL} |
854 | }; | |
855 | ||
856 | /* Calculate serialized length of metadata */ | |
857 | metadata_bytes = 4; /* map length */ | |
858 | for (i = 0; metadata[i][0] != NULL; ++i) { | |
859 | metadata_bytes += 8 + strlen(metadata[i][0]) + | |
860 | strlen(metadata[i][1]); | |
861 | metadata_key_count++; | |
862 | } | |
863 | ||
864 | /* Allocate the message */ | |
865 | msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes, | |
866 | GFP_NOFS, false); | |
867 | if (!msg) { | |
868 | pr_err("create_session_msg ENOMEM creating msg\n"); | |
869 | return NULL; | |
870 | } | |
871 | h = msg->front.iov_base; | |
872 | h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN); | |
873 | h->seq = cpu_to_le64(seq); | |
874 | ||
875 | /* | |
876 | * Serialize client metadata into waiting buffer space, using | |
877 | * the format that userspace expects for map<string, string> | |
7cfa0313 JS |
878 | * |
879 | * ClientSession messages with metadata are v2 | |
dbd0c8bf | 880 | */ |
7cfa0313 JS |
881 | msg->hdr.version = cpu_to_le16(2); |
882 | msg->hdr.compat_version = cpu_to_le16(1); | |
dbd0c8bf JS |
883 | |
884 | /* The write pointer, following the session_head structure */ | |
885 | p = msg->front.iov_base + sizeof(*h); | |
886 | ||
887 | /* Number of entries in the map */ | |
888 | ceph_encode_32(&p, metadata_key_count); | |
889 | ||
890 | /* Two length-prefixed strings for each entry in the map */ | |
891 | for (i = 0; metadata[i][0] != NULL; ++i) { | |
892 | size_t const key_len = strlen(metadata[i][0]); | |
893 | size_t const val_len = strlen(metadata[i][1]); | |
894 | ||
895 | ceph_encode_32(&p, key_len); | |
896 | memcpy(p, metadata[i][0], key_len); | |
897 | p += key_len; | |
898 | ceph_encode_32(&p, val_len); | |
899 | memcpy(p, metadata[i][1], val_len); | |
900 | p += val_len; | |
901 | } | |
902 | ||
2f2dc053 SW |
903 | return msg; |
904 | } | |
905 | ||
906 | /* | |
907 | * send session open request. | |
908 | * | |
909 | * called under mdsc->mutex | |
910 | */ | |
911 | static int __open_session(struct ceph_mds_client *mdsc, | |
912 | struct ceph_mds_session *session) | |
913 | { | |
914 | struct ceph_msg *msg; | |
915 | int mstate; | |
916 | int mds = session->s_mds; | |
2f2dc053 SW |
917 | |
918 | /* wait for mds to go active? */ | |
919 | mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds); | |
920 | dout("open_session to mds%d (%s)\n", mds, | |
921 | ceph_mds_state_name(mstate)); | |
922 | session->s_state = CEPH_MDS_SESSION_OPENING; | |
923 | session->s_renew_requested = jiffies; | |
924 | ||
925 | /* send connect message */ | |
dbd0c8bf | 926 | msg = create_session_open_msg(mdsc, session->s_seq); |
a79832f2 SW |
927 | if (!msg) |
928 | return -ENOMEM; | |
2f2dc053 | 929 | ceph_con_send(&session->s_con, msg); |
2f2dc053 SW |
930 | return 0; |
931 | } | |
932 | ||
ed0552a1 SW |
933 | /* |
934 | * open sessions for any export targets for the given mds | |
935 | * | |
936 | * called under mdsc->mutex | |
937 | */ | |
5d72d13c YZ |
938 | static struct ceph_mds_session * |
939 | __open_export_target_session(struct ceph_mds_client *mdsc, int target) | |
940 | { | |
941 | struct ceph_mds_session *session; | |
942 | ||
943 | session = __ceph_lookup_mds_session(mdsc, target); | |
944 | if (!session) { | |
945 | session = register_session(mdsc, target); | |
946 | if (IS_ERR(session)) | |
947 | return session; | |
948 | } | |
949 | if (session->s_state == CEPH_MDS_SESSION_NEW || | |
950 | session->s_state == CEPH_MDS_SESSION_CLOSING) | |
951 | __open_session(mdsc, session); | |
952 | ||
953 | return session; | |
954 | } | |
955 | ||
956 | struct ceph_mds_session * | |
957 | ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target) | |
958 | { | |
959 | struct ceph_mds_session *session; | |
960 | ||
961 | dout("open_export_target_session to mds%d\n", target); | |
962 | ||
963 | mutex_lock(&mdsc->mutex); | |
964 | session = __open_export_target_session(mdsc, target); | |
965 | mutex_unlock(&mdsc->mutex); | |
966 | ||
967 | return session; | |
968 | } | |
969 | ||
ed0552a1 SW |
970 | static void __open_export_target_sessions(struct ceph_mds_client *mdsc, |
971 | struct ceph_mds_session *session) | |
972 | { | |
973 | struct ceph_mds_info *mi; | |
974 | struct ceph_mds_session *ts; | |
975 | int i, mds = session->s_mds; | |
ed0552a1 SW |
976 | |
977 | if (mds >= mdsc->mdsmap->m_max_mds) | |
978 | return; | |
5d72d13c | 979 | |
ed0552a1 SW |
980 | mi = &mdsc->mdsmap->m_info[mds]; |
981 | dout("open_export_target_sessions for mds%d (%d targets)\n", | |
982 | session->s_mds, mi->num_export_targets); | |
983 | ||
984 | for (i = 0; i < mi->num_export_targets; i++) { | |
5d72d13c YZ |
985 | ts = __open_export_target_session(mdsc, mi->export_targets[i]); |
986 | if (!IS_ERR(ts)) | |
987 | ceph_put_mds_session(ts); | |
ed0552a1 SW |
988 | } |
989 | } | |
990 | ||
154f42c2 SW |
991 | void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc, |
992 | struct ceph_mds_session *session) | |
993 | { | |
994 | mutex_lock(&mdsc->mutex); | |
995 | __open_export_target_sessions(mdsc, session); | |
996 | mutex_unlock(&mdsc->mutex); | |
997 | } | |
998 | ||
2f2dc053 SW |
999 | /* |
1000 | * session caps | |
1001 | */ | |
1002 | ||
745a8e3b YZ |
1003 | /* caller holds s_cap_lock, we drop it */ |
1004 | static void cleanup_cap_releases(struct ceph_mds_client *mdsc, | |
1005 | struct ceph_mds_session *session) | |
1006 | __releases(session->s_cap_lock) | |
2f2dc053 | 1007 | { |
745a8e3b YZ |
1008 | LIST_HEAD(tmp_list); |
1009 | list_splice_init(&session->s_cap_releases, &tmp_list); | |
1010 | session->s_num_cap_releases = 0; | |
1011 | spin_unlock(&session->s_cap_lock); | |
2f2dc053 | 1012 | |
745a8e3b YZ |
1013 | dout("cleanup_cap_releases mds%d\n", session->s_mds); |
1014 | while (!list_empty(&tmp_list)) { | |
1015 | struct ceph_cap *cap; | |
1016 | /* zero out the in-progress message */ | |
1017 | cap = list_first_entry(&tmp_list, | |
1018 | struct ceph_cap, session_caps); | |
1019 | list_del(&cap->session_caps); | |
1020 | ceph_put_cap(mdsc, cap); | |
2f2dc053 | 1021 | } |
2f2dc053 SW |
1022 | } |
1023 | ||
1c841a96 YZ |
1024 | static void cleanup_session_requests(struct ceph_mds_client *mdsc, |
1025 | struct ceph_mds_session *session) | |
1026 | { | |
1027 | struct ceph_mds_request *req; | |
1028 | struct rb_node *p; | |
1029 | ||
1030 | dout("cleanup_session_requests mds%d\n", session->s_mds); | |
1031 | mutex_lock(&mdsc->mutex); | |
1032 | while (!list_empty(&session->s_unsafe)) { | |
1033 | req = list_first_entry(&session->s_unsafe, | |
1034 | struct ceph_mds_request, r_unsafe_item); | |
1035 | list_del_init(&req->r_unsafe_item); | |
3e0708b9 YZ |
1036 | pr_warn_ratelimited(" dropping unsafe request %llu\n", |
1037 | req->r_tid); | |
1c841a96 YZ |
1038 | __unregister_request(mdsc, req); |
1039 | } | |
1040 | /* zero r_attempts, so kick_requests() will re-send requests */ | |
1041 | p = rb_first(&mdsc->request_tree); | |
1042 | while (p) { | |
1043 | req = rb_entry(p, struct ceph_mds_request, r_node); | |
1044 | p = rb_next(p); | |
1045 | if (req->r_session && | |
1046 | req->r_session->s_mds == session->s_mds) | |
1047 | req->r_attempts = 0; | |
1048 | } | |
1049 | mutex_unlock(&mdsc->mutex); | |
1050 | } | |
1051 | ||
2f2dc053 | 1052 | /* |
f818a736 SW |
1053 | * Helper to safely iterate over all caps associated with a session, with |
1054 | * special care taken to handle a racing __ceph_remove_cap(). | |
2f2dc053 | 1055 | * |
f818a736 | 1056 | * Caller must hold session s_mutex. |
2f2dc053 SW |
1057 | */ |
1058 | static int iterate_session_caps(struct ceph_mds_session *session, | |
1059 | int (*cb)(struct inode *, struct ceph_cap *, | |
1060 | void *), void *arg) | |
1061 | { | |
7c1332b8 SW |
1062 | struct list_head *p; |
1063 | struct ceph_cap *cap; | |
1064 | struct inode *inode, *last_inode = NULL; | |
1065 | struct ceph_cap *old_cap = NULL; | |
2f2dc053 SW |
1066 | int ret; |
1067 | ||
1068 | dout("iterate_session_caps %p mds%d\n", session, session->s_mds); | |
1069 | spin_lock(&session->s_cap_lock); | |
7c1332b8 SW |
1070 | p = session->s_caps.next; |
1071 | while (p != &session->s_caps) { | |
1072 | cap = list_entry(p, struct ceph_cap, session_caps); | |
2f2dc053 | 1073 | inode = igrab(&cap->ci->vfs_inode); |
7c1332b8 SW |
1074 | if (!inode) { |
1075 | p = p->next; | |
2f2dc053 | 1076 | continue; |
7c1332b8 SW |
1077 | } |
1078 | session->s_cap_iterator = cap; | |
2f2dc053 | 1079 | spin_unlock(&session->s_cap_lock); |
7c1332b8 SW |
1080 | |
1081 | if (last_inode) { | |
1082 | iput(last_inode); | |
1083 | last_inode = NULL; | |
1084 | } | |
1085 | if (old_cap) { | |
37151668 | 1086 | ceph_put_cap(session->s_mdsc, old_cap); |
7c1332b8 SW |
1087 | old_cap = NULL; |
1088 | } | |
1089 | ||
2f2dc053 | 1090 | ret = cb(inode, cap, arg); |
7c1332b8 SW |
1091 | last_inode = inode; |
1092 | ||
2f2dc053 | 1093 | spin_lock(&session->s_cap_lock); |
7c1332b8 SW |
1094 | p = p->next; |
1095 | if (cap->ci == NULL) { | |
1096 | dout("iterate_session_caps finishing cap %p removal\n", | |
1097 | cap); | |
1098 | BUG_ON(cap->session != session); | |
745a8e3b | 1099 | cap->session = NULL; |
7c1332b8 SW |
1100 | list_del_init(&cap->session_caps); |
1101 | session->s_nr_caps--; | |
745a8e3b YZ |
1102 | if (cap->queue_release) { |
1103 | list_add_tail(&cap->session_caps, | |
1104 | &session->s_cap_releases); | |
1105 | session->s_num_cap_releases++; | |
1106 | } else { | |
1107 | old_cap = cap; /* put_cap it w/o locks held */ | |
1108 | } | |
7c1332b8 | 1109 | } |
5dacf091 SW |
1110 | if (ret < 0) |
1111 | goto out; | |
2f2dc053 | 1112 | } |
5dacf091 SW |
1113 | ret = 0; |
1114 | out: | |
7c1332b8 | 1115 | session->s_cap_iterator = NULL; |
2f2dc053 | 1116 | spin_unlock(&session->s_cap_lock); |
7c1332b8 | 1117 | |
e96a650a | 1118 | iput(last_inode); |
7c1332b8 | 1119 | if (old_cap) |
37151668 | 1120 | ceph_put_cap(session->s_mdsc, old_cap); |
7c1332b8 | 1121 | |
5dacf091 | 1122 | return ret; |
2f2dc053 SW |
1123 | } |
1124 | ||
1125 | static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap, | |
6c99f254 | 1126 | void *arg) |
2f2dc053 | 1127 | { |
6c93df5d | 1128 | struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg; |
2f2dc053 | 1129 | struct ceph_inode_info *ci = ceph_inode(inode); |
553adfd9 | 1130 | LIST_HEAD(to_remove); |
6c93df5d YZ |
1131 | bool drop = false; |
1132 | bool invalidate = false; | |
6c99f254 | 1133 | |
2f2dc053 SW |
1134 | dout("removing cap %p, ci is %p, inode is %p\n", |
1135 | cap, ci, &ci->vfs_inode); | |
be655596 | 1136 | spin_lock(&ci->i_ceph_lock); |
a096b09a | 1137 | __ceph_remove_cap(cap, false); |
571ade33 | 1138 | if (!ci->i_auth_cap) { |
553adfd9 | 1139 | struct ceph_cap_flush *cf; |
6c93df5d | 1140 | struct ceph_mds_client *mdsc = fsc->mdsc; |
6c99f254 | 1141 | |
77310320 YZ |
1142 | ci->i_ceph_flags |= CEPH_I_CAP_DROPPED; |
1143 | ||
6c93df5d YZ |
1144 | if (ci->i_wrbuffer_ref > 0 && |
1145 | ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) | |
1146 | invalidate = true; | |
1147 | ||
553adfd9 YZ |
1148 | while (true) { |
1149 | struct rb_node *n = rb_first(&ci->i_cap_flush_tree); | |
1150 | if (!n) | |
1151 | break; | |
1152 | cf = rb_entry(n, struct ceph_cap_flush, i_node); | |
1153 | rb_erase(&cf->i_node, &ci->i_cap_flush_tree); | |
1154 | list_add(&cf->list, &to_remove); | |
1155 | } | |
1156 | ||
6c99f254 | 1157 | spin_lock(&mdsc->cap_dirty_lock); |
8310b089 YZ |
1158 | |
1159 | list_for_each_entry(cf, &to_remove, list) | |
1160 | rb_erase(&cf->g_node, &mdsc->cap_flush_tree); | |
1161 | ||
6c99f254 | 1162 | if (!list_empty(&ci->i_dirty_item)) { |
3e0708b9 YZ |
1163 | pr_warn_ratelimited( |
1164 | " dropping dirty %s state for %p %lld\n", | |
6c99f254 SW |
1165 | ceph_cap_string(ci->i_dirty_caps), |
1166 | inode, ceph_ino(inode)); | |
1167 | ci->i_dirty_caps = 0; | |
1168 | list_del_init(&ci->i_dirty_item); | |
6c93df5d | 1169 | drop = true; |
6c99f254 SW |
1170 | } |
1171 | if (!list_empty(&ci->i_flushing_item)) { | |
3e0708b9 YZ |
1172 | pr_warn_ratelimited( |
1173 | " dropping dirty+flushing %s state for %p %lld\n", | |
6c99f254 SW |
1174 | ceph_cap_string(ci->i_flushing_caps), |
1175 | inode, ceph_ino(inode)); | |
1176 | ci->i_flushing_caps = 0; | |
1177 | list_del_init(&ci->i_flushing_item); | |
1178 | mdsc->num_cap_flushing--; | |
6c93df5d | 1179 | drop = true; |
6c99f254 | 1180 | } |
6c99f254 | 1181 | spin_unlock(&mdsc->cap_dirty_lock); |
553adfd9 | 1182 | |
f66fd9f0 YZ |
1183 | if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) { |
1184 | list_add(&ci->i_prealloc_cap_flush->list, &to_remove); | |
1185 | ci->i_prealloc_cap_flush = NULL; | |
1186 | } | |
6c99f254 | 1187 | } |
be655596 | 1188 | spin_unlock(&ci->i_ceph_lock); |
553adfd9 YZ |
1189 | while (!list_empty(&to_remove)) { |
1190 | struct ceph_cap_flush *cf; | |
1191 | cf = list_first_entry(&to_remove, | |
1192 | struct ceph_cap_flush, list); | |
1193 | list_del(&cf->list); | |
f66fd9f0 | 1194 | ceph_free_cap_flush(cf); |
553adfd9 | 1195 | } |
77310320 YZ |
1196 | |
1197 | wake_up_all(&ci->i_cap_wq); | |
6c93df5d YZ |
1198 | if (invalidate) |
1199 | ceph_queue_invalidate(inode); | |
77310320 | 1200 | if (drop) |
6c99f254 | 1201 | iput(inode); |
2f2dc053 SW |
1202 | return 0; |
1203 | } | |
1204 | ||
1205 | /* | |
1206 | * caller must hold session s_mutex | |
1207 | */ | |
1208 | static void remove_session_caps(struct ceph_mds_session *session) | |
1209 | { | |
6c93df5d YZ |
1210 | struct ceph_fs_client *fsc = session->s_mdsc->fsc; |
1211 | struct super_block *sb = fsc->sb; | |
2f2dc053 | 1212 | dout("remove_session_caps on %p\n", session); |
6c93df5d | 1213 | iterate_session_caps(session, remove_session_caps_cb, fsc); |
6f60f889 YZ |
1214 | |
1215 | spin_lock(&session->s_cap_lock); | |
1216 | if (session->s_nr_caps > 0) { | |
6f60f889 YZ |
1217 | struct inode *inode; |
1218 | struct ceph_cap *cap, *prev = NULL; | |
1219 | struct ceph_vino vino; | |
1220 | /* | |
1221 | * iterate_session_caps() skips inodes that are being | |
1222 | * deleted, we need to wait until deletions are complete. | |
1223 | * __wait_on_freeing_inode() is designed for the job, | |
1224 | * but it is not exported, so use lookup inode function | |
1225 | * to access it. | |
1226 | */ | |
1227 | while (!list_empty(&session->s_caps)) { | |
1228 | cap = list_entry(session->s_caps.next, | |
1229 | struct ceph_cap, session_caps); | |
1230 | if (cap == prev) | |
1231 | break; | |
1232 | prev = cap; | |
1233 | vino = cap->ci->i_vino; | |
1234 | spin_unlock(&session->s_cap_lock); | |
1235 | ||
ed284c49 | 1236 | inode = ceph_find_inode(sb, vino); |
6f60f889 YZ |
1237 | iput(inode); |
1238 | ||
1239 | spin_lock(&session->s_cap_lock); | |
1240 | } | |
1241 | } | |
745a8e3b YZ |
1242 | |
1243 | // drop cap expires and unlock s_cap_lock | |
1244 | cleanup_cap_releases(session->s_mdsc, session); | |
6f60f889 | 1245 | |
2f2dc053 | 1246 | BUG_ON(session->s_nr_caps > 0); |
6c99f254 | 1247 | BUG_ON(!list_empty(&session->s_cap_flushing)); |
2f2dc053 SW |
1248 | } |
1249 | ||
1250 | /* | |
1251 | * wake up any threads waiting on this session's caps. if the cap is | |
1252 | * old (didn't get renewed on the client reconnect), remove it now. | |
1253 | * | |
1254 | * caller must hold s_mutex. | |
1255 | */ | |
1256 | static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap, | |
1257 | void *arg) | |
1258 | { | |
0dc2570f SW |
1259 | struct ceph_inode_info *ci = ceph_inode(inode); |
1260 | ||
0dc2570f | 1261 | if (arg) { |
be655596 | 1262 | spin_lock(&ci->i_ceph_lock); |
0dc2570f SW |
1263 | ci->i_wanted_max_size = 0; |
1264 | ci->i_requested_max_size = 0; | |
be655596 | 1265 | spin_unlock(&ci->i_ceph_lock); |
0dc2570f | 1266 | } |
e5360309 | 1267 | wake_up_all(&ci->i_cap_wq); |
2f2dc053 SW |
1268 | return 0; |
1269 | } | |
1270 | ||
0dc2570f SW |
1271 | static void wake_up_session_caps(struct ceph_mds_session *session, |
1272 | int reconnect) | |
2f2dc053 SW |
1273 | { |
1274 | dout("wake_up_session_caps %p mds%d\n", session, session->s_mds); | |
0dc2570f SW |
1275 | iterate_session_caps(session, wake_up_session_cb, |
1276 | (void *)(unsigned long)reconnect); | |
2f2dc053 SW |
1277 | } |
1278 | ||
1279 | /* | |
1280 | * Send periodic message to MDS renewing all currently held caps. The | |
1281 | * ack will reset the expiration for all caps from this session. | |
1282 | * | |
1283 | * caller holds s_mutex | |
1284 | */ | |
1285 | static int send_renew_caps(struct ceph_mds_client *mdsc, | |
1286 | struct ceph_mds_session *session) | |
1287 | { | |
1288 | struct ceph_msg *msg; | |
1289 | int state; | |
1290 | ||
1291 | if (time_after_eq(jiffies, session->s_cap_ttl) && | |
1292 | time_after_eq(session->s_cap_ttl, session->s_renew_requested)) | |
1293 | pr_info("mds%d caps stale\n", session->s_mds); | |
e4cb4cb8 | 1294 | session->s_renew_requested = jiffies; |
2f2dc053 SW |
1295 | |
1296 | /* do not try to renew caps until a recovering mds has reconnected | |
1297 | * with its clients. */ | |
1298 | state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds); | |
1299 | if (state < CEPH_MDS_STATE_RECONNECT) { | |
1300 | dout("send_renew_caps ignoring mds%d (%s)\n", | |
1301 | session->s_mds, ceph_mds_state_name(state)); | |
1302 | return 0; | |
1303 | } | |
1304 | ||
1305 | dout("send_renew_caps to mds%d (%s)\n", session->s_mds, | |
1306 | ceph_mds_state_name(state)); | |
2f2dc053 SW |
1307 | msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS, |
1308 | ++session->s_renew_seq); | |
a79832f2 SW |
1309 | if (!msg) |
1310 | return -ENOMEM; | |
2f2dc053 SW |
1311 | ceph_con_send(&session->s_con, msg); |
1312 | return 0; | |
1313 | } | |
1314 | ||
186e4f7a YZ |
1315 | static int send_flushmsg_ack(struct ceph_mds_client *mdsc, |
1316 | struct ceph_mds_session *session, u64 seq) | |
1317 | { | |
1318 | struct ceph_msg *msg; | |
1319 | ||
1320 | dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n", | |
a687ecaf | 1321 | session->s_mds, ceph_session_state_name(session->s_state), seq); |
186e4f7a YZ |
1322 | msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq); |
1323 | if (!msg) | |
1324 | return -ENOMEM; | |
1325 | ceph_con_send(&session->s_con, msg); | |
1326 | return 0; | |
1327 | } | |
1328 | ||
1329 | ||
2f2dc053 SW |
1330 | /* |
1331 | * Note new cap ttl, and any transition from stale -> not stale (fresh?). | |
0dc2570f SW |
1332 | * |
1333 | * Called under session->s_mutex | |
2f2dc053 SW |
1334 | */ |
1335 | static void renewed_caps(struct ceph_mds_client *mdsc, | |
1336 | struct ceph_mds_session *session, int is_renew) | |
1337 | { | |
1338 | int was_stale; | |
1339 | int wake = 0; | |
1340 | ||
1341 | spin_lock(&session->s_cap_lock); | |
1ce208a6 | 1342 | was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl); |
2f2dc053 SW |
1343 | |
1344 | session->s_cap_ttl = session->s_renew_requested + | |
1345 | mdsc->mdsmap->m_session_timeout*HZ; | |
1346 | ||
1347 | if (was_stale) { | |
1348 | if (time_before(jiffies, session->s_cap_ttl)) { | |
1349 | pr_info("mds%d caps renewed\n", session->s_mds); | |
1350 | wake = 1; | |
1351 | } else { | |
1352 | pr_info("mds%d caps still stale\n", session->s_mds); | |
1353 | } | |
1354 | } | |
1355 | dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n", | |
1356 | session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh", | |
1357 | time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh"); | |
1358 | spin_unlock(&session->s_cap_lock); | |
1359 | ||
1360 | if (wake) | |
0dc2570f | 1361 | wake_up_session_caps(session, 0); |
2f2dc053 SW |
1362 | } |
1363 | ||
1364 | /* | |
1365 | * send a session close request | |
1366 | */ | |
1367 | static int request_close_session(struct ceph_mds_client *mdsc, | |
1368 | struct ceph_mds_session *session) | |
1369 | { | |
1370 | struct ceph_msg *msg; | |
2f2dc053 SW |
1371 | |
1372 | dout("request_close_session mds%d state %s seq %lld\n", | |
a687ecaf | 1373 | session->s_mds, ceph_session_state_name(session->s_state), |
2f2dc053 SW |
1374 | session->s_seq); |
1375 | msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq); | |
a79832f2 SW |
1376 | if (!msg) |
1377 | return -ENOMEM; | |
1378 | ceph_con_send(&session->s_con, msg); | |
1379 | return 0; | |
2f2dc053 SW |
1380 | } |
1381 | ||
1382 | /* | |
1383 | * Called with s_mutex held. | |
1384 | */ | |
1385 | static int __close_session(struct ceph_mds_client *mdsc, | |
1386 | struct ceph_mds_session *session) | |
1387 | { | |
1388 | if (session->s_state >= CEPH_MDS_SESSION_CLOSING) | |
1389 | return 0; | |
1390 | session->s_state = CEPH_MDS_SESSION_CLOSING; | |
1391 | return request_close_session(mdsc, session); | |
1392 | } | |
1393 | ||
1394 | /* | |
1395 | * Trim old(er) caps. | |
1396 | * | |
1397 | * Because we can't cache an inode without one or more caps, we do | |
1398 | * this indirectly: if a cap is unused, we prune its aliases, at which | |
1399 | * point the inode will hopefully get dropped to. | |
1400 | * | |
1401 | * Yes, this is a bit sloppy. Our only real goal here is to respond to | |
1402 | * memory pressure from the MDS, though, so it needn't be perfect. | |
1403 | */ | |
1404 | static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg) | |
1405 | { | |
1406 | struct ceph_mds_session *session = arg; | |
1407 | struct ceph_inode_info *ci = ceph_inode(inode); | |
979abfdd | 1408 | int used, wanted, oissued, mine; |
2f2dc053 SW |
1409 | |
1410 | if (session->s_trim_caps <= 0) | |
1411 | return -1; | |
1412 | ||
be655596 | 1413 | spin_lock(&ci->i_ceph_lock); |
2f2dc053 SW |
1414 | mine = cap->issued | cap->implemented; |
1415 | used = __ceph_caps_used(ci); | |
979abfdd | 1416 | wanted = __ceph_caps_file_wanted(ci); |
2f2dc053 SW |
1417 | oissued = __ceph_caps_issued_other(ci, cap); |
1418 | ||
979abfdd | 1419 | dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n", |
2f2dc053 | 1420 | inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued), |
979abfdd YZ |
1421 | ceph_cap_string(used), ceph_cap_string(wanted)); |
1422 | if (cap == ci->i_auth_cap) { | |
622f3e25 YZ |
1423 | if (ci->i_dirty_caps || ci->i_flushing_caps || |
1424 | !list_empty(&ci->i_cap_snaps)) | |
979abfdd YZ |
1425 | goto out; |
1426 | if ((used | wanted) & CEPH_CAP_ANY_WR) | |
1427 | goto out; | |
1428 | } | |
5e804ac4 YZ |
1429 | /* The inode has cached pages, but it's no longer used. |
1430 | * we can safely drop it */ | |
1431 | if (wanted == 0 && used == CEPH_CAP_FILE_CACHE && | |
1432 | !(oissued & CEPH_CAP_FILE_CACHE)) { | |
1433 | used = 0; | |
1434 | oissued = 0; | |
1435 | } | |
979abfdd | 1436 | if ((used | wanted) & ~oissued & mine) |
2f2dc053 SW |
1437 | goto out; /* we need these caps */ |
1438 | ||
1439 | session->s_trim_caps--; | |
1440 | if (oissued) { | |
1441 | /* we aren't the only cap.. just remove us */ | |
a096b09a | 1442 | __ceph_remove_cap(cap, true); |
2f2dc053 | 1443 | } else { |
5e804ac4 | 1444 | /* try dropping referring dentries */ |
be655596 | 1445 | spin_unlock(&ci->i_ceph_lock); |
2f2dc053 SW |
1446 | d_prune_aliases(inode); |
1447 | dout("trim_caps_cb %p cap %p pruned, count now %d\n", | |
1448 | inode, cap, atomic_read(&inode->i_count)); | |
1449 | return 0; | |
1450 | } | |
1451 | ||
1452 | out: | |
be655596 | 1453 | spin_unlock(&ci->i_ceph_lock); |
2f2dc053 SW |
1454 | return 0; |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * Trim session cap count down to some max number. | |
1459 | */ | |
1460 | static int trim_caps(struct ceph_mds_client *mdsc, | |
1461 | struct ceph_mds_session *session, | |
1462 | int max_caps) | |
1463 | { | |
1464 | int trim_caps = session->s_nr_caps - max_caps; | |
1465 | ||
1466 | dout("trim_caps mds%d start: %d / %d, trim %d\n", | |
1467 | session->s_mds, session->s_nr_caps, max_caps, trim_caps); | |
1468 | if (trim_caps > 0) { | |
1469 | session->s_trim_caps = trim_caps; | |
1470 | iterate_session_caps(session, trim_caps_cb, session); | |
1471 | dout("trim_caps mds%d done: %d / %d, trimmed %d\n", | |
1472 | session->s_mds, session->s_nr_caps, max_caps, | |
1473 | trim_caps - session->s_trim_caps); | |
5dacf091 | 1474 | session->s_trim_caps = 0; |
2f2dc053 | 1475 | } |
a56371d9 | 1476 | |
a56371d9 | 1477 | ceph_send_cap_releases(mdsc, session); |
2f2dc053 SW |
1478 | return 0; |
1479 | } | |
1480 | ||
8310b089 YZ |
1481 | static int check_capsnap_flush(struct ceph_inode_info *ci, |
1482 | u64 want_snap_seq) | |
d3383a8e | 1483 | { |
8310b089 | 1484 | int ret = 1; |
d3383a8e | 1485 | spin_lock(&ci->i_ceph_lock); |
affbc19a YZ |
1486 | if (want_snap_seq > 0 && !list_empty(&ci->i_cap_snaps)) { |
1487 | struct ceph_cap_snap *capsnap = | |
1488 | list_first_entry(&ci->i_cap_snaps, | |
1489 | struct ceph_cap_snap, ci_item); | |
8310b089 | 1490 | ret = capsnap->follows >= want_snap_seq; |
affbc19a | 1491 | } |
d3383a8e | 1492 | spin_unlock(&ci->i_ceph_lock); |
8310b089 YZ |
1493 | return ret; |
1494 | } | |
1495 | ||
1496 | static int check_caps_flush(struct ceph_mds_client *mdsc, | |
1497 | u64 want_flush_tid) | |
1498 | { | |
1499 | struct rb_node *n; | |
1500 | struct ceph_cap_flush *cf; | |
1501 | int ret = 1; | |
1502 | ||
1503 | spin_lock(&mdsc->cap_dirty_lock); | |
1504 | n = rb_first(&mdsc->cap_flush_tree); | |
1505 | cf = n ? rb_entry(n, struct ceph_cap_flush, g_node) : NULL; | |
1506 | if (cf && cf->tid <= want_flush_tid) { | |
1507 | dout("check_caps_flush still flushing tid %llu <= %llu\n", | |
1508 | cf->tid, want_flush_tid); | |
1509 | ret = 0; | |
1510 | } | |
1511 | spin_unlock(&mdsc->cap_dirty_lock); | |
1512 | return ret; | |
d3383a8e YZ |
1513 | } |
1514 | ||
2f2dc053 SW |
1515 | /* |
1516 | * flush all dirty inode data to disk. | |
1517 | * | |
8310b089 | 1518 | * returns true if we've flushed through want_flush_tid |
2f2dc053 | 1519 | */ |
affbc19a | 1520 | static void wait_caps_flush(struct ceph_mds_client *mdsc, |
8310b089 | 1521 | u64 want_flush_tid, u64 want_snap_seq) |
2f2dc053 | 1522 | { |
d3383a8e | 1523 | int mds; |
2f2dc053 | 1524 | |
8310b089 YZ |
1525 | dout("check_caps_flush want %llu snap want %llu\n", |
1526 | want_flush_tid, want_snap_seq); | |
2f2dc053 | 1527 | mutex_lock(&mdsc->mutex); |
affbc19a | 1528 | for (mds = 0; mds < mdsc->max_sessions; ) { |
2f2dc053 | 1529 | struct ceph_mds_session *session = mdsc->sessions[mds]; |
8310b089 | 1530 | struct inode *inode = NULL; |
2f2dc053 | 1531 | |
affbc19a YZ |
1532 | if (!session) { |
1533 | mds++; | |
2f2dc053 | 1534 | continue; |
affbc19a | 1535 | } |
2f2dc053 SW |
1536 | get_session(session); |
1537 | mutex_unlock(&mdsc->mutex); | |
1538 | ||
1539 | mutex_lock(&session->s_mutex); | |
affbc19a YZ |
1540 | if (!list_empty(&session->s_cap_snaps_flushing)) { |
1541 | struct ceph_cap_snap *capsnap = | |
1542 | list_first_entry(&session->s_cap_snaps_flushing, | |
1543 | struct ceph_cap_snap, | |
1544 | flushing_item); | |
1545 | struct ceph_inode_info *ci = capsnap->ci; | |
8310b089 | 1546 | if (!check_capsnap_flush(ci, want_snap_seq)) { |
affbc19a YZ |
1547 | dout("check_cap_flush still flushing snap %p " |
1548 | "follows %lld <= %lld to mds%d\n", | |
1549 | &ci->vfs_inode, capsnap->follows, | |
1550 | want_snap_seq, mds); | |
8310b089 | 1551 | inode = igrab(&ci->vfs_inode); |
2f2dc053 | 1552 | } |
2f2dc053 SW |
1553 | } |
1554 | mutex_unlock(&session->s_mutex); | |
1555 | ceph_put_mds_session(session); | |
1556 | ||
8310b089 | 1557 | if (inode) { |
affbc19a | 1558 | wait_event(mdsc->cap_flushing_wq, |
8310b089 YZ |
1559 | check_capsnap_flush(ceph_inode(inode), |
1560 | want_snap_seq)); | |
1561 | iput(inode); | |
1562 | } else { | |
affbc19a | 1563 | mds++; |
8310b089 | 1564 | } |
affbc19a | 1565 | |
2f2dc053 SW |
1566 | mutex_lock(&mdsc->mutex); |
1567 | } | |
2f2dc053 | 1568 | mutex_unlock(&mdsc->mutex); |
8310b089 YZ |
1569 | |
1570 | wait_event(mdsc->cap_flushing_wq, | |
1571 | check_caps_flush(mdsc, want_flush_tid)); | |
1572 | ||
1573 | dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid); | |
2f2dc053 SW |
1574 | } |
1575 | ||
1576 | /* | |
1577 | * called under s_mutex | |
1578 | */ | |
3d7ded4d SW |
1579 | void ceph_send_cap_releases(struct ceph_mds_client *mdsc, |
1580 | struct ceph_mds_session *session) | |
2f2dc053 | 1581 | { |
745a8e3b YZ |
1582 | struct ceph_msg *msg = NULL; |
1583 | struct ceph_mds_cap_release *head; | |
1584 | struct ceph_mds_cap_item *item; | |
1585 | struct ceph_cap *cap; | |
1586 | LIST_HEAD(tmp_list); | |
1587 | int num_cap_releases; | |
2f2dc053 | 1588 | |
0f8605f2 | 1589 | spin_lock(&session->s_cap_lock); |
745a8e3b YZ |
1590 | again: |
1591 | list_splice_init(&session->s_cap_releases, &tmp_list); | |
1592 | num_cap_releases = session->s_num_cap_releases; | |
1593 | session->s_num_cap_releases = 0; | |
2f2dc053 | 1594 | spin_unlock(&session->s_cap_lock); |
e01a5946 | 1595 | |
745a8e3b YZ |
1596 | while (!list_empty(&tmp_list)) { |
1597 | if (!msg) { | |
1598 | msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, | |
09cbfeaf | 1599 | PAGE_SIZE, GFP_NOFS, false); |
745a8e3b YZ |
1600 | if (!msg) |
1601 | goto out_err; | |
1602 | head = msg->front.iov_base; | |
1603 | head->num = cpu_to_le32(0); | |
1604 | msg->front.iov_len = sizeof(*head); | |
1605 | } | |
1606 | cap = list_first_entry(&tmp_list, struct ceph_cap, | |
1607 | session_caps); | |
1608 | list_del(&cap->session_caps); | |
1609 | num_cap_releases--; | |
e01a5946 | 1610 | |
00bd8edb | 1611 | head = msg->front.iov_base; |
745a8e3b YZ |
1612 | le32_add_cpu(&head->num, 1); |
1613 | item = msg->front.iov_base + msg->front.iov_len; | |
1614 | item->ino = cpu_to_le64(cap->cap_ino); | |
1615 | item->cap_id = cpu_to_le64(cap->cap_id); | |
1616 | item->migrate_seq = cpu_to_le32(cap->mseq); | |
1617 | item->seq = cpu_to_le32(cap->issue_seq); | |
1618 | msg->front.iov_len += sizeof(*item); | |
1619 | ||
1620 | ceph_put_cap(mdsc, cap); | |
1621 | ||
1622 | if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) { | |
1623 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
1624 | dout("send_cap_releases mds%d %p\n", session->s_mds, msg); | |
1625 | ceph_con_send(&session->s_con, msg); | |
1626 | msg = NULL; | |
1627 | } | |
00bd8edb | 1628 | } |
e01a5946 | 1629 | |
745a8e3b | 1630 | BUG_ON(num_cap_releases != 0); |
e01a5946 | 1631 | |
745a8e3b YZ |
1632 | spin_lock(&session->s_cap_lock); |
1633 | if (!list_empty(&session->s_cap_releases)) | |
1634 | goto again; | |
1635 | spin_unlock(&session->s_cap_lock); | |
1636 | ||
1637 | if (msg) { | |
1638 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
1639 | dout("send_cap_releases mds%d %p\n", session->s_mds, msg); | |
1640 | ceph_con_send(&session->s_con, msg); | |
e01a5946 | 1641 | } |
745a8e3b YZ |
1642 | return; |
1643 | out_err: | |
1644 | pr_err("send_cap_releases mds%d, failed to allocate message\n", | |
1645 | session->s_mds); | |
1646 | spin_lock(&session->s_cap_lock); | |
1647 | list_splice(&tmp_list, &session->s_cap_releases); | |
1648 | session->s_num_cap_releases += num_cap_releases; | |
1649 | spin_unlock(&session->s_cap_lock); | |
e01a5946 SW |
1650 | } |
1651 | ||
2f2dc053 SW |
1652 | /* |
1653 | * requests | |
1654 | */ | |
1655 | ||
54008399 YZ |
1656 | int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req, |
1657 | struct inode *dir) | |
1658 | { | |
1659 | struct ceph_inode_info *ci = ceph_inode(dir); | |
1660 | struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; | |
1661 | struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options; | |
2a5beea3 | 1662 | size_t size = sizeof(struct ceph_mds_reply_dir_entry); |
54008399 YZ |
1663 | int order, num_entries; |
1664 | ||
1665 | spin_lock(&ci->i_ceph_lock); | |
1666 | num_entries = ci->i_files + ci->i_subdirs; | |
1667 | spin_unlock(&ci->i_ceph_lock); | |
1668 | num_entries = max(num_entries, 1); | |
1669 | num_entries = min(num_entries, opt->max_readdir); | |
1670 | ||
1671 | order = get_order(size * num_entries); | |
1672 | while (order >= 0) { | |
2a5beea3 YZ |
1673 | rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL | |
1674 | __GFP_NOWARN, | |
1675 | order); | |
1676 | if (rinfo->dir_entries) | |
54008399 YZ |
1677 | break; |
1678 | order--; | |
1679 | } | |
2a5beea3 | 1680 | if (!rinfo->dir_entries) |
54008399 YZ |
1681 | return -ENOMEM; |
1682 | ||
1683 | num_entries = (PAGE_SIZE << order) / size; | |
1684 | num_entries = min(num_entries, opt->max_readdir); | |
1685 | ||
1686 | rinfo->dir_buf_size = PAGE_SIZE << order; | |
1687 | req->r_num_caps = num_entries + 1; | |
1688 | req->r_args.readdir.max_entries = cpu_to_le32(num_entries); | |
1689 | req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes); | |
1690 | return 0; | |
1691 | } | |
1692 | ||
2f2dc053 SW |
1693 | /* |
1694 | * Create an mds request. | |
1695 | */ | |
1696 | struct ceph_mds_request * | |
1697 | ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode) | |
1698 | { | |
1699 | struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS); | |
1700 | ||
1701 | if (!req) | |
1702 | return ERR_PTR(-ENOMEM); | |
1703 | ||
b4556396 | 1704 | mutex_init(&req->r_fill_mutex); |
37151668 | 1705 | req->r_mdsc = mdsc; |
2f2dc053 SW |
1706 | req->r_started = jiffies; |
1707 | req->r_resend_mds = -1; | |
1708 | INIT_LIST_HEAD(&req->r_unsafe_dir_item); | |
68cd5b4b | 1709 | INIT_LIST_HEAD(&req->r_unsafe_target_item); |
2f2dc053 | 1710 | req->r_fmode = -1; |
153c8e6b | 1711 | kref_init(&req->r_kref); |
fcd00b68 | 1712 | RB_CLEAR_NODE(&req->r_node); |
2f2dc053 SW |
1713 | INIT_LIST_HEAD(&req->r_wait); |
1714 | init_completion(&req->r_completion); | |
1715 | init_completion(&req->r_safe_completion); | |
1716 | INIT_LIST_HEAD(&req->r_unsafe_item); | |
1717 | ||
8bbd4714 | 1718 | req->r_stamp = current_fs_time(mdsc->fsc->sb); |
b8e69066 | 1719 | |
2f2dc053 SW |
1720 | req->r_op = op; |
1721 | req->r_direct_mode = mode; | |
1722 | return req; | |
1723 | } | |
1724 | ||
1725 | /* | |
44ca18f2 | 1726 | * return oldest (lowest) request, tid in request tree, 0 if none. |
2f2dc053 SW |
1727 | * |
1728 | * called under mdsc->mutex. | |
1729 | */ | |
44ca18f2 SW |
1730 | static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc) |
1731 | { | |
1732 | if (RB_EMPTY_ROOT(&mdsc->request_tree)) | |
1733 | return NULL; | |
1734 | return rb_entry(rb_first(&mdsc->request_tree), | |
1735 | struct ceph_mds_request, r_node); | |
1736 | } | |
1737 | ||
e8a7b8b1 | 1738 | static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc) |
2f2dc053 | 1739 | { |
e8a7b8b1 | 1740 | return mdsc->oldest_tid; |
2f2dc053 SW |
1741 | } |
1742 | ||
1743 | /* | |
1744 | * Build a dentry's path. Allocate on heap; caller must kfree. Based | |
1745 | * on build_path_from_dentry in fs/cifs/dir.c. | |
1746 | * | |
1747 | * If @stop_on_nosnap, generate path relative to the first non-snapped | |
1748 | * inode. | |
1749 | * | |
1750 | * Encode hidden .snap dirs as a double /, i.e. | |
1751 | * foo/.snap/bar -> foo//bar | |
1752 | */ | |
1753 | char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base, | |
1754 | int stop_on_nosnap) | |
1755 | { | |
1756 | struct dentry *temp; | |
1757 | char *path; | |
1758 | int len, pos; | |
1b71fe2e | 1759 | unsigned seq; |
2f2dc053 SW |
1760 | |
1761 | if (dentry == NULL) | |
1762 | return ERR_PTR(-EINVAL); | |
1763 | ||
1764 | retry: | |
1765 | len = 0; | |
1b71fe2e AV |
1766 | seq = read_seqbegin(&rename_lock); |
1767 | rcu_read_lock(); | |
2f2dc053 | 1768 | for (temp = dentry; !IS_ROOT(temp);) { |
2b0143b5 | 1769 | struct inode *inode = d_inode(temp); |
2f2dc053 SW |
1770 | if (inode && ceph_snap(inode) == CEPH_SNAPDIR) |
1771 | len++; /* slash only */ | |
1772 | else if (stop_on_nosnap && inode && | |
1773 | ceph_snap(inode) == CEPH_NOSNAP) | |
1774 | break; | |
1775 | else | |
1776 | len += 1 + temp->d_name.len; | |
1777 | temp = temp->d_parent; | |
2f2dc053 | 1778 | } |
1b71fe2e | 1779 | rcu_read_unlock(); |
2f2dc053 SW |
1780 | if (len) |
1781 | len--; /* no leading '/' */ | |
1782 | ||
1783 | path = kmalloc(len+1, GFP_NOFS); | |
1784 | if (path == NULL) | |
1785 | return ERR_PTR(-ENOMEM); | |
1786 | pos = len; | |
1787 | path[pos] = 0; /* trailing null */ | |
1b71fe2e | 1788 | rcu_read_lock(); |
2f2dc053 | 1789 | for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) { |
1b71fe2e | 1790 | struct inode *inode; |
2f2dc053 | 1791 | |
1b71fe2e | 1792 | spin_lock(&temp->d_lock); |
2b0143b5 | 1793 | inode = d_inode(temp); |
2f2dc053 | 1794 | if (inode && ceph_snap(inode) == CEPH_SNAPDIR) { |
104648ad | 1795 | dout("build_path path+%d: %p SNAPDIR\n", |
2f2dc053 SW |
1796 | pos, temp); |
1797 | } else if (stop_on_nosnap && inode && | |
1798 | ceph_snap(inode) == CEPH_NOSNAP) { | |
9d5a09e6 | 1799 | spin_unlock(&temp->d_lock); |
2f2dc053 SW |
1800 | break; |
1801 | } else { | |
1802 | pos -= temp->d_name.len; | |
1b71fe2e AV |
1803 | if (pos < 0) { |
1804 | spin_unlock(&temp->d_lock); | |
2f2dc053 | 1805 | break; |
1b71fe2e | 1806 | } |
2f2dc053 SW |
1807 | strncpy(path + pos, temp->d_name.name, |
1808 | temp->d_name.len); | |
2f2dc053 | 1809 | } |
1b71fe2e | 1810 | spin_unlock(&temp->d_lock); |
2f2dc053 SW |
1811 | if (pos) |
1812 | path[--pos] = '/'; | |
1813 | temp = temp->d_parent; | |
2f2dc053 | 1814 | } |
1b71fe2e AV |
1815 | rcu_read_unlock(); |
1816 | if (pos != 0 || read_seqretry(&rename_lock, seq)) { | |
104648ad | 1817 | pr_err("build_path did not end path lookup where " |
2f2dc053 SW |
1818 | "expected, namelen is %d, pos is %d\n", len, pos); |
1819 | /* presumably this is only possible if racing with a | |
1820 | rename of one of the parent directories (we can not | |
1821 | lock the dentries above us to prevent this, but | |
1822 | retrying should be harmless) */ | |
1823 | kfree(path); | |
1824 | goto retry; | |
1825 | } | |
1826 | ||
2b0143b5 | 1827 | *base = ceph_ino(d_inode(temp)); |
2f2dc053 | 1828 | *plen = len; |
104648ad | 1829 | dout("build_path on %p %d built %llx '%.*s'\n", |
84d08fa8 | 1830 | dentry, d_count(dentry), *base, len, path); |
2f2dc053 SW |
1831 | return path; |
1832 | } | |
1833 | ||
1834 | static int build_dentry_path(struct dentry *dentry, | |
1835 | const char **ppath, int *ppathlen, u64 *pino, | |
1836 | int *pfreepath) | |
1837 | { | |
1838 | char *path; | |
1839 | ||
2b0143b5 DH |
1840 | if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP) { |
1841 | *pino = ceph_ino(d_inode(dentry->d_parent)); | |
2f2dc053 SW |
1842 | *ppath = dentry->d_name.name; |
1843 | *ppathlen = dentry->d_name.len; | |
1844 | return 0; | |
1845 | } | |
1846 | path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1); | |
1847 | if (IS_ERR(path)) | |
1848 | return PTR_ERR(path); | |
1849 | *ppath = path; | |
1850 | *pfreepath = 1; | |
1851 | return 0; | |
1852 | } | |
1853 | ||
1854 | static int build_inode_path(struct inode *inode, | |
1855 | const char **ppath, int *ppathlen, u64 *pino, | |
1856 | int *pfreepath) | |
1857 | { | |
1858 | struct dentry *dentry; | |
1859 | char *path; | |
1860 | ||
1861 | if (ceph_snap(inode) == CEPH_NOSNAP) { | |
1862 | *pino = ceph_ino(inode); | |
1863 | *ppathlen = 0; | |
1864 | return 0; | |
1865 | } | |
1866 | dentry = d_find_alias(inode); | |
1867 | path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1); | |
1868 | dput(dentry); | |
1869 | if (IS_ERR(path)) | |
1870 | return PTR_ERR(path); | |
1871 | *ppath = path; | |
1872 | *pfreepath = 1; | |
1873 | return 0; | |
1874 | } | |
1875 | ||
1876 | /* | |
1877 | * request arguments may be specified via an inode *, a dentry *, or | |
1878 | * an explicit ino+path. | |
1879 | */ | |
1880 | static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry, | |
1881 | const char *rpath, u64 rino, | |
1882 | const char **ppath, int *pathlen, | |
1883 | u64 *ino, int *freepath) | |
1884 | { | |
1885 | int r = 0; | |
1886 | ||
1887 | if (rinode) { | |
1888 | r = build_inode_path(rinode, ppath, pathlen, ino, freepath); | |
1889 | dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode), | |
1890 | ceph_snap(rinode)); | |
1891 | } else if (rdentry) { | |
1892 | r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath); | |
1893 | dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen, | |
1894 | *ppath); | |
795858db | 1895 | } else if (rpath || rino) { |
2f2dc053 SW |
1896 | *ino = rino; |
1897 | *ppath = rpath; | |
b000056a | 1898 | *pathlen = rpath ? strlen(rpath) : 0; |
2f2dc053 SW |
1899 | dout(" path %.*s\n", *pathlen, rpath); |
1900 | } | |
1901 | ||
1902 | return r; | |
1903 | } | |
1904 | ||
1905 | /* | |
1906 | * called under mdsc->mutex | |
1907 | */ | |
1908 | static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc, | |
1909 | struct ceph_mds_request *req, | |
6e6f0923 | 1910 | int mds, bool drop_cap_releases) |
2f2dc053 SW |
1911 | { |
1912 | struct ceph_msg *msg; | |
1913 | struct ceph_mds_request_head *head; | |
1914 | const char *path1 = NULL; | |
1915 | const char *path2 = NULL; | |
1916 | u64 ino1 = 0, ino2 = 0; | |
1917 | int pathlen1 = 0, pathlen2 = 0; | |
1918 | int freepath1 = 0, freepath2 = 0; | |
1919 | int len; | |
1920 | u16 releases; | |
1921 | void *p, *end; | |
1922 | int ret; | |
1923 | ||
1924 | ret = set_request_path_attr(req->r_inode, req->r_dentry, | |
1925 | req->r_path1, req->r_ino1.ino, | |
1926 | &path1, &pathlen1, &ino1, &freepath1); | |
1927 | if (ret < 0) { | |
1928 | msg = ERR_PTR(ret); | |
1929 | goto out; | |
1930 | } | |
1931 | ||
1932 | ret = set_request_path_attr(NULL, req->r_old_dentry, | |
1933 | req->r_path2, req->r_ino2.ino, | |
1934 | &path2, &pathlen2, &ino2, &freepath2); | |
1935 | if (ret < 0) { | |
1936 | msg = ERR_PTR(ret); | |
1937 | goto out_free1; | |
1938 | } | |
1939 | ||
1940 | len = sizeof(*head) + | |
b8e69066 | 1941 | pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) + |
777d738a | 1942 | sizeof(struct ceph_timespec); |
2f2dc053 SW |
1943 | |
1944 | /* calculate (max) length for cap releases */ | |
1945 | len += sizeof(struct ceph_mds_request_release) * | |
1946 | (!!req->r_inode_drop + !!req->r_dentry_drop + | |
1947 | !!req->r_old_inode_drop + !!req->r_old_dentry_drop); | |
1948 | if (req->r_dentry_drop) | |
1949 | len += req->r_dentry->d_name.len; | |
1950 | if (req->r_old_dentry_drop) | |
1951 | len += req->r_old_dentry->d_name.len; | |
1952 | ||
b61c2763 | 1953 | msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false); |
a79832f2 SW |
1954 | if (!msg) { |
1955 | msg = ERR_PTR(-ENOMEM); | |
2f2dc053 | 1956 | goto out_free2; |
a79832f2 | 1957 | } |
2f2dc053 | 1958 | |
7cfa0313 | 1959 | msg->hdr.version = cpu_to_le16(2); |
6df058c0 SW |
1960 | msg->hdr.tid = cpu_to_le64(req->r_tid); |
1961 | ||
2f2dc053 SW |
1962 | head = msg->front.iov_base; |
1963 | p = msg->front.iov_base + sizeof(*head); | |
1964 | end = msg->front.iov_base + msg->front.iov_len; | |
1965 | ||
1966 | head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch); | |
1967 | head->op = cpu_to_le32(req->r_op); | |
ff3d0046 EB |
1968 | head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid)); |
1969 | head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid)); | |
2f2dc053 SW |
1970 | head->args = req->r_args; |
1971 | ||
1972 | ceph_encode_filepath(&p, end, ino1, path1); | |
1973 | ceph_encode_filepath(&p, end, ino2, path2); | |
1974 | ||
e979cf50 SW |
1975 | /* make note of release offset, in case we need to replay */ |
1976 | req->r_request_release_offset = p - msg->front.iov_base; | |
1977 | ||
2f2dc053 SW |
1978 | /* cap releases */ |
1979 | releases = 0; | |
1980 | if (req->r_inode_drop) | |
1981 | releases += ceph_encode_inode_release(&p, | |
2b0143b5 | 1982 | req->r_inode ? req->r_inode : d_inode(req->r_dentry), |
2f2dc053 SW |
1983 | mds, req->r_inode_drop, req->r_inode_unless, 0); |
1984 | if (req->r_dentry_drop) | |
1985 | releases += ceph_encode_dentry_release(&p, req->r_dentry, | |
1986 | mds, req->r_dentry_drop, req->r_dentry_unless); | |
1987 | if (req->r_old_dentry_drop) | |
1988 | releases += ceph_encode_dentry_release(&p, req->r_old_dentry, | |
1989 | mds, req->r_old_dentry_drop, req->r_old_dentry_unless); | |
1990 | if (req->r_old_inode_drop) | |
1991 | releases += ceph_encode_inode_release(&p, | |
2b0143b5 | 1992 | d_inode(req->r_old_dentry), |
2f2dc053 | 1993 | mds, req->r_old_inode_drop, req->r_old_inode_unless, 0); |
6e6f0923 YZ |
1994 | |
1995 | if (drop_cap_releases) { | |
1996 | releases = 0; | |
1997 | p = msg->front.iov_base + req->r_request_release_offset; | |
1998 | } | |
1999 | ||
2f2dc053 SW |
2000 | head->num_releases = cpu_to_le16(releases); |
2001 | ||
b8e69066 | 2002 | /* time stamp */ |
1f041a89 YZ |
2003 | { |
2004 | struct ceph_timespec ts; | |
2005 | ceph_encode_timespec(&ts, &req->r_stamp); | |
2006 | ceph_encode_copy(&p, &ts, sizeof(ts)); | |
2007 | } | |
b8e69066 | 2008 | |
2f2dc053 SW |
2009 | BUG_ON(p > end); |
2010 | msg->front.iov_len = p - msg->front.iov_base; | |
2011 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
2012 | ||
25e6bae3 YZ |
2013 | if (req->r_pagelist) { |
2014 | struct ceph_pagelist *pagelist = req->r_pagelist; | |
2015 | atomic_inc(&pagelist->refcnt); | |
2016 | ceph_msg_data_add_pagelist(msg, pagelist); | |
2017 | msg->hdr.data_len = cpu_to_le32(pagelist->length); | |
2018 | } else { | |
2019 | msg->hdr.data_len = 0; | |
ebf18f47 | 2020 | } |
02afca6c | 2021 | |
2f2dc053 SW |
2022 | msg->hdr.data_off = cpu_to_le16(0); |
2023 | ||
2024 | out_free2: | |
2025 | if (freepath2) | |
2026 | kfree((char *)path2); | |
2027 | out_free1: | |
2028 | if (freepath1) | |
2029 | kfree((char *)path1); | |
2030 | out: | |
2031 | return msg; | |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * called under mdsc->mutex if error, under no mutex if | |
2036 | * success. | |
2037 | */ | |
2038 | static void complete_request(struct ceph_mds_client *mdsc, | |
2039 | struct ceph_mds_request *req) | |
2040 | { | |
2041 | if (req->r_callback) | |
2042 | req->r_callback(mdsc, req); | |
2043 | else | |
03066f23 | 2044 | complete_all(&req->r_completion); |
2f2dc053 SW |
2045 | } |
2046 | ||
2047 | /* | |
2048 | * called under mdsc->mutex | |
2049 | */ | |
2050 | static int __prepare_send_request(struct ceph_mds_client *mdsc, | |
2051 | struct ceph_mds_request *req, | |
6e6f0923 | 2052 | int mds, bool drop_cap_releases) |
2f2dc053 SW |
2053 | { |
2054 | struct ceph_mds_request_head *rhead; | |
2055 | struct ceph_msg *msg; | |
2056 | int flags = 0; | |
2057 | ||
2f2dc053 | 2058 | req->r_attempts++; |
e55b71f8 GF |
2059 | if (req->r_inode) { |
2060 | struct ceph_cap *cap = | |
2061 | ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds); | |
2062 | ||
2063 | if (cap) | |
2064 | req->r_sent_on_mseq = cap->mseq; | |
2065 | else | |
2066 | req->r_sent_on_mseq = -1; | |
2067 | } | |
2f2dc053 SW |
2068 | dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req, |
2069 | req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts); | |
2070 | ||
01a92f17 | 2071 | if (req->r_got_unsafe) { |
c5c9a0bf | 2072 | void *p; |
01a92f17 SW |
2073 | /* |
2074 | * Replay. Do not regenerate message (and rebuild | |
2075 | * paths, etc.); just use the original message. | |
2076 | * Rebuilding paths will break for renames because | |
2077 | * d_move mangles the src name. | |
2078 | */ | |
2079 | msg = req->r_request; | |
2080 | rhead = msg->front.iov_base; | |
2081 | ||
2082 | flags = le32_to_cpu(rhead->flags); | |
2083 | flags |= CEPH_MDS_FLAG_REPLAY; | |
2084 | rhead->flags = cpu_to_le32(flags); | |
2085 | ||
2086 | if (req->r_target_inode) | |
2087 | rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode)); | |
2088 | ||
2089 | rhead->num_retry = req->r_attempts - 1; | |
e979cf50 SW |
2090 | |
2091 | /* remove cap/dentry releases from message */ | |
2092 | rhead->num_releases = 0; | |
c5c9a0bf YZ |
2093 | |
2094 | /* time stamp */ | |
2095 | p = msg->front.iov_base + req->r_request_release_offset; | |
1f041a89 YZ |
2096 | { |
2097 | struct ceph_timespec ts; | |
2098 | ceph_encode_timespec(&ts, &req->r_stamp); | |
2099 | ceph_encode_copy(&p, &ts, sizeof(ts)); | |
2100 | } | |
c5c9a0bf YZ |
2101 | |
2102 | msg->front.iov_len = p - msg->front.iov_base; | |
2103 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
01a92f17 SW |
2104 | return 0; |
2105 | } | |
2106 | ||
2f2dc053 SW |
2107 | if (req->r_request) { |
2108 | ceph_msg_put(req->r_request); | |
2109 | req->r_request = NULL; | |
2110 | } | |
6e6f0923 | 2111 | msg = create_request_message(mdsc, req, mds, drop_cap_releases); |
2f2dc053 | 2112 | if (IS_ERR(msg)) { |
e1518c7c | 2113 | req->r_err = PTR_ERR(msg); |
a79832f2 | 2114 | return PTR_ERR(msg); |
2f2dc053 SW |
2115 | } |
2116 | req->r_request = msg; | |
2117 | ||
2118 | rhead = msg->front.iov_base; | |
2f2dc053 SW |
2119 | rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc)); |
2120 | if (req->r_got_unsafe) | |
2121 | flags |= CEPH_MDS_FLAG_REPLAY; | |
2122 | if (req->r_locked_dir) | |
2123 | flags |= CEPH_MDS_FLAG_WANT_DENTRY; | |
2124 | rhead->flags = cpu_to_le32(flags); | |
2125 | rhead->num_fwd = req->r_num_fwd; | |
2126 | rhead->num_retry = req->r_attempts - 1; | |
01a92f17 | 2127 | rhead->ino = 0; |
2f2dc053 SW |
2128 | |
2129 | dout(" r_locked_dir = %p\n", req->r_locked_dir); | |
2f2dc053 SW |
2130 | return 0; |
2131 | } | |
2132 | ||
2133 | /* | |
2134 | * send request, or put it on the appropriate wait list. | |
2135 | */ | |
2136 | static int __do_request(struct ceph_mds_client *mdsc, | |
2137 | struct ceph_mds_request *req) | |
2138 | { | |
2139 | struct ceph_mds_session *session = NULL; | |
2140 | int mds = -1; | |
48fec5d0 | 2141 | int err = 0; |
2f2dc053 | 2142 | |
eb1b8af3 YZ |
2143 | if (req->r_err || req->r_got_result) { |
2144 | if (req->r_aborted) | |
2145 | __unregister_request(mdsc, req); | |
2f2dc053 | 2146 | goto out; |
eb1b8af3 | 2147 | } |
2f2dc053 SW |
2148 | |
2149 | if (req->r_timeout && | |
2150 | time_after_eq(jiffies, req->r_started + req->r_timeout)) { | |
2151 | dout("do_request timed out\n"); | |
2152 | err = -EIO; | |
2153 | goto finish; | |
2154 | } | |
48fec5d0 YZ |
2155 | if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) { |
2156 | dout("do_request forced umount\n"); | |
2157 | err = -EIO; | |
2158 | goto finish; | |
2159 | } | |
2f2dc053 | 2160 | |
dc69e2e9 SW |
2161 | put_request_session(req); |
2162 | ||
2f2dc053 SW |
2163 | mds = __choose_mds(mdsc, req); |
2164 | if (mds < 0 || | |
2165 | ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) { | |
2166 | dout("do_request no mds or not active, waiting for map\n"); | |
2167 | list_add(&req->r_wait, &mdsc->waiting_for_map); | |
2168 | goto out; | |
2169 | } | |
2170 | ||
2171 | /* get, open session */ | |
2172 | session = __ceph_lookup_mds_session(mdsc, mds); | |
9c423956 | 2173 | if (!session) { |
2f2dc053 | 2174 | session = register_session(mdsc, mds); |
9c423956 SW |
2175 | if (IS_ERR(session)) { |
2176 | err = PTR_ERR(session); | |
2177 | goto finish; | |
2178 | } | |
2179 | } | |
dc69e2e9 SW |
2180 | req->r_session = get_session(session); |
2181 | ||
2f2dc053 | 2182 | dout("do_request mds%d session %p state %s\n", mds, session, |
a687ecaf | 2183 | ceph_session_state_name(session->s_state)); |
2f2dc053 SW |
2184 | if (session->s_state != CEPH_MDS_SESSION_OPEN && |
2185 | session->s_state != CEPH_MDS_SESSION_HUNG) { | |
2186 | if (session->s_state == CEPH_MDS_SESSION_NEW || | |
2187 | session->s_state == CEPH_MDS_SESSION_CLOSING) | |
2188 | __open_session(mdsc, session); | |
2189 | list_add(&req->r_wait, &session->s_waiting); | |
2190 | goto out_session; | |
2191 | } | |
2192 | ||
2193 | /* send request */ | |
2f2dc053 SW |
2194 | req->r_resend_mds = -1; /* forget any previous mds hint */ |
2195 | ||
2196 | if (req->r_request_started == 0) /* note request start time */ | |
2197 | req->r_request_started = jiffies; | |
2198 | ||
6e6f0923 | 2199 | err = __prepare_send_request(mdsc, req, mds, false); |
2f2dc053 SW |
2200 | if (!err) { |
2201 | ceph_msg_get(req->r_request); | |
2202 | ceph_con_send(&session->s_con, req->r_request); | |
2203 | } | |
2204 | ||
2205 | out_session: | |
2206 | ceph_put_mds_session(session); | |
48fec5d0 YZ |
2207 | finish: |
2208 | if (err) { | |
2209 | dout("__do_request early error %d\n", err); | |
2210 | req->r_err = err; | |
2211 | complete_request(mdsc, req); | |
2212 | __unregister_request(mdsc, req); | |
2213 | } | |
2f2dc053 SW |
2214 | out: |
2215 | return err; | |
2f2dc053 SW |
2216 | } |
2217 | ||
2218 | /* | |
2219 | * called under mdsc->mutex | |
2220 | */ | |
2221 | static void __wake_requests(struct ceph_mds_client *mdsc, | |
2222 | struct list_head *head) | |
2223 | { | |
ed75ec2c YZ |
2224 | struct ceph_mds_request *req; |
2225 | LIST_HEAD(tmp_list); | |
2226 | ||
2227 | list_splice_init(head, &tmp_list); | |
2f2dc053 | 2228 | |
ed75ec2c YZ |
2229 | while (!list_empty(&tmp_list)) { |
2230 | req = list_entry(tmp_list.next, | |
2231 | struct ceph_mds_request, r_wait); | |
2f2dc053 | 2232 | list_del_init(&req->r_wait); |
7971bd92 | 2233 | dout(" wake request %p tid %llu\n", req, req->r_tid); |
2f2dc053 SW |
2234 | __do_request(mdsc, req); |
2235 | } | |
2236 | } | |
2237 | ||
2238 | /* | |
2239 | * Wake up threads with requests pending for @mds, so that they can | |
29790f26 | 2240 | * resubmit their requests to a possibly different mds. |
2f2dc053 | 2241 | */ |
29790f26 | 2242 | static void kick_requests(struct ceph_mds_client *mdsc, int mds) |
2f2dc053 | 2243 | { |
44ca18f2 | 2244 | struct ceph_mds_request *req; |
282c1052 | 2245 | struct rb_node *p = rb_first(&mdsc->request_tree); |
2f2dc053 SW |
2246 | |
2247 | dout("kick_requests mds%d\n", mds); | |
282c1052 | 2248 | while (p) { |
44ca18f2 | 2249 | req = rb_entry(p, struct ceph_mds_request, r_node); |
282c1052 | 2250 | p = rb_next(p); |
44ca18f2 SW |
2251 | if (req->r_got_unsafe) |
2252 | continue; | |
3de22be6 YZ |
2253 | if (req->r_attempts > 0) |
2254 | continue; /* only new requests */ | |
44ca18f2 SW |
2255 | if (req->r_session && |
2256 | req->r_session->s_mds == mds) { | |
2257 | dout(" kicking tid %llu\n", req->r_tid); | |
03974e81 | 2258 | list_del_init(&req->r_wait); |
44ca18f2 | 2259 | __do_request(mdsc, req); |
2f2dc053 SW |
2260 | } |
2261 | } | |
2262 | } | |
2263 | ||
2264 | void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, | |
2265 | struct ceph_mds_request *req) | |
2266 | { | |
2267 | dout("submit_request on %p\n", req); | |
2268 | mutex_lock(&mdsc->mutex); | |
2269 | __register_request(mdsc, req, NULL); | |
2270 | __do_request(mdsc, req); | |
2271 | mutex_unlock(&mdsc->mutex); | |
2272 | } | |
2273 | ||
2274 | /* | |
2275 | * Synchrously perform an mds request. Take care of all of the | |
2276 | * session setup, forwarding, retry details. | |
2277 | */ | |
2278 | int ceph_mdsc_do_request(struct ceph_mds_client *mdsc, | |
2279 | struct inode *dir, | |
2280 | struct ceph_mds_request *req) | |
2281 | { | |
2282 | int err; | |
2283 | ||
2284 | dout("do_request on %p\n", req); | |
2285 | ||
2286 | /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */ | |
2287 | if (req->r_inode) | |
2288 | ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN); | |
2289 | if (req->r_locked_dir) | |
2290 | ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN); | |
844d87c3 | 2291 | if (req->r_old_dentry_dir) |
41b02e1f SW |
2292 | ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir), |
2293 | CEPH_CAP_PIN); | |
2f2dc053 | 2294 | |
5ea5c5e0 YZ |
2295 | /* deny access to directories with pool_ns layouts */ |
2296 | if (req->r_inode && S_ISDIR(req->r_inode->i_mode) && | |
2297 | ceph_inode(req->r_inode)->i_pool_ns_len) | |
2298 | return -EIO; | |
2299 | if (req->r_locked_dir && | |
2300 | ceph_inode(req->r_locked_dir)->i_pool_ns_len) | |
2301 | return -EIO; | |
2302 | ||
2f2dc053 SW |
2303 | /* issue */ |
2304 | mutex_lock(&mdsc->mutex); | |
2305 | __register_request(mdsc, req, dir); | |
2306 | __do_request(mdsc, req); | |
2307 | ||
e1518c7c SW |
2308 | if (req->r_err) { |
2309 | err = req->r_err; | |
e1518c7c | 2310 | goto out; |
2f2dc053 SW |
2311 | } |
2312 | ||
e1518c7c SW |
2313 | /* wait */ |
2314 | mutex_unlock(&mdsc->mutex); | |
2315 | dout("do_request waiting\n"); | |
5be73034 | 2316 | if (!req->r_timeout && req->r_wait_for_completion) { |
9280be24 | 2317 | err = req->r_wait_for_completion(mdsc, req); |
e1518c7c | 2318 | } else { |
5be73034 ID |
2319 | long timeleft = wait_for_completion_killable_timeout( |
2320 | &req->r_completion, | |
2321 | ceph_timeout_jiffies(req->r_timeout)); | |
2322 | if (timeleft > 0) | |
2323 | err = 0; | |
2324 | else if (!timeleft) | |
2325 | err = -EIO; /* timed out */ | |
2326 | else | |
2327 | err = timeleft; /* killed */ | |
e1518c7c SW |
2328 | } |
2329 | dout("do_request waited, got %d\n", err); | |
2330 | mutex_lock(&mdsc->mutex); | |
5b1daecd | 2331 | |
e1518c7c SW |
2332 | /* only abort if we didn't race with a real reply */ |
2333 | if (req->r_got_result) { | |
2334 | err = le32_to_cpu(req->r_reply_info.head->result); | |
2335 | } else if (err < 0) { | |
2336 | dout("aborted request %lld with %d\n", req->r_tid, err); | |
b4556396 SW |
2337 | |
2338 | /* | |
2339 | * ensure we aren't running concurrently with | |
2340 | * ceph_fill_trace or ceph_readdir_prepopulate, which | |
2341 | * rely on locks (dir mutex) held by our caller. | |
2342 | */ | |
2343 | mutex_lock(&req->r_fill_mutex); | |
e1518c7c SW |
2344 | req->r_err = err; |
2345 | req->r_aborted = true; | |
b4556396 | 2346 | mutex_unlock(&req->r_fill_mutex); |
5b1daecd | 2347 | |
e1518c7c | 2348 | if (req->r_locked_dir && |
167c9e35 SW |
2349 | (req->r_op & CEPH_MDS_OP_WRITE)) |
2350 | ceph_invalidate_dir_request(req); | |
2f2dc053 | 2351 | } else { |
e1518c7c | 2352 | err = req->r_err; |
2f2dc053 | 2353 | } |
2f2dc053 | 2354 | |
e1518c7c SW |
2355 | out: |
2356 | mutex_unlock(&mdsc->mutex); | |
2f2dc053 SW |
2357 | dout("do_request %p done, result %d\n", req, err); |
2358 | return err; | |
2359 | } | |
2360 | ||
167c9e35 | 2361 | /* |
2f276c51 | 2362 | * Invalidate dir's completeness, dentry lease state on an aborted MDS |
167c9e35 SW |
2363 | * namespace request. |
2364 | */ | |
2365 | void ceph_invalidate_dir_request(struct ceph_mds_request *req) | |
2366 | { | |
2367 | struct inode *inode = req->r_locked_dir; | |
167c9e35 | 2368 | |
2f276c51 | 2369 | dout("invalidate_dir_request %p (complete, lease(s))\n", inode); |
167c9e35 | 2370 | |
2f276c51 | 2371 | ceph_dir_clear_complete(inode); |
167c9e35 SW |
2372 | if (req->r_dentry) |
2373 | ceph_invalidate_dentry_lease(req->r_dentry); | |
2374 | if (req->r_old_dentry) | |
2375 | ceph_invalidate_dentry_lease(req->r_old_dentry); | |
2376 | } | |
2377 | ||
2f2dc053 SW |
2378 | /* |
2379 | * Handle mds reply. | |
2380 | * | |
2381 | * We take the session mutex and parse and process the reply immediately. | |
2382 | * This preserves the logical ordering of replies, capabilities, etc., sent | |
2383 | * by the MDS as they are applied to our local cache. | |
2384 | */ | |
2385 | static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg) | |
2386 | { | |
2387 | struct ceph_mds_client *mdsc = session->s_mdsc; | |
2388 | struct ceph_mds_request *req; | |
2389 | struct ceph_mds_reply_head *head = msg->front.iov_base; | |
2390 | struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */ | |
982d6011 | 2391 | struct ceph_snap_realm *realm; |
2f2dc053 SW |
2392 | u64 tid; |
2393 | int err, result; | |
2600d2dd | 2394 | int mds = session->s_mds; |
2f2dc053 | 2395 | |
2f2dc053 SW |
2396 | if (msg->front.iov_len < sizeof(*head)) { |
2397 | pr_err("mdsc_handle_reply got corrupt (short) reply\n"); | |
9ec7cab1 | 2398 | ceph_msg_dump(msg); |
2f2dc053 SW |
2399 | return; |
2400 | } | |
2401 | ||
2402 | /* get request, session */ | |
6df058c0 | 2403 | tid = le64_to_cpu(msg->hdr.tid); |
2f2dc053 | 2404 | mutex_lock(&mdsc->mutex); |
fcd00b68 | 2405 | req = lookup_get_request(mdsc, tid); |
2f2dc053 SW |
2406 | if (!req) { |
2407 | dout("handle_reply on unknown tid %llu\n", tid); | |
2408 | mutex_unlock(&mdsc->mutex); | |
2409 | return; | |
2410 | } | |
2411 | dout("handle_reply %p\n", req); | |
2f2dc053 SW |
2412 | |
2413 | /* correct session? */ | |
d96d6049 | 2414 | if (req->r_session != session) { |
2f2dc053 SW |
2415 | pr_err("mdsc_handle_reply got %llu on session mds%d" |
2416 | " not mds%d\n", tid, session->s_mds, | |
2417 | req->r_session ? req->r_session->s_mds : -1); | |
2418 | mutex_unlock(&mdsc->mutex); | |
2419 | goto out; | |
2420 | } | |
2421 | ||
2422 | /* dup? */ | |
2423 | if ((req->r_got_unsafe && !head->safe) || | |
2424 | (req->r_got_safe && head->safe)) { | |
f3ae1b97 | 2425 | pr_warn("got a dup %s reply on %llu from mds%d\n", |
2f2dc053 SW |
2426 | head->safe ? "safe" : "unsafe", tid, mds); |
2427 | mutex_unlock(&mdsc->mutex); | |
2428 | goto out; | |
2429 | } | |
1550d34e | 2430 | if (req->r_got_safe) { |
f3ae1b97 | 2431 | pr_warn("got unsafe after safe on %llu from mds%d\n", |
85792d0d SW |
2432 | tid, mds); |
2433 | mutex_unlock(&mdsc->mutex); | |
2434 | goto out; | |
2435 | } | |
2f2dc053 SW |
2436 | |
2437 | result = le32_to_cpu(head->result); | |
2438 | ||
2439 | /* | |
e55b71f8 GF |
2440 | * Handle an ESTALE |
2441 | * if we're not talking to the authority, send to them | |
2442 | * if the authority has changed while we weren't looking, | |
2443 | * send to new authority | |
2444 | * Otherwise we just have to return an ESTALE | |
2f2dc053 SW |
2445 | */ |
2446 | if (result == -ESTALE) { | |
e55b71f8 | 2447 | dout("got ESTALE on request %llu", req->r_tid); |
51da8e8c | 2448 | req->r_resend_mds = -1; |
ca18bede | 2449 | if (req->r_direct_mode != USE_AUTH_MDS) { |
e55b71f8 GF |
2450 | dout("not using auth, setting for that now"); |
2451 | req->r_direct_mode = USE_AUTH_MDS; | |
2f2dc053 SW |
2452 | __do_request(mdsc, req); |
2453 | mutex_unlock(&mdsc->mutex); | |
2454 | goto out; | |
e55b71f8 | 2455 | } else { |
ca18bede YZ |
2456 | int mds = __choose_mds(mdsc, req); |
2457 | if (mds >= 0 && mds != req->r_session->s_mds) { | |
2458 | dout("but auth changed, so resending"); | |
e55b71f8 GF |
2459 | __do_request(mdsc, req); |
2460 | mutex_unlock(&mdsc->mutex); | |
2461 | goto out; | |
2462 | } | |
2f2dc053 | 2463 | } |
e55b71f8 | 2464 | dout("have to return ESTALE on request %llu", req->r_tid); |
2f2dc053 SW |
2465 | } |
2466 | ||
e55b71f8 | 2467 | |
2f2dc053 SW |
2468 | if (head->safe) { |
2469 | req->r_got_safe = true; | |
2470 | __unregister_request(mdsc, req); | |
2f2dc053 SW |
2471 | |
2472 | if (req->r_got_unsafe) { | |
2473 | /* | |
2474 | * We already handled the unsafe response, now do the | |
2475 | * cleanup. No need to examine the response; the MDS | |
2476 | * doesn't include any result info in the safe | |
2477 | * response. And even if it did, there is nothing | |
2478 | * useful we could do with a revised return value. | |
2479 | */ | |
2480 | dout("got safe reply %llu, mds%d\n", tid, mds); | |
2481 | list_del_init(&req->r_unsafe_item); | |
2482 | ||
2483 | /* last unsafe request during umount? */ | |
44ca18f2 | 2484 | if (mdsc->stopping && !__get_oldest_req(mdsc)) |
03066f23 | 2485 | complete_all(&mdsc->safe_umount_waiters); |
2f2dc053 SW |
2486 | mutex_unlock(&mdsc->mutex); |
2487 | goto out; | |
2488 | } | |
e1518c7c | 2489 | } else { |
2f2dc053 SW |
2490 | req->r_got_unsafe = true; |
2491 | list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe); | |
4c06ace8 YZ |
2492 | if (req->r_unsafe_dir) { |
2493 | struct ceph_inode_info *ci = | |
2494 | ceph_inode(req->r_unsafe_dir); | |
2495 | spin_lock(&ci->i_unsafe_lock); | |
2496 | list_add_tail(&req->r_unsafe_dir_item, | |
2497 | &ci->i_unsafe_dirops); | |
2498 | spin_unlock(&ci->i_unsafe_lock); | |
2499 | } | |
2f2dc053 SW |
2500 | } |
2501 | ||
2502 | dout("handle_reply tid %lld result %d\n", tid, result); | |
2503 | rinfo = &req->r_reply_info; | |
14303d20 | 2504 | err = parse_reply_info(msg, rinfo, session->s_con.peer_features); |
2f2dc053 SW |
2505 | mutex_unlock(&mdsc->mutex); |
2506 | ||
2507 | mutex_lock(&session->s_mutex); | |
2508 | if (err < 0) { | |
25933abd | 2509 | pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid); |
9ec7cab1 | 2510 | ceph_msg_dump(msg); |
2f2dc053 SW |
2511 | goto out_err; |
2512 | } | |
2513 | ||
2514 | /* snap trace */ | |
982d6011 | 2515 | realm = NULL; |
2f2dc053 SW |
2516 | if (rinfo->snapblob_len) { |
2517 | down_write(&mdsc->snap_rwsem); | |
2518 | ceph_update_snap_trace(mdsc, rinfo->snapblob, | |
982d6011 YZ |
2519 | rinfo->snapblob + rinfo->snapblob_len, |
2520 | le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP, | |
2521 | &realm); | |
2f2dc053 SW |
2522 | downgrade_write(&mdsc->snap_rwsem); |
2523 | } else { | |
2524 | down_read(&mdsc->snap_rwsem); | |
2525 | } | |
2526 | ||
2527 | /* insert trace into our cache */ | |
b4556396 | 2528 | mutex_lock(&req->r_fill_mutex); |
315f2408 | 2529 | current->journal_info = req; |
3d14c5d2 | 2530 | err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session); |
2f2dc053 | 2531 | if (err == 0) { |
6e8575fa | 2532 | if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR || |
81c6aea5 | 2533 | req->r_op == CEPH_MDS_OP_LSSNAP)) |
2f2dc053 | 2534 | ceph_readdir_prepopulate(req, req->r_session); |
37151668 | 2535 | ceph_unreserve_caps(mdsc, &req->r_caps_reservation); |
2f2dc053 | 2536 | } |
315f2408 | 2537 | current->journal_info = NULL; |
b4556396 | 2538 | mutex_unlock(&req->r_fill_mutex); |
2f2dc053 SW |
2539 | |
2540 | up_read(&mdsc->snap_rwsem); | |
982d6011 YZ |
2541 | if (realm) |
2542 | ceph_put_snap_realm(mdsc, realm); | |
68cd5b4b YZ |
2543 | |
2544 | if (err == 0 && req->r_got_unsafe && req->r_target_inode) { | |
2545 | struct ceph_inode_info *ci = ceph_inode(req->r_target_inode); | |
2546 | spin_lock(&ci->i_unsafe_lock); | |
2547 | list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops); | |
2548 | spin_unlock(&ci->i_unsafe_lock); | |
2549 | } | |
2f2dc053 | 2550 | out_err: |
e1518c7c SW |
2551 | mutex_lock(&mdsc->mutex); |
2552 | if (!req->r_aborted) { | |
2553 | if (err) { | |
2554 | req->r_err = err; | |
2555 | } else { | |
5fdb1389 | 2556 | req->r_reply = ceph_msg_get(msg); |
e1518c7c SW |
2557 | req->r_got_result = true; |
2558 | } | |
2f2dc053 | 2559 | } else { |
e1518c7c | 2560 | dout("reply arrived after request %lld was aborted\n", tid); |
2f2dc053 | 2561 | } |
e1518c7c | 2562 | mutex_unlock(&mdsc->mutex); |
2f2dc053 | 2563 | |
2f2dc053 SW |
2564 | mutex_unlock(&session->s_mutex); |
2565 | ||
2566 | /* kick calling process */ | |
2567 | complete_request(mdsc, req); | |
2568 | out: | |
2569 | ceph_mdsc_put_request(req); | |
2570 | return; | |
2571 | } | |
2572 | ||
2573 | ||
2574 | ||
2575 | /* | |
2576 | * handle mds notification that our request has been forwarded. | |
2577 | */ | |
2600d2dd SW |
2578 | static void handle_forward(struct ceph_mds_client *mdsc, |
2579 | struct ceph_mds_session *session, | |
2580 | struct ceph_msg *msg) | |
2f2dc053 SW |
2581 | { |
2582 | struct ceph_mds_request *req; | |
a1ea787c | 2583 | u64 tid = le64_to_cpu(msg->hdr.tid); |
2f2dc053 SW |
2584 | u32 next_mds; |
2585 | u32 fwd_seq; | |
2f2dc053 SW |
2586 | int err = -EINVAL; |
2587 | void *p = msg->front.iov_base; | |
2588 | void *end = p + msg->front.iov_len; | |
2f2dc053 | 2589 | |
a1ea787c | 2590 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); |
c89136ea SW |
2591 | next_mds = ceph_decode_32(&p); |
2592 | fwd_seq = ceph_decode_32(&p); | |
2f2dc053 SW |
2593 | |
2594 | mutex_lock(&mdsc->mutex); | |
fcd00b68 | 2595 | req = lookup_get_request(mdsc, tid); |
2f2dc053 | 2596 | if (!req) { |
2a8e5e36 | 2597 | dout("forward tid %llu to mds%d - req dne\n", tid, next_mds); |
2f2dc053 SW |
2598 | goto out; /* dup reply? */ |
2599 | } | |
2600 | ||
2a8e5e36 SW |
2601 | if (req->r_aborted) { |
2602 | dout("forward tid %llu aborted, unregistering\n", tid); | |
2603 | __unregister_request(mdsc, req); | |
2604 | } else if (fwd_seq <= req->r_num_fwd) { | |
2605 | dout("forward tid %llu to mds%d - old seq %d <= %d\n", | |
2f2dc053 SW |
2606 | tid, next_mds, req->r_num_fwd, fwd_seq); |
2607 | } else { | |
2608 | /* resend. forward race not possible; mds would drop */ | |
2a8e5e36 SW |
2609 | dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds); |
2610 | BUG_ON(req->r_err); | |
2611 | BUG_ON(req->r_got_result); | |
3de22be6 | 2612 | req->r_attempts = 0; |
2f2dc053 SW |
2613 | req->r_num_fwd = fwd_seq; |
2614 | req->r_resend_mds = next_mds; | |
2615 | put_request_session(req); | |
2616 | __do_request(mdsc, req); | |
2617 | } | |
2618 | ceph_mdsc_put_request(req); | |
2619 | out: | |
2620 | mutex_unlock(&mdsc->mutex); | |
2621 | return; | |
2622 | ||
2623 | bad: | |
2624 | pr_err("mdsc_handle_forward decode error err=%d\n", err); | |
2625 | } | |
2626 | ||
2627 | /* | |
2628 | * handle a mds session control message | |
2629 | */ | |
2630 | static void handle_session(struct ceph_mds_session *session, | |
2631 | struct ceph_msg *msg) | |
2632 | { | |
2633 | struct ceph_mds_client *mdsc = session->s_mdsc; | |
2634 | u32 op; | |
2635 | u64 seq; | |
2600d2dd | 2636 | int mds = session->s_mds; |
2f2dc053 SW |
2637 | struct ceph_mds_session_head *h = msg->front.iov_base; |
2638 | int wake = 0; | |
2639 | ||
2f2dc053 SW |
2640 | /* decode */ |
2641 | if (msg->front.iov_len != sizeof(*h)) | |
2642 | goto bad; | |
2643 | op = le32_to_cpu(h->op); | |
2644 | seq = le64_to_cpu(h->seq); | |
2645 | ||
2646 | mutex_lock(&mdsc->mutex); | |
2600d2dd SW |
2647 | if (op == CEPH_SESSION_CLOSE) |
2648 | __unregister_session(mdsc, session); | |
2f2dc053 SW |
2649 | /* FIXME: this ttl calculation is generous */ |
2650 | session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose; | |
2651 | mutex_unlock(&mdsc->mutex); | |
2652 | ||
2653 | mutex_lock(&session->s_mutex); | |
2654 | ||
2655 | dout("handle_session mds%d %s %p state %s seq %llu\n", | |
2656 | mds, ceph_session_op_name(op), session, | |
a687ecaf | 2657 | ceph_session_state_name(session->s_state), seq); |
2f2dc053 SW |
2658 | |
2659 | if (session->s_state == CEPH_MDS_SESSION_HUNG) { | |
2660 | session->s_state = CEPH_MDS_SESSION_OPEN; | |
2661 | pr_info("mds%d came back\n", session->s_mds); | |
2662 | } | |
2663 | ||
2664 | switch (op) { | |
2665 | case CEPH_SESSION_OPEN: | |
29790f26 SW |
2666 | if (session->s_state == CEPH_MDS_SESSION_RECONNECTING) |
2667 | pr_info("mds%d reconnect success\n", session->s_mds); | |
2f2dc053 SW |
2668 | session->s_state = CEPH_MDS_SESSION_OPEN; |
2669 | renewed_caps(mdsc, session, 0); | |
2670 | wake = 1; | |
2671 | if (mdsc->stopping) | |
2672 | __close_session(mdsc, session); | |
2673 | break; | |
2674 | ||
2675 | case CEPH_SESSION_RENEWCAPS: | |
2676 | if (session->s_renew_seq == seq) | |
2677 | renewed_caps(mdsc, session, 1); | |
2678 | break; | |
2679 | ||
2680 | case CEPH_SESSION_CLOSE: | |
29790f26 SW |
2681 | if (session->s_state == CEPH_MDS_SESSION_RECONNECTING) |
2682 | pr_info("mds%d reconnect denied\n", session->s_mds); | |
1c841a96 | 2683 | cleanup_session_requests(mdsc, session); |
2f2dc053 | 2684 | remove_session_caps(session); |
656e4382 | 2685 | wake = 2; /* for good measure */ |
f3c60c59 | 2686 | wake_up_all(&mdsc->session_close_wq); |
2f2dc053 SW |
2687 | break; |
2688 | ||
2689 | case CEPH_SESSION_STALE: | |
2690 | pr_info("mds%d caps went stale, renewing\n", | |
2691 | session->s_mds); | |
d8fb02ab | 2692 | spin_lock(&session->s_gen_ttl_lock); |
2f2dc053 | 2693 | session->s_cap_gen++; |
1ce208a6 | 2694 | session->s_cap_ttl = jiffies - 1; |
d8fb02ab | 2695 | spin_unlock(&session->s_gen_ttl_lock); |
2f2dc053 SW |
2696 | send_renew_caps(mdsc, session); |
2697 | break; | |
2698 | ||
2699 | case CEPH_SESSION_RECALL_STATE: | |
2700 | trim_caps(mdsc, session, le32_to_cpu(h->max_caps)); | |
2701 | break; | |
2702 | ||
186e4f7a YZ |
2703 | case CEPH_SESSION_FLUSHMSG: |
2704 | send_flushmsg_ack(mdsc, session, seq); | |
2705 | break; | |
2706 | ||
03f4fcb0 YZ |
2707 | case CEPH_SESSION_FORCE_RO: |
2708 | dout("force_session_readonly %p\n", session); | |
2709 | spin_lock(&session->s_cap_lock); | |
2710 | session->s_readonly = true; | |
2711 | spin_unlock(&session->s_cap_lock); | |
2712 | wake_up_session_caps(session, 0); | |
2713 | break; | |
2714 | ||
2f2dc053 SW |
2715 | default: |
2716 | pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds); | |
2717 | WARN_ON(1); | |
2718 | } | |
2719 | ||
2720 | mutex_unlock(&session->s_mutex); | |
2721 | if (wake) { | |
2722 | mutex_lock(&mdsc->mutex); | |
2723 | __wake_requests(mdsc, &session->s_waiting); | |
656e4382 YZ |
2724 | if (wake == 2) |
2725 | kick_requests(mdsc, mds); | |
2f2dc053 SW |
2726 | mutex_unlock(&mdsc->mutex); |
2727 | } | |
2728 | return; | |
2729 | ||
2730 | bad: | |
2731 | pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds, | |
2732 | (int)msg->front.iov_len); | |
9ec7cab1 | 2733 | ceph_msg_dump(msg); |
2f2dc053 SW |
2734 | return; |
2735 | } | |
2736 | ||
2737 | ||
2738 | /* | |
2739 | * called under session->mutex. | |
2740 | */ | |
2741 | static void replay_unsafe_requests(struct ceph_mds_client *mdsc, | |
2742 | struct ceph_mds_session *session) | |
2743 | { | |
2744 | struct ceph_mds_request *req, *nreq; | |
3de22be6 | 2745 | struct rb_node *p; |
2f2dc053 SW |
2746 | int err; |
2747 | ||
2748 | dout("replay_unsafe_requests mds%d\n", session->s_mds); | |
2749 | ||
2750 | mutex_lock(&mdsc->mutex); | |
2751 | list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) { | |
6e6f0923 | 2752 | err = __prepare_send_request(mdsc, req, session->s_mds, true); |
2f2dc053 SW |
2753 | if (!err) { |
2754 | ceph_msg_get(req->r_request); | |
2755 | ceph_con_send(&session->s_con, req->r_request); | |
2756 | } | |
2757 | } | |
3de22be6 YZ |
2758 | |
2759 | /* | |
2760 | * also re-send old requests when MDS enters reconnect stage. So that MDS | |
2761 | * can process completed request in clientreplay stage. | |
2762 | */ | |
2763 | p = rb_first(&mdsc->request_tree); | |
2764 | while (p) { | |
2765 | req = rb_entry(p, struct ceph_mds_request, r_node); | |
2766 | p = rb_next(p); | |
2767 | if (req->r_got_unsafe) | |
2768 | continue; | |
2769 | if (req->r_attempts == 0) | |
2770 | continue; /* only old requests */ | |
2771 | if (req->r_session && | |
2772 | req->r_session->s_mds == session->s_mds) { | |
6e6f0923 YZ |
2773 | err = __prepare_send_request(mdsc, req, |
2774 | session->s_mds, true); | |
3de22be6 YZ |
2775 | if (!err) { |
2776 | ceph_msg_get(req->r_request); | |
2777 | ceph_con_send(&session->s_con, req->r_request); | |
2778 | } | |
2779 | } | |
2780 | } | |
2f2dc053 SW |
2781 | mutex_unlock(&mdsc->mutex); |
2782 | } | |
2783 | ||
2784 | /* | |
2785 | * Encode information about a cap for a reconnect with the MDS. | |
2786 | */ | |
2f2dc053 SW |
2787 | static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap, |
2788 | void *arg) | |
2789 | { | |
20cb34ae SW |
2790 | union { |
2791 | struct ceph_mds_cap_reconnect v2; | |
2792 | struct ceph_mds_cap_reconnect_v1 v1; | |
2793 | } rec; | |
2794 | size_t reclen; | |
2f2dc053 | 2795 | struct ceph_inode_info *ci; |
20cb34ae SW |
2796 | struct ceph_reconnect_state *recon_state = arg; |
2797 | struct ceph_pagelist *pagelist = recon_state->pagelist; | |
2f2dc053 SW |
2798 | char *path; |
2799 | int pathlen, err; | |
2800 | u64 pathbase; | |
2801 | struct dentry *dentry; | |
2802 | ||
2803 | ci = cap->ci; | |
2804 | ||
2805 | dout(" adding %p ino %llx.%llx cap %p %lld %s\n", | |
2806 | inode, ceph_vinop(inode), cap, cap->cap_id, | |
2807 | ceph_cap_string(cap->issued)); | |
93cea5be SW |
2808 | err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode)); |
2809 | if (err) | |
2810 | return err; | |
2f2dc053 SW |
2811 | |
2812 | dentry = d_find_alias(inode); | |
2813 | if (dentry) { | |
2814 | path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0); | |
2815 | if (IS_ERR(path)) { | |
2816 | err = PTR_ERR(path); | |
e072f8aa | 2817 | goto out_dput; |
2f2dc053 SW |
2818 | } |
2819 | } else { | |
2820 | path = NULL; | |
2821 | pathlen = 0; | |
2822 | } | |
93cea5be SW |
2823 | err = ceph_pagelist_encode_string(pagelist, path, pathlen); |
2824 | if (err) | |
e072f8aa | 2825 | goto out_free; |
2f2dc053 | 2826 | |
be655596 | 2827 | spin_lock(&ci->i_ceph_lock); |
2f2dc053 SW |
2828 | cap->seq = 0; /* reset cap seq */ |
2829 | cap->issue_seq = 0; /* and issue_seq */ | |
667ca05c | 2830 | cap->mseq = 0; /* and migrate_seq */ |
99a9c273 | 2831 | cap->cap_gen = cap->session->s_cap_gen; |
20cb34ae SW |
2832 | |
2833 | if (recon_state->flock) { | |
2834 | rec.v2.cap_id = cpu_to_le64(cap->cap_id); | |
2835 | rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci)); | |
2836 | rec.v2.issued = cpu_to_le32(cap->issued); | |
2837 | rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino); | |
2838 | rec.v2.pathbase = cpu_to_le64(pathbase); | |
2839 | rec.v2.flock_len = 0; | |
2840 | reclen = sizeof(rec.v2); | |
2841 | } else { | |
2842 | rec.v1.cap_id = cpu_to_le64(cap->cap_id); | |
2843 | rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci)); | |
2844 | rec.v1.issued = cpu_to_le32(cap->issued); | |
2845 | rec.v1.size = cpu_to_le64(inode->i_size); | |
2846 | ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime); | |
2847 | ceph_encode_timespec(&rec.v1.atime, &inode->i_atime); | |
2848 | rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino); | |
2849 | rec.v1.pathbase = cpu_to_le64(pathbase); | |
2850 | reclen = sizeof(rec.v1); | |
2851 | } | |
be655596 | 2852 | spin_unlock(&ci->i_ceph_lock); |
2f2dc053 | 2853 | |
40819f6f GF |
2854 | if (recon_state->flock) { |
2855 | int num_fcntl_locks, num_flock_locks; | |
39be95e9 JS |
2856 | struct ceph_filelock *flocks; |
2857 | ||
2858 | encode_again: | |
39be95e9 | 2859 | ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks); |
39be95e9 JS |
2860 | flocks = kmalloc((num_fcntl_locks+num_flock_locks) * |
2861 | sizeof(struct ceph_filelock), GFP_NOFS); | |
2862 | if (!flocks) { | |
2863 | err = -ENOMEM; | |
2864 | goto out_free; | |
2865 | } | |
39be95e9 JS |
2866 | err = ceph_encode_locks_to_buffer(inode, flocks, |
2867 | num_fcntl_locks, | |
2868 | num_flock_locks); | |
39be95e9 JS |
2869 | if (err) { |
2870 | kfree(flocks); | |
2871 | if (err == -ENOSPC) | |
2872 | goto encode_again; | |
2873 | goto out_free; | |
2874 | } | |
2875 | /* | |
2876 | * number of encoded locks is stable, so copy to pagelist | |
2877 | */ | |
2878 | rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) + | |
2879 | (num_fcntl_locks+num_flock_locks) * | |
2880 | sizeof(struct ceph_filelock)); | |
2881 | err = ceph_pagelist_append(pagelist, &rec, reclen); | |
2882 | if (!err) | |
2883 | err = ceph_locks_to_pagelist(flocks, pagelist, | |
2884 | num_fcntl_locks, | |
2885 | num_flock_locks); | |
2886 | kfree(flocks); | |
3612abbd SW |
2887 | } else { |
2888 | err = ceph_pagelist_append(pagelist, &rec, reclen); | |
40819f6f | 2889 | } |
44c99757 YZ |
2890 | |
2891 | recon_state->nr_caps++; | |
e072f8aa | 2892 | out_free: |
2f2dc053 | 2893 | kfree(path); |
e072f8aa | 2894 | out_dput: |
2f2dc053 | 2895 | dput(dentry); |
93cea5be | 2896 | return err; |
2f2dc053 SW |
2897 | } |
2898 | ||
2899 | ||
2900 | /* | |
2901 | * If an MDS fails and recovers, clients need to reconnect in order to | |
2902 | * reestablish shared state. This includes all caps issued through | |
2903 | * this session _and_ the snap_realm hierarchy. Because it's not | |
2904 | * clear which snap realms the mds cares about, we send everything we | |
2905 | * know about.. that ensures we'll then get any new info the | |
2906 | * recovering MDS might have. | |
2907 | * | |
2908 | * This is a relatively heavyweight operation, but it's rare. | |
2909 | * | |
2910 | * called with mdsc->mutex held. | |
2911 | */ | |
34b6c855 SW |
2912 | static void send_mds_reconnect(struct ceph_mds_client *mdsc, |
2913 | struct ceph_mds_session *session) | |
2f2dc053 | 2914 | { |
2f2dc053 | 2915 | struct ceph_msg *reply; |
a105f00c | 2916 | struct rb_node *p; |
34b6c855 | 2917 | int mds = session->s_mds; |
9abf82b8 | 2918 | int err = -ENOMEM; |
44c99757 | 2919 | int s_nr_caps; |
93cea5be | 2920 | struct ceph_pagelist *pagelist; |
20cb34ae | 2921 | struct ceph_reconnect_state recon_state; |
2f2dc053 | 2922 | |
34b6c855 | 2923 | pr_info("mds%d reconnect start\n", mds); |
2f2dc053 | 2924 | |
93cea5be SW |
2925 | pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS); |
2926 | if (!pagelist) | |
2927 | goto fail_nopagelist; | |
2928 | ceph_pagelist_init(pagelist); | |
2929 | ||
b61c2763 | 2930 | reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false); |
a79832f2 | 2931 | if (!reply) |
93cea5be | 2932 | goto fail_nomsg; |
93cea5be | 2933 | |
34b6c855 SW |
2934 | mutex_lock(&session->s_mutex); |
2935 | session->s_state = CEPH_MDS_SESSION_RECONNECTING; | |
2936 | session->s_seq = 0; | |
2f2dc053 | 2937 | |
2f2dc053 | 2938 | dout("session %p state %s\n", session, |
a687ecaf | 2939 | ceph_session_state_name(session->s_state)); |
2f2dc053 | 2940 | |
99a9c273 YZ |
2941 | spin_lock(&session->s_gen_ttl_lock); |
2942 | session->s_cap_gen++; | |
2943 | spin_unlock(&session->s_gen_ttl_lock); | |
2944 | ||
2945 | spin_lock(&session->s_cap_lock); | |
03f4fcb0 YZ |
2946 | /* don't know if session is readonly */ |
2947 | session->s_readonly = 0; | |
99a9c273 YZ |
2948 | /* |
2949 | * notify __ceph_remove_cap() that we are composing cap reconnect. | |
2950 | * If a cap get released before being added to the cap reconnect, | |
2951 | * __ceph_remove_cap() should skip queuing cap release. | |
2952 | */ | |
2953 | session->s_cap_reconnect = 1; | |
e01a5946 | 2954 | /* drop old cap expires; we're about to reestablish that state */ |
745a8e3b | 2955 | cleanup_cap_releases(mdsc, session); |
e01a5946 | 2956 | |
5d23371f | 2957 | /* trim unused caps to reduce MDS's cache rejoin time */ |
c0bd50e2 YZ |
2958 | if (mdsc->fsc->sb->s_root) |
2959 | shrink_dcache_parent(mdsc->fsc->sb->s_root); | |
5d23371f YZ |
2960 | |
2961 | ceph_con_close(&session->s_con); | |
2962 | ceph_con_open(&session->s_con, | |
2963 | CEPH_ENTITY_TYPE_MDS, mds, | |
2964 | ceph_mdsmap_get_addr(mdsc->mdsmap, mds)); | |
2965 | ||
2966 | /* replay unsafe requests */ | |
2967 | replay_unsafe_requests(mdsc, session); | |
2968 | ||
2969 | down_read(&mdsc->snap_rwsem); | |
2970 | ||
2f2dc053 | 2971 | /* traverse this session's caps */ |
44c99757 YZ |
2972 | s_nr_caps = session->s_nr_caps; |
2973 | err = ceph_pagelist_encode_32(pagelist, s_nr_caps); | |
93cea5be SW |
2974 | if (err) |
2975 | goto fail; | |
20cb34ae | 2976 | |
44c99757 | 2977 | recon_state.nr_caps = 0; |
20cb34ae SW |
2978 | recon_state.pagelist = pagelist; |
2979 | recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK; | |
2980 | err = iterate_session_caps(session, encode_caps_cb, &recon_state); | |
2f2dc053 | 2981 | if (err < 0) |
9abf82b8 | 2982 | goto fail; |
2f2dc053 | 2983 | |
99a9c273 YZ |
2984 | spin_lock(&session->s_cap_lock); |
2985 | session->s_cap_reconnect = 0; | |
2986 | spin_unlock(&session->s_cap_lock); | |
2987 | ||
2f2dc053 SW |
2988 | /* |
2989 | * snaprealms. we provide mds with the ino, seq (version), and | |
2990 | * parent for all of our realms. If the mds has any newer info, | |
2991 | * it will tell us. | |
2992 | */ | |
a105f00c SW |
2993 | for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) { |
2994 | struct ceph_snap_realm *realm = | |
2995 | rb_entry(p, struct ceph_snap_realm, node); | |
93cea5be | 2996 | struct ceph_mds_snaprealm_reconnect sr_rec; |
2f2dc053 SW |
2997 | |
2998 | dout(" adding snap realm %llx seq %lld parent %llx\n", | |
2999 | realm->ino, realm->seq, realm->parent_ino); | |
93cea5be SW |
3000 | sr_rec.ino = cpu_to_le64(realm->ino); |
3001 | sr_rec.seq = cpu_to_le64(realm->seq); | |
3002 | sr_rec.parent = cpu_to_le64(realm->parent_ino); | |
3003 | err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec)); | |
3004 | if (err) | |
3005 | goto fail; | |
2f2dc053 | 3006 | } |
2f2dc053 | 3007 | |
20cb34ae SW |
3008 | if (recon_state.flock) |
3009 | reply->hdr.version = cpu_to_le16(2); | |
44c99757 YZ |
3010 | |
3011 | /* raced with cap release? */ | |
3012 | if (s_nr_caps != recon_state.nr_caps) { | |
3013 | struct page *page = list_first_entry(&pagelist->head, | |
3014 | struct page, lru); | |
3015 | __le32 *addr = kmap_atomic(page); | |
3016 | *addr = cpu_to_le32(recon_state.nr_caps); | |
3017 | kunmap_atomic(addr); | |
ebf18f47 | 3018 | } |
44c99757 YZ |
3019 | |
3020 | reply->hdr.data_len = cpu_to_le32(pagelist->length); | |
3021 | ceph_msg_data_add_pagelist(reply, pagelist); | |
e548e9b9 YZ |
3022 | |
3023 | ceph_early_kick_flushing_caps(mdsc, session); | |
3024 | ||
2f2dc053 SW |
3025 | ceph_con_send(&session->s_con, reply); |
3026 | ||
9abf82b8 SW |
3027 | mutex_unlock(&session->s_mutex); |
3028 | ||
3029 | mutex_lock(&mdsc->mutex); | |
3030 | __wake_requests(mdsc, &session->s_waiting); | |
3031 | mutex_unlock(&mdsc->mutex); | |
3032 | ||
2f2dc053 | 3033 | up_read(&mdsc->snap_rwsem); |
2f2dc053 SW |
3034 | return; |
3035 | ||
93cea5be | 3036 | fail: |
2f2dc053 | 3037 | ceph_msg_put(reply); |
9abf82b8 SW |
3038 | up_read(&mdsc->snap_rwsem); |
3039 | mutex_unlock(&session->s_mutex); | |
93cea5be SW |
3040 | fail_nomsg: |
3041 | ceph_pagelist_release(pagelist); | |
93cea5be | 3042 | fail_nopagelist: |
9abf82b8 | 3043 | pr_err("error %d preparing reconnect for mds%d\n", err, mds); |
9abf82b8 | 3044 | return; |
2f2dc053 SW |
3045 | } |
3046 | ||
3047 | ||
3048 | /* | |
3049 | * compare old and new mdsmaps, kicking requests | |
3050 | * and closing out old connections as necessary | |
3051 | * | |
3052 | * called under mdsc->mutex. | |
3053 | */ | |
3054 | static void check_new_map(struct ceph_mds_client *mdsc, | |
3055 | struct ceph_mdsmap *newmap, | |
3056 | struct ceph_mdsmap *oldmap) | |
3057 | { | |
3058 | int i; | |
3059 | int oldstate, newstate; | |
3060 | struct ceph_mds_session *s; | |
3061 | ||
3062 | dout("check_new_map new %u old %u\n", | |
3063 | newmap->m_epoch, oldmap->m_epoch); | |
3064 | ||
3065 | for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) { | |
3066 | if (mdsc->sessions[i] == NULL) | |
3067 | continue; | |
3068 | s = mdsc->sessions[i]; | |
3069 | oldstate = ceph_mdsmap_get_state(oldmap, i); | |
3070 | newstate = ceph_mdsmap_get_state(newmap, i); | |
3071 | ||
0deb01c9 | 3072 | dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n", |
2f2dc053 | 3073 | i, ceph_mds_state_name(oldstate), |
0deb01c9 | 3074 | ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "", |
2f2dc053 | 3075 | ceph_mds_state_name(newstate), |
0deb01c9 | 3076 | ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "", |
a687ecaf | 3077 | ceph_session_state_name(s->s_state)); |
2f2dc053 | 3078 | |
3e8f43a0 YZ |
3079 | if (i >= newmap->m_max_mds || |
3080 | memcmp(ceph_mdsmap_get_addr(oldmap, i), | |
2f2dc053 SW |
3081 | ceph_mdsmap_get_addr(newmap, i), |
3082 | sizeof(struct ceph_entity_addr))) { | |
3083 | if (s->s_state == CEPH_MDS_SESSION_OPENING) { | |
3084 | /* the session never opened, just close it | |
3085 | * out now */ | |
3086 | __wake_requests(mdsc, &s->s_waiting); | |
2600d2dd | 3087 | __unregister_session(mdsc, s); |
2f2dc053 SW |
3088 | } else { |
3089 | /* just close it */ | |
3090 | mutex_unlock(&mdsc->mutex); | |
3091 | mutex_lock(&s->s_mutex); | |
3092 | mutex_lock(&mdsc->mutex); | |
3093 | ceph_con_close(&s->s_con); | |
3094 | mutex_unlock(&s->s_mutex); | |
3095 | s->s_state = CEPH_MDS_SESSION_RESTARTING; | |
3096 | } | |
2f2dc053 SW |
3097 | } else if (oldstate == newstate) { |
3098 | continue; /* nothing new with this mds */ | |
3099 | } | |
3100 | ||
3101 | /* | |
3102 | * send reconnect? | |
3103 | */ | |
3104 | if (s->s_state == CEPH_MDS_SESSION_RESTARTING && | |
34b6c855 SW |
3105 | newstate >= CEPH_MDS_STATE_RECONNECT) { |
3106 | mutex_unlock(&mdsc->mutex); | |
3107 | send_mds_reconnect(mdsc, s); | |
3108 | mutex_lock(&mdsc->mutex); | |
3109 | } | |
2f2dc053 SW |
3110 | |
3111 | /* | |
29790f26 | 3112 | * kick request on any mds that has gone active. |
2f2dc053 SW |
3113 | */ |
3114 | if (oldstate < CEPH_MDS_STATE_ACTIVE && | |
3115 | newstate >= CEPH_MDS_STATE_ACTIVE) { | |
29790f26 SW |
3116 | if (oldstate != CEPH_MDS_STATE_CREATING && |
3117 | oldstate != CEPH_MDS_STATE_STARTING) | |
3118 | pr_info("mds%d recovery completed\n", s->s_mds); | |
3119 | kick_requests(mdsc, i); | |
2f2dc053 | 3120 | ceph_kick_flushing_caps(mdsc, s); |
0dc2570f | 3121 | wake_up_session_caps(s, 1); |
2f2dc053 SW |
3122 | } |
3123 | } | |
cb170a22 SW |
3124 | |
3125 | for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) { | |
3126 | s = mdsc->sessions[i]; | |
3127 | if (!s) | |
3128 | continue; | |
3129 | if (!ceph_mdsmap_is_laggy(newmap, i)) | |
3130 | continue; | |
3131 | if (s->s_state == CEPH_MDS_SESSION_OPEN || | |
3132 | s->s_state == CEPH_MDS_SESSION_HUNG || | |
3133 | s->s_state == CEPH_MDS_SESSION_CLOSING) { | |
3134 | dout(" connecting to export targets of laggy mds%d\n", | |
3135 | i); | |
3136 | __open_export_target_sessions(mdsc, s); | |
3137 | } | |
3138 | } | |
2f2dc053 SW |
3139 | } |
3140 | ||
3141 | ||
3142 | ||
3143 | /* | |
3144 | * leases | |
3145 | */ | |
3146 | ||
3147 | /* | |
3148 | * caller must hold session s_mutex, dentry->d_lock | |
3149 | */ | |
3150 | void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry) | |
3151 | { | |
3152 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
3153 | ||
3154 | ceph_put_mds_session(di->lease_session); | |
3155 | di->lease_session = NULL; | |
3156 | } | |
3157 | ||
2600d2dd SW |
3158 | static void handle_lease(struct ceph_mds_client *mdsc, |
3159 | struct ceph_mds_session *session, | |
3160 | struct ceph_msg *msg) | |
2f2dc053 | 3161 | { |
3d14c5d2 | 3162 | struct super_block *sb = mdsc->fsc->sb; |
2f2dc053 | 3163 | struct inode *inode; |
2f2dc053 SW |
3164 | struct dentry *parent, *dentry; |
3165 | struct ceph_dentry_info *di; | |
2600d2dd | 3166 | int mds = session->s_mds; |
2f2dc053 | 3167 | struct ceph_mds_lease *h = msg->front.iov_base; |
1e5ea23d | 3168 | u32 seq; |
2f2dc053 | 3169 | struct ceph_vino vino; |
2f2dc053 SW |
3170 | struct qstr dname; |
3171 | int release = 0; | |
3172 | ||
2f2dc053 SW |
3173 | dout("handle_lease from mds%d\n", mds); |
3174 | ||
3175 | /* decode */ | |
3176 | if (msg->front.iov_len < sizeof(*h) + sizeof(u32)) | |
3177 | goto bad; | |
3178 | vino.ino = le64_to_cpu(h->ino); | |
3179 | vino.snap = CEPH_NOSNAP; | |
1e5ea23d | 3180 | seq = le32_to_cpu(h->seq); |
2f2dc053 SW |
3181 | dname.name = (void *)h + sizeof(*h) + sizeof(u32); |
3182 | dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32); | |
3183 | if (dname.len != get_unaligned_le32(h+1)) | |
3184 | goto bad; | |
3185 | ||
2f2dc053 SW |
3186 | /* lookup inode */ |
3187 | inode = ceph_find_inode(sb, vino); | |
2f90b852 SW |
3188 | dout("handle_lease %s, ino %llx %p %.*s\n", |
3189 | ceph_lease_op_name(h->action), vino.ino, inode, | |
1e5ea23d | 3190 | dname.len, dname.name); |
6cd3bcad YZ |
3191 | |
3192 | mutex_lock(&session->s_mutex); | |
3193 | session->s_seq++; | |
3194 | ||
2f2dc053 SW |
3195 | if (inode == NULL) { |
3196 | dout("handle_lease no inode %llx\n", vino.ino); | |
3197 | goto release; | |
3198 | } | |
2f2dc053 SW |
3199 | |
3200 | /* dentry */ | |
3201 | parent = d_find_alias(inode); | |
3202 | if (!parent) { | |
3203 | dout("no parent dentry on inode %p\n", inode); | |
3204 | WARN_ON(1); | |
3205 | goto release; /* hrm... */ | |
3206 | } | |
3207 | dname.hash = full_name_hash(dname.name, dname.len); | |
3208 | dentry = d_lookup(parent, &dname); | |
3209 | dput(parent); | |
3210 | if (!dentry) | |
3211 | goto release; | |
3212 | ||
3213 | spin_lock(&dentry->d_lock); | |
3214 | di = ceph_dentry(dentry); | |
3215 | switch (h->action) { | |
3216 | case CEPH_MDS_LEASE_REVOKE: | |
3d8eb7a9 | 3217 | if (di->lease_session == session) { |
1e5ea23d SW |
3218 | if (ceph_seq_cmp(di->lease_seq, seq) > 0) |
3219 | h->seq = cpu_to_le32(di->lease_seq); | |
2f2dc053 SW |
3220 | __ceph_mdsc_drop_dentry_lease(dentry); |
3221 | } | |
3222 | release = 1; | |
3223 | break; | |
3224 | ||
3225 | case CEPH_MDS_LEASE_RENEW: | |
3d8eb7a9 | 3226 | if (di->lease_session == session && |
2f2dc053 SW |
3227 | di->lease_gen == session->s_cap_gen && |
3228 | di->lease_renew_from && | |
3229 | di->lease_renew_after == 0) { | |
3230 | unsigned long duration = | |
3563dbdd | 3231 | msecs_to_jiffies(le32_to_cpu(h->duration_ms)); |
2f2dc053 | 3232 | |
1e5ea23d | 3233 | di->lease_seq = seq; |
2f2dc053 SW |
3234 | dentry->d_time = di->lease_renew_from + duration; |
3235 | di->lease_renew_after = di->lease_renew_from + | |
3236 | (duration >> 1); | |
3237 | di->lease_renew_from = 0; | |
3238 | } | |
3239 | break; | |
3240 | } | |
3241 | spin_unlock(&dentry->d_lock); | |
3242 | dput(dentry); | |
3243 | ||
3244 | if (!release) | |
3245 | goto out; | |
3246 | ||
3247 | release: | |
3248 | /* let's just reuse the same message */ | |
3249 | h->action = CEPH_MDS_LEASE_REVOKE_ACK; | |
3250 | ceph_msg_get(msg); | |
3251 | ceph_con_send(&session->s_con, msg); | |
3252 | ||
3253 | out: | |
3254 | iput(inode); | |
3255 | mutex_unlock(&session->s_mutex); | |
2f2dc053 SW |
3256 | return; |
3257 | ||
3258 | bad: | |
3259 | pr_err("corrupt lease message\n"); | |
9ec7cab1 | 3260 | ceph_msg_dump(msg); |
2f2dc053 SW |
3261 | } |
3262 | ||
3263 | void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session, | |
3264 | struct inode *inode, | |
3265 | struct dentry *dentry, char action, | |
3266 | u32 seq) | |
3267 | { | |
3268 | struct ceph_msg *msg; | |
3269 | struct ceph_mds_lease *lease; | |
3270 | int len = sizeof(*lease) + sizeof(u32); | |
3271 | int dnamelen = 0; | |
3272 | ||
3273 | dout("lease_send_msg inode %p dentry %p %s to mds%d\n", | |
3274 | inode, dentry, ceph_lease_op_name(action), session->s_mds); | |
3275 | dnamelen = dentry->d_name.len; | |
3276 | len += dnamelen; | |
3277 | ||
b61c2763 | 3278 | msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false); |
a79832f2 | 3279 | if (!msg) |
2f2dc053 SW |
3280 | return; |
3281 | lease = msg->front.iov_base; | |
3282 | lease->action = action; | |
2f2dc053 SW |
3283 | lease->ino = cpu_to_le64(ceph_vino(inode).ino); |
3284 | lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap); | |
3285 | lease->seq = cpu_to_le32(seq); | |
3286 | put_unaligned_le32(dnamelen, lease + 1); | |
3287 | memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen); | |
3288 | ||
3289 | /* | |
3290 | * if this is a preemptive lease RELEASE, no need to | |
3291 | * flush request stream, since the actual request will | |
3292 | * soon follow. | |
3293 | */ | |
3294 | msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE); | |
3295 | ||
3296 | ceph_con_send(&session->s_con, msg); | |
3297 | } | |
3298 | ||
3299 | /* | |
3300 | * Preemptively release a lease we expect to invalidate anyway. | |
3301 | * Pass @inode always, @dentry is optional. | |
3302 | */ | |
3303 | void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode, | |
2f90b852 | 3304 | struct dentry *dentry) |
2f2dc053 SW |
3305 | { |
3306 | struct ceph_dentry_info *di; | |
3307 | struct ceph_mds_session *session; | |
3308 | u32 seq; | |
3309 | ||
3310 | BUG_ON(inode == NULL); | |
3311 | BUG_ON(dentry == NULL); | |
2f2dc053 SW |
3312 | |
3313 | /* is dentry lease valid? */ | |
3314 | spin_lock(&dentry->d_lock); | |
3315 | di = ceph_dentry(dentry); | |
3316 | if (!di || !di->lease_session || | |
3317 | di->lease_session->s_mds < 0 || | |
3318 | di->lease_gen != di->lease_session->s_cap_gen || | |
3319 | !time_before(jiffies, dentry->d_time)) { | |
3320 | dout("lease_release inode %p dentry %p -- " | |
2f90b852 SW |
3321 | "no lease\n", |
3322 | inode, dentry); | |
2f2dc053 SW |
3323 | spin_unlock(&dentry->d_lock); |
3324 | return; | |
3325 | } | |
3326 | ||
3327 | /* we do have a lease on this dentry; note mds and seq */ | |
3328 | session = ceph_get_mds_session(di->lease_session); | |
3329 | seq = di->lease_seq; | |
3330 | __ceph_mdsc_drop_dentry_lease(dentry); | |
3331 | spin_unlock(&dentry->d_lock); | |
3332 | ||
2f90b852 SW |
3333 | dout("lease_release inode %p dentry %p to mds%d\n", |
3334 | inode, dentry, session->s_mds); | |
2f2dc053 SW |
3335 | ceph_mdsc_lease_send_msg(session, inode, dentry, |
3336 | CEPH_MDS_LEASE_RELEASE, seq); | |
3337 | ceph_put_mds_session(session); | |
3338 | } | |
3339 | ||
3340 | /* | |
3341 | * drop all leases (and dentry refs) in preparation for umount | |
3342 | */ | |
3343 | static void drop_leases(struct ceph_mds_client *mdsc) | |
3344 | { | |
3345 | int i; | |
3346 | ||
3347 | dout("drop_leases\n"); | |
3348 | mutex_lock(&mdsc->mutex); | |
3349 | for (i = 0; i < mdsc->max_sessions; i++) { | |
3350 | struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i); | |
3351 | if (!s) | |
3352 | continue; | |
3353 | mutex_unlock(&mdsc->mutex); | |
3354 | mutex_lock(&s->s_mutex); | |
3355 | mutex_unlock(&s->s_mutex); | |
3356 | ceph_put_mds_session(s); | |
3357 | mutex_lock(&mdsc->mutex); | |
3358 | } | |
3359 | mutex_unlock(&mdsc->mutex); | |
3360 | } | |
3361 | ||
3362 | ||
3363 | ||
3364 | /* | |
3365 | * delayed work -- periodically trim expired leases, renew caps with mds | |
3366 | */ | |
3367 | static void schedule_delayed(struct ceph_mds_client *mdsc) | |
3368 | { | |
3369 | int delay = 5; | |
3370 | unsigned hz = round_jiffies_relative(HZ * delay); | |
3371 | schedule_delayed_work(&mdsc->delayed_work, hz); | |
3372 | } | |
3373 | ||
3374 | static void delayed_work(struct work_struct *work) | |
3375 | { | |
3376 | int i; | |
3377 | struct ceph_mds_client *mdsc = | |
3378 | container_of(work, struct ceph_mds_client, delayed_work.work); | |
3379 | int renew_interval; | |
3380 | int renew_caps; | |
3381 | ||
3382 | dout("mdsc delayed_work\n"); | |
afcdaea3 | 3383 | ceph_check_delayed_caps(mdsc); |
2f2dc053 SW |
3384 | |
3385 | mutex_lock(&mdsc->mutex); | |
3386 | renew_interval = mdsc->mdsmap->m_session_timeout >> 2; | |
3387 | renew_caps = time_after_eq(jiffies, HZ*renew_interval + | |
3388 | mdsc->last_renew_caps); | |
3389 | if (renew_caps) | |
3390 | mdsc->last_renew_caps = jiffies; | |
3391 | ||
3392 | for (i = 0; i < mdsc->max_sessions; i++) { | |
3393 | struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i); | |
3394 | if (s == NULL) | |
3395 | continue; | |
3396 | if (s->s_state == CEPH_MDS_SESSION_CLOSING) { | |
3397 | dout("resending session close request for mds%d\n", | |
3398 | s->s_mds); | |
3399 | request_close_session(mdsc, s); | |
3400 | ceph_put_mds_session(s); | |
3401 | continue; | |
3402 | } | |
3403 | if (s->s_ttl && time_after(jiffies, s->s_ttl)) { | |
3404 | if (s->s_state == CEPH_MDS_SESSION_OPEN) { | |
3405 | s->s_state = CEPH_MDS_SESSION_HUNG; | |
3406 | pr_info("mds%d hung\n", s->s_mds); | |
3407 | } | |
3408 | } | |
3409 | if (s->s_state < CEPH_MDS_SESSION_OPEN) { | |
3410 | /* this mds is failed or recovering, just wait */ | |
3411 | ceph_put_mds_session(s); | |
3412 | continue; | |
3413 | } | |
3414 | mutex_unlock(&mdsc->mutex); | |
3415 | ||
3416 | mutex_lock(&s->s_mutex); | |
3417 | if (renew_caps) | |
3418 | send_renew_caps(mdsc, s); | |
3419 | else | |
3420 | ceph_con_keepalive(&s->s_con); | |
aab53dd9 SW |
3421 | if (s->s_state == CEPH_MDS_SESSION_OPEN || |
3422 | s->s_state == CEPH_MDS_SESSION_HUNG) | |
3d7ded4d | 3423 | ceph_send_cap_releases(mdsc, s); |
2f2dc053 SW |
3424 | mutex_unlock(&s->s_mutex); |
3425 | ceph_put_mds_session(s); | |
3426 | ||
3427 | mutex_lock(&mdsc->mutex); | |
3428 | } | |
3429 | mutex_unlock(&mdsc->mutex); | |
3430 | ||
3431 | schedule_delayed(mdsc); | |
3432 | } | |
3433 | ||
3d14c5d2 | 3434 | int ceph_mdsc_init(struct ceph_fs_client *fsc) |
2f2dc053 | 3435 | |
2f2dc053 | 3436 | { |
3d14c5d2 YS |
3437 | struct ceph_mds_client *mdsc; |
3438 | ||
3439 | mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS); | |
3440 | if (!mdsc) | |
3441 | return -ENOMEM; | |
3442 | mdsc->fsc = fsc; | |
3443 | fsc->mdsc = mdsc; | |
2f2dc053 SW |
3444 | mutex_init(&mdsc->mutex); |
3445 | mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS); | |
fb3101b6 | 3446 | if (mdsc->mdsmap == NULL) { |
3447 | kfree(mdsc); | |
2d06eeb8 | 3448 | return -ENOMEM; |
fb3101b6 | 3449 | } |
2d06eeb8 | 3450 | |
2f2dc053 | 3451 | init_completion(&mdsc->safe_umount_waiters); |
f3c60c59 | 3452 | init_waitqueue_head(&mdsc->session_close_wq); |
2f2dc053 SW |
3453 | INIT_LIST_HEAD(&mdsc->waiting_for_map); |
3454 | mdsc->sessions = NULL; | |
86d8f67b | 3455 | atomic_set(&mdsc->num_sessions, 0); |
2f2dc053 SW |
3456 | mdsc->max_sessions = 0; |
3457 | mdsc->stopping = 0; | |
affbc19a | 3458 | mdsc->last_snap_seq = 0; |
2f2dc053 | 3459 | init_rwsem(&mdsc->snap_rwsem); |
a105f00c | 3460 | mdsc->snap_realms = RB_ROOT; |
2f2dc053 SW |
3461 | INIT_LIST_HEAD(&mdsc->snap_empty); |
3462 | spin_lock_init(&mdsc->snap_empty_lock); | |
3463 | mdsc->last_tid = 0; | |
e8a7b8b1 | 3464 | mdsc->oldest_tid = 0; |
44ca18f2 | 3465 | mdsc->request_tree = RB_ROOT; |
2f2dc053 SW |
3466 | INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work); |
3467 | mdsc->last_renew_caps = jiffies; | |
3468 | INIT_LIST_HEAD(&mdsc->cap_delay_list); | |
3469 | spin_lock_init(&mdsc->cap_delay_lock); | |
3470 | INIT_LIST_HEAD(&mdsc->snap_flush_list); | |
3471 | spin_lock_init(&mdsc->snap_flush_lock); | |
553adfd9 | 3472 | mdsc->last_cap_flush_tid = 1; |
8310b089 | 3473 | mdsc->cap_flush_tree = RB_ROOT; |
2f2dc053 | 3474 | INIT_LIST_HEAD(&mdsc->cap_dirty); |
db354052 | 3475 | INIT_LIST_HEAD(&mdsc->cap_dirty_migrating); |
2f2dc053 SW |
3476 | mdsc->num_cap_flushing = 0; |
3477 | spin_lock_init(&mdsc->cap_dirty_lock); | |
3478 | init_waitqueue_head(&mdsc->cap_flushing_wq); | |
3479 | spin_lock_init(&mdsc->dentry_lru_lock); | |
3480 | INIT_LIST_HEAD(&mdsc->dentry_lru); | |
2d06eeb8 | 3481 | |
37151668 | 3482 | ceph_caps_init(mdsc); |
3d14c5d2 | 3483 | ceph_adjust_min_caps(mdsc, fsc->min_caps); |
37151668 | 3484 | |
10183a69 YZ |
3485 | init_rwsem(&mdsc->pool_perm_rwsem); |
3486 | mdsc->pool_perm_tree = RB_ROOT; | |
3487 | ||
5f44f142 | 3488 | return 0; |
2f2dc053 SW |
3489 | } |
3490 | ||
3491 | /* | |
3492 | * Wait for safe replies on open mds requests. If we time out, drop | |
3493 | * all requests from the tree to avoid dangling dentry refs. | |
3494 | */ | |
3495 | static void wait_requests(struct ceph_mds_client *mdsc) | |
3496 | { | |
a319bf56 | 3497 | struct ceph_options *opts = mdsc->fsc->client->options; |
2f2dc053 | 3498 | struct ceph_mds_request *req; |
2f2dc053 SW |
3499 | |
3500 | mutex_lock(&mdsc->mutex); | |
44ca18f2 | 3501 | if (__get_oldest_req(mdsc)) { |
2f2dc053 | 3502 | mutex_unlock(&mdsc->mutex); |
44ca18f2 | 3503 | |
2f2dc053 SW |
3504 | dout("wait_requests waiting for requests\n"); |
3505 | wait_for_completion_timeout(&mdsc->safe_umount_waiters, | |
a319bf56 | 3506 | ceph_timeout_jiffies(opts->mount_timeout)); |
2f2dc053 SW |
3507 | |
3508 | /* tear down remaining requests */ | |
44ca18f2 SW |
3509 | mutex_lock(&mdsc->mutex); |
3510 | while ((req = __get_oldest_req(mdsc))) { | |
2f2dc053 SW |
3511 | dout("wait_requests timed out on tid %llu\n", |
3512 | req->r_tid); | |
44ca18f2 | 3513 | __unregister_request(mdsc, req); |
2f2dc053 SW |
3514 | } |
3515 | } | |
3516 | mutex_unlock(&mdsc->mutex); | |
3517 | dout("wait_requests done\n"); | |
3518 | } | |
3519 | ||
3520 | /* | |
3521 | * called before mount is ro, and before dentries are torn down. | |
3522 | * (hmm, does this still race with new lookups?) | |
3523 | */ | |
3524 | void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc) | |
3525 | { | |
3526 | dout("pre_umount\n"); | |
3527 | mdsc->stopping = 1; | |
3528 | ||
3529 | drop_leases(mdsc); | |
afcdaea3 | 3530 | ceph_flush_dirty_caps(mdsc); |
2f2dc053 | 3531 | wait_requests(mdsc); |
17c688c3 SW |
3532 | |
3533 | /* | |
3534 | * wait for reply handlers to drop their request refs and | |
3535 | * their inode/dcache refs | |
3536 | */ | |
3537 | ceph_msgr_flush(); | |
2f2dc053 SW |
3538 | } |
3539 | ||
3540 | /* | |
3541 | * wait for all write mds requests to flush. | |
3542 | */ | |
3543 | static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid) | |
3544 | { | |
80fc7314 | 3545 | struct ceph_mds_request *req = NULL, *nextreq; |
44ca18f2 | 3546 | struct rb_node *n; |
2f2dc053 SW |
3547 | |
3548 | mutex_lock(&mdsc->mutex); | |
3549 | dout("wait_unsafe_requests want %lld\n", want_tid); | |
80fc7314 | 3550 | restart: |
44ca18f2 SW |
3551 | req = __get_oldest_req(mdsc); |
3552 | while (req && req->r_tid <= want_tid) { | |
80fc7314 SW |
3553 | /* find next request */ |
3554 | n = rb_next(&req->r_node); | |
3555 | if (n) | |
3556 | nextreq = rb_entry(n, struct ceph_mds_request, r_node); | |
3557 | else | |
3558 | nextreq = NULL; | |
e8a7b8b1 YZ |
3559 | if (req->r_op != CEPH_MDS_OP_SETFILELOCK && |
3560 | (req->r_op & CEPH_MDS_OP_WRITE)) { | |
44ca18f2 SW |
3561 | /* write op */ |
3562 | ceph_mdsc_get_request(req); | |
80fc7314 SW |
3563 | if (nextreq) |
3564 | ceph_mdsc_get_request(nextreq); | |
44ca18f2 SW |
3565 | mutex_unlock(&mdsc->mutex); |
3566 | dout("wait_unsafe_requests wait on %llu (want %llu)\n", | |
3567 | req->r_tid, want_tid); | |
3568 | wait_for_completion(&req->r_safe_completion); | |
3569 | mutex_lock(&mdsc->mutex); | |
44ca18f2 | 3570 | ceph_mdsc_put_request(req); |
80fc7314 SW |
3571 | if (!nextreq) |
3572 | break; /* next dne before, so we're done! */ | |
3573 | if (RB_EMPTY_NODE(&nextreq->r_node)) { | |
3574 | /* next request was removed from tree */ | |
3575 | ceph_mdsc_put_request(nextreq); | |
3576 | goto restart; | |
3577 | } | |
3578 | ceph_mdsc_put_request(nextreq); /* won't go away */ | |
44ca18f2 | 3579 | } |
80fc7314 | 3580 | req = nextreq; |
2f2dc053 SW |
3581 | } |
3582 | mutex_unlock(&mdsc->mutex); | |
3583 | dout("wait_unsafe_requests done\n"); | |
3584 | } | |
3585 | ||
3586 | void ceph_mdsc_sync(struct ceph_mds_client *mdsc) | |
3587 | { | |
affbc19a | 3588 | u64 want_tid, want_flush, want_snap; |
2f2dc053 | 3589 | |
48fec5d0 | 3590 | if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) |
56b7cf95 SW |
3591 | return; |
3592 | ||
2f2dc053 SW |
3593 | dout("sync\n"); |
3594 | mutex_lock(&mdsc->mutex); | |
3595 | want_tid = mdsc->last_tid; | |
2f2dc053 | 3596 | mutex_unlock(&mdsc->mutex); |
2f2dc053 | 3597 | |
afcdaea3 | 3598 | ceph_flush_dirty_caps(mdsc); |
d3383a8e | 3599 | spin_lock(&mdsc->cap_dirty_lock); |
8310b089 | 3600 | want_flush = mdsc->last_cap_flush_tid; |
d3383a8e YZ |
3601 | spin_unlock(&mdsc->cap_dirty_lock); |
3602 | ||
affbc19a YZ |
3603 | down_read(&mdsc->snap_rwsem); |
3604 | want_snap = mdsc->last_snap_seq; | |
3605 | up_read(&mdsc->snap_rwsem); | |
3606 | ||
3607 | dout("sync want tid %lld flush_seq %lld snap_seq %lld\n", | |
3608 | want_tid, want_flush, want_snap); | |
2f2dc053 SW |
3609 | |
3610 | wait_unsafe_requests(mdsc, want_tid); | |
affbc19a | 3611 | wait_caps_flush(mdsc, want_flush, want_snap); |
2f2dc053 SW |
3612 | } |
3613 | ||
f3c60c59 SW |
3614 | /* |
3615 | * true if all sessions are closed, or we force unmount | |
3616 | */ | |
7fd7d101 | 3617 | static bool done_closing_sessions(struct ceph_mds_client *mdsc) |
f3c60c59 | 3618 | { |
48fec5d0 | 3619 | if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) |
f3c60c59 | 3620 | return true; |
86d8f67b | 3621 | return atomic_read(&mdsc->num_sessions) == 0; |
f3c60c59 | 3622 | } |
2f2dc053 SW |
3623 | |
3624 | /* | |
3625 | * called after sb is ro. | |
3626 | */ | |
3627 | void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc) | |
3628 | { | |
a319bf56 | 3629 | struct ceph_options *opts = mdsc->fsc->client->options; |
2f2dc053 SW |
3630 | struct ceph_mds_session *session; |
3631 | int i; | |
2f2dc053 SW |
3632 | |
3633 | dout("close_sessions\n"); | |
3634 | ||
2f2dc053 | 3635 | /* close sessions */ |
f3c60c59 SW |
3636 | mutex_lock(&mdsc->mutex); |
3637 | for (i = 0; i < mdsc->max_sessions; i++) { | |
3638 | session = __ceph_lookup_mds_session(mdsc, i); | |
3639 | if (!session) | |
3640 | continue; | |
2f2dc053 | 3641 | mutex_unlock(&mdsc->mutex); |
f3c60c59 SW |
3642 | mutex_lock(&session->s_mutex); |
3643 | __close_session(mdsc, session); | |
3644 | mutex_unlock(&session->s_mutex); | |
3645 | ceph_put_mds_session(session); | |
2f2dc053 SW |
3646 | mutex_lock(&mdsc->mutex); |
3647 | } | |
f3c60c59 SW |
3648 | mutex_unlock(&mdsc->mutex); |
3649 | ||
3650 | dout("waiting for sessions to close\n"); | |
3651 | wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc), | |
a319bf56 | 3652 | ceph_timeout_jiffies(opts->mount_timeout)); |
2f2dc053 SW |
3653 | |
3654 | /* tear down remaining sessions */ | |
f3c60c59 | 3655 | mutex_lock(&mdsc->mutex); |
2f2dc053 SW |
3656 | for (i = 0; i < mdsc->max_sessions; i++) { |
3657 | if (mdsc->sessions[i]) { | |
3658 | session = get_session(mdsc->sessions[i]); | |
2600d2dd | 3659 | __unregister_session(mdsc, session); |
2f2dc053 SW |
3660 | mutex_unlock(&mdsc->mutex); |
3661 | mutex_lock(&session->s_mutex); | |
3662 | remove_session_caps(session); | |
3663 | mutex_unlock(&session->s_mutex); | |
3664 | ceph_put_mds_session(session); | |
3665 | mutex_lock(&mdsc->mutex); | |
3666 | } | |
3667 | } | |
2f2dc053 | 3668 | WARN_ON(!list_empty(&mdsc->cap_delay_list)); |
2f2dc053 SW |
3669 | mutex_unlock(&mdsc->mutex); |
3670 | ||
3671 | ceph_cleanup_empty_realms(mdsc); | |
3672 | ||
3673 | cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */ | |
3674 | ||
3675 | dout("stopped\n"); | |
3676 | } | |
3677 | ||
48fec5d0 YZ |
3678 | void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc) |
3679 | { | |
3680 | struct ceph_mds_session *session; | |
3681 | int mds; | |
3682 | ||
3683 | dout("force umount\n"); | |
3684 | ||
3685 | mutex_lock(&mdsc->mutex); | |
3686 | for (mds = 0; mds < mdsc->max_sessions; mds++) { | |
3687 | session = __ceph_lookup_mds_session(mdsc, mds); | |
3688 | if (!session) | |
3689 | continue; | |
3690 | mutex_unlock(&mdsc->mutex); | |
3691 | mutex_lock(&session->s_mutex); | |
3692 | __close_session(mdsc, session); | |
3693 | if (session->s_state == CEPH_MDS_SESSION_CLOSING) { | |
3694 | cleanup_session_requests(mdsc, session); | |
3695 | remove_session_caps(session); | |
3696 | } | |
3697 | mutex_unlock(&session->s_mutex); | |
3698 | ceph_put_mds_session(session); | |
3699 | mutex_lock(&mdsc->mutex); | |
3700 | kick_requests(mdsc, mds); | |
3701 | } | |
3702 | __wake_requests(mdsc, &mdsc->waiting_for_map); | |
3703 | mutex_unlock(&mdsc->mutex); | |
3704 | } | |
3705 | ||
3d14c5d2 | 3706 | static void ceph_mdsc_stop(struct ceph_mds_client *mdsc) |
2f2dc053 SW |
3707 | { |
3708 | dout("stop\n"); | |
3709 | cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */ | |
3710 | if (mdsc->mdsmap) | |
3711 | ceph_mdsmap_destroy(mdsc->mdsmap); | |
3712 | kfree(mdsc->sessions); | |
37151668 | 3713 | ceph_caps_finalize(mdsc); |
10183a69 | 3714 | ceph_pool_perm_destroy(mdsc); |
2f2dc053 SW |
3715 | } |
3716 | ||
3d14c5d2 YS |
3717 | void ceph_mdsc_destroy(struct ceph_fs_client *fsc) |
3718 | { | |
3719 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
3720 | ||
ef550f6f | 3721 | dout("mdsc_destroy %p\n", mdsc); |
3d14c5d2 | 3722 | ceph_mdsc_stop(mdsc); |
ef550f6f SW |
3723 | |
3724 | /* flush out any connection work with references to us */ | |
3725 | ceph_msgr_flush(); | |
3726 | ||
3d14c5d2 YS |
3727 | fsc->mdsc = NULL; |
3728 | kfree(mdsc); | |
ef550f6f | 3729 | dout("mdsc_destroy %p done\n", mdsc); |
3d14c5d2 YS |
3730 | } |
3731 | ||
2f2dc053 SW |
3732 | |
3733 | /* | |
3734 | * handle mds map update. | |
3735 | */ | |
3736 | void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg) | |
3737 | { | |
3738 | u32 epoch; | |
3739 | u32 maplen; | |
3740 | void *p = msg->front.iov_base; | |
3741 | void *end = p + msg->front.iov_len; | |
3742 | struct ceph_mdsmap *newmap, *oldmap; | |
3743 | struct ceph_fsid fsid; | |
3744 | int err = -EINVAL; | |
3745 | ||
3746 | ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad); | |
3747 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
3d14c5d2 | 3748 | if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0) |
0743304d | 3749 | return; |
c89136ea SW |
3750 | epoch = ceph_decode_32(&p); |
3751 | maplen = ceph_decode_32(&p); | |
2f2dc053 SW |
3752 | dout("handle_map epoch %u len %d\n", epoch, (int)maplen); |
3753 | ||
3754 | /* do we need it? */ | |
2f2dc053 SW |
3755 | mutex_lock(&mdsc->mutex); |
3756 | if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) { | |
3757 | dout("handle_map epoch %u <= our %u\n", | |
3758 | epoch, mdsc->mdsmap->m_epoch); | |
3759 | mutex_unlock(&mdsc->mutex); | |
3760 | return; | |
3761 | } | |
3762 | ||
3763 | newmap = ceph_mdsmap_decode(&p, end); | |
3764 | if (IS_ERR(newmap)) { | |
3765 | err = PTR_ERR(newmap); | |
3766 | goto bad_unlock; | |
3767 | } | |
3768 | ||
3769 | /* swap into place */ | |
3770 | if (mdsc->mdsmap) { | |
3771 | oldmap = mdsc->mdsmap; | |
3772 | mdsc->mdsmap = newmap; | |
3773 | check_new_map(mdsc, newmap, oldmap); | |
3774 | ceph_mdsmap_destroy(oldmap); | |
3775 | } else { | |
3776 | mdsc->mdsmap = newmap; /* first mds map */ | |
3777 | } | |
3d14c5d2 | 3778 | mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size; |
2f2dc053 SW |
3779 | |
3780 | __wake_requests(mdsc, &mdsc->waiting_for_map); | |
82dcabad ID |
3781 | ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP, |
3782 | mdsc->mdsmap->m_epoch); | |
2f2dc053 SW |
3783 | |
3784 | mutex_unlock(&mdsc->mutex); | |
3785 | schedule_delayed(mdsc); | |
3786 | return; | |
3787 | ||
3788 | bad_unlock: | |
3789 | mutex_unlock(&mdsc->mutex); | |
3790 | bad: | |
3791 | pr_err("error decoding mdsmap %d\n", err); | |
3792 | return; | |
3793 | } | |
3794 | ||
3795 | static struct ceph_connection *con_get(struct ceph_connection *con) | |
3796 | { | |
3797 | struct ceph_mds_session *s = con->private; | |
3798 | ||
3799 | if (get_session(s)) { | |
2600d2dd | 3800 | dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref)); |
2f2dc053 SW |
3801 | return con; |
3802 | } | |
3803 | dout("mdsc con_get %p FAIL\n", s); | |
3804 | return NULL; | |
3805 | } | |
3806 | ||
3807 | static void con_put(struct ceph_connection *con) | |
3808 | { | |
3809 | struct ceph_mds_session *s = con->private; | |
3810 | ||
7d8e18a6 | 3811 | dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1); |
2f2dc053 SW |
3812 | ceph_put_mds_session(s); |
3813 | } | |
3814 | ||
3815 | /* | |
3816 | * if the client is unresponsive for long enough, the mds will kill | |
3817 | * the session entirely. | |
3818 | */ | |
3819 | static void peer_reset(struct ceph_connection *con) | |
3820 | { | |
3821 | struct ceph_mds_session *s = con->private; | |
7e70f0ed | 3822 | struct ceph_mds_client *mdsc = s->s_mdsc; |
2f2dc053 | 3823 | |
f3ae1b97 | 3824 | pr_warn("mds%d closed our session\n", s->s_mds); |
7e70f0ed | 3825 | send_mds_reconnect(mdsc, s); |
2f2dc053 SW |
3826 | } |
3827 | ||
3828 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
3829 | { | |
3830 | struct ceph_mds_session *s = con->private; | |
3831 | struct ceph_mds_client *mdsc = s->s_mdsc; | |
3832 | int type = le16_to_cpu(msg->hdr.type); | |
3833 | ||
2600d2dd SW |
3834 | mutex_lock(&mdsc->mutex); |
3835 | if (__verify_registered_session(mdsc, s) < 0) { | |
3836 | mutex_unlock(&mdsc->mutex); | |
3837 | goto out; | |
3838 | } | |
3839 | mutex_unlock(&mdsc->mutex); | |
3840 | ||
2f2dc053 SW |
3841 | switch (type) { |
3842 | case CEPH_MSG_MDS_MAP: | |
3843 | ceph_mdsc_handle_map(mdsc, msg); | |
3844 | break; | |
3845 | case CEPH_MSG_CLIENT_SESSION: | |
3846 | handle_session(s, msg); | |
3847 | break; | |
3848 | case CEPH_MSG_CLIENT_REPLY: | |
3849 | handle_reply(s, msg); | |
3850 | break; | |
3851 | case CEPH_MSG_CLIENT_REQUEST_FORWARD: | |
2600d2dd | 3852 | handle_forward(mdsc, s, msg); |
2f2dc053 SW |
3853 | break; |
3854 | case CEPH_MSG_CLIENT_CAPS: | |
3855 | ceph_handle_caps(s, msg); | |
3856 | break; | |
3857 | case CEPH_MSG_CLIENT_SNAP: | |
2600d2dd | 3858 | ceph_handle_snap(mdsc, s, msg); |
2f2dc053 SW |
3859 | break; |
3860 | case CEPH_MSG_CLIENT_LEASE: | |
2600d2dd | 3861 | handle_lease(mdsc, s, msg); |
2f2dc053 SW |
3862 | break; |
3863 | ||
3864 | default: | |
3865 | pr_err("received unknown message type %d %s\n", type, | |
3866 | ceph_msg_type_name(type)); | |
3867 | } | |
2600d2dd | 3868 | out: |
2f2dc053 SW |
3869 | ceph_msg_put(msg); |
3870 | } | |
3871 | ||
4e7a5dcd SW |
3872 | /* |
3873 | * authentication | |
3874 | */ | |
a3530df3 AE |
3875 | |
3876 | /* | |
3877 | * Note: returned pointer is the address of a structure that's | |
3878 | * managed separately. Caller must *not* attempt to free it. | |
3879 | */ | |
3880 | static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, | |
8f43fb53 | 3881 | int *proto, int force_new) |
4e7a5dcd SW |
3882 | { |
3883 | struct ceph_mds_session *s = con->private; | |
3884 | struct ceph_mds_client *mdsc = s->s_mdsc; | |
3d14c5d2 | 3885 | struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth; |
74f1869f | 3886 | struct ceph_auth_handshake *auth = &s->s_auth; |
4e7a5dcd | 3887 | |
74f1869f | 3888 | if (force_new && auth->authorizer) { |
6c1ea260 | 3889 | ceph_auth_destroy_authorizer(auth->authorizer); |
74f1869f | 3890 | auth->authorizer = NULL; |
4e7a5dcd | 3891 | } |
27859f97 SW |
3892 | if (!auth->authorizer) { |
3893 | int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS, | |
3894 | auth); | |
0bed9b5c SW |
3895 | if (ret) |
3896 | return ERR_PTR(ret); | |
27859f97 SW |
3897 | } else { |
3898 | int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS, | |
3899 | auth); | |
a255651d | 3900 | if (ret) |
a3530df3 | 3901 | return ERR_PTR(ret); |
4e7a5dcd | 3902 | } |
4e7a5dcd | 3903 | *proto = ac->protocol; |
74f1869f | 3904 | |
a3530df3 | 3905 | return auth; |
4e7a5dcd SW |
3906 | } |
3907 | ||
3908 | ||
3909 | static int verify_authorizer_reply(struct ceph_connection *con, int len) | |
3910 | { | |
3911 | struct ceph_mds_session *s = con->private; | |
3912 | struct ceph_mds_client *mdsc = s->s_mdsc; | |
3d14c5d2 | 3913 | struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth; |
4e7a5dcd | 3914 | |
27859f97 | 3915 | return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len); |
4e7a5dcd SW |
3916 | } |
3917 | ||
9bd2e6f8 SW |
3918 | static int invalidate_authorizer(struct ceph_connection *con) |
3919 | { | |
3920 | struct ceph_mds_session *s = con->private; | |
3921 | struct ceph_mds_client *mdsc = s->s_mdsc; | |
3d14c5d2 | 3922 | struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth; |
9bd2e6f8 | 3923 | |
27859f97 | 3924 | ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS); |
9bd2e6f8 | 3925 | |
3d14c5d2 | 3926 | return ceph_monc_validate_auth(&mdsc->fsc->client->monc); |
9bd2e6f8 SW |
3927 | } |
3928 | ||
53ded495 AE |
3929 | static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con, |
3930 | struct ceph_msg_header *hdr, int *skip) | |
3931 | { | |
3932 | struct ceph_msg *msg; | |
3933 | int type = (int) le16_to_cpu(hdr->type); | |
3934 | int front_len = (int) le32_to_cpu(hdr->front_len); | |
3935 | ||
3936 | if (con->in_msg) | |
3937 | return con->in_msg; | |
3938 | ||
3939 | *skip = 0; | |
3940 | msg = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
3941 | if (!msg) { | |
3942 | pr_err("unable to allocate msg type %d len %d\n", | |
3943 | type, front_len); | |
3944 | return NULL; | |
3945 | } | |
53ded495 AE |
3946 | |
3947 | return msg; | |
3948 | } | |
3949 | ||
79dbd1ba | 3950 | static int mds_sign_message(struct ceph_msg *msg) |
33d07337 | 3951 | { |
79dbd1ba | 3952 | struct ceph_mds_session *s = msg->con->private; |
33d07337 | 3953 | struct ceph_auth_handshake *auth = &s->s_auth; |
79dbd1ba | 3954 | |
33d07337 YZ |
3955 | return ceph_auth_sign_message(auth, msg); |
3956 | } | |
3957 | ||
79dbd1ba | 3958 | static int mds_check_message_signature(struct ceph_msg *msg) |
33d07337 | 3959 | { |
79dbd1ba | 3960 | struct ceph_mds_session *s = msg->con->private; |
33d07337 | 3961 | struct ceph_auth_handshake *auth = &s->s_auth; |
79dbd1ba | 3962 | |
33d07337 YZ |
3963 | return ceph_auth_check_message_signature(auth, msg); |
3964 | } | |
3965 | ||
9e32789f | 3966 | static const struct ceph_connection_operations mds_con_ops = { |
2f2dc053 SW |
3967 | .get = con_get, |
3968 | .put = con_put, | |
3969 | .dispatch = dispatch, | |
4e7a5dcd SW |
3970 | .get_authorizer = get_authorizer, |
3971 | .verify_authorizer_reply = verify_authorizer_reply, | |
9bd2e6f8 | 3972 | .invalidate_authorizer = invalidate_authorizer, |
2f2dc053 | 3973 | .peer_reset = peer_reset, |
53ded495 | 3974 | .alloc_msg = mds_alloc_msg, |
79dbd1ba ID |
3975 | .sign_message = mds_sign_message, |
3976 | .check_message_signature = mds_check_message_signature, | |
2f2dc053 SW |
3977 | }; |
3978 | ||
2f2dc053 | 3979 | /* eof */ |