blob: 8f8f5c043c37640e12bd9c896e5a51ad3cd964ab [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil2f2dc052009-10-06 11:31:09 -07002
Sage Weil496e5952010-09-22 19:57:10 -07003#include <linux/fs.h>
Sage Weil2f2dc052009-10-06 11:31:09 -07004#include <linux/wait.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09005#include <linux/slab.h>
Sage Weil2f2dc052009-10-06 11:31:09 -07006#include <linux/sched.h>
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07007#include <linux/debugfs.h>
8#include <linux/seq_file.h>
Sage Weil2f2dc052009-10-06 11:31:09 -07009
Sage Weil2f2dc052009-10-06 11:31:09 -070010#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070011#include "mds_client.h"
12
Sage Weil1fe60e52012-07-30 16:23:22 -070013#include <linux/ceph/ceph_features.h>
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070014#include <linux/ceph/messenger.h>
15#include <linux/ceph/decode.h>
16#include <linux/ceph/pagelist.h>
17#include <linux/ceph/auth.h>
18#include <linux/ceph/debugfs.h>
Sage Weil2f2dc052009-10-06 11:31:09 -070019
20/*
21 * A cluster of MDS (metadata server) daemons is responsible for
22 * managing the file system namespace (the directory hierarchy and
23 * inodes) and for coordinating shared access to storage. Metadata is
24 * partitioning hierarchically across a number of servers, and that
25 * partition varies over time as the cluster adjusts the distribution
26 * in order to balance load.
27 *
28 * The MDS client is primarily responsible to managing synchronous
29 * metadata requests for operations like open, unlink, and so forth.
30 * If there is a MDS failure, we find out about it when we (possibly
31 * request and) receive a new MDS map, and can resubmit affected
32 * requests.
33 *
34 * For the most part, though, we take advantage of a lossless
35 * communications channel to the MDS, and do not need to worry about
36 * timing out or resubmitting requests.
37 *
38 * We maintain a stateful "session" with each MDS we interact with.
39 * Within each session, we sent periodic heartbeat messages to ensure
40 * any capabilities or leases we have been issues remain valid. If
41 * the session times out and goes stale, our leases and capabilities
42 * are no longer valid.
43 */
44
Sage Weil20cb34a2010-05-12 15:21:32 -070045struct ceph_reconnect_state {
46 struct ceph_pagelist *pagelist;
47 bool flock;
48};
49
Sage Weil2f2dc052009-10-06 11:31:09 -070050static void __wake_requests(struct ceph_mds_client *mdsc,
51 struct list_head *head);
52
Tobias Klauser9e327892010-05-20 10:40:19 +020053static const struct ceph_connection_operations mds_con_ops;
Sage Weil2f2dc052009-10-06 11:31:09 -070054
55
56/*
57 * mds reply parsing
58 */
59
60/*
61 * parse individual inode info
62 */
63static int parse_reply_info_in(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -080064 struct ceph_mds_reply_info_in *info,
65 int features)
Sage Weil2f2dc052009-10-06 11:31:09 -070066{
67 int err = -EIO;
68
69 info->in = *p;
70 *p += sizeof(struct ceph_mds_reply_inode) +
71 sizeof(*info->in->fragtree.splits) *
72 le32_to_cpu(info->in->fragtree.nsplits);
73
74 ceph_decode_32_safe(p, end, info->symlink_len, bad);
75 ceph_decode_need(p, end, info->symlink_len, bad);
76 info->symlink = *p;
77 *p += info->symlink_len;
78
Sage Weil14303d22010-12-14 17:37:52 -080079 if (features & CEPH_FEATURE_DIRLAYOUTHASH)
80 ceph_decode_copy_safe(p, end, &info->dir_layout,
81 sizeof(info->dir_layout), bad);
82 else
83 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
84
Sage Weil2f2dc052009-10-06 11:31:09 -070085 ceph_decode_32_safe(p, end, info->xattr_len, bad);
86 ceph_decode_need(p, end, info->xattr_len, bad);
87 info->xattr_data = *p;
88 *p += info->xattr_len;
89 return 0;
90bad:
91 return err;
92}
93
94/*
95 * parse a normal reply, which may contain a (dir+)dentry and/or a
96 * target inode.
97 */
98static int parse_reply_info_trace(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -080099 struct ceph_mds_reply_info_parsed *info,
100 int features)
Sage Weil2f2dc052009-10-06 11:31:09 -0700101{
102 int err;
103
104 if (info->head->is_dentry) {
Sage Weil14303d22010-12-14 17:37:52 -0800105 err = parse_reply_info_in(p, end, &info->diri, features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700106 if (err < 0)
107 goto out_bad;
108
109 if (unlikely(*p + sizeof(*info->dirfrag) > end))
110 goto bad;
111 info->dirfrag = *p;
112 *p += sizeof(*info->dirfrag) +
113 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
114 if (unlikely(*p > end))
115 goto bad;
116
117 ceph_decode_32_safe(p, end, info->dname_len, bad);
118 ceph_decode_need(p, end, info->dname_len, bad);
119 info->dname = *p;
120 *p += info->dname_len;
121 info->dlease = *p;
122 *p += sizeof(*info->dlease);
123 }
124
125 if (info->head->is_target) {
Sage Weil14303d22010-12-14 17:37:52 -0800126 err = parse_reply_info_in(p, end, &info->targeti, features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700127 if (err < 0)
128 goto out_bad;
129 }
130
131 if (unlikely(*p != end))
132 goto bad;
133 return 0;
134
135bad:
136 err = -EIO;
137out_bad:
138 pr_err("problem parsing mds trace %d\n", err);
139 return err;
140}
141
142/*
143 * parse readdir results
144 */
145static int parse_reply_info_dir(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -0800146 struct ceph_mds_reply_info_parsed *info,
147 int features)
Sage Weil2f2dc052009-10-06 11:31:09 -0700148{
149 u32 num, i = 0;
150 int err;
151
152 info->dir_dir = *p;
153 if (*p + sizeof(*info->dir_dir) > end)
154 goto bad;
155 *p += sizeof(*info->dir_dir) +
156 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
157 if (*p > end)
158 goto bad;
159
160 ceph_decode_need(p, end, sizeof(num) + 2, bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700161 num = ceph_decode_32(p);
162 info->dir_end = ceph_decode_8(p);
163 info->dir_complete = ceph_decode_8(p);
Sage Weil2f2dc052009-10-06 11:31:09 -0700164 if (num == 0)
165 goto done;
166
167 /* alloc large array */
168 info->dir_nr = num;
169 info->dir_in = kcalloc(num, sizeof(*info->dir_in) +
170 sizeof(*info->dir_dname) +
171 sizeof(*info->dir_dname_len) +
172 sizeof(*info->dir_dlease),
173 GFP_NOFS);
174 if (info->dir_in == NULL) {
175 err = -ENOMEM;
176 goto out_bad;
177 }
178 info->dir_dname = (void *)(info->dir_in + num);
179 info->dir_dname_len = (void *)(info->dir_dname + num);
180 info->dir_dlease = (void *)(info->dir_dname_len + num);
181
182 while (num) {
183 /* dentry */
184 ceph_decode_need(p, end, sizeof(u32)*2, bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700185 info->dir_dname_len[i] = ceph_decode_32(p);
Sage Weil2f2dc052009-10-06 11:31:09 -0700186 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
187 info->dir_dname[i] = *p;
188 *p += info->dir_dname_len[i];
189 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
190 info->dir_dname[i]);
191 info->dir_dlease[i] = *p;
192 *p += sizeof(struct ceph_mds_reply_lease);
193
194 /* inode */
Sage Weil14303d22010-12-14 17:37:52 -0800195 err = parse_reply_info_in(p, end, &info->dir_in[i], features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700196 if (err < 0)
197 goto out_bad;
198 i++;
199 num--;
200 }
201
202done:
203 if (*p != end)
204 goto bad;
205 return 0;
206
207bad:
208 err = -EIO;
209out_bad:
210 pr_err("problem parsing dir contents %d\n", err);
211 return err;
212}
213
214/*
Herb Shiu25933ab2010-12-01 14:14:38 -0800215 * parse fcntl F_GETLK results
216 */
217static int parse_reply_info_filelock(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -0800218 struct ceph_mds_reply_info_parsed *info,
219 int features)
Herb Shiu25933ab2010-12-01 14:14:38 -0800220{
221 if (*p + sizeof(*info->filelock_reply) > end)
222 goto bad;
223
224 info->filelock_reply = *p;
225 *p += sizeof(*info->filelock_reply);
226
227 if (unlikely(*p != end))
228 goto bad;
229 return 0;
230
231bad:
232 return -EIO;
233}
234
235/*
Sam Lang6e8575f2012-12-28 09:56:46 -0800236 * parse create results
237 */
238static int parse_reply_info_create(void **p, void *end,
239 struct ceph_mds_reply_info_parsed *info,
240 int features)
241{
242 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
243 if (*p == end) {
244 info->has_create_ino = false;
245 } else {
246 info->has_create_ino = true;
247 info->ino = ceph_decode_64(p);
248 }
249 }
250
251 if (unlikely(*p != end))
252 goto bad;
253 return 0;
254
255bad:
256 return -EIO;
257}
258
259/*
Herb Shiu25933ab2010-12-01 14:14:38 -0800260 * parse extra results
261 */
262static int parse_reply_info_extra(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -0800263 struct ceph_mds_reply_info_parsed *info,
264 int features)
Herb Shiu25933ab2010-12-01 14:14:38 -0800265{
266 if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
Sage Weil14303d22010-12-14 17:37:52 -0800267 return parse_reply_info_filelock(p, end, info, features);
Yan, Zheng8a034492013-02-21 13:43:55 +0800268 else if (info->head->op == CEPH_MDS_OP_READDIR ||
269 info->head->op == CEPH_MDS_OP_LSSNAP)
Sage Weil14303d22010-12-14 17:37:52 -0800270 return parse_reply_info_dir(p, end, info, features);
Sam Lang6e8575f2012-12-28 09:56:46 -0800271 else if (info->head->op == CEPH_MDS_OP_CREATE)
272 return parse_reply_info_create(p, end, info, features);
273 else
274 return -EIO;
Herb Shiu25933ab2010-12-01 14:14:38 -0800275}
276
277/*
Sage Weil2f2dc052009-10-06 11:31:09 -0700278 * parse entire mds reply
279 */
280static int parse_reply_info(struct ceph_msg *msg,
Sage Weil14303d22010-12-14 17:37:52 -0800281 struct ceph_mds_reply_info_parsed *info,
282 int features)
Sage Weil2f2dc052009-10-06 11:31:09 -0700283{
284 void *p, *end;
285 u32 len;
286 int err;
287
288 info->head = msg->front.iov_base;
289 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
290 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
291
292 /* trace */
293 ceph_decode_32_safe(&p, end, len, bad);
294 if (len > 0) {
Xi Wang32852a82012-01-14 22:20:59 -0500295 ceph_decode_need(&p, end, len, bad);
Sage Weil14303d22010-12-14 17:37:52 -0800296 err = parse_reply_info_trace(&p, p+len, info, features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700297 if (err < 0)
298 goto out_bad;
299 }
300
Herb Shiu25933ab2010-12-01 14:14:38 -0800301 /* extra */
Sage Weil2f2dc052009-10-06 11:31:09 -0700302 ceph_decode_32_safe(&p, end, len, bad);
303 if (len > 0) {
Xi Wang32852a82012-01-14 22:20:59 -0500304 ceph_decode_need(&p, end, len, bad);
Sage Weil14303d22010-12-14 17:37:52 -0800305 err = parse_reply_info_extra(&p, p+len, info, features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700306 if (err < 0)
307 goto out_bad;
308 }
309
310 /* snap blob */
311 ceph_decode_32_safe(&p, end, len, bad);
312 info->snapblob_len = len;
313 info->snapblob = p;
314 p += len;
315
316 if (p != end)
317 goto bad;
318 return 0;
319
320bad:
321 err = -EIO;
322out_bad:
323 pr_err("mds parse_reply err %d\n", err);
324 return err;
325}
326
327static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
328{
329 kfree(info->dir_in);
330}
331
332
333/*
334 * sessions
335 */
336static const char *session_state_name(int s)
337{
338 switch (s) {
339 case CEPH_MDS_SESSION_NEW: return "new";
340 case CEPH_MDS_SESSION_OPENING: return "opening";
341 case CEPH_MDS_SESSION_OPEN: return "open";
342 case CEPH_MDS_SESSION_HUNG: return "hung";
343 case CEPH_MDS_SESSION_CLOSING: return "closing";
Sage Weil44ca18f2010-02-15 12:08:46 -0800344 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
Sage Weil2f2dc052009-10-06 11:31:09 -0700345 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
346 default: return "???";
347 }
348}
349
350static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
351{
352 if (atomic_inc_not_zero(&s->s_ref)) {
353 dout("mdsc get_session %p %d -> %d\n", s,
354 atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
355 return s;
356 } else {
357 dout("mdsc get_session %p 0 -- FAIL", s);
358 return NULL;
359 }
360}
361
362void ceph_put_mds_session(struct ceph_mds_session *s)
363{
364 dout("mdsc put_session %p %d -> %d\n", s,
365 atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800366 if (atomic_dec_and_test(&s->s_ref)) {
Alex Elder6c4a1912012-05-16 15:16:38 -0500367 if (s->s_auth.authorizer)
Sage Weil27859f92013-03-25 10:26:14 -0700368 ceph_auth_destroy_authorizer(
369 s->s_mdsc->fsc->client->monc.auth,
370 s->s_auth.authorizer);
Sage Weil2f2dc052009-10-06 11:31:09 -0700371 kfree(s);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800372 }
Sage Weil2f2dc052009-10-06 11:31:09 -0700373}
374
375/*
376 * called under mdsc->mutex
377 */
378struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
379 int mds)
380{
381 struct ceph_mds_session *session;
382
383 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
384 return NULL;
385 session = mdsc->sessions[mds];
386 dout("lookup_mds_session %p %d\n", session,
387 atomic_read(&session->s_ref));
388 get_session(session);
389 return session;
390}
391
392static bool __have_session(struct ceph_mds_client *mdsc, int mds)
393{
394 if (mds >= mdsc->max_sessions)
395 return false;
396 return mdsc->sessions[mds];
397}
398
Sage Weil2600d2d2010-02-22 15:12:16 -0800399static int __verify_registered_session(struct ceph_mds_client *mdsc,
400 struct ceph_mds_session *s)
401{
402 if (s->s_mds >= mdsc->max_sessions ||
403 mdsc->sessions[s->s_mds] != s)
404 return -ENOENT;
405 return 0;
406}
407
Sage Weil2f2dc052009-10-06 11:31:09 -0700408/*
409 * create+register a new session for given mds.
410 * called under mdsc->mutex.
411 */
412static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
413 int mds)
414{
415 struct ceph_mds_session *s;
416
Nathaniel Yazdanic338c072013-08-04 21:04:30 -0700417 if (mds >= mdsc->mdsmap->m_max_mds)
418 return ERR_PTR(-EINVAL);
419
Sage Weil2f2dc052009-10-06 11:31:09 -0700420 s = kzalloc(sizeof(*s), GFP_NOFS);
Dan Carpenter4736b002010-03-20 15:30:16 +0300421 if (!s)
422 return ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -0700423 s->s_mdsc = mdsc;
424 s->s_mds = mds;
425 s->s_state = CEPH_MDS_SESSION_NEW;
426 s->s_ttl = 0;
427 s->s_seq = 0;
428 mutex_init(&s->s_mutex);
429
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700430 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
Sage Weil2f2dc052009-10-06 11:31:09 -0700431
Alex Elderd8fb02a2012-01-12 17:48:10 -0800432 spin_lock_init(&s->s_gen_ttl_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700433 s->s_cap_gen = 0;
Alex Elder1ce208a2012-01-12 17:48:11 -0800434 s->s_cap_ttl = jiffies - 1;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800435
436 spin_lock_init(&s->s_cap_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700437 s->s_renew_requested = 0;
438 s->s_renew_seq = 0;
439 INIT_LIST_HEAD(&s->s_caps);
440 s->s_nr_caps = 0;
Sage Weil5dacf092009-12-21 20:40:34 -0800441 s->s_trim_caps = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -0700442 atomic_set(&s->s_ref, 1);
443 INIT_LIST_HEAD(&s->s_waiting);
444 INIT_LIST_HEAD(&s->s_unsafe);
445 s->s_num_cap_releases = 0;
Sage Weil7c1332b2010-02-16 11:39:45 -0800446 s->s_cap_iterator = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700447 INIT_LIST_HEAD(&s->s_cap_releases);
448 INIT_LIST_HEAD(&s->s_cap_releases_done);
449 INIT_LIST_HEAD(&s->s_cap_flushing);
450 INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
451
452 dout("register_session mds%d\n", mds);
453 if (mds >= mdsc->max_sessions) {
454 int newmax = 1 << get_count_order(mds+1);
455 struct ceph_mds_session **sa;
456
457 dout("register_session realloc to %d\n", newmax);
458 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
459 if (sa == NULL)
Sage Weil42ce56e2009-11-18 11:22:36 -0800460 goto fail_realloc;
Sage Weil2f2dc052009-10-06 11:31:09 -0700461 if (mdsc->sessions) {
462 memcpy(sa, mdsc->sessions,
463 mdsc->max_sessions * sizeof(void *));
464 kfree(mdsc->sessions);
465 }
466 mdsc->sessions = sa;
467 mdsc->max_sessions = newmax;
468 }
469 mdsc->sessions[mds] = s;
470 atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */
Sage Weil42ce56e2009-11-18 11:22:36 -0800471
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700472 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
473 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
Sage Weil42ce56e2009-11-18 11:22:36 -0800474
Sage Weil2f2dc052009-10-06 11:31:09 -0700475 return s;
Sage Weil42ce56e2009-11-18 11:22:36 -0800476
477fail_realloc:
478 kfree(s);
479 return ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -0700480}
481
482/*
483 * called under mdsc->mutex
484 */
Sage Weil2600d2d2010-02-22 15:12:16 -0800485static void __unregister_session(struct ceph_mds_client *mdsc,
Sage Weil42ce56e2009-11-18 11:22:36 -0800486 struct ceph_mds_session *s)
Sage Weil2f2dc052009-10-06 11:31:09 -0700487{
Sage Weil2600d2d2010-02-22 15:12:16 -0800488 dout("__unregister_session mds%d %p\n", s->s_mds, s);
489 BUG_ON(mdsc->sessions[s->s_mds] != s);
Sage Weil42ce56e2009-11-18 11:22:36 -0800490 mdsc->sessions[s->s_mds] = NULL;
491 ceph_con_close(&s->s_con);
492 ceph_put_mds_session(s);
Sage Weil2f2dc052009-10-06 11:31:09 -0700493}
494
495/*
496 * drop session refs in request.
497 *
498 * should be last request ref, or hold mdsc->mutex
499 */
500static void put_request_session(struct ceph_mds_request *req)
501{
502 if (req->r_session) {
503 ceph_put_mds_session(req->r_session);
504 req->r_session = NULL;
505 }
506}
507
Sage Weil153c8e62009-12-07 12:31:09 -0800508void ceph_mdsc_release_request(struct kref *kref)
Sage Weil2f2dc052009-10-06 11:31:09 -0700509{
Sage Weil153c8e62009-12-07 12:31:09 -0800510 struct ceph_mds_request *req = container_of(kref,
511 struct ceph_mds_request,
512 r_kref);
513 if (req->r_request)
514 ceph_msg_put(req->r_request);
515 if (req->r_reply) {
516 ceph_msg_put(req->r_reply);
517 destroy_reply_info(&req->r_reply_info);
Sage Weil2f2dc052009-10-06 11:31:09 -0700518 }
Sage Weil153c8e62009-12-07 12:31:09 -0800519 if (req->r_inode) {
Sage Weil41b02e12011-07-26 11:31:14 -0700520 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
Sage Weil153c8e62009-12-07 12:31:09 -0800521 iput(req->r_inode);
522 }
523 if (req->r_locked_dir)
Sage Weil41b02e12011-07-26 11:31:14 -0700524 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
Sage Weil153c8e62009-12-07 12:31:09 -0800525 if (req->r_target_inode)
526 iput(req->r_target_inode);
527 if (req->r_dentry)
528 dput(req->r_dentry);
529 if (req->r_old_dentry) {
Sage Weil41b02e12011-07-26 11:31:14 -0700530 /*
531 * track (and drop pins for) r_old_dentry_dir
532 * separately, since r_old_dentry's d_parent may have
533 * changed between the dir mutex being dropped and
534 * this request being freed.
535 */
536 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
537 CEPH_CAP_PIN);
Sage Weil153c8e62009-12-07 12:31:09 -0800538 dput(req->r_old_dentry);
Sage Weil41b02e12011-07-26 11:31:14 -0700539 iput(req->r_old_dentry_dir);
Sage Weil153c8e62009-12-07 12:31:09 -0800540 }
541 kfree(req->r_path1);
542 kfree(req->r_path2);
543 put_request_session(req);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700544 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
Sage Weil153c8e62009-12-07 12:31:09 -0800545 kfree(req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700546}
547
548/*
549 * lookup session, bump ref if found.
550 *
551 * called under mdsc->mutex.
552 */
553static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
554 u64 tid)
555{
556 struct ceph_mds_request *req;
Sage Weil44ca18f2010-02-15 12:08:46 -0800557 struct rb_node *n = mdsc->request_tree.rb_node;
558
559 while (n) {
560 req = rb_entry(n, struct ceph_mds_request, r_node);
561 if (tid < req->r_tid)
562 n = n->rb_left;
563 else if (tid > req->r_tid)
564 n = n->rb_right;
565 else {
566 ceph_mdsc_get_request(req);
567 return req;
568 }
569 }
570 return NULL;
571}
572
573static void __insert_request(struct ceph_mds_client *mdsc,
574 struct ceph_mds_request *new)
575{
576 struct rb_node **p = &mdsc->request_tree.rb_node;
577 struct rb_node *parent = NULL;
578 struct ceph_mds_request *req = NULL;
579
580 while (*p) {
581 parent = *p;
582 req = rb_entry(parent, struct ceph_mds_request, r_node);
583 if (new->r_tid < req->r_tid)
584 p = &(*p)->rb_left;
585 else if (new->r_tid > req->r_tid)
586 p = &(*p)->rb_right;
587 else
588 BUG();
589 }
590
591 rb_link_node(&new->r_node, parent, p);
592 rb_insert_color(&new->r_node, &mdsc->request_tree);
Sage Weil2f2dc052009-10-06 11:31:09 -0700593}
594
595/*
596 * Register an in-flight request, and assign a tid. Link to directory
597 * are modifying (if any).
598 *
599 * Called under mdsc->mutex.
600 */
601static void __register_request(struct ceph_mds_client *mdsc,
602 struct ceph_mds_request *req,
603 struct inode *dir)
604{
605 req->r_tid = ++mdsc->last_tid;
606 if (req->r_num_caps)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700607 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
608 req->r_num_caps);
Sage Weil2f2dc052009-10-06 11:31:09 -0700609 dout("__register_request %p tid %lld\n", req, req->r_tid);
610 ceph_mdsc_get_request(req);
Sage Weil44ca18f2010-02-15 12:08:46 -0800611 __insert_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700612
Sage Weilcb4276c2010-11-08 07:28:52 -0800613 req->r_uid = current_fsuid();
614 req->r_gid = current_fsgid();
615
Sage Weil2f2dc052009-10-06 11:31:09 -0700616 if (dir) {
617 struct ceph_inode_info *ci = ceph_inode(dir);
618
Sage Weil3b663782011-05-18 16:12:12 -0700619 ihold(dir);
Sage Weil2f2dc052009-10-06 11:31:09 -0700620 spin_lock(&ci->i_unsafe_lock);
621 req->r_unsafe_dir = dir;
622 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
623 spin_unlock(&ci->i_unsafe_lock);
624 }
625}
626
627static void __unregister_request(struct ceph_mds_client *mdsc,
628 struct ceph_mds_request *req)
629{
630 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
Sage Weil44ca18f2010-02-15 12:08:46 -0800631 rb_erase(&req->r_node, &mdsc->request_tree);
Sage Weil80fc7312010-03-16 15:28:54 -0700632 RB_CLEAR_NODE(&req->r_node);
Sage Weil2f2dc052009-10-06 11:31:09 -0700633
634 if (req->r_unsafe_dir) {
635 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
636
637 spin_lock(&ci->i_unsafe_lock);
638 list_del_init(&req->r_unsafe_dir_item);
639 spin_unlock(&ci->i_unsafe_lock);
Sage Weil3b663782011-05-18 16:12:12 -0700640
641 iput(req->r_unsafe_dir);
642 req->r_unsafe_dir = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700643 }
Sage Weil94aa8ae2010-03-28 21:22:50 -0700644
645 ceph_mdsc_put_request(req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700646}
647
648/*
649 * Choose mds to send request to next. If there is a hint set in the
650 * request (e.g., due to a prior forward hint from the mds), use that.
651 * Otherwise, consult frag tree and/or caps to identify the
652 * appropriate mds. If all else fails, choose randomly.
653 *
654 * Called under mdsc->mutex.
655 */
H Hartley Sweeten7fd7d102011-09-23 13:22:11 -0700656static struct dentry *get_nonsnap_parent(struct dentry *dentry)
Sage Weileb6bb1c2010-08-16 09:21:27 -0700657{
Sage Weild79698d2011-07-26 11:31:26 -0700658 /*
659 * we don't need to worry about protecting the d_parent access
660 * here because we never renaming inside the snapped namespace
661 * except to resplice to another snapdir, and either the old or new
662 * result is a valid result.
663 */
Sage Weileb6bb1c2010-08-16 09:21:27 -0700664 while (!IS_ROOT(dentry) && ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
665 dentry = dentry->d_parent;
666 return dentry;
667}
668
Sage Weil2f2dc052009-10-06 11:31:09 -0700669static int __choose_mds(struct ceph_mds_client *mdsc,
670 struct ceph_mds_request *req)
671{
672 struct inode *inode;
673 struct ceph_inode_info *ci;
674 struct ceph_cap *cap;
675 int mode = req->r_direct_mode;
676 int mds = -1;
677 u32 hash = req->r_direct_hash;
678 bool is_hash = req->r_direct_is_hash;
679
680 /*
681 * is there a specific mds we should try? ignore hint if we have
682 * no session and the mds is not up (active or recovering).
683 */
684 if (req->r_resend_mds >= 0 &&
685 (__have_session(mdsc, req->r_resend_mds) ||
686 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
687 dout("choose_mds using resend_mds mds%d\n",
688 req->r_resend_mds);
689 return req->r_resend_mds;
690 }
691
692 if (mode == USE_RANDOM_MDS)
693 goto random;
694
695 inode = NULL;
696 if (req->r_inode) {
697 inode = req->r_inode;
698 } else if (req->r_dentry) {
Sage Weild79698d2011-07-26 11:31:26 -0700699 /* ignore race with rename; old or new d_parent is okay */
700 struct dentry *parent = req->r_dentry->d_parent;
701 struct inode *dir = parent->d_inode;
Sage Weileb6bb1c2010-08-16 09:21:27 -0700702
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700703 if (dir->i_sb != mdsc->fsc->sb) {
Sage Weileb6bb1c2010-08-16 09:21:27 -0700704 /* not this fs! */
705 inode = req->r_dentry->d_inode;
706 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
707 /* direct snapped/virtual snapdir requests
708 * based on parent dir inode */
Sage Weild79698d2011-07-26 11:31:26 -0700709 struct dentry *dn = get_nonsnap_parent(parent);
Sage Weileb6bb1c2010-08-16 09:21:27 -0700710 inode = dn->d_inode;
711 dout("__choose_mds using nonsnap parent %p\n", inode);
712 } else if (req->r_dentry->d_inode) {
713 /* dentry target */
Sage Weil2f2dc052009-10-06 11:31:09 -0700714 inode = req->r_dentry->d_inode;
715 } else {
Sage Weileb6bb1c2010-08-16 09:21:27 -0700716 /* dir + name */
717 inode = dir;
Sage Weile5f86dc2011-07-26 11:30:55 -0700718 hash = ceph_dentry_hash(dir, req->r_dentry);
Sage Weil2f2dc052009-10-06 11:31:09 -0700719 is_hash = true;
720 }
721 }
Sage Weileb6bb1c2010-08-16 09:21:27 -0700722
Sage Weil2f2dc052009-10-06 11:31:09 -0700723 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
724 (int)hash, mode);
725 if (!inode)
726 goto random;
727 ci = ceph_inode(inode);
728
729 if (is_hash && S_ISDIR(inode->i_mode)) {
730 struct ceph_inode_frag frag;
731 int found;
732
733 ceph_choose_frag(ci, hash, &frag, &found);
734 if (found) {
735 if (mode == USE_ANY_MDS && frag.ndist > 0) {
736 u8 r;
737
738 /* choose a random replica */
739 get_random_bytes(&r, 1);
740 r %= frag.ndist;
741 mds = frag.dist[r];
742 dout("choose_mds %p %llx.%llx "
743 "frag %u mds%d (%d/%d)\n",
744 inode, ceph_vinop(inode),
Sage Weild66bbd42011-01-21 21:16:46 -0800745 frag.frag, mds,
Sage Weil2f2dc052009-10-06 11:31:09 -0700746 (int)r, frag.ndist);
Sage Weild66bbd42011-01-21 21:16:46 -0800747 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
748 CEPH_MDS_STATE_ACTIVE)
749 return mds;
Sage Weil2f2dc052009-10-06 11:31:09 -0700750 }
751
752 /* since this file/dir wasn't known to be
753 * replicated, then we want to look for the
754 * authoritative mds. */
755 mode = USE_AUTH_MDS;
756 if (frag.mds >= 0) {
757 /* choose auth mds */
758 mds = frag.mds;
759 dout("choose_mds %p %llx.%llx "
760 "frag %u mds%d (auth)\n",
761 inode, ceph_vinop(inode), frag.frag, mds);
Sage Weild66bbd42011-01-21 21:16:46 -0800762 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
763 CEPH_MDS_STATE_ACTIVE)
764 return mds;
Sage Weil2f2dc052009-10-06 11:31:09 -0700765 }
766 }
767 }
768
Sage Weilbe655592011-11-30 09:47:09 -0800769 spin_lock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700770 cap = NULL;
771 if (mode == USE_AUTH_MDS)
772 cap = ci->i_auth_cap;
773 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
774 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
775 if (!cap) {
Sage Weilbe655592011-11-30 09:47:09 -0800776 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700777 goto random;
778 }
779 mds = cap->session->s_mds;
780 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
781 inode, ceph_vinop(inode), mds,
782 cap == ci->i_auth_cap ? "auth " : "", cap);
Sage Weilbe655592011-11-30 09:47:09 -0800783 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700784 return mds;
785
786random:
787 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
788 dout("choose_mds chose random mds%d\n", mds);
789 return mds;
790}
791
792
793/*
794 * session messages
795 */
796static struct ceph_msg *create_session_msg(u32 op, u64 seq)
797{
798 struct ceph_msg *msg;
799 struct ceph_mds_session_head *h;
800
Sage Weilb61c2762011-08-09 15:03:46 -0700801 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
802 false);
Sage Weila79832f2010-04-01 16:06:19 -0700803 if (!msg) {
Sage Weil2f2dc052009-10-06 11:31:09 -0700804 pr_err("create_session_msg ENOMEM creating msg\n");
Sage Weila79832f2010-04-01 16:06:19 -0700805 return NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700806 }
807 h = msg->front.iov_base;
808 h->op = cpu_to_le32(op);
809 h->seq = cpu_to_le64(seq);
810 return msg;
811}
812
813/*
814 * send session open request.
815 *
816 * called under mdsc->mutex
817 */
818static int __open_session(struct ceph_mds_client *mdsc,
819 struct ceph_mds_session *session)
820{
821 struct ceph_msg *msg;
822 int mstate;
823 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -0700824
825 /* wait for mds to go active? */
826 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
827 dout("open_session to mds%d (%s)\n", mds,
828 ceph_mds_state_name(mstate));
829 session->s_state = CEPH_MDS_SESSION_OPENING;
830 session->s_renew_requested = jiffies;
831
832 /* send connect message */
833 msg = create_session_msg(CEPH_SESSION_REQUEST_OPEN, session->s_seq);
Sage Weila79832f2010-04-01 16:06:19 -0700834 if (!msg)
835 return -ENOMEM;
Sage Weil2f2dc052009-10-06 11:31:09 -0700836 ceph_con_send(&session->s_con, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -0700837 return 0;
838}
839
840/*
Sage Weiled0552a2010-06-21 13:38:25 -0700841 * open sessions for any export targets for the given mds
842 *
843 * called under mdsc->mutex
844 */
845static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
846 struct ceph_mds_session *session)
847{
848 struct ceph_mds_info *mi;
849 struct ceph_mds_session *ts;
850 int i, mds = session->s_mds;
851 int target;
852
853 if (mds >= mdsc->mdsmap->m_max_mds)
854 return;
855 mi = &mdsc->mdsmap->m_info[mds];
856 dout("open_export_target_sessions for mds%d (%d targets)\n",
857 session->s_mds, mi->num_export_targets);
858
859 for (i = 0; i < mi->num_export_targets; i++) {
860 target = mi->export_targets[i];
861 ts = __ceph_lookup_mds_session(mdsc, target);
862 if (!ts) {
863 ts = register_session(mdsc, target);
864 if (IS_ERR(ts))
865 return;
866 }
867 if (session->s_state == CEPH_MDS_SESSION_NEW ||
868 session->s_state == CEPH_MDS_SESSION_CLOSING)
869 __open_session(mdsc, session);
870 else
871 dout(" mds%d target mds%d %p is %s\n", session->s_mds,
872 i, ts, session_state_name(ts->s_state));
873 ceph_put_mds_session(ts);
874 }
875}
876
Sage Weil154f42c2010-06-21 13:45:04 -0700877void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
878 struct ceph_mds_session *session)
879{
880 mutex_lock(&mdsc->mutex);
881 __open_export_target_sessions(mdsc, session);
882 mutex_unlock(&mdsc->mutex);
883}
884
Sage Weiled0552a2010-06-21 13:38:25 -0700885/*
Sage Weil2f2dc052009-10-06 11:31:09 -0700886 * session caps
887 */
888
889/*
890 * Free preallocated cap messages assigned to this session
891 */
892static void cleanup_cap_releases(struct ceph_mds_session *session)
893{
894 struct ceph_msg *msg;
895
896 spin_lock(&session->s_cap_lock);
897 while (!list_empty(&session->s_cap_releases)) {
898 msg = list_first_entry(&session->s_cap_releases,
899 struct ceph_msg, list_head);
900 list_del_init(&msg->list_head);
901 ceph_msg_put(msg);
902 }
903 while (!list_empty(&session->s_cap_releases_done)) {
904 msg = list_first_entry(&session->s_cap_releases_done,
905 struct ceph_msg, list_head);
906 list_del_init(&msg->list_head);
907 ceph_msg_put(msg);
908 }
909 spin_unlock(&session->s_cap_lock);
910}
911
912/*
Sage Weilf818a732010-05-11 20:56:31 -0700913 * Helper to safely iterate over all caps associated with a session, with
914 * special care taken to handle a racing __ceph_remove_cap().
Sage Weil2f2dc052009-10-06 11:31:09 -0700915 *
Sage Weilf818a732010-05-11 20:56:31 -0700916 * Caller must hold session s_mutex.
Sage Weil2f2dc052009-10-06 11:31:09 -0700917 */
918static int iterate_session_caps(struct ceph_mds_session *session,
919 int (*cb)(struct inode *, struct ceph_cap *,
920 void *), void *arg)
921{
Sage Weil7c1332b2010-02-16 11:39:45 -0800922 struct list_head *p;
923 struct ceph_cap *cap;
924 struct inode *inode, *last_inode = NULL;
925 struct ceph_cap *old_cap = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700926 int ret;
927
928 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
929 spin_lock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800930 p = session->s_caps.next;
931 while (p != &session->s_caps) {
932 cap = list_entry(p, struct ceph_cap, session_caps);
Sage Weil2f2dc052009-10-06 11:31:09 -0700933 inode = igrab(&cap->ci->vfs_inode);
Sage Weil7c1332b2010-02-16 11:39:45 -0800934 if (!inode) {
935 p = p->next;
Sage Weil2f2dc052009-10-06 11:31:09 -0700936 continue;
Sage Weil7c1332b2010-02-16 11:39:45 -0800937 }
938 session->s_cap_iterator = cap;
Sage Weil2f2dc052009-10-06 11:31:09 -0700939 spin_unlock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800940
941 if (last_inode) {
942 iput(last_inode);
943 last_inode = NULL;
944 }
945 if (old_cap) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700946 ceph_put_cap(session->s_mdsc, old_cap);
Sage Weil7c1332b2010-02-16 11:39:45 -0800947 old_cap = NULL;
948 }
949
Sage Weil2f2dc052009-10-06 11:31:09 -0700950 ret = cb(inode, cap, arg);
Sage Weil7c1332b2010-02-16 11:39:45 -0800951 last_inode = inode;
952
Sage Weil2f2dc052009-10-06 11:31:09 -0700953 spin_lock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800954 p = p->next;
955 if (cap->ci == NULL) {
956 dout("iterate_session_caps finishing cap %p removal\n",
957 cap);
958 BUG_ON(cap->session != session);
959 list_del_init(&cap->session_caps);
960 session->s_nr_caps--;
961 cap->session = NULL;
962 old_cap = cap; /* put_cap it w/o locks held */
963 }
Sage Weil5dacf092009-12-21 20:40:34 -0800964 if (ret < 0)
965 goto out;
Sage Weil2f2dc052009-10-06 11:31:09 -0700966 }
Sage Weil5dacf092009-12-21 20:40:34 -0800967 ret = 0;
968out:
Sage Weil7c1332b2010-02-16 11:39:45 -0800969 session->s_cap_iterator = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700970 spin_unlock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800971
972 if (last_inode)
973 iput(last_inode);
974 if (old_cap)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700975 ceph_put_cap(session->s_mdsc, old_cap);
Sage Weil7c1332b2010-02-16 11:39:45 -0800976
Sage Weil5dacf092009-12-21 20:40:34 -0800977 return ret;
Sage Weil2f2dc052009-10-06 11:31:09 -0700978}
979
980static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
Sage Weil6c99f252010-05-10 16:12:25 -0700981 void *arg)
Sage Weil2f2dc052009-10-06 11:31:09 -0700982{
983 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil6c99f252010-05-10 16:12:25 -0700984 int drop = 0;
985
Sage Weil2f2dc052009-10-06 11:31:09 -0700986 dout("removing cap %p, ci is %p, inode is %p\n",
987 cap, ci, &ci->vfs_inode);
Sage Weilbe655592011-11-30 09:47:09 -0800988 spin_lock(&ci->i_ceph_lock);
Yan, Zhenga096b092013-09-22 10:15:58 +0800989 __ceph_remove_cap(cap, false);
Sage Weil6c99f252010-05-10 16:12:25 -0700990 if (!__ceph_is_any_real_caps(ci)) {
991 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700992 ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weil6c99f252010-05-10 16:12:25 -0700993
994 spin_lock(&mdsc->cap_dirty_lock);
995 if (!list_empty(&ci->i_dirty_item)) {
996 pr_info(" dropping dirty %s state for %p %lld\n",
997 ceph_cap_string(ci->i_dirty_caps),
998 inode, ceph_ino(inode));
999 ci->i_dirty_caps = 0;
1000 list_del_init(&ci->i_dirty_item);
1001 drop = 1;
1002 }
1003 if (!list_empty(&ci->i_flushing_item)) {
1004 pr_info(" dropping dirty+flushing %s state for %p %lld\n",
1005 ceph_cap_string(ci->i_flushing_caps),
1006 inode, ceph_ino(inode));
1007 ci->i_flushing_caps = 0;
1008 list_del_init(&ci->i_flushing_item);
1009 mdsc->num_cap_flushing--;
1010 drop = 1;
1011 }
1012 if (drop && ci->i_wrbuffer_ref) {
1013 pr_info(" dropping dirty data for %p %lld\n",
1014 inode, ceph_ino(inode));
1015 ci->i_wrbuffer_ref = 0;
1016 ci->i_wrbuffer_ref_head = 0;
1017 drop++;
1018 }
1019 spin_unlock(&mdsc->cap_dirty_lock);
1020 }
Sage Weilbe655592011-11-30 09:47:09 -08001021 spin_unlock(&ci->i_ceph_lock);
Sage Weil6c99f252010-05-10 16:12:25 -07001022 while (drop--)
1023 iput(inode);
Sage Weil2f2dc052009-10-06 11:31:09 -07001024 return 0;
1025}
1026
1027/*
1028 * caller must hold session s_mutex
1029 */
1030static void remove_session_caps(struct ceph_mds_session *session)
1031{
1032 dout("remove_session_caps on %p\n", session);
1033 iterate_session_caps(session, remove_session_caps_cb, NULL);
Yan, Zheng6f60f882013-07-24 12:22:11 +08001034
1035 spin_lock(&session->s_cap_lock);
1036 if (session->s_nr_caps > 0) {
1037 struct super_block *sb = session->s_mdsc->fsc->sb;
1038 struct inode *inode;
1039 struct ceph_cap *cap, *prev = NULL;
1040 struct ceph_vino vino;
1041 /*
1042 * iterate_session_caps() skips inodes that are being
1043 * deleted, we need to wait until deletions are complete.
1044 * __wait_on_freeing_inode() is designed for the job,
1045 * but it is not exported, so use lookup inode function
1046 * to access it.
1047 */
1048 while (!list_empty(&session->s_caps)) {
1049 cap = list_entry(session->s_caps.next,
1050 struct ceph_cap, session_caps);
1051 if (cap == prev)
1052 break;
1053 prev = cap;
1054 vino = cap->ci->i_vino;
1055 spin_unlock(&session->s_cap_lock);
1056
Yan, Zhenged284c42013-09-02 15:19:53 +08001057 inode = ceph_find_inode(sb, vino);
Yan, Zheng6f60f882013-07-24 12:22:11 +08001058 iput(inode);
1059
1060 spin_lock(&session->s_cap_lock);
1061 }
1062 }
1063 spin_unlock(&session->s_cap_lock);
1064
Sage Weil2f2dc052009-10-06 11:31:09 -07001065 BUG_ON(session->s_nr_caps > 0);
Sage Weil6c99f252010-05-10 16:12:25 -07001066 BUG_ON(!list_empty(&session->s_cap_flushing));
Sage Weil2f2dc052009-10-06 11:31:09 -07001067 cleanup_cap_releases(session);
1068}
1069
1070/*
1071 * wake up any threads waiting on this session's caps. if the cap is
1072 * old (didn't get renewed on the client reconnect), remove it now.
1073 *
1074 * caller must hold s_mutex.
1075 */
1076static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1077 void *arg)
1078{
Sage Weil0dc25702009-11-20 13:43:45 -08001079 struct ceph_inode_info *ci = ceph_inode(inode);
1080
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001081 wake_up_all(&ci->i_cap_wq);
Sage Weil0dc25702009-11-20 13:43:45 -08001082 if (arg) {
Sage Weilbe655592011-11-30 09:47:09 -08001083 spin_lock(&ci->i_ceph_lock);
Sage Weil0dc25702009-11-20 13:43:45 -08001084 ci->i_wanted_max_size = 0;
1085 ci->i_requested_max_size = 0;
Sage Weilbe655592011-11-30 09:47:09 -08001086 spin_unlock(&ci->i_ceph_lock);
Sage Weil0dc25702009-11-20 13:43:45 -08001087 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001088 return 0;
1089}
1090
Sage Weil0dc25702009-11-20 13:43:45 -08001091static void wake_up_session_caps(struct ceph_mds_session *session,
1092 int reconnect)
Sage Weil2f2dc052009-10-06 11:31:09 -07001093{
1094 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
Sage Weil0dc25702009-11-20 13:43:45 -08001095 iterate_session_caps(session, wake_up_session_cb,
1096 (void *)(unsigned long)reconnect);
Sage Weil2f2dc052009-10-06 11:31:09 -07001097}
1098
1099/*
1100 * Send periodic message to MDS renewing all currently held caps. The
1101 * ack will reset the expiration for all caps from this session.
1102 *
1103 * caller holds s_mutex
1104 */
1105static int send_renew_caps(struct ceph_mds_client *mdsc,
1106 struct ceph_mds_session *session)
1107{
1108 struct ceph_msg *msg;
1109 int state;
1110
1111 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1112 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1113 pr_info("mds%d caps stale\n", session->s_mds);
Sage Weile4cb4cb2010-03-18 13:43:09 -07001114 session->s_renew_requested = jiffies;
Sage Weil2f2dc052009-10-06 11:31:09 -07001115
1116 /* do not try to renew caps until a recovering mds has reconnected
1117 * with its clients. */
1118 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1119 if (state < CEPH_MDS_STATE_RECONNECT) {
1120 dout("send_renew_caps ignoring mds%d (%s)\n",
1121 session->s_mds, ceph_mds_state_name(state));
1122 return 0;
1123 }
1124
1125 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1126 ceph_mds_state_name(state));
Sage Weil2f2dc052009-10-06 11:31:09 -07001127 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1128 ++session->s_renew_seq);
Sage Weila79832f2010-04-01 16:06:19 -07001129 if (!msg)
1130 return -ENOMEM;
Sage Weil2f2dc052009-10-06 11:31:09 -07001131 ceph_con_send(&session->s_con, msg);
1132 return 0;
1133}
1134
1135/*
1136 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
Sage Weil0dc25702009-11-20 13:43:45 -08001137 *
1138 * Called under session->s_mutex
Sage Weil2f2dc052009-10-06 11:31:09 -07001139 */
1140static void renewed_caps(struct ceph_mds_client *mdsc,
1141 struct ceph_mds_session *session, int is_renew)
1142{
1143 int was_stale;
1144 int wake = 0;
1145
1146 spin_lock(&session->s_cap_lock);
Alex Elder1ce208a2012-01-12 17:48:11 -08001147 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
Sage Weil2f2dc052009-10-06 11:31:09 -07001148
1149 session->s_cap_ttl = session->s_renew_requested +
1150 mdsc->mdsmap->m_session_timeout*HZ;
1151
1152 if (was_stale) {
1153 if (time_before(jiffies, session->s_cap_ttl)) {
1154 pr_info("mds%d caps renewed\n", session->s_mds);
1155 wake = 1;
1156 } else {
1157 pr_info("mds%d caps still stale\n", session->s_mds);
1158 }
1159 }
1160 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1161 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1162 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1163 spin_unlock(&session->s_cap_lock);
1164
1165 if (wake)
Sage Weil0dc25702009-11-20 13:43:45 -08001166 wake_up_session_caps(session, 0);
Sage Weil2f2dc052009-10-06 11:31:09 -07001167}
1168
1169/*
1170 * send a session close request
1171 */
1172static int request_close_session(struct ceph_mds_client *mdsc,
1173 struct ceph_mds_session *session)
1174{
1175 struct ceph_msg *msg;
Sage Weil2f2dc052009-10-06 11:31:09 -07001176
1177 dout("request_close_session mds%d state %s seq %lld\n",
1178 session->s_mds, session_state_name(session->s_state),
1179 session->s_seq);
1180 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
Sage Weila79832f2010-04-01 16:06:19 -07001181 if (!msg)
1182 return -ENOMEM;
1183 ceph_con_send(&session->s_con, msg);
1184 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001185}
1186
1187/*
1188 * Called with s_mutex held.
1189 */
1190static int __close_session(struct ceph_mds_client *mdsc,
1191 struct ceph_mds_session *session)
1192{
1193 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1194 return 0;
1195 session->s_state = CEPH_MDS_SESSION_CLOSING;
1196 return request_close_session(mdsc, session);
1197}
1198
1199/*
1200 * Trim old(er) caps.
1201 *
1202 * Because we can't cache an inode without one or more caps, we do
1203 * this indirectly: if a cap is unused, we prune its aliases, at which
1204 * point the inode will hopefully get dropped to.
1205 *
1206 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1207 * memory pressure from the MDS, though, so it needn't be perfect.
1208 */
1209static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1210{
1211 struct ceph_mds_session *session = arg;
1212 struct ceph_inode_info *ci = ceph_inode(inode);
1213 int used, oissued, mine;
1214
1215 if (session->s_trim_caps <= 0)
1216 return -1;
1217
Sage Weilbe655592011-11-30 09:47:09 -08001218 spin_lock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001219 mine = cap->issued | cap->implemented;
1220 used = __ceph_caps_used(ci);
1221 oissued = __ceph_caps_issued_other(ci, cap);
1222
1223 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s\n",
1224 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1225 ceph_cap_string(used));
1226 if (ci->i_dirty_caps)
1227 goto out; /* dirty caps */
1228 if ((used & ~oissued) & mine)
1229 goto out; /* we need these caps */
1230
1231 session->s_trim_caps--;
1232 if (oissued) {
1233 /* we aren't the only cap.. just remove us */
Yan, Zhenga096b092013-09-22 10:15:58 +08001234 __ceph_remove_cap(cap, true);
Sage Weil2f2dc052009-10-06 11:31:09 -07001235 } else {
1236 /* try to drop referring dentries */
Sage Weilbe655592011-11-30 09:47:09 -08001237 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001238 d_prune_aliases(inode);
1239 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1240 inode, cap, atomic_read(&inode->i_count));
1241 return 0;
1242 }
1243
1244out:
Sage Weilbe655592011-11-30 09:47:09 -08001245 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001246 return 0;
1247}
1248
1249/*
1250 * Trim session cap count down to some max number.
1251 */
1252static int trim_caps(struct ceph_mds_client *mdsc,
1253 struct ceph_mds_session *session,
1254 int max_caps)
1255{
1256 int trim_caps = session->s_nr_caps - max_caps;
1257
1258 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1259 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1260 if (trim_caps > 0) {
1261 session->s_trim_caps = trim_caps;
1262 iterate_session_caps(session, trim_caps_cb, session);
1263 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1264 session->s_mds, session->s_nr_caps, max_caps,
1265 trim_caps - session->s_trim_caps);
Sage Weil5dacf092009-12-21 20:40:34 -08001266 session->s_trim_caps = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001267 }
1268 return 0;
1269}
1270
1271/*
1272 * Allocate cap_release messages. If there is a partially full message
1273 * in the queue, try to allocate enough to cover it's remainder, so that
1274 * we can send it immediately.
1275 *
1276 * Called under s_mutex.
1277 */
Sage Weil2b2300d2010-06-09 16:52:04 -07001278int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
Sage Weilee6b2722010-06-10 12:55:52 -07001279 struct ceph_mds_session *session)
Sage Weil2f2dc052009-10-06 11:31:09 -07001280{
Sage Weil38e8883e2010-06-10 12:57:11 -07001281 struct ceph_msg *msg, *partial = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -07001282 struct ceph_mds_cap_release *head;
1283 int err = -ENOMEM;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001284 int extra = mdsc->fsc->mount_options->cap_release_safety;
Sage Weil38e8883e2010-06-10 12:57:11 -07001285 int num;
Sage Weil2f2dc052009-10-06 11:31:09 -07001286
Sage Weil38e8883e2010-06-10 12:57:11 -07001287 dout("add_cap_releases %p mds%d extra %d\n", session, session->s_mds,
1288 extra);
Sage Weil2f2dc052009-10-06 11:31:09 -07001289
1290 spin_lock(&session->s_cap_lock);
1291
1292 if (!list_empty(&session->s_cap_releases)) {
1293 msg = list_first_entry(&session->s_cap_releases,
1294 struct ceph_msg,
1295 list_head);
1296 head = msg->front.iov_base;
Sage Weil38e8883e2010-06-10 12:57:11 -07001297 num = le32_to_cpu(head->num);
1298 if (num) {
1299 dout(" partial %p with (%d/%d)\n", msg, num,
1300 (int)CEPH_CAPS_PER_RELEASE);
1301 extra += CEPH_CAPS_PER_RELEASE - num;
1302 partial = msg;
1303 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001304 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001305 while (session->s_num_cap_releases < session->s_nr_caps + extra) {
1306 spin_unlock(&session->s_cap_lock);
Yehuda Sadeh34d23762010-04-06 14:33:58 -07001307 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
Sage Weilb61c2762011-08-09 15:03:46 -07001308 GFP_NOFS, false);
Sage Weil2f2dc052009-10-06 11:31:09 -07001309 if (!msg)
1310 goto out_unlocked;
1311 dout("add_cap_releases %p msg %p now %d\n", session, msg,
1312 (int)msg->front.iov_len);
1313 head = msg->front.iov_base;
1314 head->num = cpu_to_le32(0);
1315 msg->front.iov_len = sizeof(*head);
1316 spin_lock(&session->s_cap_lock);
1317 list_add(&msg->list_head, &session->s_cap_releases);
1318 session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE;
1319 }
1320
Sage Weil38e8883e2010-06-10 12:57:11 -07001321 if (partial) {
1322 head = partial->front.iov_base;
1323 num = le32_to_cpu(head->num);
1324 dout(" queueing partial %p with %d/%d\n", partial, num,
1325 (int)CEPH_CAPS_PER_RELEASE);
1326 list_move_tail(&partial->list_head,
1327 &session->s_cap_releases_done);
1328 session->s_num_cap_releases -= CEPH_CAPS_PER_RELEASE - num;
Sage Weil2f2dc052009-10-06 11:31:09 -07001329 }
1330 err = 0;
1331 spin_unlock(&session->s_cap_lock);
1332out_unlocked:
1333 return err;
1334}
1335
1336/*
1337 * flush all dirty inode data to disk.
1338 *
1339 * returns true if we've flushed through want_flush_seq
1340 */
1341static int check_cap_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq)
1342{
1343 int mds, ret = 1;
1344
1345 dout("check_cap_flush want %lld\n", want_flush_seq);
1346 mutex_lock(&mdsc->mutex);
1347 for (mds = 0; ret && mds < mdsc->max_sessions; mds++) {
1348 struct ceph_mds_session *session = mdsc->sessions[mds];
1349
1350 if (!session)
1351 continue;
1352 get_session(session);
1353 mutex_unlock(&mdsc->mutex);
1354
1355 mutex_lock(&session->s_mutex);
1356 if (!list_empty(&session->s_cap_flushing)) {
1357 struct ceph_inode_info *ci =
1358 list_entry(session->s_cap_flushing.next,
1359 struct ceph_inode_info,
1360 i_flushing_item);
1361 struct inode *inode = &ci->vfs_inode;
1362
Sage Weilbe655592011-11-30 09:47:09 -08001363 spin_lock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001364 if (ci->i_cap_flush_seq <= want_flush_seq) {
1365 dout("check_cap_flush still flushing %p "
1366 "seq %lld <= %lld to mds%d\n", inode,
1367 ci->i_cap_flush_seq, want_flush_seq,
1368 session->s_mds);
1369 ret = 0;
1370 }
Sage Weilbe655592011-11-30 09:47:09 -08001371 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001372 }
1373 mutex_unlock(&session->s_mutex);
1374 ceph_put_mds_session(session);
1375
1376 if (!ret)
1377 return ret;
1378 mutex_lock(&mdsc->mutex);
1379 }
1380
1381 mutex_unlock(&mdsc->mutex);
1382 dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
1383 return ret;
1384}
1385
1386/*
1387 * called under s_mutex
1388 */
Sage Weil3d7ded42010-06-09 16:47:10 -07001389void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1390 struct ceph_mds_session *session)
Sage Weil2f2dc052009-10-06 11:31:09 -07001391{
1392 struct ceph_msg *msg;
1393
1394 dout("send_cap_releases mds%d\n", session->s_mds);
Sage Weil0f8605f2010-05-05 15:51:35 -07001395 spin_lock(&session->s_cap_lock);
1396 while (!list_empty(&session->s_cap_releases_done)) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001397 msg = list_first_entry(&session->s_cap_releases_done,
1398 struct ceph_msg, list_head);
1399 list_del_init(&msg->list_head);
1400 spin_unlock(&session->s_cap_lock);
1401 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1402 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1403 ceph_con_send(&session->s_con, msg);
Sage Weil0f8605f2010-05-05 15:51:35 -07001404 spin_lock(&session->s_cap_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001405 }
1406 spin_unlock(&session->s_cap_lock);
1407}
1408
Sage Weile01a5942010-05-10 15:36:44 -07001409static void discard_cap_releases(struct ceph_mds_client *mdsc,
1410 struct ceph_mds_session *session)
1411{
1412 struct ceph_msg *msg;
1413 struct ceph_mds_cap_release *head;
1414 unsigned num;
1415
1416 dout("discard_cap_releases mds%d\n", session->s_mds);
1417 spin_lock(&session->s_cap_lock);
1418
1419 /* zero out the in-progress message */
1420 msg = list_first_entry(&session->s_cap_releases,
1421 struct ceph_msg, list_head);
1422 head = msg->front.iov_base;
1423 num = le32_to_cpu(head->num);
1424 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg, num);
1425 head->num = cpu_to_le32(0);
Yan, Zheng3803da42013-05-31 16:26:44 +08001426 msg->front.iov_len = sizeof(*head);
Sage Weile01a5942010-05-10 15:36:44 -07001427 session->s_num_cap_releases += num;
1428
1429 /* requeue completed messages */
1430 while (!list_empty(&session->s_cap_releases_done)) {
1431 msg = list_first_entry(&session->s_cap_releases_done,
1432 struct ceph_msg, list_head);
1433 list_del_init(&msg->list_head);
1434
1435 head = msg->front.iov_base;
1436 num = le32_to_cpu(head->num);
1437 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg,
1438 num);
1439 session->s_num_cap_releases += num;
1440 head->num = cpu_to_le32(0);
1441 msg->front.iov_len = sizeof(*head);
1442 list_add(&msg->list_head, &session->s_cap_releases);
1443 }
1444
1445 spin_unlock(&session->s_cap_lock);
1446}
1447
Sage Weil2f2dc052009-10-06 11:31:09 -07001448/*
1449 * requests
1450 */
1451
1452/*
1453 * Create an mds request.
1454 */
1455struct ceph_mds_request *
1456ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1457{
1458 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1459
1460 if (!req)
1461 return ERR_PTR(-ENOMEM);
1462
Sage Weilb4556392010-05-13 12:01:13 -07001463 mutex_init(&req->r_fill_mutex);
Yehuda Sadeh37151662010-06-17 16:16:12 -07001464 req->r_mdsc = mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07001465 req->r_started = jiffies;
1466 req->r_resend_mds = -1;
1467 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1468 req->r_fmode = -1;
Sage Weil153c8e62009-12-07 12:31:09 -08001469 kref_init(&req->r_kref);
Sage Weil2f2dc052009-10-06 11:31:09 -07001470 INIT_LIST_HEAD(&req->r_wait);
1471 init_completion(&req->r_completion);
1472 init_completion(&req->r_safe_completion);
1473 INIT_LIST_HEAD(&req->r_unsafe_item);
1474
1475 req->r_op = op;
1476 req->r_direct_mode = mode;
1477 return req;
1478}
1479
1480/*
Sage Weil44ca18f2010-02-15 12:08:46 -08001481 * return oldest (lowest) request, tid in request tree, 0 if none.
Sage Weil2f2dc052009-10-06 11:31:09 -07001482 *
1483 * called under mdsc->mutex.
1484 */
Sage Weil44ca18f2010-02-15 12:08:46 -08001485static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1486{
1487 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1488 return NULL;
1489 return rb_entry(rb_first(&mdsc->request_tree),
1490 struct ceph_mds_request, r_node);
1491}
1492
Sage Weil2f2dc052009-10-06 11:31:09 -07001493static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1494{
Sage Weil44ca18f2010-02-15 12:08:46 -08001495 struct ceph_mds_request *req = __get_oldest_req(mdsc);
1496
1497 if (req)
1498 return req->r_tid;
1499 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001500}
1501
1502/*
1503 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1504 * on build_path_from_dentry in fs/cifs/dir.c.
1505 *
1506 * If @stop_on_nosnap, generate path relative to the first non-snapped
1507 * inode.
1508 *
1509 * Encode hidden .snap dirs as a double /, i.e.
1510 * foo/.snap/bar -> foo//bar
1511 */
1512char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1513 int stop_on_nosnap)
1514{
1515 struct dentry *temp;
1516 char *path;
1517 int len, pos;
Al Viro1b71fe22011-07-16 23:43:58 -04001518 unsigned seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07001519
1520 if (dentry == NULL)
1521 return ERR_PTR(-EINVAL);
1522
1523retry:
1524 len = 0;
Al Viro1b71fe22011-07-16 23:43:58 -04001525 seq = read_seqbegin(&rename_lock);
1526 rcu_read_lock();
Sage Weil2f2dc052009-10-06 11:31:09 -07001527 for (temp = dentry; !IS_ROOT(temp);) {
1528 struct inode *inode = temp->d_inode;
1529 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1530 len++; /* slash only */
1531 else if (stop_on_nosnap && inode &&
1532 ceph_snap(inode) == CEPH_NOSNAP)
1533 break;
1534 else
1535 len += 1 + temp->d_name.len;
1536 temp = temp->d_parent;
Sage Weil2f2dc052009-10-06 11:31:09 -07001537 }
Al Viro1b71fe22011-07-16 23:43:58 -04001538 rcu_read_unlock();
Sage Weil2f2dc052009-10-06 11:31:09 -07001539 if (len)
1540 len--; /* no leading '/' */
1541
1542 path = kmalloc(len+1, GFP_NOFS);
1543 if (path == NULL)
1544 return ERR_PTR(-ENOMEM);
1545 pos = len;
1546 path[pos] = 0; /* trailing null */
Al Viro1b71fe22011-07-16 23:43:58 -04001547 rcu_read_lock();
Sage Weil2f2dc052009-10-06 11:31:09 -07001548 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
Al Viro1b71fe22011-07-16 23:43:58 -04001549 struct inode *inode;
Sage Weil2f2dc052009-10-06 11:31:09 -07001550
Al Viro1b71fe22011-07-16 23:43:58 -04001551 spin_lock(&temp->d_lock);
1552 inode = temp->d_inode;
Sage Weil2f2dc052009-10-06 11:31:09 -07001553 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
Sage Weil104648a2010-03-18 10:14:30 -07001554 dout("build_path path+%d: %p SNAPDIR\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07001555 pos, temp);
1556 } else if (stop_on_nosnap && inode &&
1557 ceph_snap(inode) == CEPH_NOSNAP) {
Yehuda Sadeh9d5a09e2011-12-13 09:57:44 -08001558 spin_unlock(&temp->d_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001559 break;
1560 } else {
1561 pos -= temp->d_name.len;
Al Viro1b71fe22011-07-16 23:43:58 -04001562 if (pos < 0) {
1563 spin_unlock(&temp->d_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001564 break;
Al Viro1b71fe22011-07-16 23:43:58 -04001565 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001566 strncpy(path + pos, temp->d_name.name,
1567 temp->d_name.len);
Sage Weil2f2dc052009-10-06 11:31:09 -07001568 }
Al Viro1b71fe22011-07-16 23:43:58 -04001569 spin_unlock(&temp->d_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001570 if (pos)
1571 path[--pos] = '/';
1572 temp = temp->d_parent;
Sage Weil2f2dc052009-10-06 11:31:09 -07001573 }
Al Viro1b71fe22011-07-16 23:43:58 -04001574 rcu_read_unlock();
1575 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
Sage Weil104648a2010-03-18 10:14:30 -07001576 pr_err("build_path did not end path lookup where "
Sage Weil2f2dc052009-10-06 11:31:09 -07001577 "expected, namelen is %d, pos is %d\n", len, pos);
1578 /* presumably this is only possible if racing with a
1579 rename of one of the parent directories (we can not
1580 lock the dentries above us to prevent this, but
1581 retrying should be harmless) */
1582 kfree(path);
1583 goto retry;
1584 }
1585
1586 *base = ceph_ino(temp->d_inode);
1587 *plen = len;
Sage Weil104648a2010-03-18 10:14:30 -07001588 dout("build_path on %p %d built %llx '%.*s'\n",
Al Viro84d08fa2013-07-05 18:59:33 +04001589 dentry, d_count(dentry), *base, len, path);
Sage Weil2f2dc052009-10-06 11:31:09 -07001590 return path;
1591}
1592
1593static int build_dentry_path(struct dentry *dentry,
1594 const char **ppath, int *ppathlen, u64 *pino,
1595 int *pfreepath)
1596{
1597 char *path;
1598
1599 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) {
1600 *pino = ceph_ino(dentry->d_parent->d_inode);
1601 *ppath = dentry->d_name.name;
1602 *ppathlen = dentry->d_name.len;
1603 return 0;
1604 }
1605 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1606 if (IS_ERR(path))
1607 return PTR_ERR(path);
1608 *ppath = path;
1609 *pfreepath = 1;
1610 return 0;
1611}
1612
1613static int build_inode_path(struct inode *inode,
1614 const char **ppath, int *ppathlen, u64 *pino,
1615 int *pfreepath)
1616{
1617 struct dentry *dentry;
1618 char *path;
1619
1620 if (ceph_snap(inode) == CEPH_NOSNAP) {
1621 *pino = ceph_ino(inode);
1622 *ppathlen = 0;
1623 return 0;
1624 }
1625 dentry = d_find_alias(inode);
1626 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1627 dput(dentry);
1628 if (IS_ERR(path))
1629 return PTR_ERR(path);
1630 *ppath = path;
1631 *pfreepath = 1;
1632 return 0;
1633}
1634
1635/*
1636 * request arguments may be specified via an inode *, a dentry *, or
1637 * an explicit ino+path.
1638 */
1639static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1640 const char *rpath, u64 rino,
1641 const char **ppath, int *pathlen,
1642 u64 *ino, int *freepath)
1643{
1644 int r = 0;
1645
1646 if (rinode) {
1647 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1648 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1649 ceph_snap(rinode));
1650 } else if (rdentry) {
1651 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1652 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1653 *ppath);
Sage Weil795858d2011-08-15 13:02:37 -07001654 } else if (rpath || rino) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001655 *ino = rino;
1656 *ppath = rpath;
David Zafmanb0000562012-10-25 10:23:46 -07001657 *pathlen = rpath ? strlen(rpath) : 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001658 dout(" path %.*s\n", *pathlen, rpath);
1659 }
1660
1661 return r;
1662}
1663
1664/*
1665 * called under mdsc->mutex
1666 */
1667static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1668 struct ceph_mds_request *req,
1669 int mds)
1670{
1671 struct ceph_msg *msg;
1672 struct ceph_mds_request_head *head;
1673 const char *path1 = NULL;
1674 const char *path2 = NULL;
1675 u64 ino1 = 0, ino2 = 0;
1676 int pathlen1 = 0, pathlen2 = 0;
1677 int freepath1 = 0, freepath2 = 0;
1678 int len;
1679 u16 releases;
1680 void *p, *end;
1681 int ret;
1682
1683 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1684 req->r_path1, req->r_ino1.ino,
1685 &path1, &pathlen1, &ino1, &freepath1);
1686 if (ret < 0) {
1687 msg = ERR_PTR(ret);
1688 goto out;
1689 }
1690
1691 ret = set_request_path_attr(NULL, req->r_old_dentry,
1692 req->r_path2, req->r_ino2.ino,
1693 &path2, &pathlen2, &ino2, &freepath2);
1694 if (ret < 0) {
1695 msg = ERR_PTR(ret);
1696 goto out_free1;
1697 }
1698
1699 len = sizeof(*head) +
Sage Weilac8839d2010-01-27 14:28:10 -08001700 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64));
Sage Weil2f2dc052009-10-06 11:31:09 -07001701
1702 /* calculate (max) length for cap releases */
1703 len += sizeof(struct ceph_mds_request_release) *
1704 (!!req->r_inode_drop + !!req->r_dentry_drop +
1705 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1706 if (req->r_dentry_drop)
1707 len += req->r_dentry->d_name.len;
1708 if (req->r_old_dentry_drop)
1709 len += req->r_old_dentry->d_name.len;
1710
Sage Weilb61c2762011-08-09 15:03:46 -07001711 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07001712 if (!msg) {
1713 msg = ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -07001714 goto out_free2;
Sage Weila79832f2010-04-01 16:06:19 -07001715 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001716
Sage Weil6df058c2009-12-22 11:24:33 -08001717 msg->hdr.tid = cpu_to_le64(req->r_tid);
1718
Sage Weil2f2dc052009-10-06 11:31:09 -07001719 head = msg->front.iov_base;
1720 p = msg->front.iov_base + sizeof(*head);
1721 end = msg->front.iov_base + msg->front.iov_len;
1722
1723 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1724 head->op = cpu_to_le32(req->r_op);
Eric W. Biedermanff3d0042013-01-31 04:01:53 -08001725 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1726 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
Sage Weil2f2dc052009-10-06 11:31:09 -07001727 head->args = req->r_args;
1728
1729 ceph_encode_filepath(&p, end, ino1, path1);
1730 ceph_encode_filepath(&p, end, ino2, path2);
1731
Sage Weile979cf52010-07-15 14:58:39 -07001732 /* make note of release offset, in case we need to replay */
1733 req->r_request_release_offset = p - msg->front.iov_base;
1734
Sage Weil2f2dc052009-10-06 11:31:09 -07001735 /* cap releases */
1736 releases = 0;
1737 if (req->r_inode_drop)
1738 releases += ceph_encode_inode_release(&p,
1739 req->r_inode ? req->r_inode : req->r_dentry->d_inode,
1740 mds, req->r_inode_drop, req->r_inode_unless, 0);
1741 if (req->r_dentry_drop)
1742 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1743 mds, req->r_dentry_drop, req->r_dentry_unless);
1744 if (req->r_old_dentry_drop)
1745 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1746 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1747 if (req->r_old_inode_drop)
1748 releases += ceph_encode_inode_release(&p,
1749 req->r_old_dentry->d_inode,
1750 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1751 head->num_releases = cpu_to_le16(releases);
1752
1753 BUG_ON(p > end);
1754 msg->front.iov_len = p - msg->front.iov_base;
1755 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1756
Alex Elderebf18f42013-03-04 22:29:57 -06001757 if (req->r_data_len) {
1758 /* outbound data set only by ceph_sync_setxattr() */
1759 BUG_ON(!req->r_pages);
Alex Elder90af3602013-04-05 14:46:01 -05001760 ceph_msg_data_add_pages(msg, req->r_pages, req->r_data_len, 0);
Alex Elderebf18f42013-03-04 22:29:57 -06001761 }
Alex Elder02afca62013-02-14 12:16:43 -06001762
Sage Weil2f2dc052009-10-06 11:31:09 -07001763 msg->hdr.data_len = cpu_to_le32(req->r_data_len);
1764 msg->hdr.data_off = cpu_to_le16(0);
1765
1766out_free2:
1767 if (freepath2)
1768 kfree((char *)path2);
1769out_free1:
1770 if (freepath1)
1771 kfree((char *)path1);
1772out:
1773 return msg;
1774}
1775
1776/*
1777 * called under mdsc->mutex if error, under no mutex if
1778 * success.
1779 */
1780static void complete_request(struct ceph_mds_client *mdsc,
1781 struct ceph_mds_request *req)
1782{
1783 if (req->r_callback)
1784 req->r_callback(mdsc, req);
1785 else
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001786 complete_all(&req->r_completion);
Sage Weil2f2dc052009-10-06 11:31:09 -07001787}
1788
1789/*
1790 * called under mdsc->mutex
1791 */
1792static int __prepare_send_request(struct ceph_mds_client *mdsc,
1793 struct ceph_mds_request *req,
1794 int mds)
1795{
1796 struct ceph_mds_request_head *rhead;
1797 struct ceph_msg *msg;
1798 int flags = 0;
1799
Sage Weil2f2dc052009-10-06 11:31:09 -07001800 req->r_attempts++;
Greg Farnume55b71f2010-06-22 15:58:01 -07001801 if (req->r_inode) {
1802 struct ceph_cap *cap =
1803 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
1804
1805 if (cap)
1806 req->r_sent_on_mseq = cap->mseq;
1807 else
1808 req->r_sent_on_mseq = -1;
1809 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001810 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
1811 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
1812
Sage Weil01a92f12010-07-15 13:24:32 -07001813 if (req->r_got_unsafe) {
1814 /*
1815 * Replay. Do not regenerate message (and rebuild
1816 * paths, etc.); just use the original message.
1817 * Rebuilding paths will break for renames because
1818 * d_move mangles the src name.
1819 */
1820 msg = req->r_request;
1821 rhead = msg->front.iov_base;
1822
1823 flags = le32_to_cpu(rhead->flags);
1824 flags |= CEPH_MDS_FLAG_REPLAY;
1825 rhead->flags = cpu_to_le32(flags);
1826
1827 if (req->r_target_inode)
1828 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
1829
1830 rhead->num_retry = req->r_attempts - 1;
Sage Weile979cf52010-07-15 14:58:39 -07001831
1832 /* remove cap/dentry releases from message */
1833 rhead->num_releases = 0;
1834 msg->hdr.front_len = cpu_to_le32(req->r_request_release_offset);
1835 msg->front.iov_len = req->r_request_release_offset;
Sage Weil01a92f12010-07-15 13:24:32 -07001836 return 0;
1837 }
1838
Sage Weil2f2dc052009-10-06 11:31:09 -07001839 if (req->r_request) {
1840 ceph_msg_put(req->r_request);
1841 req->r_request = NULL;
1842 }
1843 msg = create_request_message(mdsc, req, mds);
1844 if (IS_ERR(msg)) {
Sage Weile1518c72010-05-13 11:19:06 -07001845 req->r_err = PTR_ERR(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07001846 complete_request(mdsc, req);
Sage Weila79832f2010-04-01 16:06:19 -07001847 return PTR_ERR(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07001848 }
1849 req->r_request = msg;
1850
1851 rhead = msg->front.iov_base;
Sage Weil2f2dc052009-10-06 11:31:09 -07001852 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
1853 if (req->r_got_unsafe)
1854 flags |= CEPH_MDS_FLAG_REPLAY;
1855 if (req->r_locked_dir)
1856 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
1857 rhead->flags = cpu_to_le32(flags);
1858 rhead->num_fwd = req->r_num_fwd;
1859 rhead->num_retry = req->r_attempts - 1;
Sage Weil01a92f12010-07-15 13:24:32 -07001860 rhead->ino = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001861
1862 dout(" r_locked_dir = %p\n", req->r_locked_dir);
Sage Weil2f2dc052009-10-06 11:31:09 -07001863 return 0;
1864}
1865
1866/*
1867 * send request, or put it on the appropriate wait list.
1868 */
1869static int __do_request(struct ceph_mds_client *mdsc,
1870 struct ceph_mds_request *req)
1871{
1872 struct ceph_mds_session *session = NULL;
1873 int mds = -1;
1874 int err = -EAGAIN;
1875
Sage Weile1518c72010-05-13 11:19:06 -07001876 if (req->r_err || req->r_got_result)
Sage Weil2f2dc052009-10-06 11:31:09 -07001877 goto out;
1878
1879 if (req->r_timeout &&
1880 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
1881 dout("do_request timed out\n");
1882 err = -EIO;
1883 goto finish;
1884 }
1885
Sage Weildc69e2e92010-11-02 13:49:00 -07001886 put_request_session(req);
1887
Sage Weil2f2dc052009-10-06 11:31:09 -07001888 mds = __choose_mds(mdsc, req);
1889 if (mds < 0 ||
1890 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
1891 dout("do_request no mds or not active, waiting for map\n");
1892 list_add(&req->r_wait, &mdsc->waiting_for_map);
1893 goto out;
1894 }
1895
1896 /* get, open session */
1897 session = __ceph_lookup_mds_session(mdsc, mds);
Sage Weil9c4239562010-03-20 20:43:28 -07001898 if (!session) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001899 session = register_session(mdsc, mds);
Sage Weil9c4239562010-03-20 20:43:28 -07001900 if (IS_ERR(session)) {
1901 err = PTR_ERR(session);
1902 goto finish;
1903 }
1904 }
Sage Weildc69e2e92010-11-02 13:49:00 -07001905 req->r_session = get_session(session);
1906
Sage Weil2f2dc052009-10-06 11:31:09 -07001907 dout("do_request mds%d session %p state %s\n", mds, session,
1908 session_state_name(session->s_state));
1909 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
1910 session->s_state != CEPH_MDS_SESSION_HUNG) {
1911 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1912 session->s_state == CEPH_MDS_SESSION_CLOSING)
1913 __open_session(mdsc, session);
1914 list_add(&req->r_wait, &session->s_waiting);
1915 goto out_session;
1916 }
1917
1918 /* send request */
Sage Weil2f2dc052009-10-06 11:31:09 -07001919 req->r_resend_mds = -1; /* forget any previous mds hint */
1920
1921 if (req->r_request_started == 0) /* note request start time */
1922 req->r_request_started = jiffies;
1923
1924 err = __prepare_send_request(mdsc, req, mds);
1925 if (!err) {
1926 ceph_msg_get(req->r_request);
1927 ceph_con_send(&session->s_con, req->r_request);
1928 }
1929
1930out_session:
1931 ceph_put_mds_session(session);
1932out:
1933 return err;
1934
1935finish:
Sage Weile1518c72010-05-13 11:19:06 -07001936 req->r_err = err;
Sage Weil2f2dc052009-10-06 11:31:09 -07001937 complete_request(mdsc, req);
1938 goto out;
1939}
1940
1941/*
1942 * called under mdsc->mutex
1943 */
1944static void __wake_requests(struct ceph_mds_client *mdsc,
1945 struct list_head *head)
1946{
Yan, Zhenged75ec22012-11-19 10:49:06 +08001947 struct ceph_mds_request *req;
1948 LIST_HEAD(tmp_list);
Sage Weil2f2dc052009-10-06 11:31:09 -07001949
Yan, Zhenged75ec22012-11-19 10:49:06 +08001950 list_splice_init(head, &tmp_list);
1951
1952 while (!list_empty(&tmp_list)) {
1953 req = list_entry(tmp_list.next,
1954 struct ceph_mds_request, r_wait);
Sage Weil2f2dc052009-10-06 11:31:09 -07001955 list_del_init(&req->r_wait);
Sage Weil7971bd92013-05-01 21:15:58 -07001956 dout(" wake request %p tid %llu\n", req, req->r_tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07001957 __do_request(mdsc, req);
1958 }
1959}
1960
1961/*
1962 * Wake up threads with requests pending for @mds, so that they can
Sage Weil29790f22010-03-18 14:45:05 -07001963 * resubmit their requests to a possibly different mds.
Sage Weil2f2dc052009-10-06 11:31:09 -07001964 */
Sage Weil29790f22010-03-18 14:45:05 -07001965static void kick_requests(struct ceph_mds_client *mdsc, int mds)
Sage Weil2f2dc052009-10-06 11:31:09 -07001966{
Sage Weil44ca18f2010-02-15 12:08:46 -08001967 struct ceph_mds_request *req;
1968 struct rb_node *p;
Sage Weil2f2dc052009-10-06 11:31:09 -07001969
1970 dout("kick_requests mds%d\n", mds);
Sage Weil44ca18f2010-02-15 12:08:46 -08001971 for (p = rb_first(&mdsc->request_tree); p; p = rb_next(p)) {
1972 req = rb_entry(p, struct ceph_mds_request, r_node);
1973 if (req->r_got_unsafe)
1974 continue;
1975 if (req->r_session &&
1976 req->r_session->s_mds == mds) {
1977 dout(" kicking tid %llu\n", req->r_tid);
Sage Weil44ca18f2010-02-15 12:08:46 -08001978 __do_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -07001979 }
1980 }
1981}
1982
1983void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
1984 struct ceph_mds_request *req)
1985{
1986 dout("submit_request on %p\n", req);
1987 mutex_lock(&mdsc->mutex);
1988 __register_request(mdsc, req, NULL);
1989 __do_request(mdsc, req);
1990 mutex_unlock(&mdsc->mutex);
1991}
1992
1993/*
1994 * Synchrously perform an mds request. Take care of all of the
1995 * session setup, forwarding, retry details.
1996 */
1997int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
1998 struct inode *dir,
1999 struct ceph_mds_request *req)
2000{
2001 int err;
2002
2003 dout("do_request on %p\n", req);
2004
2005 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2006 if (req->r_inode)
2007 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2008 if (req->r_locked_dir)
2009 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
2010 if (req->r_old_dentry)
Sage Weil41b02e12011-07-26 11:31:14 -07002011 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2012 CEPH_CAP_PIN);
Sage Weil2f2dc052009-10-06 11:31:09 -07002013
2014 /* issue */
2015 mutex_lock(&mdsc->mutex);
2016 __register_request(mdsc, req, dir);
2017 __do_request(mdsc, req);
2018
Sage Weile1518c72010-05-13 11:19:06 -07002019 if (req->r_err) {
Sage Weil2f2dc052009-10-06 11:31:09 -07002020 err = req->r_err;
Sage Weile1518c72010-05-13 11:19:06 -07002021 __unregister_request(mdsc, req);
2022 dout("do_request early error %d\n", err);
2023 goto out;
Sage Weil2f2dc052009-10-06 11:31:09 -07002024 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002025
Sage Weile1518c72010-05-13 11:19:06 -07002026 /* wait */
2027 mutex_unlock(&mdsc->mutex);
2028 dout("do_request waiting\n");
2029 if (req->r_timeout) {
Sage Weilaa916472010-05-24 11:15:51 -07002030 err = (long)wait_for_completion_killable_timeout(
Sage Weile1518c72010-05-13 11:19:06 -07002031 &req->r_completion, req->r_timeout);
2032 if (err == 0)
2033 err = -EIO;
2034 } else {
Sage Weilaa916472010-05-24 11:15:51 -07002035 err = wait_for_completion_killable(&req->r_completion);
Sage Weile1518c72010-05-13 11:19:06 -07002036 }
2037 dout("do_request waited, got %d\n", err);
2038 mutex_lock(&mdsc->mutex);
2039
2040 /* only abort if we didn't race with a real reply */
2041 if (req->r_got_result) {
2042 err = le32_to_cpu(req->r_reply_info.head->result);
2043 } else if (err < 0) {
2044 dout("aborted request %lld with %d\n", req->r_tid, err);
Sage Weilb4556392010-05-13 12:01:13 -07002045
2046 /*
2047 * ensure we aren't running concurrently with
2048 * ceph_fill_trace or ceph_readdir_prepopulate, which
2049 * rely on locks (dir mutex) held by our caller.
2050 */
2051 mutex_lock(&req->r_fill_mutex);
Sage Weile1518c72010-05-13 11:19:06 -07002052 req->r_err = err;
2053 req->r_aborted = true;
Sage Weilb4556392010-05-13 12:01:13 -07002054 mutex_unlock(&req->r_fill_mutex);
Sage Weile1518c72010-05-13 11:19:06 -07002055
2056 if (req->r_locked_dir &&
Sage Weil167c9e32010-05-14 10:02:57 -07002057 (req->r_op & CEPH_MDS_OP_WRITE))
2058 ceph_invalidate_dir_request(req);
Sage Weile1518c72010-05-13 11:19:06 -07002059 } else {
2060 err = req->r_err;
2061 }
2062
2063out:
2064 mutex_unlock(&mdsc->mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002065 dout("do_request %p done, result %d\n", req, err);
2066 return err;
2067}
2068
2069/*
Yan, Zheng2f276c52013-03-13 19:44:32 +08002070 * Invalidate dir's completeness, dentry lease state on an aborted MDS
Sage Weil167c9e32010-05-14 10:02:57 -07002071 * namespace request.
2072 */
2073void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2074{
2075 struct inode *inode = req->r_locked_dir;
Sage Weil167c9e32010-05-14 10:02:57 -07002076
Yan, Zheng2f276c52013-03-13 19:44:32 +08002077 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
Sage Weil167c9e32010-05-14 10:02:57 -07002078
Yan, Zheng2f276c52013-03-13 19:44:32 +08002079 ceph_dir_clear_complete(inode);
Sage Weil167c9e32010-05-14 10:02:57 -07002080 if (req->r_dentry)
2081 ceph_invalidate_dentry_lease(req->r_dentry);
2082 if (req->r_old_dentry)
2083 ceph_invalidate_dentry_lease(req->r_old_dentry);
2084}
2085
2086/*
Sage Weil2f2dc052009-10-06 11:31:09 -07002087 * Handle mds reply.
2088 *
2089 * We take the session mutex and parse and process the reply immediately.
2090 * This preserves the logical ordering of replies, capabilities, etc., sent
2091 * by the MDS as they are applied to our local cache.
2092 */
2093static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2094{
2095 struct ceph_mds_client *mdsc = session->s_mdsc;
2096 struct ceph_mds_request *req;
2097 struct ceph_mds_reply_head *head = msg->front.iov_base;
2098 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
2099 u64 tid;
2100 int err, result;
Sage Weil2600d2d2010-02-22 15:12:16 -08002101 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07002102
Sage Weil2f2dc052009-10-06 11:31:09 -07002103 if (msg->front.iov_len < sizeof(*head)) {
2104 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002105 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002106 return;
2107 }
2108
2109 /* get request, session */
Sage Weil6df058c2009-12-22 11:24:33 -08002110 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002111 mutex_lock(&mdsc->mutex);
2112 req = __lookup_request(mdsc, tid);
2113 if (!req) {
2114 dout("handle_reply on unknown tid %llu\n", tid);
2115 mutex_unlock(&mdsc->mutex);
2116 return;
2117 }
2118 dout("handle_reply %p\n", req);
Sage Weil2f2dc052009-10-06 11:31:09 -07002119
2120 /* correct session? */
Sage Weild96d6042010-03-20 20:50:58 -07002121 if (req->r_session != session) {
Sage Weil2f2dc052009-10-06 11:31:09 -07002122 pr_err("mdsc_handle_reply got %llu on session mds%d"
2123 " not mds%d\n", tid, session->s_mds,
2124 req->r_session ? req->r_session->s_mds : -1);
2125 mutex_unlock(&mdsc->mutex);
2126 goto out;
2127 }
2128
2129 /* dup? */
2130 if ((req->r_got_unsafe && !head->safe) ||
2131 (req->r_got_safe && head->safe)) {
2132 pr_warning("got a dup %s reply on %llu from mds%d\n",
2133 head->safe ? "safe" : "unsafe", tid, mds);
2134 mutex_unlock(&mdsc->mutex);
2135 goto out;
2136 }
Sage Weil85792d0d2010-05-13 09:06:02 -07002137 if (req->r_got_safe && !head->safe) {
2138 pr_warning("got unsafe after safe on %llu from mds%d\n",
2139 tid, mds);
2140 mutex_unlock(&mdsc->mutex);
2141 goto out;
2142 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002143
2144 result = le32_to_cpu(head->result);
2145
2146 /*
Greg Farnume55b71f2010-06-22 15:58:01 -07002147 * Handle an ESTALE
2148 * if we're not talking to the authority, send to them
2149 * if the authority has changed while we weren't looking,
2150 * send to new authority
2151 * Otherwise we just have to return an ESTALE
Sage Weil2f2dc052009-10-06 11:31:09 -07002152 */
2153 if (result == -ESTALE) {
Greg Farnume55b71f2010-06-22 15:58:01 -07002154 dout("got ESTALE on request %llu", req->r_tid);
Sage Weil213c99e2010-08-03 10:25:11 -07002155 if (!req->r_inode) {
2156 /* do nothing; not an authority problem */
2157 } else if (req->r_direct_mode != USE_AUTH_MDS) {
Greg Farnume55b71f2010-06-22 15:58:01 -07002158 dout("not using auth, setting for that now");
2159 req->r_direct_mode = USE_AUTH_MDS;
Sage Weil2f2dc052009-10-06 11:31:09 -07002160 __do_request(mdsc, req);
2161 mutex_unlock(&mdsc->mutex);
2162 goto out;
Greg Farnume55b71f2010-06-22 15:58:01 -07002163 } else {
2164 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
Sage Weil4af25fd2010-11-02 13:41:47 -07002165 struct ceph_cap *cap = NULL;
2166
2167 if (req->r_session)
2168 cap = ceph_get_cap_for_mds(ci,
2169 req->r_session->s_mds);
Greg Farnume55b71f2010-06-22 15:58:01 -07002170
2171 dout("already using auth");
2172 if ((!cap || cap != ci->i_auth_cap) ||
2173 (cap->mseq != req->r_sent_on_mseq)) {
2174 dout("but cap changed, so resending");
2175 __do_request(mdsc, req);
2176 mutex_unlock(&mdsc->mutex);
2177 goto out;
2178 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002179 }
Greg Farnume55b71f2010-06-22 15:58:01 -07002180 dout("have to return ESTALE on request %llu", req->r_tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002181 }
2182
Greg Farnume55b71f2010-06-22 15:58:01 -07002183
Sage Weil2f2dc052009-10-06 11:31:09 -07002184 if (head->safe) {
2185 req->r_got_safe = true;
2186 __unregister_request(mdsc, req);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002187 complete_all(&req->r_safe_completion);
Sage Weil2f2dc052009-10-06 11:31:09 -07002188
2189 if (req->r_got_unsafe) {
2190 /*
2191 * We already handled the unsafe response, now do the
2192 * cleanup. No need to examine the response; the MDS
2193 * doesn't include any result info in the safe
2194 * response. And even if it did, there is nothing
2195 * useful we could do with a revised return value.
2196 */
2197 dout("got safe reply %llu, mds%d\n", tid, mds);
2198 list_del_init(&req->r_unsafe_item);
2199
2200 /* last unsafe request during umount? */
Sage Weil44ca18f2010-02-15 12:08:46 -08002201 if (mdsc->stopping && !__get_oldest_req(mdsc))
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002202 complete_all(&mdsc->safe_umount_waiters);
Sage Weil2f2dc052009-10-06 11:31:09 -07002203 mutex_unlock(&mdsc->mutex);
2204 goto out;
2205 }
Sage Weile1518c72010-05-13 11:19:06 -07002206 } else {
Sage Weil2f2dc052009-10-06 11:31:09 -07002207 req->r_got_unsafe = true;
2208 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2209 }
2210
2211 dout("handle_reply tid %lld result %d\n", tid, result);
2212 rinfo = &req->r_reply_info;
Sage Weil14303d22010-12-14 17:37:52 -08002213 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
Sage Weil2f2dc052009-10-06 11:31:09 -07002214 mutex_unlock(&mdsc->mutex);
2215
2216 mutex_lock(&session->s_mutex);
2217 if (err < 0) {
Herb Shiu25933ab2010-12-01 14:14:38 -08002218 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
Sage Weil9ec7cab2009-12-14 15:13:47 -08002219 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002220 goto out_err;
2221 }
2222
2223 /* snap trace */
2224 if (rinfo->snapblob_len) {
2225 down_write(&mdsc->snap_rwsem);
2226 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2227 rinfo->snapblob + rinfo->snapblob_len,
2228 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP);
2229 downgrade_write(&mdsc->snap_rwsem);
2230 } else {
2231 down_read(&mdsc->snap_rwsem);
2232 }
2233
2234 /* insert trace into our cache */
Sage Weilb4556392010-05-13 12:01:13 -07002235 mutex_lock(&req->r_fill_mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002236 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002237 if (err == 0) {
Sam Lang6e8575f2012-12-28 09:56:46 -08002238 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
Yan, Zheng81c6aea2013-09-18 09:44:13 +08002239 req->r_op == CEPH_MDS_OP_LSSNAP))
Sage Weil2f2dc052009-10-06 11:31:09 -07002240 ceph_readdir_prepopulate(req, req->r_session);
Yehuda Sadeh37151662010-06-17 16:16:12 -07002241 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
Sage Weil2f2dc052009-10-06 11:31:09 -07002242 }
Sage Weilb4556392010-05-13 12:01:13 -07002243 mutex_unlock(&req->r_fill_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002244
2245 up_read(&mdsc->snap_rwsem);
2246out_err:
Sage Weile1518c72010-05-13 11:19:06 -07002247 mutex_lock(&mdsc->mutex);
2248 if (!req->r_aborted) {
2249 if (err) {
2250 req->r_err = err;
2251 } else {
2252 req->r_reply = msg;
2253 ceph_msg_get(msg);
2254 req->r_got_result = true;
2255 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002256 } else {
Sage Weile1518c72010-05-13 11:19:06 -07002257 dout("reply arrived after request %lld was aborted\n", tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002258 }
Sage Weile1518c72010-05-13 11:19:06 -07002259 mutex_unlock(&mdsc->mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002260
Sage Weilee6b2722010-06-10 12:55:52 -07002261 ceph_add_cap_releases(mdsc, req->r_session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002262 mutex_unlock(&session->s_mutex);
2263
2264 /* kick calling process */
2265 complete_request(mdsc, req);
2266out:
2267 ceph_mdsc_put_request(req);
2268 return;
2269}
2270
2271
2272
2273/*
2274 * handle mds notification that our request has been forwarded.
2275 */
Sage Weil2600d2d2010-02-22 15:12:16 -08002276static void handle_forward(struct ceph_mds_client *mdsc,
2277 struct ceph_mds_session *session,
2278 struct ceph_msg *msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07002279{
2280 struct ceph_mds_request *req;
Sage Weila1ea7872010-02-23 14:02:44 -08002281 u64 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002282 u32 next_mds;
2283 u32 fwd_seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002284 int err = -EINVAL;
2285 void *p = msg->front.iov_base;
2286 void *end = p + msg->front.iov_len;
Sage Weil2f2dc052009-10-06 11:31:09 -07002287
Sage Weila1ea7872010-02-23 14:02:44 -08002288 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07002289 next_mds = ceph_decode_32(&p);
2290 fwd_seq = ceph_decode_32(&p);
Sage Weil2f2dc052009-10-06 11:31:09 -07002291
2292 mutex_lock(&mdsc->mutex);
2293 req = __lookup_request(mdsc, tid);
2294 if (!req) {
Sage Weil2a8e5e32010-05-28 16:43:16 -07002295 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002296 goto out; /* dup reply? */
2297 }
2298
Sage Weil2a8e5e32010-05-28 16:43:16 -07002299 if (req->r_aborted) {
2300 dout("forward tid %llu aborted, unregistering\n", tid);
2301 __unregister_request(mdsc, req);
2302 } else if (fwd_seq <= req->r_num_fwd) {
2303 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07002304 tid, next_mds, req->r_num_fwd, fwd_seq);
2305 } else {
2306 /* resend. forward race not possible; mds would drop */
Sage Weil2a8e5e32010-05-28 16:43:16 -07002307 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2308 BUG_ON(req->r_err);
2309 BUG_ON(req->r_got_result);
Sage Weil2f2dc052009-10-06 11:31:09 -07002310 req->r_num_fwd = fwd_seq;
2311 req->r_resend_mds = next_mds;
2312 put_request_session(req);
2313 __do_request(mdsc, req);
2314 }
2315 ceph_mdsc_put_request(req);
2316out:
2317 mutex_unlock(&mdsc->mutex);
2318 return;
2319
2320bad:
2321 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2322}
2323
2324/*
2325 * handle a mds session control message
2326 */
2327static void handle_session(struct ceph_mds_session *session,
2328 struct ceph_msg *msg)
2329{
2330 struct ceph_mds_client *mdsc = session->s_mdsc;
2331 u32 op;
2332 u64 seq;
Sage Weil2600d2d2010-02-22 15:12:16 -08002333 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07002334 struct ceph_mds_session_head *h = msg->front.iov_base;
2335 int wake = 0;
2336
Sage Weil2f2dc052009-10-06 11:31:09 -07002337 /* decode */
2338 if (msg->front.iov_len != sizeof(*h))
2339 goto bad;
2340 op = le32_to_cpu(h->op);
2341 seq = le64_to_cpu(h->seq);
2342
2343 mutex_lock(&mdsc->mutex);
Sage Weil2600d2d2010-02-22 15:12:16 -08002344 if (op == CEPH_SESSION_CLOSE)
2345 __unregister_session(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002346 /* FIXME: this ttl calculation is generous */
2347 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2348 mutex_unlock(&mdsc->mutex);
2349
2350 mutex_lock(&session->s_mutex);
2351
2352 dout("handle_session mds%d %s %p state %s seq %llu\n",
2353 mds, ceph_session_op_name(op), session,
2354 session_state_name(session->s_state), seq);
2355
2356 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2357 session->s_state = CEPH_MDS_SESSION_OPEN;
2358 pr_info("mds%d came back\n", session->s_mds);
2359 }
2360
2361 switch (op) {
2362 case CEPH_SESSION_OPEN:
Sage Weil29790f22010-03-18 14:45:05 -07002363 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2364 pr_info("mds%d reconnect success\n", session->s_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002365 session->s_state = CEPH_MDS_SESSION_OPEN;
2366 renewed_caps(mdsc, session, 0);
2367 wake = 1;
2368 if (mdsc->stopping)
2369 __close_session(mdsc, session);
2370 break;
2371
2372 case CEPH_SESSION_RENEWCAPS:
2373 if (session->s_renew_seq == seq)
2374 renewed_caps(mdsc, session, 1);
2375 break;
2376
2377 case CEPH_SESSION_CLOSE:
Sage Weil29790f22010-03-18 14:45:05 -07002378 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2379 pr_info("mds%d reconnect denied\n", session->s_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002380 remove_session_caps(session);
2381 wake = 1; /* for good measure */
Sage Weilf3c60c52010-08-11 14:51:23 -07002382 wake_up_all(&mdsc->session_close_wq);
Sage Weil29790f22010-03-18 14:45:05 -07002383 kick_requests(mdsc, mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002384 break;
2385
2386 case CEPH_SESSION_STALE:
2387 pr_info("mds%d caps went stale, renewing\n",
2388 session->s_mds);
Alex Elderd8fb02a2012-01-12 17:48:10 -08002389 spin_lock(&session->s_gen_ttl_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07002390 session->s_cap_gen++;
Alex Elder1ce208a2012-01-12 17:48:11 -08002391 session->s_cap_ttl = jiffies - 1;
Alex Elderd8fb02a2012-01-12 17:48:10 -08002392 spin_unlock(&session->s_gen_ttl_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07002393 send_renew_caps(mdsc, session);
2394 break;
2395
2396 case CEPH_SESSION_RECALL_STATE:
2397 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2398 break;
2399
2400 default:
2401 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2402 WARN_ON(1);
2403 }
2404
2405 mutex_unlock(&session->s_mutex);
2406 if (wake) {
2407 mutex_lock(&mdsc->mutex);
2408 __wake_requests(mdsc, &session->s_waiting);
2409 mutex_unlock(&mdsc->mutex);
2410 }
2411 return;
2412
2413bad:
2414 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2415 (int)msg->front.iov_len);
Sage Weil9ec7cab2009-12-14 15:13:47 -08002416 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002417 return;
2418}
2419
2420
2421/*
2422 * called under session->mutex.
2423 */
2424static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2425 struct ceph_mds_session *session)
2426{
2427 struct ceph_mds_request *req, *nreq;
2428 int err;
2429
2430 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2431
2432 mutex_lock(&mdsc->mutex);
2433 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2434 err = __prepare_send_request(mdsc, req, session->s_mds);
2435 if (!err) {
2436 ceph_msg_get(req->r_request);
2437 ceph_con_send(&session->s_con, req->r_request);
2438 }
2439 }
2440 mutex_unlock(&mdsc->mutex);
2441}
2442
2443/*
2444 * Encode information about a cap for a reconnect with the MDS.
2445 */
Sage Weil2f2dc052009-10-06 11:31:09 -07002446static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2447 void *arg)
2448{
Sage Weil20cb34a2010-05-12 15:21:32 -07002449 union {
2450 struct ceph_mds_cap_reconnect v2;
2451 struct ceph_mds_cap_reconnect_v1 v1;
2452 } rec;
2453 size_t reclen;
Sage Weil2f2dc052009-10-06 11:31:09 -07002454 struct ceph_inode_info *ci;
Sage Weil20cb34a2010-05-12 15:21:32 -07002455 struct ceph_reconnect_state *recon_state = arg;
2456 struct ceph_pagelist *pagelist = recon_state->pagelist;
Sage Weil2f2dc052009-10-06 11:31:09 -07002457 char *path;
2458 int pathlen, err;
2459 u64 pathbase;
2460 struct dentry *dentry;
2461
2462 ci = cap->ci;
2463
2464 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2465 inode, ceph_vinop(inode), cap, cap->cap_id,
2466 ceph_cap_string(cap->issued));
Sage Weil93cea5b2009-12-23 12:21:51 -08002467 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2468 if (err)
2469 return err;
Sage Weil2f2dc052009-10-06 11:31:09 -07002470
2471 dentry = d_find_alias(inode);
2472 if (dentry) {
2473 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2474 if (IS_ERR(path)) {
2475 err = PTR_ERR(path);
Sage Weile072f8a2010-08-26 09:26:37 -07002476 goto out_dput;
Sage Weil2f2dc052009-10-06 11:31:09 -07002477 }
2478 } else {
2479 path = NULL;
2480 pathlen = 0;
2481 }
Sage Weil93cea5b2009-12-23 12:21:51 -08002482 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2483 if (err)
Sage Weile072f8a2010-08-26 09:26:37 -07002484 goto out_free;
Sage Weil2f2dc052009-10-06 11:31:09 -07002485
Sage Weilbe655592011-11-30 09:47:09 -08002486 spin_lock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07002487 cap->seq = 0; /* reset cap seq */
2488 cap->issue_seq = 0; /* and issue_seq */
Yan, Zheng667ca052013-05-31 16:25:36 +08002489 cap->mseq = 0; /* and migrate_seq */
Sage Weil20cb34a2010-05-12 15:21:32 -07002490
2491 if (recon_state->flock) {
2492 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2493 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2494 rec.v2.issued = cpu_to_le32(cap->issued);
2495 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2496 rec.v2.pathbase = cpu_to_le64(pathbase);
2497 rec.v2.flock_len = 0;
2498 reclen = sizeof(rec.v2);
2499 } else {
2500 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2501 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2502 rec.v1.issued = cpu_to_le32(cap->issued);
2503 rec.v1.size = cpu_to_le64(inode->i_size);
2504 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2505 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2506 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2507 rec.v1.pathbase = cpu_to_le64(pathbase);
2508 reclen = sizeof(rec.v1);
2509 }
Sage Weilbe655592011-11-30 09:47:09 -08002510 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07002511
Greg Farnum40819f62010-08-02 15:34:23 -07002512 if (recon_state->flock) {
2513 int num_fcntl_locks, num_flock_locks;
Jim Schutt39be95e2013-05-15 13:03:35 -05002514 struct ceph_filelock *flocks;
Greg Farnum40819f62010-08-02 15:34:23 -07002515
Jim Schutt39be95e2013-05-15 13:03:35 -05002516encode_again:
Jeff Layton1c8c6012013-06-21 08:58:15 -04002517 spin_lock(&inode->i_lock);
Jim Schutt39be95e2013-05-15 13:03:35 -05002518 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
Jeff Layton1c8c6012013-06-21 08:58:15 -04002519 spin_unlock(&inode->i_lock);
Jim Schutt39be95e2013-05-15 13:03:35 -05002520 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2521 sizeof(struct ceph_filelock), GFP_NOFS);
2522 if (!flocks) {
2523 err = -ENOMEM;
2524 goto out_free;
2525 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002526 spin_lock(&inode->i_lock);
Jim Schutt39be95e2013-05-15 13:03:35 -05002527 err = ceph_encode_locks_to_buffer(inode, flocks,
2528 num_fcntl_locks,
2529 num_flock_locks);
Jeff Layton1c8c6012013-06-21 08:58:15 -04002530 spin_unlock(&inode->i_lock);
Jim Schutt39be95e2013-05-15 13:03:35 -05002531 if (err) {
2532 kfree(flocks);
2533 if (err == -ENOSPC)
2534 goto encode_again;
2535 goto out_free;
2536 }
2537 /*
2538 * number of encoded locks is stable, so copy to pagelist
2539 */
2540 rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2541 (num_fcntl_locks+num_flock_locks) *
2542 sizeof(struct ceph_filelock));
2543 err = ceph_pagelist_append(pagelist, &rec, reclen);
2544 if (!err)
2545 err = ceph_locks_to_pagelist(flocks, pagelist,
2546 num_fcntl_locks,
2547 num_flock_locks);
2548 kfree(flocks);
Sage Weil3612abb2010-09-07 15:59:27 -07002549 } else {
2550 err = ceph_pagelist_append(pagelist, &rec, reclen);
Greg Farnum40819f62010-08-02 15:34:23 -07002551 }
Sage Weile072f8a2010-08-26 09:26:37 -07002552out_free:
Sage Weil2f2dc052009-10-06 11:31:09 -07002553 kfree(path);
Sage Weile072f8a2010-08-26 09:26:37 -07002554out_dput:
Sage Weil2f2dc052009-10-06 11:31:09 -07002555 dput(dentry);
Sage Weil93cea5b2009-12-23 12:21:51 -08002556 return err;
Sage Weil2f2dc052009-10-06 11:31:09 -07002557}
2558
2559
2560/*
2561 * If an MDS fails and recovers, clients need to reconnect in order to
2562 * reestablish shared state. This includes all caps issued through
2563 * this session _and_ the snap_realm hierarchy. Because it's not
2564 * clear which snap realms the mds cares about, we send everything we
2565 * know about.. that ensures we'll then get any new info the
2566 * recovering MDS might have.
2567 *
2568 * This is a relatively heavyweight operation, but it's rare.
2569 *
2570 * called with mdsc->mutex held.
2571 */
Sage Weil34b6c852010-05-10 16:31:25 -07002572static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2573 struct ceph_mds_session *session)
Sage Weil2f2dc052009-10-06 11:31:09 -07002574{
Sage Weil2f2dc052009-10-06 11:31:09 -07002575 struct ceph_msg *reply;
Sage Weila105f002010-02-15 14:37:55 -08002576 struct rb_node *p;
Sage Weil34b6c852010-05-10 16:31:25 -07002577 int mds = session->s_mds;
Sage Weil9abf82b2010-05-10 21:58:38 -07002578 int err = -ENOMEM;
Sage Weil93cea5b2009-12-23 12:21:51 -08002579 struct ceph_pagelist *pagelist;
Sage Weil20cb34a2010-05-12 15:21:32 -07002580 struct ceph_reconnect_state recon_state;
Sage Weil2f2dc052009-10-06 11:31:09 -07002581
Sage Weil34b6c852010-05-10 16:31:25 -07002582 pr_info("mds%d reconnect start\n", mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002583
Sage Weil93cea5b2009-12-23 12:21:51 -08002584 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2585 if (!pagelist)
2586 goto fail_nopagelist;
2587 ceph_pagelist_init(pagelist);
2588
Sage Weilb61c2762011-08-09 15:03:46 -07002589 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07002590 if (!reply)
Sage Weil93cea5b2009-12-23 12:21:51 -08002591 goto fail_nomsg;
Sage Weil93cea5b2009-12-23 12:21:51 -08002592
Sage Weil34b6c852010-05-10 16:31:25 -07002593 mutex_lock(&session->s_mutex);
2594 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2595 session->s_seq = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07002596
Sage Weila53aab62012-07-30 16:21:17 -07002597 ceph_con_close(&session->s_con);
Sage Weil34b6c852010-05-10 16:31:25 -07002598 ceph_con_open(&session->s_con,
Sage Weilb7a9e5d2012-06-27 12:24:08 -07002599 CEPH_ENTITY_TYPE_MDS, mds,
Sage Weil34b6c852010-05-10 16:31:25 -07002600 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
Sage Weil2f2dc052009-10-06 11:31:09 -07002601
Sage Weil34b6c852010-05-10 16:31:25 -07002602 /* replay unsafe requests */
2603 replay_unsafe_requests(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002604
2605 down_read(&mdsc->snap_rwsem);
2606
Sage Weil2f2dc052009-10-06 11:31:09 -07002607 dout("session %p state %s\n", session,
2608 session_state_name(session->s_state));
2609
Sage Weile01a5942010-05-10 15:36:44 -07002610 /* drop old cap expires; we're about to reestablish that state */
2611 discard_cap_releases(mdsc, session);
2612
Sage Weil2f2dc052009-10-06 11:31:09 -07002613 /* traverse this session's caps */
Sage Weil93cea5b2009-12-23 12:21:51 -08002614 err = ceph_pagelist_encode_32(pagelist, session->s_nr_caps);
2615 if (err)
2616 goto fail;
Sage Weil20cb34a2010-05-12 15:21:32 -07002617
2618 recon_state.pagelist = pagelist;
2619 recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2620 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
Sage Weil2f2dc052009-10-06 11:31:09 -07002621 if (err < 0)
Sage Weil9abf82b2010-05-10 21:58:38 -07002622 goto fail;
Sage Weil2f2dc052009-10-06 11:31:09 -07002623
2624 /*
2625 * snaprealms. we provide mds with the ino, seq (version), and
2626 * parent for all of our realms. If the mds has any newer info,
2627 * it will tell us.
2628 */
Sage Weila105f002010-02-15 14:37:55 -08002629 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2630 struct ceph_snap_realm *realm =
2631 rb_entry(p, struct ceph_snap_realm, node);
Sage Weil93cea5b2009-12-23 12:21:51 -08002632 struct ceph_mds_snaprealm_reconnect sr_rec;
Sage Weil2f2dc052009-10-06 11:31:09 -07002633
2634 dout(" adding snap realm %llx seq %lld parent %llx\n",
2635 realm->ino, realm->seq, realm->parent_ino);
Sage Weil93cea5b2009-12-23 12:21:51 -08002636 sr_rec.ino = cpu_to_le64(realm->ino);
2637 sr_rec.seq = cpu_to_le64(realm->seq);
2638 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2639 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2640 if (err)
2641 goto fail;
Sage Weil2f2dc052009-10-06 11:31:09 -07002642 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002643
Sage Weil20cb34a2010-05-12 15:21:32 -07002644 if (recon_state.flock)
2645 reply->hdr.version = cpu_to_le16(2);
Alex Elderebf18f42013-03-04 22:29:57 -06002646 if (pagelist->length) {
2647 /* set up outbound data if we have any */
2648 reply->hdr.data_len = cpu_to_le32(pagelist->length);
Alex Elder90af3602013-04-05 14:46:01 -05002649 ceph_msg_data_add_pagelist(reply, pagelist);
Alex Elderebf18f42013-03-04 22:29:57 -06002650 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002651 ceph_con_send(&session->s_con, reply);
2652
Sage Weil9abf82b2010-05-10 21:58:38 -07002653 mutex_unlock(&session->s_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002654
Sage Weil9abf82b2010-05-10 21:58:38 -07002655 mutex_lock(&mdsc->mutex);
2656 __wake_requests(mdsc, &session->s_waiting);
2657 mutex_unlock(&mdsc->mutex);
2658
Sage Weil2f2dc052009-10-06 11:31:09 -07002659 up_read(&mdsc->snap_rwsem);
Sage Weil2f2dc052009-10-06 11:31:09 -07002660 return;
2661
Sage Weil93cea5b2009-12-23 12:21:51 -08002662fail:
Sage Weil2f2dc052009-10-06 11:31:09 -07002663 ceph_msg_put(reply);
Sage Weil9abf82b2010-05-10 21:58:38 -07002664 up_read(&mdsc->snap_rwsem);
2665 mutex_unlock(&session->s_mutex);
Sage Weil93cea5b2009-12-23 12:21:51 -08002666fail_nomsg:
2667 ceph_pagelist_release(pagelist);
2668 kfree(pagelist);
2669fail_nopagelist:
Sage Weil9abf82b2010-05-10 21:58:38 -07002670 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
Sage Weil9abf82b2010-05-10 21:58:38 -07002671 return;
Sage Weil2f2dc052009-10-06 11:31:09 -07002672}
2673
2674
2675/*
2676 * compare old and new mdsmaps, kicking requests
2677 * and closing out old connections as necessary
2678 *
2679 * called under mdsc->mutex.
2680 */
2681static void check_new_map(struct ceph_mds_client *mdsc,
2682 struct ceph_mdsmap *newmap,
2683 struct ceph_mdsmap *oldmap)
2684{
2685 int i;
2686 int oldstate, newstate;
2687 struct ceph_mds_session *s;
2688
2689 dout("check_new_map new %u old %u\n",
2690 newmap->m_epoch, oldmap->m_epoch);
2691
2692 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2693 if (mdsc->sessions[i] == NULL)
2694 continue;
2695 s = mdsc->sessions[i];
2696 oldstate = ceph_mdsmap_get_state(oldmap, i);
2697 newstate = ceph_mdsmap_get_state(newmap, i);
2698
Sage Weil0deb01c2010-06-17 14:19:01 -07002699 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07002700 i, ceph_mds_state_name(oldstate),
Sage Weil0deb01c2010-06-17 14:19:01 -07002701 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
Sage Weil2f2dc052009-10-06 11:31:09 -07002702 ceph_mds_state_name(newstate),
Sage Weil0deb01c2010-06-17 14:19:01 -07002703 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
Sage Weil2f2dc052009-10-06 11:31:09 -07002704 session_state_name(s->s_state));
2705
Yan, Zheng3e8f43a2012-09-20 17:42:25 +08002706 if (i >= newmap->m_max_mds ||
2707 memcmp(ceph_mdsmap_get_addr(oldmap, i),
Sage Weil2f2dc052009-10-06 11:31:09 -07002708 ceph_mdsmap_get_addr(newmap, i),
2709 sizeof(struct ceph_entity_addr))) {
2710 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2711 /* the session never opened, just close it
2712 * out now */
2713 __wake_requests(mdsc, &s->s_waiting);
Sage Weil2600d2d2010-02-22 15:12:16 -08002714 __unregister_session(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07002715 } else {
2716 /* just close it */
2717 mutex_unlock(&mdsc->mutex);
2718 mutex_lock(&s->s_mutex);
2719 mutex_lock(&mdsc->mutex);
2720 ceph_con_close(&s->s_con);
2721 mutex_unlock(&s->s_mutex);
2722 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2723 }
2724
2725 /* kick any requests waiting on the recovering mds */
Sage Weil29790f22010-03-18 14:45:05 -07002726 kick_requests(mdsc, i);
Sage Weil2f2dc052009-10-06 11:31:09 -07002727 } else if (oldstate == newstate) {
2728 continue; /* nothing new with this mds */
2729 }
2730
2731 /*
2732 * send reconnect?
2733 */
2734 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
Sage Weil34b6c852010-05-10 16:31:25 -07002735 newstate >= CEPH_MDS_STATE_RECONNECT) {
2736 mutex_unlock(&mdsc->mutex);
2737 send_mds_reconnect(mdsc, s);
2738 mutex_lock(&mdsc->mutex);
2739 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002740
2741 /*
Sage Weil29790f22010-03-18 14:45:05 -07002742 * kick request on any mds that has gone active.
Sage Weil2f2dc052009-10-06 11:31:09 -07002743 */
2744 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2745 newstate >= CEPH_MDS_STATE_ACTIVE) {
Sage Weil29790f22010-03-18 14:45:05 -07002746 if (oldstate != CEPH_MDS_STATE_CREATING &&
2747 oldstate != CEPH_MDS_STATE_STARTING)
2748 pr_info("mds%d recovery completed\n", s->s_mds);
2749 kick_requests(mdsc, i);
Sage Weil2f2dc052009-10-06 11:31:09 -07002750 ceph_kick_flushing_caps(mdsc, s);
Sage Weil0dc25702009-11-20 13:43:45 -08002751 wake_up_session_caps(s, 1);
Sage Weil2f2dc052009-10-06 11:31:09 -07002752 }
2753 }
Sage Weilcb170a22010-06-21 13:38:35 -07002754
2755 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
2756 s = mdsc->sessions[i];
2757 if (!s)
2758 continue;
2759 if (!ceph_mdsmap_is_laggy(newmap, i))
2760 continue;
2761 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2762 s->s_state == CEPH_MDS_SESSION_HUNG ||
2763 s->s_state == CEPH_MDS_SESSION_CLOSING) {
2764 dout(" connecting to export targets of laggy mds%d\n",
2765 i);
2766 __open_export_target_sessions(mdsc, s);
2767 }
2768 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002769}
2770
2771
2772
2773/*
2774 * leases
2775 */
2776
2777/*
2778 * caller must hold session s_mutex, dentry->d_lock
2779 */
2780void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
2781{
2782 struct ceph_dentry_info *di = ceph_dentry(dentry);
2783
2784 ceph_put_mds_session(di->lease_session);
2785 di->lease_session = NULL;
2786}
2787
Sage Weil2600d2d2010-02-22 15:12:16 -08002788static void handle_lease(struct ceph_mds_client *mdsc,
2789 struct ceph_mds_session *session,
2790 struct ceph_msg *msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07002791{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002792 struct super_block *sb = mdsc->fsc->sb;
Sage Weil2f2dc052009-10-06 11:31:09 -07002793 struct inode *inode;
Sage Weil2f2dc052009-10-06 11:31:09 -07002794 struct dentry *parent, *dentry;
2795 struct ceph_dentry_info *di;
Sage Weil2600d2d2010-02-22 15:12:16 -08002796 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07002797 struct ceph_mds_lease *h = msg->front.iov_base;
Sage Weil1e5ea232010-06-04 10:05:40 -07002798 u32 seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002799 struct ceph_vino vino;
Sage Weil2f2dc052009-10-06 11:31:09 -07002800 struct qstr dname;
2801 int release = 0;
2802
Sage Weil2f2dc052009-10-06 11:31:09 -07002803 dout("handle_lease from mds%d\n", mds);
2804
2805 /* decode */
2806 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
2807 goto bad;
2808 vino.ino = le64_to_cpu(h->ino);
2809 vino.snap = CEPH_NOSNAP;
Sage Weil1e5ea232010-06-04 10:05:40 -07002810 seq = le32_to_cpu(h->seq);
Sage Weil2f2dc052009-10-06 11:31:09 -07002811 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
2812 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
2813 if (dname.len != get_unaligned_le32(h+1))
2814 goto bad;
2815
Sage Weil2f2dc052009-10-06 11:31:09 -07002816 mutex_lock(&session->s_mutex);
2817 session->s_seq++;
2818
2819 /* lookup inode */
2820 inode = ceph_find_inode(sb, vino);
Sage Weil2f90b852011-07-26 11:28:25 -07002821 dout("handle_lease %s, ino %llx %p %.*s\n",
2822 ceph_lease_op_name(h->action), vino.ino, inode,
Sage Weil1e5ea232010-06-04 10:05:40 -07002823 dname.len, dname.name);
Sage Weil2f2dc052009-10-06 11:31:09 -07002824 if (inode == NULL) {
2825 dout("handle_lease no inode %llx\n", vino.ino);
2826 goto release;
2827 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002828
2829 /* dentry */
2830 parent = d_find_alias(inode);
2831 if (!parent) {
2832 dout("no parent dentry on inode %p\n", inode);
2833 WARN_ON(1);
2834 goto release; /* hrm... */
2835 }
2836 dname.hash = full_name_hash(dname.name, dname.len);
2837 dentry = d_lookup(parent, &dname);
2838 dput(parent);
2839 if (!dentry)
2840 goto release;
2841
2842 spin_lock(&dentry->d_lock);
2843 di = ceph_dentry(dentry);
2844 switch (h->action) {
2845 case CEPH_MDS_LEASE_REVOKE:
Sage Weil3d8eb7a2011-11-11 09:48:53 -08002846 if (di->lease_session == session) {
Sage Weil1e5ea232010-06-04 10:05:40 -07002847 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
2848 h->seq = cpu_to_le32(di->lease_seq);
Sage Weil2f2dc052009-10-06 11:31:09 -07002849 __ceph_mdsc_drop_dentry_lease(dentry);
2850 }
2851 release = 1;
2852 break;
2853
2854 case CEPH_MDS_LEASE_RENEW:
Sage Weil3d8eb7a2011-11-11 09:48:53 -08002855 if (di->lease_session == session &&
Sage Weil2f2dc052009-10-06 11:31:09 -07002856 di->lease_gen == session->s_cap_gen &&
2857 di->lease_renew_from &&
2858 di->lease_renew_after == 0) {
2859 unsigned long duration =
2860 le32_to_cpu(h->duration_ms) * HZ / 1000;
2861
Sage Weil1e5ea232010-06-04 10:05:40 -07002862 di->lease_seq = seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002863 dentry->d_time = di->lease_renew_from + duration;
2864 di->lease_renew_after = di->lease_renew_from +
2865 (duration >> 1);
2866 di->lease_renew_from = 0;
2867 }
2868 break;
2869 }
2870 spin_unlock(&dentry->d_lock);
2871 dput(dentry);
2872
2873 if (!release)
2874 goto out;
2875
2876release:
2877 /* let's just reuse the same message */
2878 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
2879 ceph_msg_get(msg);
2880 ceph_con_send(&session->s_con, msg);
2881
2882out:
2883 iput(inode);
2884 mutex_unlock(&session->s_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002885 return;
2886
2887bad:
2888 pr_err("corrupt lease message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002889 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002890}
2891
2892void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
2893 struct inode *inode,
2894 struct dentry *dentry, char action,
2895 u32 seq)
2896{
2897 struct ceph_msg *msg;
2898 struct ceph_mds_lease *lease;
2899 int len = sizeof(*lease) + sizeof(u32);
2900 int dnamelen = 0;
2901
2902 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
2903 inode, dentry, ceph_lease_op_name(action), session->s_mds);
2904 dnamelen = dentry->d_name.len;
2905 len += dnamelen;
2906
Sage Weilb61c2762011-08-09 15:03:46 -07002907 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07002908 if (!msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07002909 return;
2910 lease = msg->front.iov_base;
2911 lease->action = action;
Sage Weil2f2dc052009-10-06 11:31:09 -07002912 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
2913 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
2914 lease->seq = cpu_to_le32(seq);
2915 put_unaligned_le32(dnamelen, lease + 1);
2916 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
2917
2918 /*
2919 * if this is a preemptive lease RELEASE, no need to
2920 * flush request stream, since the actual request will
2921 * soon follow.
2922 */
2923 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
2924
2925 ceph_con_send(&session->s_con, msg);
2926}
2927
2928/*
2929 * Preemptively release a lease we expect to invalidate anyway.
2930 * Pass @inode always, @dentry is optional.
2931 */
2932void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
Sage Weil2f90b852011-07-26 11:28:25 -07002933 struct dentry *dentry)
Sage Weil2f2dc052009-10-06 11:31:09 -07002934{
2935 struct ceph_dentry_info *di;
2936 struct ceph_mds_session *session;
2937 u32 seq;
2938
2939 BUG_ON(inode == NULL);
2940 BUG_ON(dentry == NULL);
Sage Weil2f2dc052009-10-06 11:31:09 -07002941
2942 /* is dentry lease valid? */
2943 spin_lock(&dentry->d_lock);
2944 di = ceph_dentry(dentry);
2945 if (!di || !di->lease_session ||
2946 di->lease_session->s_mds < 0 ||
2947 di->lease_gen != di->lease_session->s_cap_gen ||
2948 !time_before(jiffies, dentry->d_time)) {
2949 dout("lease_release inode %p dentry %p -- "
Sage Weil2f90b852011-07-26 11:28:25 -07002950 "no lease\n",
2951 inode, dentry);
Sage Weil2f2dc052009-10-06 11:31:09 -07002952 spin_unlock(&dentry->d_lock);
2953 return;
2954 }
2955
2956 /* we do have a lease on this dentry; note mds and seq */
2957 session = ceph_get_mds_session(di->lease_session);
2958 seq = di->lease_seq;
2959 __ceph_mdsc_drop_dentry_lease(dentry);
2960 spin_unlock(&dentry->d_lock);
2961
Sage Weil2f90b852011-07-26 11:28:25 -07002962 dout("lease_release inode %p dentry %p to mds%d\n",
2963 inode, dentry, session->s_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002964 ceph_mdsc_lease_send_msg(session, inode, dentry,
2965 CEPH_MDS_LEASE_RELEASE, seq);
2966 ceph_put_mds_session(session);
2967}
2968
2969/*
2970 * drop all leases (and dentry refs) in preparation for umount
2971 */
2972static void drop_leases(struct ceph_mds_client *mdsc)
2973{
2974 int i;
2975
2976 dout("drop_leases\n");
2977 mutex_lock(&mdsc->mutex);
2978 for (i = 0; i < mdsc->max_sessions; i++) {
2979 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2980 if (!s)
2981 continue;
2982 mutex_unlock(&mdsc->mutex);
2983 mutex_lock(&s->s_mutex);
2984 mutex_unlock(&s->s_mutex);
2985 ceph_put_mds_session(s);
2986 mutex_lock(&mdsc->mutex);
2987 }
2988 mutex_unlock(&mdsc->mutex);
2989}
2990
2991
2992
2993/*
2994 * delayed work -- periodically trim expired leases, renew caps with mds
2995 */
2996static void schedule_delayed(struct ceph_mds_client *mdsc)
2997{
2998 int delay = 5;
2999 unsigned hz = round_jiffies_relative(HZ * delay);
3000 schedule_delayed_work(&mdsc->delayed_work, hz);
3001}
3002
3003static void delayed_work(struct work_struct *work)
3004{
3005 int i;
3006 struct ceph_mds_client *mdsc =
3007 container_of(work, struct ceph_mds_client, delayed_work.work);
3008 int renew_interval;
3009 int renew_caps;
3010
3011 dout("mdsc delayed_work\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07003012 ceph_check_delayed_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07003013
3014 mutex_lock(&mdsc->mutex);
3015 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3016 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3017 mdsc->last_renew_caps);
3018 if (renew_caps)
3019 mdsc->last_renew_caps = jiffies;
3020
3021 for (i = 0; i < mdsc->max_sessions; i++) {
3022 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3023 if (s == NULL)
3024 continue;
3025 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3026 dout("resending session close request for mds%d\n",
3027 s->s_mds);
3028 request_close_session(mdsc, s);
3029 ceph_put_mds_session(s);
3030 continue;
3031 }
3032 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3033 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3034 s->s_state = CEPH_MDS_SESSION_HUNG;
3035 pr_info("mds%d hung\n", s->s_mds);
3036 }
3037 }
3038 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3039 /* this mds is failed or recovering, just wait */
3040 ceph_put_mds_session(s);
3041 continue;
3042 }
3043 mutex_unlock(&mdsc->mutex);
3044
3045 mutex_lock(&s->s_mutex);
3046 if (renew_caps)
3047 send_renew_caps(mdsc, s);
3048 else
3049 ceph_con_keepalive(&s->s_con);
Sage Weilee6b2722010-06-10 12:55:52 -07003050 ceph_add_cap_releases(mdsc, s);
Sage Weilaab53dd2010-03-17 16:30:21 -07003051 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3052 s->s_state == CEPH_MDS_SESSION_HUNG)
Sage Weil3d7ded42010-06-09 16:47:10 -07003053 ceph_send_cap_releases(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07003054 mutex_unlock(&s->s_mutex);
3055 ceph_put_mds_session(s);
3056
3057 mutex_lock(&mdsc->mutex);
3058 }
3059 mutex_unlock(&mdsc->mutex);
3060
3061 schedule_delayed(mdsc);
3062}
3063
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003064int ceph_mdsc_init(struct ceph_fs_client *fsc)
Sage Weil2f2dc052009-10-06 11:31:09 -07003065
Sage Weil2f2dc052009-10-06 11:31:09 -07003066{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003067 struct ceph_mds_client *mdsc;
3068
3069 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3070 if (!mdsc)
3071 return -ENOMEM;
3072 mdsc->fsc = fsc;
3073 fsc->mdsc = mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07003074 mutex_init(&mdsc->mutex);
3075 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
majianpengfb3101b2013-06-25 14:48:19 +08003076 if (mdsc->mdsmap == NULL) {
3077 kfree(mdsc);
Cheng Renquan2d06eeb2010-03-26 18:04:40 +08003078 return -ENOMEM;
majianpengfb3101b2013-06-25 14:48:19 +08003079 }
Cheng Renquan2d06eeb2010-03-26 18:04:40 +08003080
Sage Weil2f2dc052009-10-06 11:31:09 -07003081 init_completion(&mdsc->safe_umount_waiters);
Sage Weilf3c60c52010-08-11 14:51:23 -07003082 init_waitqueue_head(&mdsc->session_close_wq);
Sage Weil2f2dc052009-10-06 11:31:09 -07003083 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3084 mdsc->sessions = NULL;
3085 mdsc->max_sessions = 0;
3086 mdsc->stopping = 0;
3087 init_rwsem(&mdsc->snap_rwsem);
Sage Weila105f002010-02-15 14:37:55 -08003088 mdsc->snap_realms = RB_ROOT;
Sage Weil2f2dc052009-10-06 11:31:09 -07003089 INIT_LIST_HEAD(&mdsc->snap_empty);
3090 spin_lock_init(&mdsc->snap_empty_lock);
3091 mdsc->last_tid = 0;
Sage Weil44ca18f2010-02-15 12:08:46 -08003092 mdsc->request_tree = RB_ROOT;
Sage Weil2f2dc052009-10-06 11:31:09 -07003093 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3094 mdsc->last_renew_caps = jiffies;
3095 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3096 spin_lock_init(&mdsc->cap_delay_lock);
3097 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3098 spin_lock_init(&mdsc->snap_flush_lock);
3099 mdsc->cap_flush_seq = 0;
3100 INIT_LIST_HEAD(&mdsc->cap_dirty);
Sage Weildb354052011-05-24 11:46:31 -07003101 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
Sage Weil2f2dc052009-10-06 11:31:09 -07003102 mdsc->num_cap_flushing = 0;
3103 spin_lock_init(&mdsc->cap_dirty_lock);
3104 init_waitqueue_head(&mdsc->cap_flushing_wq);
3105 spin_lock_init(&mdsc->dentry_lru_lock);
3106 INIT_LIST_HEAD(&mdsc->dentry_lru);
Cheng Renquan2d06eeb2010-03-26 18:04:40 +08003107
Yehuda Sadeh37151662010-06-17 16:16:12 -07003108 ceph_caps_init(mdsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003109 ceph_adjust_min_caps(mdsc, fsc->min_caps);
Yehuda Sadeh37151662010-06-17 16:16:12 -07003110
Sage Weil5f44f142009-11-18 14:52:18 -08003111 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07003112}
3113
3114/*
3115 * Wait for safe replies on open mds requests. If we time out, drop
3116 * all requests from the tree to avoid dangling dentry refs.
3117 */
3118static void wait_requests(struct ceph_mds_client *mdsc)
3119{
3120 struct ceph_mds_request *req;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003121 struct ceph_fs_client *fsc = mdsc->fsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07003122
3123 mutex_lock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08003124 if (__get_oldest_req(mdsc)) {
Sage Weil2f2dc052009-10-06 11:31:09 -07003125 mutex_unlock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08003126
Sage Weil2f2dc052009-10-06 11:31:09 -07003127 dout("wait_requests waiting for requests\n");
3128 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003129 fsc->client->options->mount_timeout * HZ);
Sage Weil2f2dc052009-10-06 11:31:09 -07003130
3131 /* tear down remaining requests */
Sage Weil44ca18f2010-02-15 12:08:46 -08003132 mutex_lock(&mdsc->mutex);
3133 while ((req = __get_oldest_req(mdsc))) {
Sage Weil2f2dc052009-10-06 11:31:09 -07003134 dout("wait_requests timed out on tid %llu\n",
3135 req->r_tid);
Sage Weil44ca18f2010-02-15 12:08:46 -08003136 __unregister_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -07003137 }
3138 }
3139 mutex_unlock(&mdsc->mutex);
3140 dout("wait_requests done\n");
3141}
3142
3143/*
3144 * called before mount is ro, and before dentries are torn down.
3145 * (hmm, does this still race with new lookups?)
3146 */
3147void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3148{
3149 dout("pre_umount\n");
3150 mdsc->stopping = 1;
3151
3152 drop_leases(mdsc);
Sage Weilafcdaea2009-10-14 14:27:38 -07003153 ceph_flush_dirty_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07003154 wait_requests(mdsc);
Sage Weil17c688c2010-06-21 16:12:26 -07003155
3156 /*
3157 * wait for reply handlers to drop their request refs and
3158 * their inode/dcache refs
3159 */
3160 ceph_msgr_flush();
Sage Weil2f2dc052009-10-06 11:31:09 -07003161}
3162
3163/*
3164 * wait for all write mds requests to flush.
3165 */
3166static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3167{
Sage Weil80fc7312010-03-16 15:28:54 -07003168 struct ceph_mds_request *req = NULL, *nextreq;
Sage Weil44ca18f2010-02-15 12:08:46 -08003169 struct rb_node *n;
Sage Weil2f2dc052009-10-06 11:31:09 -07003170
3171 mutex_lock(&mdsc->mutex);
3172 dout("wait_unsafe_requests want %lld\n", want_tid);
Sage Weil80fc7312010-03-16 15:28:54 -07003173restart:
Sage Weil44ca18f2010-02-15 12:08:46 -08003174 req = __get_oldest_req(mdsc);
3175 while (req && req->r_tid <= want_tid) {
Sage Weil80fc7312010-03-16 15:28:54 -07003176 /* find next request */
3177 n = rb_next(&req->r_node);
3178 if (n)
3179 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3180 else
3181 nextreq = NULL;
Sage Weil44ca18f2010-02-15 12:08:46 -08003182 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
3183 /* write op */
3184 ceph_mdsc_get_request(req);
Sage Weil80fc7312010-03-16 15:28:54 -07003185 if (nextreq)
3186 ceph_mdsc_get_request(nextreq);
Sage Weil44ca18f2010-02-15 12:08:46 -08003187 mutex_unlock(&mdsc->mutex);
3188 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3189 req->r_tid, want_tid);
3190 wait_for_completion(&req->r_safe_completion);
3191 mutex_lock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08003192 ceph_mdsc_put_request(req);
Sage Weil80fc7312010-03-16 15:28:54 -07003193 if (!nextreq)
3194 break; /* next dne before, so we're done! */
3195 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3196 /* next request was removed from tree */
3197 ceph_mdsc_put_request(nextreq);
3198 goto restart;
3199 }
3200 ceph_mdsc_put_request(nextreq); /* won't go away */
Sage Weil44ca18f2010-02-15 12:08:46 -08003201 }
Sage Weil80fc7312010-03-16 15:28:54 -07003202 req = nextreq;
Sage Weil2f2dc052009-10-06 11:31:09 -07003203 }
3204 mutex_unlock(&mdsc->mutex);
3205 dout("wait_unsafe_requests done\n");
3206}
3207
3208void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3209{
3210 u64 want_tid, want_flush;
3211
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003212 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
Sage Weil56b7cf92010-05-03 15:22:00 -07003213 return;
3214
Sage Weil2f2dc052009-10-06 11:31:09 -07003215 dout("sync\n");
3216 mutex_lock(&mdsc->mutex);
3217 want_tid = mdsc->last_tid;
3218 want_flush = mdsc->cap_flush_seq;
3219 mutex_unlock(&mdsc->mutex);
3220 dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
3221
Sage Weilafcdaea2009-10-14 14:27:38 -07003222 ceph_flush_dirty_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07003223
3224 wait_unsafe_requests(mdsc, want_tid);
3225 wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush));
3226}
3227
Sage Weilf3c60c52010-08-11 14:51:23 -07003228/*
3229 * true if all sessions are closed, or we force unmount
3230 */
H Hartley Sweeten7fd7d102011-09-23 13:22:11 -07003231static bool done_closing_sessions(struct ceph_mds_client *mdsc)
Sage Weilf3c60c52010-08-11 14:51:23 -07003232{
3233 int i, n = 0;
3234
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003235 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
Sage Weilf3c60c52010-08-11 14:51:23 -07003236 return true;
3237
3238 mutex_lock(&mdsc->mutex);
3239 for (i = 0; i < mdsc->max_sessions; i++)
3240 if (mdsc->sessions[i])
3241 n++;
3242 mutex_unlock(&mdsc->mutex);
3243 return n == 0;
3244}
Sage Weil2f2dc052009-10-06 11:31:09 -07003245
3246/*
3247 * called after sb is ro.
3248 */
3249void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3250{
3251 struct ceph_mds_session *session;
3252 int i;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003253 struct ceph_fs_client *fsc = mdsc->fsc;
3254 unsigned long timeout = fsc->client->options->mount_timeout * HZ;
Sage Weil2f2dc052009-10-06 11:31:09 -07003255
3256 dout("close_sessions\n");
3257
Sage Weil2f2dc052009-10-06 11:31:09 -07003258 /* close sessions */
Sage Weilf3c60c52010-08-11 14:51:23 -07003259 mutex_lock(&mdsc->mutex);
3260 for (i = 0; i < mdsc->max_sessions; i++) {
3261 session = __ceph_lookup_mds_session(mdsc, i);
3262 if (!session)
3263 continue;
Sage Weil2f2dc052009-10-06 11:31:09 -07003264 mutex_unlock(&mdsc->mutex);
Sage Weilf3c60c52010-08-11 14:51:23 -07003265 mutex_lock(&session->s_mutex);
3266 __close_session(mdsc, session);
3267 mutex_unlock(&session->s_mutex);
3268 ceph_put_mds_session(session);
Sage Weil2f2dc052009-10-06 11:31:09 -07003269 mutex_lock(&mdsc->mutex);
3270 }
Sage Weilf3c60c52010-08-11 14:51:23 -07003271 mutex_unlock(&mdsc->mutex);
3272
3273 dout("waiting for sessions to close\n");
3274 wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3275 timeout);
Sage Weil2f2dc052009-10-06 11:31:09 -07003276
3277 /* tear down remaining sessions */
Sage Weilf3c60c52010-08-11 14:51:23 -07003278 mutex_lock(&mdsc->mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07003279 for (i = 0; i < mdsc->max_sessions; i++) {
3280 if (mdsc->sessions[i]) {
3281 session = get_session(mdsc->sessions[i]);
Sage Weil2600d2d2010-02-22 15:12:16 -08003282 __unregister_session(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07003283 mutex_unlock(&mdsc->mutex);
3284 mutex_lock(&session->s_mutex);
3285 remove_session_caps(session);
3286 mutex_unlock(&session->s_mutex);
3287 ceph_put_mds_session(session);
3288 mutex_lock(&mdsc->mutex);
3289 }
3290 }
Sage Weil2f2dc052009-10-06 11:31:09 -07003291 WARN_ON(!list_empty(&mdsc->cap_delay_list));
Sage Weil2f2dc052009-10-06 11:31:09 -07003292 mutex_unlock(&mdsc->mutex);
3293
3294 ceph_cleanup_empty_realms(mdsc);
3295
3296 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3297
3298 dout("stopped\n");
3299}
3300
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003301static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
Sage Weil2f2dc052009-10-06 11:31:09 -07003302{
3303 dout("stop\n");
3304 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3305 if (mdsc->mdsmap)
3306 ceph_mdsmap_destroy(mdsc->mdsmap);
3307 kfree(mdsc->sessions);
Yehuda Sadeh37151662010-06-17 16:16:12 -07003308 ceph_caps_finalize(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07003309}
3310
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003311void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3312{
3313 struct ceph_mds_client *mdsc = fsc->mdsc;
3314
Sage Weilef550f62011-03-25 13:27:48 -07003315 dout("mdsc_destroy %p\n", mdsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003316 ceph_mdsc_stop(mdsc);
Sage Weilef550f62011-03-25 13:27:48 -07003317
3318 /* flush out any connection work with references to us */
3319 ceph_msgr_flush();
3320
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003321 fsc->mdsc = NULL;
3322 kfree(mdsc);
Sage Weilef550f62011-03-25 13:27:48 -07003323 dout("mdsc_destroy %p done\n", mdsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003324}
3325
Sage Weil2f2dc052009-10-06 11:31:09 -07003326
3327/*
3328 * handle mds map update.
3329 */
3330void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3331{
3332 u32 epoch;
3333 u32 maplen;
3334 void *p = msg->front.iov_base;
3335 void *end = p + msg->front.iov_len;
3336 struct ceph_mdsmap *newmap, *oldmap;
3337 struct ceph_fsid fsid;
3338 int err = -EINVAL;
3339
3340 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3341 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003342 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
Sage Weil07433042009-11-18 16:50:41 -08003343 return;
Sage Weilc89136e2009-10-14 09:59:09 -07003344 epoch = ceph_decode_32(&p);
3345 maplen = ceph_decode_32(&p);
Sage Weil2f2dc052009-10-06 11:31:09 -07003346 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3347
3348 /* do we need it? */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003349 ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
Sage Weil2f2dc052009-10-06 11:31:09 -07003350 mutex_lock(&mdsc->mutex);
3351 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3352 dout("handle_map epoch %u <= our %u\n",
3353 epoch, mdsc->mdsmap->m_epoch);
3354 mutex_unlock(&mdsc->mutex);
3355 return;
3356 }
3357
3358 newmap = ceph_mdsmap_decode(&p, end);
3359 if (IS_ERR(newmap)) {
3360 err = PTR_ERR(newmap);
3361 goto bad_unlock;
3362 }
3363
3364 /* swap into place */
3365 if (mdsc->mdsmap) {
3366 oldmap = mdsc->mdsmap;
3367 mdsc->mdsmap = newmap;
3368 check_new_map(mdsc, newmap, oldmap);
3369 ceph_mdsmap_destroy(oldmap);
3370 } else {
3371 mdsc->mdsmap = newmap; /* first mds map */
3372 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003373 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
Sage Weil2f2dc052009-10-06 11:31:09 -07003374
3375 __wake_requests(mdsc, &mdsc->waiting_for_map);
3376
3377 mutex_unlock(&mdsc->mutex);
3378 schedule_delayed(mdsc);
3379 return;
3380
3381bad_unlock:
3382 mutex_unlock(&mdsc->mutex);
3383bad:
3384 pr_err("error decoding mdsmap %d\n", err);
3385 return;
3386}
3387
3388static struct ceph_connection *con_get(struct ceph_connection *con)
3389{
3390 struct ceph_mds_session *s = con->private;
3391
3392 if (get_session(s)) {
Sage Weil2600d2d2010-02-22 15:12:16 -08003393 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
Sage Weil2f2dc052009-10-06 11:31:09 -07003394 return con;
3395 }
3396 dout("mdsc con_get %p FAIL\n", s);
3397 return NULL;
3398}
3399
3400static void con_put(struct ceph_connection *con)
3401{
3402 struct ceph_mds_session *s = con->private;
3403
Henry C Chang7d8e18a2011-05-11 10:29:52 +00003404 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
Sage Weil2f2dc052009-10-06 11:31:09 -07003405 ceph_put_mds_session(s);
3406}
3407
3408/*
3409 * if the client is unresponsive for long enough, the mds will kill
3410 * the session entirely.
3411 */
3412static void peer_reset(struct ceph_connection *con)
3413{
3414 struct ceph_mds_session *s = con->private;
Sage Weil7e70f0e2010-03-18 13:59:12 -07003415 struct ceph_mds_client *mdsc = s->s_mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07003416
Sage Weil7e70f0e2010-03-18 13:59:12 -07003417 pr_warning("mds%d closed our session\n", s->s_mds);
3418 send_mds_reconnect(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07003419}
3420
3421static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3422{
3423 struct ceph_mds_session *s = con->private;
3424 struct ceph_mds_client *mdsc = s->s_mdsc;
3425 int type = le16_to_cpu(msg->hdr.type);
3426
Sage Weil2600d2d2010-02-22 15:12:16 -08003427 mutex_lock(&mdsc->mutex);
3428 if (__verify_registered_session(mdsc, s) < 0) {
3429 mutex_unlock(&mdsc->mutex);
3430 goto out;
3431 }
3432 mutex_unlock(&mdsc->mutex);
3433
Sage Weil2f2dc052009-10-06 11:31:09 -07003434 switch (type) {
3435 case CEPH_MSG_MDS_MAP:
3436 ceph_mdsc_handle_map(mdsc, msg);
3437 break;
3438 case CEPH_MSG_CLIENT_SESSION:
3439 handle_session(s, msg);
3440 break;
3441 case CEPH_MSG_CLIENT_REPLY:
3442 handle_reply(s, msg);
3443 break;
3444 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
Sage Weil2600d2d2010-02-22 15:12:16 -08003445 handle_forward(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003446 break;
3447 case CEPH_MSG_CLIENT_CAPS:
3448 ceph_handle_caps(s, msg);
3449 break;
3450 case CEPH_MSG_CLIENT_SNAP:
Sage Weil2600d2d2010-02-22 15:12:16 -08003451 ceph_handle_snap(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003452 break;
3453 case CEPH_MSG_CLIENT_LEASE:
Sage Weil2600d2d2010-02-22 15:12:16 -08003454 handle_lease(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003455 break;
3456
3457 default:
3458 pr_err("received unknown message type %d %s\n", type,
3459 ceph_msg_type_name(type));
3460 }
Sage Weil2600d2d2010-02-22 15:12:16 -08003461out:
Sage Weil2f2dc052009-10-06 11:31:09 -07003462 ceph_msg_put(msg);
3463}
3464
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003465/*
3466 * authentication
3467 */
Alex Eldera3530df2012-05-16 15:16:39 -05003468
3469/*
3470 * Note: returned pointer is the address of a structure that's
3471 * managed separately. Caller must *not* attempt to free it.
3472 */
3473static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
Alex Elder8f43fb52012-05-16 15:16:39 -05003474 int *proto, int force_new)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003475{
3476 struct ceph_mds_session *s = con->private;
3477 struct ceph_mds_client *mdsc = s->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003478 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
Alex Elder74f18692012-05-16 15:16:39 -05003479 struct ceph_auth_handshake *auth = &s->s_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003480
Alex Elder74f18692012-05-16 15:16:39 -05003481 if (force_new && auth->authorizer) {
Sage Weil27859f92013-03-25 10:26:14 -07003482 ceph_auth_destroy_authorizer(ac, auth->authorizer);
Alex Elder74f18692012-05-16 15:16:39 -05003483 auth->authorizer = NULL;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003484 }
Sage Weil27859f92013-03-25 10:26:14 -07003485 if (!auth->authorizer) {
3486 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3487 auth);
Sage Weil0bed9b52013-03-25 10:26:01 -07003488 if (ret)
3489 return ERR_PTR(ret);
Sage Weil27859f92013-03-25 10:26:14 -07003490 } else {
3491 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3492 auth);
Alex Eldera2556512012-05-16 15:16:39 -05003493 if (ret)
Alex Eldera3530df2012-05-16 15:16:39 -05003494 return ERR_PTR(ret);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003495 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003496 *proto = ac->protocol;
Alex Elder74f18692012-05-16 15:16:39 -05003497
Alex Eldera3530df2012-05-16 15:16:39 -05003498 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003499}
3500
3501
3502static int verify_authorizer_reply(struct ceph_connection *con, int len)
3503{
3504 struct ceph_mds_session *s = con->private;
3505 struct ceph_mds_client *mdsc = s->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003506 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003507
Sage Weil27859f92013-03-25 10:26:14 -07003508 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003509}
3510
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003511static int invalidate_authorizer(struct ceph_connection *con)
3512{
3513 struct ceph_mds_session *s = con->private;
3514 struct ceph_mds_client *mdsc = s->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003515 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003516
Sage Weil27859f92013-03-25 10:26:14 -07003517 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003518
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003519 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003520}
3521
Alex Elder53ded492013-03-01 18:00:14 -06003522static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3523 struct ceph_msg_header *hdr, int *skip)
3524{
3525 struct ceph_msg *msg;
3526 int type = (int) le16_to_cpu(hdr->type);
3527 int front_len = (int) le32_to_cpu(hdr->front_len);
3528
3529 if (con->in_msg)
3530 return con->in_msg;
3531
3532 *skip = 0;
3533 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3534 if (!msg) {
3535 pr_err("unable to allocate msg type %d len %d\n",
3536 type, front_len);
3537 return NULL;
3538 }
Alex Elder53ded492013-03-01 18:00:14 -06003539
3540 return msg;
3541}
3542
Tobias Klauser9e327892010-05-20 10:40:19 +02003543static const struct ceph_connection_operations mds_con_ops = {
Sage Weil2f2dc052009-10-06 11:31:09 -07003544 .get = con_get,
3545 .put = con_put,
3546 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003547 .get_authorizer = get_authorizer,
3548 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003549 .invalidate_authorizer = invalidate_authorizer,
Sage Weil2f2dc052009-10-06 11:31:09 -07003550 .peer_reset = peer_reset,
Alex Elder53ded492013-03-01 18:00:14 -06003551 .alloc_msg = mds_alloc_msg,
Sage Weil2f2dc052009-10-06 11:31:09 -07003552};
3553
Sage Weil2f2dc052009-10-06 11:31:09 -07003554/* eof */