blob: 2b4d093d056319424369f01321c710f2c9d27185 [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>
Yan, Zheng54008392014-03-29 13:41:15 +08006#include <linux/gfp.h>
Sage Weil2f2dc052009-10-06 11:31:09 -07007#include <linux/sched.h>
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07008#include <linux/debugfs.h>
9#include <linux/seq_file.h>
Sage Weil2f2dc052009-10-06 11:31:09 -070010
Sage Weil2f2dc052009-10-06 11:31:09 -070011#include "super.h"
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070012#include "mds_client.h"
13
Sage Weil1fe60e52012-07-30 16:23:22 -070014#include <linux/ceph/ceph_features.h>
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070015#include <linux/ceph/messenger.h>
16#include <linux/ceph/decode.h>
17#include <linux/ceph/pagelist.h>
18#include <linux/ceph/auth.h>
19#include <linux/ceph/debugfs.h>
Sage Weil2f2dc052009-10-06 11:31:09 -070020
21/*
22 * A cluster of MDS (metadata server) daemons is responsible for
23 * managing the file system namespace (the directory hierarchy and
24 * inodes) and for coordinating shared access to storage. Metadata is
25 * partitioning hierarchically across a number of servers, and that
26 * partition varies over time as the cluster adjusts the distribution
27 * in order to balance load.
28 *
29 * The MDS client is primarily responsible to managing synchronous
30 * metadata requests for operations like open, unlink, and so forth.
31 * If there is a MDS failure, we find out about it when we (possibly
32 * request and) receive a new MDS map, and can resubmit affected
33 * requests.
34 *
35 * For the most part, though, we take advantage of a lossless
36 * communications channel to the MDS, and do not need to worry about
37 * timing out or resubmitting requests.
38 *
39 * We maintain a stateful "session" with each MDS we interact with.
40 * Within each session, we sent periodic heartbeat messages to ensure
41 * any capabilities or leases we have been issues remain valid. If
42 * the session times out and goes stale, our leases and capabilities
43 * are no longer valid.
44 */
45
Sage Weil20cb34a2010-05-12 15:21:32 -070046struct ceph_reconnect_state {
Yan, Zheng44c99752013-09-22 10:28:10 +080047 int nr_caps;
Sage Weil20cb34a2010-05-12 15:21:32 -070048 struct ceph_pagelist *pagelist;
49 bool flock;
50};
51
Sage Weil2f2dc052009-10-06 11:31:09 -070052static void __wake_requests(struct ceph_mds_client *mdsc,
53 struct list_head *head);
54
Tobias Klauser9e327892010-05-20 10:40:19 +020055static const struct ceph_connection_operations mds_con_ops;
Sage Weil2f2dc052009-10-06 11:31:09 -070056
57
58/*
59 * mds reply parsing
60 */
61
62/*
63 * parse individual inode info
64 */
65static int parse_reply_info_in(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -080066 struct ceph_mds_reply_info_in *info,
Ilya Dryomov12b46292013-12-24 21:19:23 +020067 u64 features)
Sage Weil2f2dc052009-10-06 11:31:09 -070068{
69 int err = -EIO;
70
71 info->in = *p;
72 *p += sizeof(struct ceph_mds_reply_inode) +
73 sizeof(*info->in->fragtree.splits) *
74 le32_to_cpu(info->in->fragtree.nsplits);
75
76 ceph_decode_32_safe(p, end, info->symlink_len, bad);
77 ceph_decode_need(p, end, info->symlink_len, bad);
78 info->symlink = *p;
79 *p += info->symlink_len;
80
Sage Weil14303d22010-12-14 17:37:52 -080081 if (features & CEPH_FEATURE_DIRLAYOUTHASH)
82 ceph_decode_copy_safe(p, end, &info->dir_layout,
83 sizeof(info->dir_layout), bad);
84 else
85 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
86
Sage Weil2f2dc052009-10-06 11:31:09 -070087 ceph_decode_32_safe(p, end, info->xattr_len, bad);
88 ceph_decode_need(p, end, info->xattr_len, bad);
89 info->xattr_data = *p;
90 *p += info->xattr_len;
91 return 0;
92bad:
93 return err;
94}
95
96/*
97 * parse a normal reply, which may contain a (dir+)dentry and/or a
98 * target inode.
99 */
100static int parse_reply_info_trace(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -0800101 struct ceph_mds_reply_info_parsed *info,
Ilya Dryomov12b46292013-12-24 21:19:23 +0200102 u64 features)
Sage Weil2f2dc052009-10-06 11:31:09 -0700103{
104 int err;
105
106 if (info->head->is_dentry) {
Sage Weil14303d22010-12-14 17:37:52 -0800107 err = parse_reply_info_in(p, end, &info->diri, features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700108 if (err < 0)
109 goto out_bad;
110
111 if (unlikely(*p + sizeof(*info->dirfrag) > end))
112 goto bad;
113 info->dirfrag = *p;
114 *p += sizeof(*info->dirfrag) +
115 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
116 if (unlikely(*p > end))
117 goto bad;
118
119 ceph_decode_32_safe(p, end, info->dname_len, bad);
120 ceph_decode_need(p, end, info->dname_len, bad);
121 info->dname = *p;
122 *p += info->dname_len;
123 info->dlease = *p;
124 *p += sizeof(*info->dlease);
125 }
126
127 if (info->head->is_target) {
Sage Weil14303d22010-12-14 17:37:52 -0800128 err = parse_reply_info_in(p, end, &info->targeti, features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700129 if (err < 0)
130 goto out_bad;
131 }
132
133 if (unlikely(*p != end))
134 goto bad;
135 return 0;
136
137bad:
138 err = -EIO;
139out_bad:
140 pr_err("problem parsing mds trace %d\n", err);
141 return err;
142}
143
144/*
145 * parse readdir results
146 */
147static int parse_reply_info_dir(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -0800148 struct ceph_mds_reply_info_parsed *info,
Ilya Dryomov12b46292013-12-24 21:19:23 +0200149 u64 features)
Sage Weil2f2dc052009-10-06 11:31:09 -0700150{
151 u32 num, i = 0;
152 int err;
153
154 info->dir_dir = *p;
155 if (*p + sizeof(*info->dir_dir) > end)
156 goto bad;
157 *p += sizeof(*info->dir_dir) +
158 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
159 if (*p > end)
160 goto bad;
161
162 ceph_decode_need(p, end, sizeof(num) + 2, bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700163 num = ceph_decode_32(p);
164 info->dir_end = ceph_decode_8(p);
165 info->dir_complete = ceph_decode_8(p);
Sage Weil2f2dc052009-10-06 11:31:09 -0700166 if (num == 0)
167 goto done;
168
Yan, Zheng54008392014-03-29 13:41:15 +0800169 BUG_ON(!info->dir_in);
Sage Weil2f2dc052009-10-06 11:31:09 -0700170 info->dir_dname = (void *)(info->dir_in + num);
171 info->dir_dname_len = (void *)(info->dir_dname + num);
172 info->dir_dlease = (void *)(info->dir_dname_len + num);
Yan, Zheng54008392014-03-29 13:41:15 +0800173 if ((unsigned long)(info->dir_dlease + num) >
174 (unsigned long)info->dir_in + info->dir_buf_size) {
175 pr_err("dir contents are larger than expected\n");
176 WARN_ON(1);
177 goto bad;
178 }
Sage Weil2f2dc052009-10-06 11:31:09 -0700179
Yan, Zheng54008392014-03-29 13:41:15 +0800180 info->dir_nr = num;
Sage Weil2f2dc052009-10-06 11:31:09 -0700181 while (num) {
182 /* dentry */
183 ceph_decode_need(p, end, sizeof(u32)*2, bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700184 info->dir_dname_len[i] = ceph_decode_32(p);
Sage Weil2f2dc052009-10-06 11:31:09 -0700185 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
186 info->dir_dname[i] = *p;
187 *p += info->dir_dname_len[i];
188 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
189 info->dir_dname[i]);
190 info->dir_dlease[i] = *p;
191 *p += sizeof(struct ceph_mds_reply_lease);
192
193 /* inode */
Sage Weil14303d22010-12-14 17:37:52 -0800194 err = parse_reply_info_in(p, end, &info->dir_in[i], features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700195 if (err < 0)
196 goto out_bad;
197 i++;
198 num--;
199 }
200
201done:
202 if (*p != end)
203 goto bad;
204 return 0;
205
206bad:
207 err = -EIO;
208out_bad:
209 pr_err("problem parsing dir contents %d\n", err);
210 return err;
211}
212
213/*
Herb Shiu25933ab2010-12-01 14:14:38 -0800214 * parse fcntl F_GETLK results
215 */
216static int parse_reply_info_filelock(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -0800217 struct ceph_mds_reply_info_parsed *info,
Ilya Dryomov12b46292013-12-24 21:19:23 +0200218 u64 features)
Herb Shiu25933ab2010-12-01 14:14:38 -0800219{
220 if (*p + sizeof(*info->filelock_reply) > end)
221 goto bad;
222
223 info->filelock_reply = *p;
224 *p += sizeof(*info->filelock_reply);
225
226 if (unlikely(*p != end))
227 goto bad;
228 return 0;
229
230bad:
231 return -EIO;
232}
233
234/*
Sam Lang6e8575f2012-12-28 09:56:46 -0800235 * parse create results
236 */
237static int parse_reply_info_create(void **p, void *end,
238 struct ceph_mds_reply_info_parsed *info,
Ilya Dryomov12b46292013-12-24 21:19:23 +0200239 u64 features)
Sam Lang6e8575f2012-12-28 09:56:46 -0800240{
241 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
242 if (*p == end) {
243 info->has_create_ino = false;
244 } else {
245 info->has_create_ino = true;
246 info->ino = ceph_decode_64(p);
247 }
248 }
249
250 if (unlikely(*p != end))
251 goto bad;
252 return 0;
253
254bad:
255 return -EIO;
256}
257
258/*
Herb Shiu25933ab2010-12-01 14:14:38 -0800259 * parse extra results
260 */
261static int parse_reply_info_extra(void **p, void *end,
Sage Weil14303d22010-12-14 17:37:52 -0800262 struct ceph_mds_reply_info_parsed *info,
Ilya Dryomov12b46292013-12-24 21:19:23 +0200263 u64 features)
Herb Shiu25933ab2010-12-01 14:14:38 -0800264{
265 if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
Sage Weil14303d22010-12-14 17:37:52 -0800266 return parse_reply_info_filelock(p, end, info, features);
Yan, Zheng8a034492013-02-21 13:43:55 +0800267 else if (info->head->op == CEPH_MDS_OP_READDIR ||
268 info->head->op == CEPH_MDS_OP_LSSNAP)
Sage Weil14303d22010-12-14 17:37:52 -0800269 return parse_reply_info_dir(p, end, info, features);
Sam Lang6e8575f2012-12-28 09:56:46 -0800270 else if (info->head->op == CEPH_MDS_OP_CREATE)
271 return parse_reply_info_create(p, end, info, features);
272 else
273 return -EIO;
Herb Shiu25933ab2010-12-01 14:14:38 -0800274}
275
276/*
Sage Weil2f2dc052009-10-06 11:31:09 -0700277 * parse entire mds reply
278 */
279static int parse_reply_info(struct ceph_msg *msg,
Sage Weil14303d22010-12-14 17:37:52 -0800280 struct ceph_mds_reply_info_parsed *info,
Ilya Dryomov12b46292013-12-24 21:19:23 +0200281 u64 features)
Sage Weil2f2dc052009-10-06 11:31:09 -0700282{
283 void *p, *end;
284 u32 len;
285 int err;
286
287 info->head = msg->front.iov_base;
288 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
289 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
290
291 /* trace */
292 ceph_decode_32_safe(&p, end, len, bad);
293 if (len > 0) {
Xi Wang32852a82012-01-14 22:20:59 -0500294 ceph_decode_need(&p, end, len, bad);
Sage Weil14303d22010-12-14 17:37:52 -0800295 err = parse_reply_info_trace(&p, p+len, info, features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700296 if (err < 0)
297 goto out_bad;
298 }
299
Herb Shiu25933ab2010-12-01 14:14:38 -0800300 /* extra */
Sage Weil2f2dc052009-10-06 11:31:09 -0700301 ceph_decode_32_safe(&p, end, len, bad);
302 if (len > 0) {
Xi Wang32852a82012-01-14 22:20:59 -0500303 ceph_decode_need(&p, end, len, bad);
Sage Weil14303d22010-12-14 17:37:52 -0800304 err = parse_reply_info_extra(&p, p+len, info, features);
Sage Weil2f2dc052009-10-06 11:31:09 -0700305 if (err < 0)
306 goto out_bad;
307 }
308
309 /* snap blob */
310 ceph_decode_32_safe(&p, end, len, bad);
311 info->snapblob_len = len;
312 info->snapblob = p;
313 p += len;
314
315 if (p != end)
316 goto bad;
317 return 0;
318
319bad:
320 err = -EIO;
321out_bad:
322 pr_err("mds parse_reply err %d\n", err);
323 return err;
324}
325
326static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
327{
Yan, Zheng54008392014-03-29 13:41:15 +0800328 if (!info->dir_in)
329 return;
330 free_pages((unsigned long)info->dir_in, get_order(info->dir_buf_size));
Sage Weil2f2dc052009-10-06 11:31:09 -0700331}
332
333
334/*
335 * sessions
336 */
337static const char *session_state_name(int s)
338{
339 switch (s) {
340 case CEPH_MDS_SESSION_NEW: return "new";
341 case CEPH_MDS_SESSION_OPENING: return "opening";
342 case CEPH_MDS_SESSION_OPEN: return "open";
343 case CEPH_MDS_SESSION_HUNG: return "hung";
344 case CEPH_MDS_SESSION_CLOSING: return "closing";
Sage Weil44ca18f2010-02-15 12:08:46 -0800345 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
Sage Weil2f2dc052009-10-06 11:31:09 -0700346 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
347 default: return "???";
348 }
349}
350
351static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
352{
353 if (atomic_inc_not_zero(&s->s_ref)) {
354 dout("mdsc get_session %p %d -> %d\n", s,
355 atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
356 return s;
357 } else {
358 dout("mdsc get_session %p 0 -- FAIL", s);
359 return NULL;
360 }
361}
362
363void ceph_put_mds_session(struct ceph_mds_session *s)
364{
365 dout("mdsc put_session %p %d -> %d\n", s,
366 atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800367 if (atomic_dec_and_test(&s->s_ref)) {
Alex Elder6c4a1912012-05-16 15:16:38 -0500368 if (s->s_auth.authorizer)
Sage Weil27859f92013-03-25 10:26:14 -0700369 ceph_auth_destroy_authorizer(
370 s->s_mdsc->fsc->client->monc.auth,
371 s->s_auth.authorizer);
Sage Weil2f2dc052009-10-06 11:31:09 -0700372 kfree(s);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800373 }
Sage Weil2f2dc052009-10-06 11:31:09 -0700374}
375
376/*
377 * called under mdsc->mutex
378 */
379struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
380 int mds)
381{
382 struct ceph_mds_session *session;
383
384 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
385 return NULL;
386 session = mdsc->sessions[mds];
387 dout("lookup_mds_session %p %d\n", session,
388 atomic_read(&session->s_ref));
389 get_session(session);
390 return session;
391}
392
393static bool __have_session(struct ceph_mds_client *mdsc, int mds)
394{
395 if (mds >= mdsc->max_sessions)
396 return false;
397 return mdsc->sessions[mds];
398}
399
Sage Weil2600d2d2010-02-22 15:12:16 -0800400static int __verify_registered_session(struct ceph_mds_client *mdsc,
401 struct ceph_mds_session *s)
402{
403 if (s->s_mds >= mdsc->max_sessions ||
404 mdsc->sessions[s->s_mds] != s)
405 return -ENOENT;
406 return 0;
407}
408
Sage Weil2f2dc052009-10-06 11:31:09 -0700409/*
410 * create+register a new session for given mds.
411 * called under mdsc->mutex.
412 */
413static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
414 int mds)
415{
416 struct ceph_mds_session *s;
417
Nathaniel Yazdanic338c072013-08-04 21:04:30 -0700418 if (mds >= mdsc->mdsmap->m_max_mds)
419 return ERR_PTR(-EINVAL);
420
Sage Weil2f2dc052009-10-06 11:31:09 -0700421 s = kzalloc(sizeof(*s), GFP_NOFS);
Dan Carpenter4736b002010-03-20 15:30:16 +0300422 if (!s)
423 return ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -0700424 s->s_mdsc = mdsc;
425 s->s_mds = mds;
426 s->s_state = CEPH_MDS_SESSION_NEW;
427 s->s_ttl = 0;
428 s->s_seq = 0;
429 mutex_init(&s->s_mutex);
430
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700431 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
Sage Weil2f2dc052009-10-06 11:31:09 -0700432
Alex Elderd8fb02a2012-01-12 17:48:10 -0800433 spin_lock_init(&s->s_gen_ttl_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700434 s->s_cap_gen = 0;
Alex Elder1ce208a2012-01-12 17:48:11 -0800435 s->s_cap_ttl = jiffies - 1;
Alex Elderd8fb02a2012-01-12 17:48:10 -0800436
437 spin_lock_init(&s->s_cap_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700438 s->s_renew_requested = 0;
439 s->s_renew_seq = 0;
440 INIT_LIST_HEAD(&s->s_caps);
441 s->s_nr_caps = 0;
Sage Weil5dacf092009-12-21 20:40:34 -0800442 s->s_trim_caps = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -0700443 atomic_set(&s->s_ref, 1);
444 INIT_LIST_HEAD(&s->s_waiting);
445 INIT_LIST_HEAD(&s->s_unsafe);
446 s->s_num_cap_releases = 0;
Yan, Zheng99a9c272013-09-22 11:08:14 +0800447 s->s_cap_reconnect = 0;
Sage Weil7c1332b2010-02-16 11:39:45 -0800448 s->s_cap_iterator = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700449 INIT_LIST_HEAD(&s->s_cap_releases);
450 INIT_LIST_HEAD(&s->s_cap_releases_done);
451 INIT_LIST_HEAD(&s->s_cap_flushing);
452 INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
453
454 dout("register_session mds%d\n", mds);
455 if (mds >= mdsc->max_sessions) {
456 int newmax = 1 << get_count_order(mds+1);
457 struct ceph_mds_session **sa;
458
459 dout("register_session realloc to %d\n", newmax);
460 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
461 if (sa == NULL)
Sage Weil42ce56e2009-11-18 11:22:36 -0800462 goto fail_realloc;
Sage Weil2f2dc052009-10-06 11:31:09 -0700463 if (mdsc->sessions) {
464 memcpy(sa, mdsc->sessions,
465 mdsc->max_sessions * sizeof(void *));
466 kfree(mdsc->sessions);
467 }
468 mdsc->sessions = sa;
469 mdsc->max_sessions = newmax;
470 }
471 mdsc->sessions[mds] = s;
472 atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */
Sage Weil42ce56e2009-11-18 11:22:36 -0800473
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700474 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
475 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
Sage Weil42ce56e2009-11-18 11:22:36 -0800476
Sage Weil2f2dc052009-10-06 11:31:09 -0700477 return s;
Sage Weil42ce56e2009-11-18 11:22:36 -0800478
479fail_realloc:
480 kfree(s);
481 return ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -0700482}
483
484/*
485 * called under mdsc->mutex
486 */
Sage Weil2600d2d2010-02-22 15:12:16 -0800487static void __unregister_session(struct ceph_mds_client *mdsc,
Sage Weil42ce56e2009-11-18 11:22:36 -0800488 struct ceph_mds_session *s)
Sage Weil2f2dc052009-10-06 11:31:09 -0700489{
Sage Weil2600d2d2010-02-22 15:12:16 -0800490 dout("__unregister_session mds%d %p\n", s->s_mds, s);
491 BUG_ON(mdsc->sessions[s->s_mds] != s);
Sage Weil42ce56e2009-11-18 11:22:36 -0800492 mdsc->sessions[s->s_mds] = NULL;
493 ceph_con_close(&s->s_con);
494 ceph_put_mds_session(s);
Sage Weil2f2dc052009-10-06 11:31:09 -0700495}
496
497/*
498 * drop session refs in request.
499 *
500 * should be last request ref, or hold mdsc->mutex
501 */
502static void put_request_session(struct ceph_mds_request *req)
503{
504 if (req->r_session) {
505 ceph_put_mds_session(req->r_session);
506 req->r_session = NULL;
507 }
508}
509
Sage Weil153c8e62009-12-07 12:31:09 -0800510void ceph_mdsc_release_request(struct kref *kref)
Sage Weil2f2dc052009-10-06 11:31:09 -0700511{
Sage Weil153c8e62009-12-07 12:31:09 -0800512 struct ceph_mds_request *req = container_of(kref,
513 struct ceph_mds_request,
514 r_kref);
Yan, Zheng54008392014-03-29 13:41:15 +0800515 destroy_reply_info(&req->r_reply_info);
Sage Weil153c8e62009-12-07 12:31:09 -0800516 if (req->r_request)
517 ceph_msg_put(req->r_request);
Yan, Zheng54008392014-03-29 13:41:15 +0800518 if (req->r_reply)
Sage Weil153c8e62009-12-07 12:31:09 -0800519 ceph_msg_put(req->r_reply);
Sage Weil153c8e62009-12-07 12:31:09 -0800520 if (req->r_inode) {
Sage Weil41b02e12011-07-26 11:31:14 -0700521 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
Sage Weil153c8e62009-12-07 12:31:09 -0800522 iput(req->r_inode);
523 }
524 if (req->r_locked_dir)
Sage Weil41b02e12011-07-26 11:31:14 -0700525 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
Sage Weil153c8e62009-12-07 12:31:09 -0800526 if (req->r_target_inode)
527 iput(req->r_target_inode);
528 if (req->r_dentry)
529 dput(req->r_dentry);
Sage Weil844d87c2013-02-05 13:40:09 -0800530 if (req->r_old_dentry)
531 dput(req->r_old_dentry);
532 if (req->r_old_dentry_dir) {
Sage Weil41b02e12011-07-26 11:31:14 -0700533 /*
534 * track (and drop pins for) r_old_dentry_dir
535 * separately, since r_old_dentry's d_parent may have
536 * changed between the dir mutex being dropped and
537 * this request being freed.
538 */
539 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
540 CEPH_CAP_PIN);
Sage Weil41b02e12011-07-26 11:31:14 -0700541 iput(req->r_old_dentry_dir);
Sage Weil153c8e62009-12-07 12:31:09 -0800542 }
543 kfree(req->r_path1);
544 kfree(req->r_path2);
545 put_request_session(req);
Yehuda Sadeh37151662010-06-17 16:16:12 -0700546 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
Sage Weil153c8e62009-12-07 12:31:09 -0800547 kfree(req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700548}
549
550/*
551 * lookup session, bump ref if found.
552 *
553 * called under mdsc->mutex.
554 */
555static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
556 u64 tid)
557{
558 struct ceph_mds_request *req;
Sage Weil44ca18f2010-02-15 12:08:46 -0800559 struct rb_node *n = mdsc->request_tree.rb_node;
560
561 while (n) {
562 req = rb_entry(n, struct ceph_mds_request, r_node);
563 if (tid < req->r_tid)
564 n = n->rb_left;
565 else if (tid > req->r_tid)
566 n = n->rb_right;
567 else {
568 ceph_mdsc_get_request(req);
569 return req;
570 }
571 }
572 return NULL;
573}
574
575static void __insert_request(struct ceph_mds_client *mdsc,
576 struct ceph_mds_request *new)
577{
578 struct rb_node **p = &mdsc->request_tree.rb_node;
579 struct rb_node *parent = NULL;
580 struct ceph_mds_request *req = NULL;
581
582 while (*p) {
583 parent = *p;
584 req = rb_entry(parent, struct ceph_mds_request, r_node);
585 if (new->r_tid < req->r_tid)
586 p = &(*p)->rb_left;
587 else if (new->r_tid > req->r_tid)
588 p = &(*p)->rb_right;
589 else
590 BUG();
591 }
592
593 rb_link_node(&new->r_node, parent, p);
594 rb_insert_color(&new->r_node, &mdsc->request_tree);
Sage Weil2f2dc052009-10-06 11:31:09 -0700595}
596
597/*
598 * Register an in-flight request, and assign a tid. Link to directory
599 * are modifying (if any).
600 *
601 * Called under mdsc->mutex.
602 */
603static void __register_request(struct ceph_mds_client *mdsc,
604 struct ceph_mds_request *req,
605 struct inode *dir)
606{
607 req->r_tid = ++mdsc->last_tid;
608 if (req->r_num_caps)
Yehuda Sadeh37151662010-06-17 16:16:12 -0700609 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
610 req->r_num_caps);
Sage Weil2f2dc052009-10-06 11:31:09 -0700611 dout("__register_request %p tid %lld\n", req, req->r_tid);
612 ceph_mdsc_get_request(req);
Sage Weil44ca18f2010-02-15 12:08:46 -0800613 __insert_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700614
Sage Weilcb4276c2010-11-08 07:28:52 -0800615 req->r_uid = current_fsuid();
616 req->r_gid = current_fsgid();
617
Sage Weil2f2dc052009-10-06 11:31:09 -0700618 if (dir) {
619 struct ceph_inode_info *ci = ceph_inode(dir);
620
Sage Weil3b663782011-05-18 16:12:12 -0700621 ihold(dir);
Sage Weil2f2dc052009-10-06 11:31:09 -0700622 spin_lock(&ci->i_unsafe_lock);
623 req->r_unsafe_dir = dir;
624 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
625 spin_unlock(&ci->i_unsafe_lock);
626 }
627}
628
629static void __unregister_request(struct ceph_mds_client *mdsc,
630 struct ceph_mds_request *req)
631{
632 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
Sage Weil44ca18f2010-02-15 12:08:46 -0800633 rb_erase(&req->r_node, &mdsc->request_tree);
Sage Weil80fc7312010-03-16 15:28:54 -0700634 RB_CLEAR_NODE(&req->r_node);
Sage Weil2f2dc052009-10-06 11:31:09 -0700635
636 if (req->r_unsafe_dir) {
637 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
638
639 spin_lock(&ci->i_unsafe_lock);
640 list_del_init(&req->r_unsafe_dir_item);
641 spin_unlock(&ci->i_unsafe_lock);
Sage Weil3b663782011-05-18 16:12:12 -0700642
643 iput(req->r_unsafe_dir);
644 req->r_unsafe_dir = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700645 }
Sage Weil94aa8ae2010-03-28 21:22:50 -0700646
Yan, Zhengfc55d2c2013-10-31 09:10:47 +0800647 complete_all(&req->r_safe_completion);
648
Sage Weil94aa8ae2010-03-28 21:22:50 -0700649 ceph_mdsc_put_request(req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700650}
651
652/*
653 * Choose mds to send request to next. If there is a hint set in the
654 * request (e.g., due to a prior forward hint from the mds), use that.
655 * Otherwise, consult frag tree and/or caps to identify the
656 * appropriate mds. If all else fails, choose randomly.
657 *
658 * Called under mdsc->mutex.
659 */
H Hartley Sweeten7fd7d102011-09-23 13:22:11 -0700660static struct dentry *get_nonsnap_parent(struct dentry *dentry)
Sage Weileb6bb1c2010-08-16 09:21:27 -0700661{
Sage Weild79698d2011-07-26 11:31:26 -0700662 /*
663 * we don't need to worry about protecting the d_parent access
664 * here because we never renaming inside the snapped namespace
665 * except to resplice to another snapdir, and either the old or new
666 * result is a valid result.
667 */
Sage Weileb6bb1c2010-08-16 09:21:27 -0700668 while (!IS_ROOT(dentry) && ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
669 dentry = dentry->d_parent;
670 return dentry;
671}
672
Sage Weil2f2dc052009-10-06 11:31:09 -0700673static int __choose_mds(struct ceph_mds_client *mdsc,
674 struct ceph_mds_request *req)
675{
676 struct inode *inode;
677 struct ceph_inode_info *ci;
678 struct ceph_cap *cap;
679 int mode = req->r_direct_mode;
680 int mds = -1;
681 u32 hash = req->r_direct_hash;
682 bool is_hash = req->r_direct_is_hash;
683
684 /*
685 * is there a specific mds we should try? ignore hint if we have
686 * no session and the mds is not up (active or recovering).
687 */
688 if (req->r_resend_mds >= 0 &&
689 (__have_session(mdsc, req->r_resend_mds) ||
690 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
691 dout("choose_mds using resend_mds mds%d\n",
692 req->r_resend_mds);
693 return req->r_resend_mds;
694 }
695
696 if (mode == USE_RANDOM_MDS)
697 goto random;
698
699 inode = NULL;
700 if (req->r_inode) {
701 inode = req->r_inode;
702 } else if (req->r_dentry) {
Sage Weild79698d2011-07-26 11:31:26 -0700703 /* ignore race with rename; old or new d_parent is okay */
704 struct dentry *parent = req->r_dentry->d_parent;
705 struct inode *dir = parent->d_inode;
Sage Weileb6bb1c2010-08-16 09:21:27 -0700706
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700707 if (dir->i_sb != mdsc->fsc->sb) {
Sage Weileb6bb1c2010-08-16 09:21:27 -0700708 /* not this fs! */
709 inode = req->r_dentry->d_inode;
710 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
711 /* direct snapped/virtual snapdir requests
712 * based on parent dir inode */
Sage Weild79698d2011-07-26 11:31:26 -0700713 struct dentry *dn = get_nonsnap_parent(parent);
Sage Weileb6bb1c2010-08-16 09:21:27 -0700714 inode = dn->d_inode;
715 dout("__choose_mds using nonsnap parent %p\n", inode);
Yan, Zhengca18bed2013-11-22 14:21:44 +0800716 } else {
Sage Weileb6bb1c2010-08-16 09:21:27 -0700717 /* dentry target */
Sage Weil2f2dc052009-10-06 11:31:09 -0700718 inode = req->r_dentry->d_inode;
Yan, Zhengca18bed2013-11-22 14:21:44 +0800719 if (!inode || mode == USE_AUTH_MDS) {
720 /* dir + name */
721 inode = dir;
722 hash = ceph_dentry_hash(dir, req->r_dentry);
723 is_hash = true;
724 }
Sage Weil2f2dc052009-10-06 11:31:09 -0700725 }
726 }
Sage Weileb6bb1c2010-08-16 09:21:27 -0700727
Sage Weil2f2dc052009-10-06 11:31:09 -0700728 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
729 (int)hash, mode);
730 if (!inode)
731 goto random;
732 ci = ceph_inode(inode);
733
734 if (is_hash && S_ISDIR(inode->i_mode)) {
735 struct ceph_inode_frag frag;
736 int found;
737
738 ceph_choose_frag(ci, hash, &frag, &found);
739 if (found) {
740 if (mode == USE_ANY_MDS && frag.ndist > 0) {
741 u8 r;
742
743 /* choose a random replica */
744 get_random_bytes(&r, 1);
745 r %= frag.ndist;
746 mds = frag.dist[r];
747 dout("choose_mds %p %llx.%llx "
748 "frag %u mds%d (%d/%d)\n",
749 inode, ceph_vinop(inode),
Sage Weild66bbd42011-01-21 21:16:46 -0800750 frag.frag, mds,
Sage Weil2f2dc052009-10-06 11:31:09 -0700751 (int)r, frag.ndist);
Sage Weild66bbd42011-01-21 21:16:46 -0800752 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
753 CEPH_MDS_STATE_ACTIVE)
754 return mds;
Sage Weil2f2dc052009-10-06 11:31:09 -0700755 }
756
757 /* since this file/dir wasn't known to be
758 * replicated, then we want to look for the
759 * authoritative mds. */
760 mode = USE_AUTH_MDS;
761 if (frag.mds >= 0) {
762 /* choose auth mds */
763 mds = frag.mds;
764 dout("choose_mds %p %llx.%llx "
765 "frag %u mds%d (auth)\n",
766 inode, ceph_vinop(inode), frag.frag, mds);
Sage Weild66bbd42011-01-21 21:16:46 -0800767 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
768 CEPH_MDS_STATE_ACTIVE)
769 return mds;
Sage Weil2f2dc052009-10-06 11:31:09 -0700770 }
771 }
772 }
773
Sage Weilbe655592011-11-30 09:47:09 -0800774 spin_lock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700775 cap = NULL;
776 if (mode == USE_AUTH_MDS)
777 cap = ci->i_auth_cap;
778 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
779 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
780 if (!cap) {
Sage Weilbe655592011-11-30 09:47:09 -0800781 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700782 goto random;
783 }
784 mds = cap->session->s_mds;
785 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
786 inode, ceph_vinop(inode), mds,
787 cap == ci->i_auth_cap ? "auth " : "", cap);
Sage Weilbe655592011-11-30 09:47:09 -0800788 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -0700789 return mds;
790
791random:
792 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
793 dout("choose_mds chose random mds%d\n", mds);
794 return mds;
795}
796
797
798/*
799 * session messages
800 */
801static struct ceph_msg *create_session_msg(u32 op, u64 seq)
802{
803 struct ceph_msg *msg;
804 struct ceph_mds_session_head *h;
805
Sage Weilb61c2762011-08-09 15:03:46 -0700806 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
807 false);
Sage Weila79832f2010-04-01 16:06:19 -0700808 if (!msg) {
Sage Weil2f2dc052009-10-06 11:31:09 -0700809 pr_err("create_session_msg ENOMEM creating msg\n");
Sage Weila79832f2010-04-01 16:06:19 -0700810 return NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700811 }
812 h = msg->front.iov_base;
813 h->op = cpu_to_le32(op);
814 h->seq = cpu_to_le64(seq);
815 return msg;
816}
817
818/*
819 * send session open request.
820 *
821 * called under mdsc->mutex
822 */
823static int __open_session(struct ceph_mds_client *mdsc,
824 struct ceph_mds_session *session)
825{
826 struct ceph_msg *msg;
827 int mstate;
828 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -0700829
830 /* wait for mds to go active? */
831 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
832 dout("open_session to mds%d (%s)\n", mds,
833 ceph_mds_state_name(mstate));
834 session->s_state = CEPH_MDS_SESSION_OPENING;
835 session->s_renew_requested = jiffies;
836
837 /* send connect message */
838 msg = create_session_msg(CEPH_SESSION_REQUEST_OPEN, session->s_seq);
Sage Weila79832f2010-04-01 16:06:19 -0700839 if (!msg)
840 return -ENOMEM;
Sage Weil2f2dc052009-10-06 11:31:09 -0700841 ceph_con_send(&session->s_con, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -0700842 return 0;
843}
844
845/*
Sage Weiled0552a2010-06-21 13:38:25 -0700846 * open sessions for any export targets for the given mds
847 *
848 * called under mdsc->mutex
849 */
Yan, Zheng5d72d132013-11-24 14:33:01 +0800850static struct ceph_mds_session *
851__open_export_target_session(struct ceph_mds_client *mdsc, int target)
852{
853 struct ceph_mds_session *session;
854
855 session = __ceph_lookup_mds_session(mdsc, target);
856 if (!session) {
857 session = register_session(mdsc, target);
858 if (IS_ERR(session))
859 return session;
860 }
861 if (session->s_state == CEPH_MDS_SESSION_NEW ||
862 session->s_state == CEPH_MDS_SESSION_CLOSING)
863 __open_session(mdsc, session);
864
865 return session;
866}
867
868struct ceph_mds_session *
869ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
870{
871 struct ceph_mds_session *session;
872
873 dout("open_export_target_session to mds%d\n", target);
874
875 mutex_lock(&mdsc->mutex);
876 session = __open_export_target_session(mdsc, target);
877 mutex_unlock(&mdsc->mutex);
878
879 return session;
880}
881
Sage Weiled0552a2010-06-21 13:38:25 -0700882static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
883 struct ceph_mds_session *session)
884{
885 struct ceph_mds_info *mi;
886 struct ceph_mds_session *ts;
887 int i, mds = session->s_mds;
Sage Weiled0552a2010-06-21 13:38:25 -0700888
889 if (mds >= mdsc->mdsmap->m_max_mds)
890 return;
Yan, Zheng5d72d132013-11-24 14:33:01 +0800891
Sage Weiled0552a2010-06-21 13:38:25 -0700892 mi = &mdsc->mdsmap->m_info[mds];
893 dout("open_export_target_sessions for mds%d (%d targets)\n",
894 session->s_mds, mi->num_export_targets);
895
896 for (i = 0; i < mi->num_export_targets; i++) {
Yan, Zheng5d72d132013-11-24 14:33:01 +0800897 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
898 if (!IS_ERR(ts))
899 ceph_put_mds_session(ts);
Sage Weiled0552a2010-06-21 13:38:25 -0700900 }
901}
902
Sage Weil154f42c2010-06-21 13:45:04 -0700903void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
904 struct ceph_mds_session *session)
905{
906 mutex_lock(&mdsc->mutex);
907 __open_export_target_sessions(mdsc, session);
908 mutex_unlock(&mdsc->mutex);
909}
910
Sage Weiled0552a2010-06-21 13:38:25 -0700911/*
Sage Weil2f2dc052009-10-06 11:31:09 -0700912 * session caps
913 */
914
915/*
916 * Free preallocated cap messages assigned to this session
917 */
918static void cleanup_cap_releases(struct ceph_mds_session *session)
919{
920 struct ceph_msg *msg;
921
922 spin_lock(&session->s_cap_lock);
923 while (!list_empty(&session->s_cap_releases)) {
924 msg = list_first_entry(&session->s_cap_releases,
925 struct ceph_msg, list_head);
926 list_del_init(&msg->list_head);
927 ceph_msg_put(msg);
928 }
929 while (!list_empty(&session->s_cap_releases_done)) {
930 msg = list_first_entry(&session->s_cap_releases_done,
931 struct ceph_msg, list_head);
932 list_del_init(&msg->list_head);
933 ceph_msg_put(msg);
934 }
935 spin_unlock(&session->s_cap_lock);
936}
937
938/*
Sage Weilf818a732010-05-11 20:56:31 -0700939 * Helper to safely iterate over all caps associated with a session, with
940 * special care taken to handle a racing __ceph_remove_cap().
Sage Weil2f2dc052009-10-06 11:31:09 -0700941 *
Sage Weilf818a732010-05-11 20:56:31 -0700942 * Caller must hold session s_mutex.
Sage Weil2f2dc052009-10-06 11:31:09 -0700943 */
944static int iterate_session_caps(struct ceph_mds_session *session,
945 int (*cb)(struct inode *, struct ceph_cap *,
946 void *), void *arg)
947{
Sage Weil7c1332b2010-02-16 11:39:45 -0800948 struct list_head *p;
949 struct ceph_cap *cap;
950 struct inode *inode, *last_inode = NULL;
951 struct ceph_cap *old_cap = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700952 int ret;
953
954 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
955 spin_lock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800956 p = session->s_caps.next;
957 while (p != &session->s_caps) {
958 cap = list_entry(p, struct ceph_cap, session_caps);
Sage Weil2f2dc052009-10-06 11:31:09 -0700959 inode = igrab(&cap->ci->vfs_inode);
Sage Weil7c1332b2010-02-16 11:39:45 -0800960 if (!inode) {
961 p = p->next;
Sage Weil2f2dc052009-10-06 11:31:09 -0700962 continue;
Sage Weil7c1332b2010-02-16 11:39:45 -0800963 }
964 session->s_cap_iterator = cap;
Sage Weil2f2dc052009-10-06 11:31:09 -0700965 spin_unlock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800966
967 if (last_inode) {
968 iput(last_inode);
969 last_inode = NULL;
970 }
971 if (old_cap) {
Yehuda Sadeh37151662010-06-17 16:16:12 -0700972 ceph_put_cap(session->s_mdsc, old_cap);
Sage Weil7c1332b2010-02-16 11:39:45 -0800973 old_cap = NULL;
974 }
975
Sage Weil2f2dc052009-10-06 11:31:09 -0700976 ret = cb(inode, cap, arg);
Sage Weil7c1332b2010-02-16 11:39:45 -0800977 last_inode = inode;
978
Sage Weil2f2dc052009-10-06 11:31:09 -0700979 spin_lock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800980 p = p->next;
981 if (cap->ci == NULL) {
982 dout("iterate_session_caps finishing cap %p removal\n",
983 cap);
984 BUG_ON(cap->session != session);
985 list_del_init(&cap->session_caps);
986 session->s_nr_caps--;
987 cap->session = NULL;
988 old_cap = cap; /* put_cap it w/o locks held */
989 }
Sage Weil5dacf092009-12-21 20:40:34 -0800990 if (ret < 0)
991 goto out;
Sage Weil2f2dc052009-10-06 11:31:09 -0700992 }
Sage Weil5dacf092009-12-21 20:40:34 -0800993 ret = 0;
994out:
Sage Weil7c1332b2010-02-16 11:39:45 -0800995 session->s_cap_iterator = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700996 spin_unlock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800997
998 if (last_inode)
999 iput(last_inode);
1000 if (old_cap)
Yehuda Sadeh37151662010-06-17 16:16:12 -07001001 ceph_put_cap(session->s_mdsc, old_cap);
Sage Weil7c1332b2010-02-16 11:39:45 -08001002
Sage Weil5dacf092009-12-21 20:40:34 -08001003 return ret;
Sage Weil2f2dc052009-10-06 11:31:09 -07001004}
1005
1006static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
Sage Weil6c99f252010-05-10 16:12:25 -07001007 void *arg)
Sage Weil2f2dc052009-10-06 11:31:09 -07001008{
1009 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil6c99f252010-05-10 16:12:25 -07001010 int drop = 0;
1011
Sage Weil2f2dc052009-10-06 11:31:09 -07001012 dout("removing cap %p, ci is %p, inode is %p\n",
1013 cap, ci, &ci->vfs_inode);
Sage Weilbe655592011-11-30 09:47:09 -08001014 spin_lock(&ci->i_ceph_lock);
Yan, Zhenga096b092013-09-22 10:15:58 +08001015 __ceph_remove_cap(cap, false);
Sage Weil6c99f252010-05-10 16:12:25 -07001016 if (!__ceph_is_any_real_caps(ci)) {
1017 struct ceph_mds_client *mdsc =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001018 ceph_sb_to_client(inode->i_sb)->mdsc;
Sage Weil6c99f252010-05-10 16:12:25 -07001019
1020 spin_lock(&mdsc->cap_dirty_lock);
1021 if (!list_empty(&ci->i_dirty_item)) {
1022 pr_info(" dropping dirty %s state for %p %lld\n",
1023 ceph_cap_string(ci->i_dirty_caps),
1024 inode, ceph_ino(inode));
1025 ci->i_dirty_caps = 0;
1026 list_del_init(&ci->i_dirty_item);
1027 drop = 1;
1028 }
1029 if (!list_empty(&ci->i_flushing_item)) {
1030 pr_info(" dropping dirty+flushing %s state for %p %lld\n",
1031 ceph_cap_string(ci->i_flushing_caps),
1032 inode, ceph_ino(inode));
1033 ci->i_flushing_caps = 0;
1034 list_del_init(&ci->i_flushing_item);
1035 mdsc->num_cap_flushing--;
1036 drop = 1;
1037 }
1038 if (drop && ci->i_wrbuffer_ref) {
1039 pr_info(" dropping dirty data for %p %lld\n",
1040 inode, ceph_ino(inode));
1041 ci->i_wrbuffer_ref = 0;
1042 ci->i_wrbuffer_ref_head = 0;
1043 drop++;
1044 }
1045 spin_unlock(&mdsc->cap_dirty_lock);
1046 }
Sage Weilbe655592011-11-30 09:47:09 -08001047 spin_unlock(&ci->i_ceph_lock);
Sage Weil6c99f252010-05-10 16:12:25 -07001048 while (drop--)
1049 iput(inode);
Sage Weil2f2dc052009-10-06 11:31:09 -07001050 return 0;
1051}
1052
1053/*
1054 * caller must hold session s_mutex
1055 */
1056static void remove_session_caps(struct ceph_mds_session *session)
1057{
1058 dout("remove_session_caps on %p\n", session);
1059 iterate_session_caps(session, remove_session_caps_cb, NULL);
Yan, Zheng6f60f882013-07-24 12:22:11 +08001060
1061 spin_lock(&session->s_cap_lock);
1062 if (session->s_nr_caps > 0) {
1063 struct super_block *sb = session->s_mdsc->fsc->sb;
1064 struct inode *inode;
1065 struct ceph_cap *cap, *prev = NULL;
1066 struct ceph_vino vino;
1067 /*
1068 * iterate_session_caps() skips inodes that are being
1069 * deleted, we need to wait until deletions are complete.
1070 * __wait_on_freeing_inode() is designed for the job,
1071 * but it is not exported, so use lookup inode function
1072 * to access it.
1073 */
1074 while (!list_empty(&session->s_caps)) {
1075 cap = list_entry(session->s_caps.next,
1076 struct ceph_cap, session_caps);
1077 if (cap == prev)
1078 break;
1079 prev = cap;
1080 vino = cap->ci->i_vino;
1081 spin_unlock(&session->s_cap_lock);
1082
Yan, Zhenged284c42013-09-02 15:19:53 +08001083 inode = ceph_find_inode(sb, vino);
Yan, Zheng6f60f882013-07-24 12:22:11 +08001084 iput(inode);
1085
1086 spin_lock(&session->s_cap_lock);
1087 }
1088 }
1089 spin_unlock(&session->s_cap_lock);
1090
Sage Weil2f2dc052009-10-06 11:31:09 -07001091 BUG_ON(session->s_nr_caps > 0);
Sage Weil6c99f252010-05-10 16:12:25 -07001092 BUG_ON(!list_empty(&session->s_cap_flushing));
Sage Weil2f2dc052009-10-06 11:31:09 -07001093 cleanup_cap_releases(session);
1094}
1095
1096/*
1097 * wake up any threads waiting on this session's caps. if the cap is
1098 * old (didn't get renewed on the client reconnect), remove it now.
1099 *
1100 * caller must hold s_mutex.
1101 */
1102static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1103 void *arg)
1104{
Sage Weil0dc25702009-11-20 13:43:45 -08001105 struct ceph_inode_info *ci = ceph_inode(inode);
1106
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001107 wake_up_all(&ci->i_cap_wq);
Sage Weil0dc25702009-11-20 13:43:45 -08001108 if (arg) {
Sage Weilbe655592011-11-30 09:47:09 -08001109 spin_lock(&ci->i_ceph_lock);
Sage Weil0dc25702009-11-20 13:43:45 -08001110 ci->i_wanted_max_size = 0;
1111 ci->i_requested_max_size = 0;
Sage Weilbe655592011-11-30 09:47:09 -08001112 spin_unlock(&ci->i_ceph_lock);
Sage Weil0dc25702009-11-20 13:43:45 -08001113 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001114 return 0;
1115}
1116
Sage Weil0dc25702009-11-20 13:43:45 -08001117static void wake_up_session_caps(struct ceph_mds_session *session,
1118 int reconnect)
Sage Weil2f2dc052009-10-06 11:31:09 -07001119{
1120 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
Sage Weil0dc25702009-11-20 13:43:45 -08001121 iterate_session_caps(session, wake_up_session_cb,
1122 (void *)(unsigned long)reconnect);
Sage Weil2f2dc052009-10-06 11:31:09 -07001123}
1124
1125/*
1126 * Send periodic message to MDS renewing all currently held caps. The
1127 * ack will reset the expiration for all caps from this session.
1128 *
1129 * caller holds s_mutex
1130 */
1131static int send_renew_caps(struct ceph_mds_client *mdsc,
1132 struct ceph_mds_session *session)
1133{
1134 struct ceph_msg *msg;
1135 int state;
1136
1137 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1138 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1139 pr_info("mds%d caps stale\n", session->s_mds);
Sage Weile4cb4cb2010-03-18 13:43:09 -07001140 session->s_renew_requested = jiffies;
Sage Weil2f2dc052009-10-06 11:31:09 -07001141
1142 /* do not try to renew caps until a recovering mds has reconnected
1143 * with its clients. */
1144 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1145 if (state < CEPH_MDS_STATE_RECONNECT) {
1146 dout("send_renew_caps ignoring mds%d (%s)\n",
1147 session->s_mds, ceph_mds_state_name(state));
1148 return 0;
1149 }
1150
1151 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1152 ceph_mds_state_name(state));
Sage Weil2f2dc052009-10-06 11:31:09 -07001153 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1154 ++session->s_renew_seq);
Sage Weila79832f2010-04-01 16:06:19 -07001155 if (!msg)
1156 return -ENOMEM;
Sage Weil2f2dc052009-10-06 11:31:09 -07001157 ceph_con_send(&session->s_con, msg);
1158 return 0;
1159}
1160
Yan, Zheng186e4f72013-11-22 14:48:37 +08001161static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1162 struct ceph_mds_session *session, u64 seq)
1163{
1164 struct ceph_msg *msg;
1165
1166 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1167 session->s_mds, session_state_name(session->s_state), seq);
1168 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1169 if (!msg)
1170 return -ENOMEM;
1171 ceph_con_send(&session->s_con, msg);
1172 return 0;
1173}
1174
1175
Sage Weil2f2dc052009-10-06 11:31:09 -07001176/*
1177 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
Sage Weil0dc25702009-11-20 13:43:45 -08001178 *
1179 * Called under session->s_mutex
Sage Weil2f2dc052009-10-06 11:31:09 -07001180 */
1181static void renewed_caps(struct ceph_mds_client *mdsc,
1182 struct ceph_mds_session *session, int is_renew)
1183{
1184 int was_stale;
1185 int wake = 0;
1186
1187 spin_lock(&session->s_cap_lock);
Alex Elder1ce208a2012-01-12 17:48:11 -08001188 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
Sage Weil2f2dc052009-10-06 11:31:09 -07001189
1190 session->s_cap_ttl = session->s_renew_requested +
1191 mdsc->mdsmap->m_session_timeout*HZ;
1192
1193 if (was_stale) {
1194 if (time_before(jiffies, session->s_cap_ttl)) {
1195 pr_info("mds%d caps renewed\n", session->s_mds);
1196 wake = 1;
1197 } else {
1198 pr_info("mds%d caps still stale\n", session->s_mds);
1199 }
1200 }
1201 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1202 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1203 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1204 spin_unlock(&session->s_cap_lock);
1205
1206 if (wake)
Sage Weil0dc25702009-11-20 13:43:45 -08001207 wake_up_session_caps(session, 0);
Sage Weil2f2dc052009-10-06 11:31:09 -07001208}
1209
1210/*
1211 * send a session close request
1212 */
1213static int request_close_session(struct ceph_mds_client *mdsc,
1214 struct ceph_mds_session *session)
1215{
1216 struct ceph_msg *msg;
Sage Weil2f2dc052009-10-06 11:31:09 -07001217
1218 dout("request_close_session mds%d state %s seq %lld\n",
1219 session->s_mds, session_state_name(session->s_state),
1220 session->s_seq);
1221 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
Sage Weila79832f2010-04-01 16:06:19 -07001222 if (!msg)
1223 return -ENOMEM;
1224 ceph_con_send(&session->s_con, msg);
1225 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001226}
1227
1228/*
1229 * Called with s_mutex held.
1230 */
1231static int __close_session(struct ceph_mds_client *mdsc,
1232 struct ceph_mds_session *session)
1233{
1234 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1235 return 0;
1236 session->s_state = CEPH_MDS_SESSION_CLOSING;
1237 return request_close_session(mdsc, session);
1238}
1239
1240/*
1241 * Trim old(er) caps.
1242 *
1243 * Because we can't cache an inode without one or more caps, we do
1244 * this indirectly: if a cap is unused, we prune its aliases, at which
1245 * point the inode will hopefully get dropped to.
1246 *
1247 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1248 * memory pressure from the MDS, though, so it needn't be perfect.
1249 */
1250static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1251{
1252 struct ceph_mds_session *session = arg;
1253 struct ceph_inode_info *ci = ceph_inode(inode);
Yan, Zheng979abfd2013-11-22 13:56:24 +08001254 int used, wanted, oissued, mine;
Sage Weil2f2dc052009-10-06 11:31:09 -07001255
1256 if (session->s_trim_caps <= 0)
1257 return -1;
1258
Sage Weilbe655592011-11-30 09:47:09 -08001259 spin_lock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001260 mine = cap->issued | cap->implemented;
1261 used = __ceph_caps_used(ci);
Yan, Zheng979abfd2013-11-22 13:56:24 +08001262 wanted = __ceph_caps_file_wanted(ci);
Sage Weil2f2dc052009-10-06 11:31:09 -07001263 oissued = __ceph_caps_issued_other(ci, cap);
1264
Yan, Zheng979abfd2013-11-22 13:56:24 +08001265 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07001266 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
Yan, Zheng979abfd2013-11-22 13:56:24 +08001267 ceph_cap_string(used), ceph_cap_string(wanted));
1268 if (cap == ci->i_auth_cap) {
1269 if (ci->i_dirty_caps | ci->i_flushing_caps)
1270 goto out;
1271 if ((used | wanted) & CEPH_CAP_ANY_WR)
1272 goto out;
1273 }
1274 if ((used | wanted) & ~oissued & mine)
Sage Weil2f2dc052009-10-06 11:31:09 -07001275 goto out; /* we need these caps */
1276
1277 session->s_trim_caps--;
1278 if (oissued) {
1279 /* we aren't the only cap.. just remove us */
Yan, Zhenga096b092013-09-22 10:15:58 +08001280 __ceph_remove_cap(cap, true);
Sage Weil2f2dc052009-10-06 11:31:09 -07001281 } else {
1282 /* try to drop referring dentries */
Sage Weilbe655592011-11-30 09:47:09 -08001283 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001284 d_prune_aliases(inode);
1285 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1286 inode, cap, atomic_read(&inode->i_count));
1287 return 0;
1288 }
1289
1290out:
Sage Weilbe655592011-11-30 09:47:09 -08001291 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001292 return 0;
1293}
1294
1295/*
1296 * Trim session cap count down to some max number.
1297 */
1298static int trim_caps(struct ceph_mds_client *mdsc,
1299 struct ceph_mds_session *session,
1300 int max_caps)
1301{
1302 int trim_caps = session->s_nr_caps - max_caps;
1303
1304 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1305 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1306 if (trim_caps > 0) {
1307 session->s_trim_caps = trim_caps;
1308 iterate_session_caps(session, trim_caps_cb, session);
1309 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1310 session->s_mds, session->s_nr_caps, max_caps,
1311 trim_caps - session->s_trim_caps);
Sage Weil5dacf092009-12-21 20:40:34 -08001312 session->s_trim_caps = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001313 }
Yan, Zhenga56371d2014-04-01 20:34:56 +08001314
1315 ceph_add_cap_releases(mdsc, session);
1316 ceph_send_cap_releases(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07001317 return 0;
1318}
1319
1320/*
1321 * Allocate cap_release messages. If there is a partially full message
1322 * in the queue, try to allocate enough to cover it's remainder, so that
1323 * we can send it immediately.
1324 *
1325 * Called under s_mutex.
1326 */
Sage Weil2b2300d2010-06-09 16:52:04 -07001327int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
Sage Weilee6b2722010-06-10 12:55:52 -07001328 struct ceph_mds_session *session)
Sage Weil2f2dc052009-10-06 11:31:09 -07001329{
Sage Weil38e8883e2010-06-10 12:57:11 -07001330 struct ceph_msg *msg, *partial = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -07001331 struct ceph_mds_cap_release *head;
1332 int err = -ENOMEM;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001333 int extra = mdsc->fsc->mount_options->cap_release_safety;
Sage Weil38e8883e2010-06-10 12:57:11 -07001334 int num;
Sage Weil2f2dc052009-10-06 11:31:09 -07001335
Sage Weil38e8883e2010-06-10 12:57:11 -07001336 dout("add_cap_releases %p mds%d extra %d\n", session, session->s_mds,
1337 extra);
Sage Weil2f2dc052009-10-06 11:31:09 -07001338
1339 spin_lock(&session->s_cap_lock);
1340
1341 if (!list_empty(&session->s_cap_releases)) {
1342 msg = list_first_entry(&session->s_cap_releases,
1343 struct ceph_msg,
1344 list_head);
1345 head = msg->front.iov_base;
Sage Weil38e8883e2010-06-10 12:57:11 -07001346 num = le32_to_cpu(head->num);
1347 if (num) {
1348 dout(" partial %p with (%d/%d)\n", msg, num,
1349 (int)CEPH_CAPS_PER_RELEASE);
1350 extra += CEPH_CAPS_PER_RELEASE - num;
1351 partial = msg;
1352 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001353 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001354 while (session->s_num_cap_releases < session->s_nr_caps + extra) {
1355 spin_unlock(&session->s_cap_lock);
Yehuda Sadeh34d23762010-04-06 14:33:58 -07001356 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
Sage Weilb61c2762011-08-09 15:03:46 -07001357 GFP_NOFS, false);
Sage Weil2f2dc052009-10-06 11:31:09 -07001358 if (!msg)
1359 goto out_unlocked;
1360 dout("add_cap_releases %p msg %p now %d\n", session, msg,
1361 (int)msg->front.iov_len);
1362 head = msg->front.iov_base;
1363 head->num = cpu_to_le32(0);
1364 msg->front.iov_len = sizeof(*head);
1365 spin_lock(&session->s_cap_lock);
1366 list_add(&msg->list_head, &session->s_cap_releases);
1367 session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE;
1368 }
1369
Sage Weil38e8883e2010-06-10 12:57:11 -07001370 if (partial) {
1371 head = partial->front.iov_base;
1372 num = le32_to_cpu(head->num);
1373 dout(" queueing partial %p with %d/%d\n", partial, num,
1374 (int)CEPH_CAPS_PER_RELEASE);
1375 list_move_tail(&partial->list_head,
1376 &session->s_cap_releases_done);
1377 session->s_num_cap_releases -= CEPH_CAPS_PER_RELEASE - num;
Sage Weil2f2dc052009-10-06 11:31:09 -07001378 }
1379 err = 0;
1380 spin_unlock(&session->s_cap_lock);
1381out_unlocked:
1382 return err;
1383}
1384
1385/*
1386 * flush all dirty inode data to disk.
1387 *
1388 * returns true if we've flushed through want_flush_seq
1389 */
1390static int check_cap_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq)
1391{
1392 int mds, ret = 1;
1393
1394 dout("check_cap_flush want %lld\n", want_flush_seq);
1395 mutex_lock(&mdsc->mutex);
1396 for (mds = 0; ret && mds < mdsc->max_sessions; mds++) {
1397 struct ceph_mds_session *session = mdsc->sessions[mds];
1398
1399 if (!session)
1400 continue;
1401 get_session(session);
1402 mutex_unlock(&mdsc->mutex);
1403
1404 mutex_lock(&session->s_mutex);
1405 if (!list_empty(&session->s_cap_flushing)) {
1406 struct ceph_inode_info *ci =
1407 list_entry(session->s_cap_flushing.next,
1408 struct ceph_inode_info,
1409 i_flushing_item);
1410 struct inode *inode = &ci->vfs_inode;
1411
Sage Weilbe655592011-11-30 09:47:09 -08001412 spin_lock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001413 if (ci->i_cap_flush_seq <= want_flush_seq) {
1414 dout("check_cap_flush still flushing %p "
1415 "seq %lld <= %lld to mds%d\n", inode,
1416 ci->i_cap_flush_seq, want_flush_seq,
1417 session->s_mds);
1418 ret = 0;
1419 }
Sage Weilbe655592011-11-30 09:47:09 -08001420 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001421 }
1422 mutex_unlock(&session->s_mutex);
1423 ceph_put_mds_session(session);
1424
1425 if (!ret)
1426 return ret;
1427 mutex_lock(&mdsc->mutex);
1428 }
1429
1430 mutex_unlock(&mdsc->mutex);
1431 dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
1432 return ret;
1433}
1434
1435/*
1436 * called under s_mutex
1437 */
Sage Weil3d7ded42010-06-09 16:47:10 -07001438void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1439 struct ceph_mds_session *session)
Sage Weil2f2dc052009-10-06 11:31:09 -07001440{
1441 struct ceph_msg *msg;
1442
1443 dout("send_cap_releases mds%d\n", session->s_mds);
Sage Weil0f8605f2010-05-05 15:51:35 -07001444 spin_lock(&session->s_cap_lock);
1445 while (!list_empty(&session->s_cap_releases_done)) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001446 msg = list_first_entry(&session->s_cap_releases_done,
1447 struct ceph_msg, list_head);
1448 list_del_init(&msg->list_head);
1449 spin_unlock(&session->s_cap_lock);
1450 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1451 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1452 ceph_con_send(&session->s_con, msg);
Sage Weil0f8605f2010-05-05 15:51:35 -07001453 spin_lock(&session->s_cap_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001454 }
1455 spin_unlock(&session->s_cap_lock);
1456}
1457
Sage Weile01a5942010-05-10 15:36:44 -07001458static void discard_cap_releases(struct ceph_mds_client *mdsc,
1459 struct ceph_mds_session *session)
1460{
1461 struct ceph_msg *msg;
1462 struct ceph_mds_cap_release *head;
1463 unsigned num;
1464
1465 dout("discard_cap_releases mds%d\n", session->s_mds);
Sage Weile01a5942010-05-10 15:36:44 -07001466
Yan, Zheng00bd8ed2014-03-24 09:56:43 +08001467 if (!list_empty(&session->s_cap_releases)) {
1468 /* zero out the in-progress message */
1469 msg = list_first_entry(&session->s_cap_releases,
1470 struct ceph_msg, list_head);
1471 head = msg->front.iov_base;
1472 num = le32_to_cpu(head->num);
1473 dout("discard_cap_releases mds%d %p %u\n",
1474 session->s_mds, msg, num);
1475 head->num = cpu_to_le32(0);
1476 msg->front.iov_len = sizeof(*head);
1477 session->s_num_cap_releases += num;
1478 }
Sage Weile01a5942010-05-10 15:36:44 -07001479
1480 /* requeue completed messages */
1481 while (!list_empty(&session->s_cap_releases_done)) {
1482 msg = list_first_entry(&session->s_cap_releases_done,
1483 struct ceph_msg, list_head);
1484 list_del_init(&msg->list_head);
1485
1486 head = msg->front.iov_base;
1487 num = le32_to_cpu(head->num);
1488 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg,
1489 num);
1490 session->s_num_cap_releases += num;
1491 head->num = cpu_to_le32(0);
1492 msg->front.iov_len = sizeof(*head);
1493 list_add(&msg->list_head, &session->s_cap_releases);
1494 }
Sage Weile01a5942010-05-10 15:36:44 -07001495}
1496
Sage Weil2f2dc052009-10-06 11:31:09 -07001497/*
1498 * requests
1499 */
1500
Yan, Zheng54008392014-03-29 13:41:15 +08001501int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1502 struct inode *dir)
1503{
1504 struct ceph_inode_info *ci = ceph_inode(dir);
1505 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1506 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
1507 size_t size = sizeof(*rinfo->dir_in) + sizeof(*rinfo->dir_dname_len) +
1508 sizeof(*rinfo->dir_dname) + sizeof(*rinfo->dir_dlease);
1509 int order, num_entries;
1510
1511 spin_lock(&ci->i_ceph_lock);
1512 num_entries = ci->i_files + ci->i_subdirs;
1513 spin_unlock(&ci->i_ceph_lock);
1514 num_entries = max(num_entries, 1);
1515 num_entries = min(num_entries, opt->max_readdir);
1516
1517 order = get_order(size * num_entries);
1518 while (order >= 0) {
1519 rinfo->dir_in = (void*)__get_free_pages(GFP_NOFS | __GFP_NOWARN,
1520 order);
1521 if (rinfo->dir_in)
1522 break;
1523 order--;
1524 }
1525 if (!rinfo->dir_in)
1526 return -ENOMEM;
1527
1528 num_entries = (PAGE_SIZE << order) / size;
1529 num_entries = min(num_entries, opt->max_readdir);
1530
1531 rinfo->dir_buf_size = PAGE_SIZE << order;
1532 req->r_num_caps = num_entries + 1;
1533 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1534 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1535 return 0;
1536}
1537
Sage Weil2f2dc052009-10-06 11:31:09 -07001538/*
1539 * Create an mds request.
1540 */
1541struct ceph_mds_request *
1542ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1543{
1544 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1545
1546 if (!req)
1547 return ERR_PTR(-ENOMEM);
1548
Sage Weilb4556392010-05-13 12:01:13 -07001549 mutex_init(&req->r_fill_mutex);
Yehuda Sadeh37151662010-06-17 16:16:12 -07001550 req->r_mdsc = mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07001551 req->r_started = jiffies;
1552 req->r_resend_mds = -1;
1553 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1554 req->r_fmode = -1;
Sage Weil153c8e62009-12-07 12:31:09 -08001555 kref_init(&req->r_kref);
Sage Weil2f2dc052009-10-06 11:31:09 -07001556 INIT_LIST_HEAD(&req->r_wait);
1557 init_completion(&req->r_completion);
1558 init_completion(&req->r_safe_completion);
1559 INIT_LIST_HEAD(&req->r_unsafe_item);
1560
1561 req->r_op = op;
1562 req->r_direct_mode = mode;
1563 return req;
1564}
1565
1566/*
Sage Weil44ca18f2010-02-15 12:08:46 -08001567 * return oldest (lowest) request, tid in request tree, 0 if none.
Sage Weil2f2dc052009-10-06 11:31:09 -07001568 *
1569 * called under mdsc->mutex.
1570 */
Sage Weil44ca18f2010-02-15 12:08:46 -08001571static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1572{
1573 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1574 return NULL;
1575 return rb_entry(rb_first(&mdsc->request_tree),
1576 struct ceph_mds_request, r_node);
1577}
1578
Sage Weil2f2dc052009-10-06 11:31:09 -07001579static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1580{
Sage Weil44ca18f2010-02-15 12:08:46 -08001581 struct ceph_mds_request *req = __get_oldest_req(mdsc);
1582
1583 if (req)
1584 return req->r_tid;
1585 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001586}
1587
1588/*
1589 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1590 * on build_path_from_dentry in fs/cifs/dir.c.
1591 *
1592 * If @stop_on_nosnap, generate path relative to the first non-snapped
1593 * inode.
1594 *
1595 * Encode hidden .snap dirs as a double /, i.e.
1596 * foo/.snap/bar -> foo//bar
1597 */
1598char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1599 int stop_on_nosnap)
1600{
1601 struct dentry *temp;
1602 char *path;
1603 int len, pos;
Al Viro1b71fe22011-07-16 23:43:58 -04001604 unsigned seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07001605
1606 if (dentry == NULL)
1607 return ERR_PTR(-EINVAL);
1608
1609retry:
1610 len = 0;
Al Viro1b71fe22011-07-16 23:43:58 -04001611 seq = read_seqbegin(&rename_lock);
1612 rcu_read_lock();
Sage Weil2f2dc052009-10-06 11:31:09 -07001613 for (temp = dentry; !IS_ROOT(temp);) {
1614 struct inode *inode = temp->d_inode;
1615 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1616 len++; /* slash only */
1617 else if (stop_on_nosnap && inode &&
1618 ceph_snap(inode) == CEPH_NOSNAP)
1619 break;
1620 else
1621 len += 1 + temp->d_name.len;
1622 temp = temp->d_parent;
Sage Weil2f2dc052009-10-06 11:31:09 -07001623 }
Al Viro1b71fe22011-07-16 23:43:58 -04001624 rcu_read_unlock();
Sage Weil2f2dc052009-10-06 11:31:09 -07001625 if (len)
1626 len--; /* no leading '/' */
1627
1628 path = kmalloc(len+1, GFP_NOFS);
1629 if (path == NULL)
1630 return ERR_PTR(-ENOMEM);
1631 pos = len;
1632 path[pos] = 0; /* trailing null */
Al Viro1b71fe22011-07-16 23:43:58 -04001633 rcu_read_lock();
Sage Weil2f2dc052009-10-06 11:31:09 -07001634 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
Al Viro1b71fe22011-07-16 23:43:58 -04001635 struct inode *inode;
Sage Weil2f2dc052009-10-06 11:31:09 -07001636
Al Viro1b71fe22011-07-16 23:43:58 -04001637 spin_lock(&temp->d_lock);
1638 inode = temp->d_inode;
Sage Weil2f2dc052009-10-06 11:31:09 -07001639 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
Sage Weil104648a2010-03-18 10:14:30 -07001640 dout("build_path path+%d: %p SNAPDIR\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07001641 pos, temp);
1642 } else if (stop_on_nosnap && inode &&
1643 ceph_snap(inode) == CEPH_NOSNAP) {
Yehuda Sadeh9d5a09e2011-12-13 09:57:44 -08001644 spin_unlock(&temp->d_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001645 break;
1646 } else {
1647 pos -= temp->d_name.len;
Al Viro1b71fe22011-07-16 23:43:58 -04001648 if (pos < 0) {
1649 spin_unlock(&temp->d_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001650 break;
Al Viro1b71fe22011-07-16 23:43:58 -04001651 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001652 strncpy(path + pos, temp->d_name.name,
1653 temp->d_name.len);
Sage Weil2f2dc052009-10-06 11:31:09 -07001654 }
Al Viro1b71fe22011-07-16 23:43:58 -04001655 spin_unlock(&temp->d_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001656 if (pos)
1657 path[--pos] = '/';
1658 temp = temp->d_parent;
Sage Weil2f2dc052009-10-06 11:31:09 -07001659 }
Al Viro1b71fe22011-07-16 23:43:58 -04001660 rcu_read_unlock();
1661 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
Sage Weil104648a2010-03-18 10:14:30 -07001662 pr_err("build_path did not end path lookup where "
Sage Weil2f2dc052009-10-06 11:31:09 -07001663 "expected, namelen is %d, pos is %d\n", len, pos);
1664 /* presumably this is only possible if racing with a
1665 rename of one of the parent directories (we can not
1666 lock the dentries above us to prevent this, but
1667 retrying should be harmless) */
1668 kfree(path);
1669 goto retry;
1670 }
1671
1672 *base = ceph_ino(temp->d_inode);
1673 *plen = len;
Sage Weil104648a2010-03-18 10:14:30 -07001674 dout("build_path on %p %d built %llx '%.*s'\n",
Al Viro84d08fa2013-07-05 18:59:33 +04001675 dentry, d_count(dentry), *base, len, path);
Sage Weil2f2dc052009-10-06 11:31:09 -07001676 return path;
1677}
1678
1679static int build_dentry_path(struct dentry *dentry,
1680 const char **ppath, int *ppathlen, u64 *pino,
1681 int *pfreepath)
1682{
1683 char *path;
1684
1685 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) {
1686 *pino = ceph_ino(dentry->d_parent->d_inode);
1687 *ppath = dentry->d_name.name;
1688 *ppathlen = dentry->d_name.len;
1689 return 0;
1690 }
1691 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1692 if (IS_ERR(path))
1693 return PTR_ERR(path);
1694 *ppath = path;
1695 *pfreepath = 1;
1696 return 0;
1697}
1698
1699static int build_inode_path(struct inode *inode,
1700 const char **ppath, int *ppathlen, u64 *pino,
1701 int *pfreepath)
1702{
1703 struct dentry *dentry;
1704 char *path;
1705
1706 if (ceph_snap(inode) == CEPH_NOSNAP) {
1707 *pino = ceph_ino(inode);
1708 *ppathlen = 0;
1709 return 0;
1710 }
1711 dentry = d_find_alias(inode);
1712 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1713 dput(dentry);
1714 if (IS_ERR(path))
1715 return PTR_ERR(path);
1716 *ppath = path;
1717 *pfreepath = 1;
1718 return 0;
1719}
1720
1721/*
1722 * request arguments may be specified via an inode *, a dentry *, or
1723 * an explicit ino+path.
1724 */
1725static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1726 const char *rpath, u64 rino,
1727 const char **ppath, int *pathlen,
1728 u64 *ino, int *freepath)
1729{
1730 int r = 0;
1731
1732 if (rinode) {
1733 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1734 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1735 ceph_snap(rinode));
1736 } else if (rdentry) {
1737 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1738 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1739 *ppath);
Sage Weil795858d2011-08-15 13:02:37 -07001740 } else if (rpath || rino) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001741 *ino = rino;
1742 *ppath = rpath;
David Zafmanb0000562012-10-25 10:23:46 -07001743 *pathlen = rpath ? strlen(rpath) : 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001744 dout(" path %.*s\n", *pathlen, rpath);
1745 }
1746
1747 return r;
1748}
1749
1750/*
1751 * called under mdsc->mutex
1752 */
1753static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1754 struct ceph_mds_request *req,
1755 int mds)
1756{
1757 struct ceph_msg *msg;
1758 struct ceph_mds_request_head *head;
1759 const char *path1 = NULL;
1760 const char *path2 = NULL;
1761 u64 ino1 = 0, ino2 = 0;
1762 int pathlen1 = 0, pathlen2 = 0;
1763 int freepath1 = 0, freepath2 = 0;
1764 int len;
1765 u16 releases;
1766 void *p, *end;
1767 int ret;
1768
1769 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1770 req->r_path1, req->r_ino1.ino,
1771 &path1, &pathlen1, &ino1, &freepath1);
1772 if (ret < 0) {
1773 msg = ERR_PTR(ret);
1774 goto out;
1775 }
1776
1777 ret = set_request_path_attr(NULL, req->r_old_dentry,
1778 req->r_path2, req->r_ino2.ino,
1779 &path2, &pathlen2, &ino2, &freepath2);
1780 if (ret < 0) {
1781 msg = ERR_PTR(ret);
1782 goto out_free1;
1783 }
1784
1785 len = sizeof(*head) +
Sage Weilac8839d2010-01-27 14:28:10 -08001786 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64));
Sage Weil2f2dc052009-10-06 11:31:09 -07001787
1788 /* calculate (max) length for cap releases */
1789 len += sizeof(struct ceph_mds_request_release) *
1790 (!!req->r_inode_drop + !!req->r_dentry_drop +
1791 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1792 if (req->r_dentry_drop)
1793 len += req->r_dentry->d_name.len;
1794 if (req->r_old_dentry_drop)
1795 len += req->r_old_dentry->d_name.len;
1796
Sage Weilb61c2762011-08-09 15:03:46 -07001797 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07001798 if (!msg) {
1799 msg = ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -07001800 goto out_free2;
Sage Weila79832f2010-04-01 16:06:19 -07001801 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001802
Sage Weil6df058c2009-12-22 11:24:33 -08001803 msg->hdr.tid = cpu_to_le64(req->r_tid);
1804
Sage Weil2f2dc052009-10-06 11:31:09 -07001805 head = msg->front.iov_base;
1806 p = msg->front.iov_base + sizeof(*head);
1807 end = msg->front.iov_base + msg->front.iov_len;
1808
1809 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1810 head->op = cpu_to_le32(req->r_op);
Eric W. Biedermanff3d0042013-01-31 04:01:53 -08001811 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1812 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
Sage Weil2f2dc052009-10-06 11:31:09 -07001813 head->args = req->r_args;
1814
1815 ceph_encode_filepath(&p, end, ino1, path1);
1816 ceph_encode_filepath(&p, end, ino2, path2);
1817
Sage Weile979cf52010-07-15 14:58:39 -07001818 /* make note of release offset, in case we need to replay */
1819 req->r_request_release_offset = p - msg->front.iov_base;
1820
Sage Weil2f2dc052009-10-06 11:31:09 -07001821 /* cap releases */
1822 releases = 0;
1823 if (req->r_inode_drop)
1824 releases += ceph_encode_inode_release(&p,
1825 req->r_inode ? req->r_inode : req->r_dentry->d_inode,
1826 mds, req->r_inode_drop, req->r_inode_unless, 0);
1827 if (req->r_dentry_drop)
1828 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1829 mds, req->r_dentry_drop, req->r_dentry_unless);
1830 if (req->r_old_dentry_drop)
1831 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1832 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1833 if (req->r_old_inode_drop)
1834 releases += ceph_encode_inode_release(&p,
1835 req->r_old_dentry->d_inode,
1836 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1837 head->num_releases = cpu_to_le16(releases);
1838
1839 BUG_ON(p > end);
1840 msg->front.iov_len = p - msg->front.iov_base;
1841 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1842
Alex Elderebf18f42013-03-04 22:29:57 -06001843 if (req->r_data_len) {
1844 /* outbound data set only by ceph_sync_setxattr() */
1845 BUG_ON(!req->r_pages);
Alex Elder90af3602013-04-05 14:46:01 -05001846 ceph_msg_data_add_pages(msg, req->r_pages, req->r_data_len, 0);
Alex Elderebf18f42013-03-04 22:29:57 -06001847 }
Alex Elder02afca62013-02-14 12:16:43 -06001848
Sage Weil2f2dc052009-10-06 11:31:09 -07001849 msg->hdr.data_len = cpu_to_le32(req->r_data_len);
1850 msg->hdr.data_off = cpu_to_le16(0);
1851
1852out_free2:
1853 if (freepath2)
1854 kfree((char *)path2);
1855out_free1:
1856 if (freepath1)
1857 kfree((char *)path1);
1858out:
1859 return msg;
1860}
1861
1862/*
1863 * called under mdsc->mutex if error, under no mutex if
1864 * success.
1865 */
1866static void complete_request(struct ceph_mds_client *mdsc,
1867 struct ceph_mds_request *req)
1868{
1869 if (req->r_callback)
1870 req->r_callback(mdsc, req);
1871 else
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001872 complete_all(&req->r_completion);
Sage Weil2f2dc052009-10-06 11:31:09 -07001873}
1874
1875/*
1876 * called under mdsc->mutex
1877 */
1878static int __prepare_send_request(struct ceph_mds_client *mdsc,
1879 struct ceph_mds_request *req,
1880 int mds)
1881{
1882 struct ceph_mds_request_head *rhead;
1883 struct ceph_msg *msg;
1884 int flags = 0;
1885
Sage Weil2f2dc052009-10-06 11:31:09 -07001886 req->r_attempts++;
Greg Farnume55b71f2010-06-22 15:58:01 -07001887 if (req->r_inode) {
1888 struct ceph_cap *cap =
1889 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
1890
1891 if (cap)
1892 req->r_sent_on_mseq = cap->mseq;
1893 else
1894 req->r_sent_on_mseq = -1;
1895 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001896 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
1897 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
1898
Sage Weil01a92f12010-07-15 13:24:32 -07001899 if (req->r_got_unsafe) {
1900 /*
1901 * Replay. Do not regenerate message (and rebuild
1902 * paths, etc.); just use the original message.
1903 * Rebuilding paths will break for renames because
1904 * d_move mangles the src name.
1905 */
1906 msg = req->r_request;
1907 rhead = msg->front.iov_base;
1908
1909 flags = le32_to_cpu(rhead->flags);
1910 flags |= CEPH_MDS_FLAG_REPLAY;
1911 rhead->flags = cpu_to_le32(flags);
1912
1913 if (req->r_target_inode)
1914 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
1915
1916 rhead->num_retry = req->r_attempts - 1;
Sage Weile979cf52010-07-15 14:58:39 -07001917
1918 /* remove cap/dentry releases from message */
1919 rhead->num_releases = 0;
1920 msg->hdr.front_len = cpu_to_le32(req->r_request_release_offset);
1921 msg->front.iov_len = req->r_request_release_offset;
Sage Weil01a92f12010-07-15 13:24:32 -07001922 return 0;
1923 }
1924
Sage Weil2f2dc052009-10-06 11:31:09 -07001925 if (req->r_request) {
1926 ceph_msg_put(req->r_request);
1927 req->r_request = NULL;
1928 }
1929 msg = create_request_message(mdsc, req, mds);
1930 if (IS_ERR(msg)) {
Sage Weile1518c72010-05-13 11:19:06 -07001931 req->r_err = PTR_ERR(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07001932 complete_request(mdsc, req);
Sage Weila79832f2010-04-01 16:06:19 -07001933 return PTR_ERR(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07001934 }
1935 req->r_request = msg;
1936
1937 rhead = msg->front.iov_base;
Sage Weil2f2dc052009-10-06 11:31:09 -07001938 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
1939 if (req->r_got_unsafe)
1940 flags |= CEPH_MDS_FLAG_REPLAY;
1941 if (req->r_locked_dir)
1942 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
1943 rhead->flags = cpu_to_le32(flags);
1944 rhead->num_fwd = req->r_num_fwd;
1945 rhead->num_retry = req->r_attempts - 1;
Sage Weil01a92f12010-07-15 13:24:32 -07001946 rhead->ino = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001947
1948 dout(" r_locked_dir = %p\n", req->r_locked_dir);
Sage Weil2f2dc052009-10-06 11:31:09 -07001949 return 0;
1950}
1951
1952/*
1953 * send request, or put it on the appropriate wait list.
1954 */
1955static int __do_request(struct ceph_mds_client *mdsc,
1956 struct ceph_mds_request *req)
1957{
1958 struct ceph_mds_session *session = NULL;
1959 int mds = -1;
1960 int err = -EAGAIN;
1961
Yan, Zhengeb1b8af2013-09-26 14:25:36 +08001962 if (req->r_err || req->r_got_result) {
1963 if (req->r_aborted)
1964 __unregister_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -07001965 goto out;
Yan, Zhengeb1b8af2013-09-26 14:25:36 +08001966 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001967
1968 if (req->r_timeout &&
1969 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
1970 dout("do_request timed out\n");
1971 err = -EIO;
1972 goto finish;
1973 }
1974
Sage Weildc69e2e2010-11-02 13:49:00 -07001975 put_request_session(req);
1976
Sage Weil2f2dc052009-10-06 11:31:09 -07001977 mds = __choose_mds(mdsc, req);
1978 if (mds < 0 ||
1979 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
1980 dout("do_request no mds or not active, waiting for map\n");
1981 list_add(&req->r_wait, &mdsc->waiting_for_map);
1982 goto out;
1983 }
1984
1985 /* get, open session */
1986 session = __ceph_lookup_mds_session(mdsc, mds);
Sage Weil9c4239562010-03-20 20:43:28 -07001987 if (!session) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001988 session = register_session(mdsc, mds);
Sage Weil9c4239562010-03-20 20:43:28 -07001989 if (IS_ERR(session)) {
1990 err = PTR_ERR(session);
1991 goto finish;
1992 }
1993 }
Sage Weildc69e2e2010-11-02 13:49:00 -07001994 req->r_session = get_session(session);
1995
Sage Weil2f2dc052009-10-06 11:31:09 -07001996 dout("do_request mds%d session %p state %s\n", mds, session,
1997 session_state_name(session->s_state));
1998 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
1999 session->s_state != CEPH_MDS_SESSION_HUNG) {
2000 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2001 session->s_state == CEPH_MDS_SESSION_CLOSING)
2002 __open_session(mdsc, session);
2003 list_add(&req->r_wait, &session->s_waiting);
2004 goto out_session;
2005 }
2006
2007 /* send request */
Sage Weil2f2dc052009-10-06 11:31:09 -07002008 req->r_resend_mds = -1; /* forget any previous mds hint */
2009
2010 if (req->r_request_started == 0) /* note request start time */
2011 req->r_request_started = jiffies;
2012
2013 err = __prepare_send_request(mdsc, req, mds);
2014 if (!err) {
2015 ceph_msg_get(req->r_request);
2016 ceph_con_send(&session->s_con, req->r_request);
2017 }
2018
2019out_session:
2020 ceph_put_mds_session(session);
2021out:
2022 return err;
2023
2024finish:
Sage Weile1518c72010-05-13 11:19:06 -07002025 req->r_err = err;
Sage Weil2f2dc052009-10-06 11:31:09 -07002026 complete_request(mdsc, req);
2027 goto out;
2028}
2029
2030/*
2031 * called under mdsc->mutex
2032 */
2033static void __wake_requests(struct ceph_mds_client *mdsc,
2034 struct list_head *head)
2035{
Yan, Zhenged75ec22012-11-19 10:49:06 +08002036 struct ceph_mds_request *req;
2037 LIST_HEAD(tmp_list);
Sage Weil2f2dc052009-10-06 11:31:09 -07002038
Yan, Zhenged75ec22012-11-19 10:49:06 +08002039 list_splice_init(head, &tmp_list);
2040
2041 while (!list_empty(&tmp_list)) {
2042 req = list_entry(tmp_list.next,
2043 struct ceph_mds_request, r_wait);
Sage Weil2f2dc052009-10-06 11:31:09 -07002044 list_del_init(&req->r_wait);
Sage Weil7971bd92013-05-01 21:15:58 -07002045 dout(" wake request %p tid %llu\n", req, req->r_tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002046 __do_request(mdsc, req);
2047 }
2048}
2049
2050/*
2051 * Wake up threads with requests pending for @mds, so that they can
Sage Weil29790f22010-03-18 14:45:05 -07002052 * resubmit their requests to a possibly different mds.
Sage Weil2f2dc052009-10-06 11:31:09 -07002053 */
Sage Weil29790f22010-03-18 14:45:05 -07002054static void kick_requests(struct ceph_mds_client *mdsc, int mds)
Sage Weil2f2dc052009-10-06 11:31:09 -07002055{
Sage Weil44ca18f2010-02-15 12:08:46 -08002056 struct ceph_mds_request *req;
2057 struct rb_node *p;
Sage Weil2f2dc052009-10-06 11:31:09 -07002058
2059 dout("kick_requests mds%d\n", mds);
Sage Weil44ca18f2010-02-15 12:08:46 -08002060 for (p = rb_first(&mdsc->request_tree); p; p = rb_next(p)) {
2061 req = rb_entry(p, struct ceph_mds_request, r_node);
2062 if (req->r_got_unsafe)
2063 continue;
2064 if (req->r_session &&
2065 req->r_session->s_mds == mds) {
2066 dout(" kicking tid %llu\n", req->r_tid);
Sage Weil44ca18f2010-02-15 12:08:46 -08002067 __do_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -07002068 }
2069 }
2070}
2071
2072void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2073 struct ceph_mds_request *req)
2074{
2075 dout("submit_request on %p\n", req);
2076 mutex_lock(&mdsc->mutex);
2077 __register_request(mdsc, req, NULL);
2078 __do_request(mdsc, req);
2079 mutex_unlock(&mdsc->mutex);
2080}
2081
2082/*
2083 * Synchrously perform an mds request. Take care of all of the
2084 * session setup, forwarding, retry details.
2085 */
2086int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2087 struct inode *dir,
2088 struct ceph_mds_request *req)
2089{
2090 int err;
2091
2092 dout("do_request on %p\n", req);
2093
2094 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2095 if (req->r_inode)
2096 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2097 if (req->r_locked_dir)
2098 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
Sage Weil844d87c2013-02-05 13:40:09 -08002099 if (req->r_old_dentry_dir)
Sage Weil41b02e12011-07-26 11:31:14 -07002100 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2101 CEPH_CAP_PIN);
Sage Weil2f2dc052009-10-06 11:31:09 -07002102
2103 /* issue */
2104 mutex_lock(&mdsc->mutex);
2105 __register_request(mdsc, req, dir);
2106 __do_request(mdsc, req);
2107
Sage Weile1518c72010-05-13 11:19:06 -07002108 if (req->r_err) {
Sage Weil2f2dc052009-10-06 11:31:09 -07002109 err = req->r_err;
Sage Weile1518c72010-05-13 11:19:06 -07002110 __unregister_request(mdsc, req);
2111 dout("do_request early error %d\n", err);
2112 goto out;
Sage Weil2f2dc052009-10-06 11:31:09 -07002113 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002114
Sage Weile1518c72010-05-13 11:19:06 -07002115 /* wait */
2116 mutex_unlock(&mdsc->mutex);
2117 dout("do_request waiting\n");
2118 if (req->r_timeout) {
Sage Weilaa916472010-05-24 11:15:51 -07002119 err = (long)wait_for_completion_killable_timeout(
Sage Weile1518c72010-05-13 11:19:06 -07002120 &req->r_completion, req->r_timeout);
2121 if (err == 0)
2122 err = -EIO;
2123 } else {
Sage Weilaa916472010-05-24 11:15:51 -07002124 err = wait_for_completion_killable(&req->r_completion);
Sage Weile1518c72010-05-13 11:19:06 -07002125 }
2126 dout("do_request waited, got %d\n", err);
2127 mutex_lock(&mdsc->mutex);
2128
2129 /* only abort if we didn't race with a real reply */
2130 if (req->r_got_result) {
2131 err = le32_to_cpu(req->r_reply_info.head->result);
2132 } else if (err < 0) {
2133 dout("aborted request %lld with %d\n", req->r_tid, err);
Sage Weilb4556392010-05-13 12:01:13 -07002134
2135 /*
2136 * ensure we aren't running concurrently with
2137 * ceph_fill_trace or ceph_readdir_prepopulate, which
2138 * rely on locks (dir mutex) held by our caller.
2139 */
2140 mutex_lock(&req->r_fill_mutex);
Sage Weile1518c72010-05-13 11:19:06 -07002141 req->r_err = err;
2142 req->r_aborted = true;
Sage Weilb4556392010-05-13 12:01:13 -07002143 mutex_unlock(&req->r_fill_mutex);
Sage Weile1518c72010-05-13 11:19:06 -07002144
2145 if (req->r_locked_dir &&
Sage Weil167c9e32010-05-14 10:02:57 -07002146 (req->r_op & CEPH_MDS_OP_WRITE))
2147 ceph_invalidate_dir_request(req);
Sage Weile1518c72010-05-13 11:19:06 -07002148 } else {
2149 err = req->r_err;
2150 }
2151
2152out:
2153 mutex_unlock(&mdsc->mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002154 dout("do_request %p done, result %d\n", req, err);
2155 return err;
2156}
2157
2158/*
Yan, Zheng2f276c52013-03-13 19:44:32 +08002159 * Invalidate dir's completeness, dentry lease state on an aborted MDS
Sage Weil167c9e32010-05-14 10:02:57 -07002160 * namespace request.
2161 */
2162void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2163{
2164 struct inode *inode = req->r_locked_dir;
Sage Weil167c9e32010-05-14 10:02:57 -07002165
Yan, Zheng2f276c52013-03-13 19:44:32 +08002166 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
Sage Weil167c9e32010-05-14 10:02:57 -07002167
Yan, Zheng2f276c52013-03-13 19:44:32 +08002168 ceph_dir_clear_complete(inode);
Sage Weil167c9e32010-05-14 10:02:57 -07002169 if (req->r_dentry)
2170 ceph_invalidate_dentry_lease(req->r_dentry);
2171 if (req->r_old_dentry)
2172 ceph_invalidate_dentry_lease(req->r_old_dentry);
2173}
2174
2175/*
Sage Weil2f2dc052009-10-06 11:31:09 -07002176 * Handle mds reply.
2177 *
2178 * We take the session mutex and parse and process the reply immediately.
2179 * This preserves the logical ordering of replies, capabilities, etc., sent
2180 * by the MDS as they are applied to our local cache.
2181 */
2182static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2183{
2184 struct ceph_mds_client *mdsc = session->s_mdsc;
2185 struct ceph_mds_request *req;
2186 struct ceph_mds_reply_head *head = msg->front.iov_base;
2187 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
2188 u64 tid;
2189 int err, result;
Sage Weil2600d2d2010-02-22 15:12:16 -08002190 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07002191
Sage Weil2f2dc052009-10-06 11:31:09 -07002192 if (msg->front.iov_len < sizeof(*head)) {
2193 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002194 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002195 return;
2196 }
2197
2198 /* get request, session */
Sage Weil6df058c2009-12-22 11:24:33 -08002199 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002200 mutex_lock(&mdsc->mutex);
2201 req = __lookup_request(mdsc, tid);
2202 if (!req) {
2203 dout("handle_reply on unknown tid %llu\n", tid);
2204 mutex_unlock(&mdsc->mutex);
2205 return;
2206 }
2207 dout("handle_reply %p\n", req);
Sage Weil2f2dc052009-10-06 11:31:09 -07002208
2209 /* correct session? */
Sage Weild96d6042010-03-20 20:50:58 -07002210 if (req->r_session != session) {
Sage Weil2f2dc052009-10-06 11:31:09 -07002211 pr_err("mdsc_handle_reply got %llu on session mds%d"
2212 " not mds%d\n", tid, session->s_mds,
2213 req->r_session ? req->r_session->s_mds : -1);
2214 mutex_unlock(&mdsc->mutex);
2215 goto out;
2216 }
2217
2218 /* dup? */
2219 if ((req->r_got_unsafe && !head->safe) ||
2220 (req->r_got_safe && head->safe)) {
2221 pr_warning("got a dup %s reply on %llu from mds%d\n",
2222 head->safe ? "safe" : "unsafe", tid, mds);
2223 mutex_unlock(&mdsc->mutex);
2224 goto out;
2225 }
Sage Weil85792d0d2010-05-13 09:06:02 -07002226 if (req->r_got_safe && !head->safe) {
2227 pr_warning("got unsafe after safe on %llu from mds%d\n",
2228 tid, mds);
2229 mutex_unlock(&mdsc->mutex);
2230 goto out;
2231 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002232
2233 result = le32_to_cpu(head->result);
2234
2235 /*
Greg Farnume55b71f2010-06-22 15:58:01 -07002236 * Handle an ESTALE
2237 * if we're not talking to the authority, send to them
2238 * if the authority has changed while we weren't looking,
2239 * send to new authority
2240 * Otherwise we just have to return an ESTALE
Sage Weil2f2dc052009-10-06 11:31:09 -07002241 */
2242 if (result == -ESTALE) {
Greg Farnume55b71f2010-06-22 15:58:01 -07002243 dout("got ESTALE on request %llu", req->r_tid);
Yan, Zhengca18bed2013-11-22 14:21:44 +08002244 if (req->r_direct_mode != USE_AUTH_MDS) {
Greg Farnume55b71f2010-06-22 15:58:01 -07002245 dout("not using auth, setting for that now");
2246 req->r_direct_mode = USE_AUTH_MDS;
Sage Weil2f2dc052009-10-06 11:31:09 -07002247 __do_request(mdsc, req);
2248 mutex_unlock(&mdsc->mutex);
2249 goto out;
Greg Farnume55b71f2010-06-22 15:58:01 -07002250 } else {
Yan, Zhengca18bed2013-11-22 14:21:44 +08002251 int mds = __choose_mds(mdsc, req);
2252 if (mds >= 0 && mds != req->r_session->s_mds) {
2253 dout("but auth changed, so resending");
Greg Farnume55b71f2010-06-22 15:58:01 -07002254 __do_request(mdsc, req);
2255 mutex_unlock(&mdsc->mutex);
2256 goto out;
2257 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002258 }
Greg Farnume55b71f2010-06-22 15:58:01 -07002259 dout("have to return ESTALE on request %llu", req->r_tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002260 }
2261
Greg Farnume55b71f2010-06-22 15:58:01 -07002262
Sage Weil2f2dc052009-10-06 11:31:09 -07002263 if (head->safe) {
2264 req->r_got_safe = true;
2265 __unregister_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -07002266
2267 if (req->r_got_unsafe) {
2268 /*
2269 * We already handled the unsafe response, now do the
2270 * cleanup. No need to examine the response; the MDS
2271 * doesn't include any result info in the safe
2272 * response. And even if it did, there is nothing
2273 * useful we could do with a revised return value.
2274 */
2275 dout("got safe reply %llu, mds%d\n", tid, mds);
2276 list_del_init(&req->r_unsafe_item);
2277
2278 /* last unsafe request during umount? */
Sage Weil44ca18f2010-02-15 12:08:46 -08002279 if (mdsc->stopping && !__get_oldest_req(mdsc))
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002280 complete_all(&mdsc->safe_umount_waiters);
Sage Weil2f2dc052009-10-06 11:31:09 -07002281 mutex_unlock(&mdsc->mutex);
2282 goto out;
2283 }
Sage Weile1518c72010-05-13 11:19:06 -07002284 } else {
Sage Weil2f2dc052009-10-06 11:31:09 -07002285 req->r_got_unsafe = true;
2286 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2287 }
2288
2289 dout("handle_reply tid %lld result %d\n", tid, result);
2290 rinfo = &req->r_reply_info;
Sage Weil14303d22010-12-14 17:37:52 -08002291 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
Sage Weil2f2dc052009-10-06 11:31:09 -07002292 mutex_unlock(&mdsc->mutex);
2293
2294 mutex_lock(&session->s_mutex);
2295 if (err < 0) {
Herb Shiu25933ab2010-12-01 14:14:38 -08002296 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
Sage Weil9ec7cab2009-12-14 15:13:47 -08002297 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002298 goto out_err;
2299 }
2300
2301 /* snap trace */
2302 if (rinfo->snapblob_len) {
2303 down_write(&mdsc->snap_rwsem);
2304 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2305 rinfo->snapblob + rinfo->snapblob_len,
2306 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP);
2307 downgrade_write(&mdsc->snap_rwsem);
2308 } else {
2309 down_read(&mdsc->snap_rwsem);
2310 }
2311
2312 /* insert trace into our cache */
Sage Weilb4556392010-05-13 12:01:13 -07002313 mutex_lock(&req->r_fill_mutex);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002314 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002315 if (err == 0) {
Sam Lang6e8575f2012-12-28 09:56:46 -08002316 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
Yan, Zheng81c6aea2013-09-18 09:44:13 +08002317 req->r_op == CEPH_MDS_OP_LSSNAP))
Sage Weil2f2dc052009-10-06 11:31:09 -07002318 ceph_readdir_prepopulate(req, req->r_session);
Yehuda Sadeh37151662010-06-17 16:16:12 -07002319 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
Sage Weil2f2dc052009-10-06 11:31:09 -07002320 }
Sage Weilb4556392010-05-13 12:01:13 -07002321 mutex_unlock(&req->r_fill_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002322
2323 up_read(&mdsc->snap_rwsem);
2324out_err:
Sage Weile1518c72010-05-13 11:19:06 -07002325 mutex_lock(&mdsc->mutex);
2326 if (!req->r_aborted) {
2327 if (err) {
2328 req->r_err = err;
2329 } else {
2330 req->r_reply = msg;
2331 ceph_msg_get(msg);
2332 req->r_got_result = true;
2333 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002334 } else {
Sage Weile1518c72010-05-13 11:19:06 -07002335 dout("reply arrived after request %lld was aborted\n", tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002336 }
Sage Weile1518c72010-05-13 11:19:06 -07002337 mutex_unlock(&mdsc->mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002338
Sage Weilee6b2722010-06-10 12:55:52 -07002339 ceph_add_cap_releases(mdsc, req->r_session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002340 mutex_unlock(&session->s_mutex);
2341
2342 /* kick calling process */
2343 complete_request(mdsc, req);
2344out:
2345 ceph_mdsc_put_request(req);
2346 return;
2347}
2348
2349
2350
2351/*
2352 * handle mds notification that our request has been forwarded.
2353 */
Sage Weil2600d2d2010-02-22 15:12:16 -08002354static void handle_forward(struct ceph_mds_client *mdsc,
2355 struct ceph_mds_session *session,
2356 struct ceph_msg *msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07002357{
2358 struct ceph_mds_request *req;
Sage Weila1ea7872010-02-23 14:02:44 -08002359 u64 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002360 u32 next_mds;
2361 u32 fwd_seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002362 int err = -EINVAL;
2363 void *p = msg->front.iov_base;
2364 void *end = p + msg->front.iov_len;
Sage Weil2f2dc052009-10-06 11:31:09 -07002365
Sage Weila1ea7872010-02-23 14:02:44 -08002366 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07002367 next_mds = ceph_decode_32(&p);
2368 fwd_seq = ceph_decode_32(&p);
Sage Weil2f2dc052009-10-06 11:31:09 -07002369
2370 mutex_lock(&mdsc->mutex);
2371 req = __lookup_request(mdsc, tid);
2372 if (!req) {
Sage Weil2a8e5e32010-05-28 16:43:16 -07002373 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002374 goto out; /* dup reply? */
2375 }
2376
Sage Weil2a8e5e32010-05-28 16:43:16 -07002377 if (req->r_aborted) {
2378 dout("forward tid %llu aborted, unregistering\n", tid);
2379 __unregister_request(mdsc, req);
2380 } else if (fwd_seq <= req->r_num_fwd) {
2381 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07002382 tid, next_mds, req->r_num_fwd, fwd_seq);
2383 } else {
2384 /* resend. forward race not possible; mds would drop */
Sage Weil2a8e5e32010-05-28 16:43:16 -07002385 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2386 BUG_ON(req->r_err);
2387 BUG_ON(req->r_got_result);
Sage Weil2f2dc052009-10-06 11:31:09 -07002388 req->r_num_fwd = fwd_seq;
2389 req->r_resend_mds = next_mds;
2390 put_request_session(req);
2391 __do_request(mdsc, req);
2392 }
2393 ceph_mdsc_put_request(req);
2394out:
2395 mutex_unlock(&mdsc->mutex);
2396 return;
2397
2398bad:
2399 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2400}
2401
2402/*
2403 * handle a mds session control message
2404 */
2405static void handle_session(struct ceph_mds_session *session,
2406 struct ceph_msg *msg)
2407{
2408 struct ceph_mds_client *mdsc = session->s_mdsc;
2409 u32 op;
2410 u64 seq;
Sage Weil2600d2d2010-02-22 15:12:16 -08002411 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07002412 struct ceph_mds_session_head *h = msg->front.iov_base;
2413 int wake = 0;
2414
Sage Weil2f2dc052009-10-06 11:31:09 -07002415 /* decode */
2416 if (msg->front.iov_len != sizeof(*h))
2417 goto bad;
2418 op = le32_to_cpu(h->op);
2419 seq = le64_to_cpu(h->seq);
2420
2421 mutex_lock(&mdsc->mutex);
Sage Weil2600d2d2010-02-22 15:12:16 -08002422 if (op == CEPH_SESSION_CLOSE)
2423 __unregister_session(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002424 /* FIXME: this ttl calculation is generous */
2425 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2426 mutex_unlock(&mdsc->mutex);
2427
2428 mutex_lock(&session->s_mutex);
2429
2430 dout("handle_session mds%d %s %p state %s seq %llu\n",
2431 mds, ceph_session_op_name(op), session,
2432 session_state_name(session->s_state), seq);
2433
2434 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2435 session->s_state = CEPH_MDS_SESSION_OPEN;
2436 pr_info("mds%d came back\n", session->s_mds);
2437 }
2438
2439 switch (op) {
2440 case CEPH_SESSION_OPEN:
Sage Weil29790f22010-03-18 14:45:05 -07002441 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2442 pr_info("mds%d reconnect success\n", session->s_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002443 session->s_state = CEPH_MDS_SESSION_OPEN;
2444 renewed_caps(mdsc, session, 0);
2445 wake = 1;
2446 if (mdsc->stopping)
2447 __close_session(mdsc, session);
2448 break;
2449
2450 case CEPH_SESSION_RENEWCAPS:
2451 if (session->s_renew_seq == seq)
2452 renewed_caps(mdsc, session, 1);
2453 break;
2454
2455 case CEPH_SESSION_CLOSE:
Sage Weil29790f22010-03-18 14:45:05 -07002456 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2457 pr_info("mds%d reconnect denied\n", session->s_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002458 remove_session_caps(session);
2459 wake = 1; /* for good measure */
Sage Weilf3c60c52010-08-11 14:51:23 -07002460 wake_up_all(&mdsc->session_close_wq);
Sage Weil29790f22010-03-18 14:45:05 -07002461 kick_requests(mdsc, mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002462 break;
2463
2464 case CEPH_SESSION_STALE:
2465 pr_info("mds%d caps went stale, renewing\n",
2466 session->s_mds);
Alex Elderd8fb02a2012-01-12 17:48:10 -08002467 spin_lock(&session->s_gen_ttl_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07002468 session->s_cap_gen++;
Alex Elder1ce208a2012-01-12 17:48:11 -08002469 session->s_cap_ttl = jiffies - 1;
Alex Elderd8fb02a2012-01-12 17:48:10 -08002470 spin_unlock(&session->s_gen_ttl_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07002471 send_renew_caps(mdsc, session);
2472 break;
2473
2474 case CEPH_SESSION_RECALL_STATE:
2475 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2476 break;
2477
Yan, Zheng186e4f72013-11-22 14:48:37 +08002478 case CEPH_SESSION_FLUSHMSG:
2479 send_flushmsg_ack(mdsc, session, seq);
2480 break;
2481
Sage Weil2f2dc052009-10-06 11:31:09 -07002482 default:
2483 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2484 WARN_ON(1);
2485 }
2486
2487 mutex_unlock(&session->s_mutex);
2488 if (wake) {
2489 mutex_lock(&mdsc->mutex);
2490 __wake_requests(mdsc, &session->s_waiting);
2491 mutex_unlock(&mdsc->mutex);
2492 }
2493 return;
2494
2495bad:
2496 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2497 (int)msg->front.iov_len);
Sage Weil9ec7cab2009-12-14 15:13:47 -08002498 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002499 return;
2500}
2501
2502
2503/*
2504 * called under session->mutex.
2505 */
2506static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2507 struct ceph_mds_session *session)
2508{
2509 struct ceph_mds_request *req, *nreq;
2510 int err;
2511
2512 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2513
2514 mutex_lock(&mdsc->mutex);
2515 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2516 err = __prepare_send_request(mdsc, req, session->s_mds);
2517 if (!err) {
2518 ceph_msg_get(req->r_request);
2519 ceph_con_send(&session->s_con, req->r_request);
2520 }
2521 }
2522 mutex_unlock(&mdsc->mutex);
2523}
2524
2525/*
2526 * Encode information about a cap for a reconnect with the MDS.
2527 */
Sage Weil2f2dc052009-10-06 11:31:09 -07002528static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2529 void *arg)
2530{
Sage Weil20cb34a2010-05-12 15:21:32 -07002531 union {
2532 struct ceph_mds_cap_reconnect v2;
2533 struct ceph_mds_cap_reconnect_v1 v1;
2534 } rec;
2535 size_t reclen;
Sage Weil2f2dc052009-10-06 11:31:09 -07002536 struct ceph_inode_info *ci;
Sage Weil20cb34a2010-05-12 15:21:32 -07002537 struct ceph_reconnect_state *recon_state = arg;
2538 struct ceph_pagelist *pagelist = recon_state->pagelist;
Sage Weil2f2dc052009-10-06 11:31:09 -07002539 char *path;
2540 int pathlen, err;
2541 u64 pathbase;
2542 struct dentry *dentry;
2543
2544 ci = cap->ci;
2545
2546 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2547 inode, ceph_vinop(inode), cap, cap->cap_id,
2548 ceph_cap_string(cap->issued));
Sage Weil93cea5b2009-12-23 12:21:51 -08002549 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2550 if (err)
2551 return err;
Sage Weil2f2dc052009-10-06 11:31:09 -07002552
2553 dentry = d_find_alias(inode);
2554 if (dentry) {
2555 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2556 if (IS_ERR(path)) {
2557 err = PTR_ERR(path);
Sage Weile072f8a2010-08-26 09:26:37 -07002558 goto out_dput;
Sage Weil2f2dc052009-10-06 11:31:09 -07002559 }
2560 } else {
2561 path = NULL;
2562 pathlen = 0;
2563 }
Sage Weil93cea5b2009-12-23 12:21:51 -08002564 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2565 if (err)
Sage Weile072f8a2010-08-26 09:26:37 -07002566 goto out_free;
Sage Weil2f2dc052009-10-06 11:31:09 -07002567
Sage Weilbe655592011-11-30 09:47:09 -08002568 spin_lock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07002569 cap->seq = 0; /* reset cap seq */
2570 cap->issue_seq = 0; /* and issue_seq */
Yan, Zheng667ca052013-05-31 16:25:36 +08002571 cap->mseq = 0; /* and migrate_seq */
Yan, Zheng99a9c272013-09-22 11:08:14 +08002572 cap->cap_gen = cap->session->s_cap_gen;
Sage Weil20cb34a2010-05-12 15:21:32 -07002573
2574 if (recon_state->flock) {
2575 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2576 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2577 rec.v2.issued = cpu_to_le32(cap->issued);
2578 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2579 rec.v2.pathbase = cpu_to_le64(pathbase);
2580 rec.v2.flock_len = 0;
2581 reclen = sizeof(rec.v2);
2582 } else {
2583 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2584 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2585 rec.v1.issued = cpu_to_le32(cap->issued);
2586 rec.v1.size = cpu_to_le64(inode->i_size);
2587 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2588 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2589 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2590 rec.v1.pathbase = cpu_to_le64(pathbase);
2591 reclen = sizeof(rec.v1);
2592 }
Sage Weilbe655592011-11-30 09:47:09 -08002593 spin_unlock(&ci->i_ceph_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07002594
Greg Farnum40819f62010-08-02 15:34:23 -07002595 if (recon_state->flock) {
2596 int num_fcntl_locks, num_flock_locks;
Jim Schutt39be95e2013-05-15 13:03:35 -05002597 struct ceph_filelock *flocks;
Greg Farnum40819f62010-08-02 15:34:23 -07002598
Jim Schutt39be95e2013-05-15 13:03:35 -05002599encode_again:
Jeff Layton1c8c6012013-06-21 08:58:15 -04002600 spin_lock(&inode->i_lock);
Jim Schutt39be95e2013-05-15 13:03:35 -05002601 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
Jeff Layton1c8c6012013-06-21 08:58:15 -04002602 spin_unlock(&inode->i_lock);
Jim Schutt39be95e2013-05-15 13:03:35 -05002603 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2604 sizeof(struct ceph_filelock), GFP_NOFS);
2605 if (!flocks) {
2606 err = -ENOMEM;
2607 goto out_free;
2608 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002609 spin_lock(&inode->i_lock);
Jim Schutt39be95e2013-05-15 13:03:35 -05002610 err = ceph_encode_locks_to_buffer(inode, flocks,
2611 num_fcntl_locks,
2612 num_flock_locks);
Jeff Layton1c8c6012013-06-21 08:58:15 -04002613 spin_unlock(&inode->i_lock);
Jim Schutt39be95e2013-05-15 13:03:35 -05002614 if (err) {
2615 kfree(flocks);
2616 if (err == -ENOSPC)
2617 goto encode_again;
2618 goto out_free;
2619 }
2620 /*
2621 * number of encoded locks is stable, so copy to pagelist
2622 */
2623 rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2624 (num_fcntl_locks+num_flock_locks) *
2625 sizeof(struct ceph_filelock));
2626 err = ceph_pagelist_append(pagelist, &rec, reclen);
2627 if (!err)
2628 err = ceph_locks_to_pagelist(flocks, pagelist,
2629 num_fcntl_locks,
2630 num_flock_locks);
2631 kfree(flocks);
Sage Weil3612abb2010-09-07 15:59:27 -07002632 } else {
2633 err = ceph_pagelist_append(pagelist, &rec, reclen);
Greg Farnum40819f62010-08-02 15:34:23 -07002634 }
Yan, Zheng44c99752013-09-22 10:28:10 +08002635
2636 recon_state->nr_caps++;
Sage Weile072f8a2010-08-26 09:26:37 -07002637out_free:
Sage Weil2f2dc052009-10-06 11:31:09 -07002638 kfree(path);
Sage Weile072f8a2010-08-26 09:26:37 -07002639out_dput:
Sage Weil2f2dc052009-10-06 11:31:09 -07002640 dput(dentry);
Sage Weil93cea5b2009-12-23 12:21:51 -08002641 return err;
Sage Weil2f2dc052009-10-06 11:31:09 -07002642}
2643
2644
2645/*
2646 * If an MDS fails and recovers, clients need to reconnect in order to
2647 * reestablish shared state. This includes all caps issued through
2648 * this session _and_ the snap_realm hierarchy. Because it's not
2649 * clear which snap realms the mds cares about, we send everything we
2650 * know about.. that ensures we'll then get any new info the
2651 * recovering MDS might have.
2652 *
2653 * This is a relatively heavyweight operation, but it's rare.
2654 *
2655 * called with mdsc->mutex held.
2656 */
Sage Weil34b6c852010-05-10 16:31:25 -07002657static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2658 struct ceph_mds_session *session)
Sage Weil2f2dc052009-10-06 11:31:09 -07002659{
Sage Weil2f2dc052009-10-06 11:31:09 -07002660 struct ceph_msg *reply;
Sage Weila105f002010-02-15 14:37:55 -08002661 struct rb_node *p;
Sage Weil34b6c852010-05-10 16:31:25 -07002662 int mds = session->s_mds;
Sage Weil9abf82b2010-05-10 21:58:38 -07002663 int err = -ENOMEM;
Yan, Zheng44c99752013-09-22 10:28:10 +08002664 int s_nr_caps;
Sage Weil93cea5b2009-12-23 12:21:51 -08002665 struct ceph_pagelist *pagelist;
Sage Weil20cb34a2010-05-12 15:21:32 -07002666 struct ceph_reconnect_state recon_state;
Sage Weil2f2dc052009-10-06 11:31:09 -07002667
Sage Weil34b6c852010-05-10 16:31:25 -07002668 pr_info("mds%d reconnect start\n", mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002669
Sage Weil93cea5b2009-12-23 12:21:51 -08002670 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2671 if (!pagelist)
2672 goto fail_nopagelist;
2673 ceph_pagelist_init(pagelist);
2674
Sage Weilb61c2762011-08-09 15:03:46 -07002675 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07002676 if (!reply)
Sage Weil93cea5b2009-12-23 12:21:51 -08002677 goto fail_nomsg;
Sage Weil93cea5b2009-12-23 12:21:51 -08002678
Sage Weil34b6c852010-05-10 16:31:25 -07002679 mutex_lock(&session->s_mutex);
2680 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2681 session->s_seq = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07002682
Sage Weila53aab62012-07-30 16:21:17 -07002683 ceph_con_close(&session->s_con);
Sage Weil34b6c852010-05-10 16:31:25 -07002684 ceph_con_open(&session->s_con,
Sage Weilb7a9e5d2012-06-27 12:24:08 -07002685 CEPH_ENTITY_TYPE_MDS, mds,
Sage Weil34b6c852010-05-10 16:31:25 -07002686 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
Sage Weil2f2dc052009-10-06 11:31:09 -07002687
Sage Weil34b6c852010-05-10 16:31:25 -07002688 /* replay unsafe requests */
2689 replay_unsafe_requests(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002690
2691 down_read(&mdsc->snap_rwsem);
2692
Sage Weil2f2dc052009-10-06 11:31:09 -07002693 dout("session %p state %s\n", session,
2694 session_state_name(session->s_state));
2695
Yan, Zheng99a9c272013-09-22 11:08:14 +08002696 spin_lock(&session->s_gen_ttl_lock);
2697 session->s_cap_gen++;
2698 spin_unlock(&session->s_gen_ttl_lock);
2699
2700 spin_lock(&session->s_cap_lock);
2701 /*
2702 * notify __ceph_remove_cap() that we are composing cap reconnect.
2703 * If a cap get released before being added to the cap reconnect,
2704 * __ceph_remove_cap() should skip queuing cap release.
2705 */
2706 session->s_cap_reconnect = 1;
Sage Weile01a5942010-05-10 15:36:44 -07002707 /* drop old cap expires; we're about to reestablish that state */
2708 discard_cap_releases(mdsc, session);
Yan, Zheng99a9c272013-09-22 11:08:14 +08002709 spin_unlock(&session->s_cap_lock);
Sage Weile01a5942010-05-10 15:36:44 -07002710
Sage Weil2f2dc052009-10-06 11:31:09 -07002711 /* traverse this session's caps */
Yan, Zheng44c99752013-09-22 10:28:10 +08002712 s_nr_caps = session->s_nr_caps;
2713 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
Sage Weil93cea5b2009-12-23 12:21:51 -08002714 if (err)
2715 goto fail;
Sage Weil20cb34a2010-05-12 15:21:32 -07002716
Yan, Zheng44c99752013-09-22 10:28:10 +08002717 recon_state.nr_caps = 0;
Sage Weil20cb34a2010-05-12 15:21:32 -07002718 recon_state.pagelist = pagelist;
2719 recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2720 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
Sage Weil2f2dc052009-10-06 11:31:09 -07002721 if (err < 0)
Sage Weil9abf82b2010-05-10 21:58:38 -07002722 goto fail;
Sage Weil2f2dc052009-10-06 11:31:09 -07002723
Yan, Zheng99a9c272013-09-22 11:08:14 +08002724 spin_lock(&session->s_cap_lock);
2725 session->s_cap_reconnect = 0;
2726 spin_unlock(&session->s_cap_lock);
2727
Sage Weil2f2dc052009-10-06 11:31:09 -07002728 /*
2729 * snaprealms. we provide mds with the ino, seq (version), and
2730 * parent for all of our realms. If the mds has any newer info,
2731 * it will tell us.
2732 */
Sage Weila105f002010-02-15 14:37:55 -08002733 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2734 struct ceph_snap_realm *realm =
2735 rb_entry(p, struct ceph_snap_realm, node);
Sage Weil93cea5b2009-12-23 12:21:51 -08002736 struct ceph_mds_snaprealm_reconnect sr_rec;
Sage Weil2f2dc052009-10-06 11:31:09 -07002737
2738 dout(" adding snap realm %llx seq %lld parent %llx\n",
2739 realm->ino, realm->seq, realm->parent_ino);
Sage Weil93cea5b2009-12-23 12:21:51 -08002740 sr_rec.ino = cpu_to_le64(realm->ino);
2741 sr_rec.seq = cpu_to_le64(realm->seq);
2742 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2743 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2744 if (err)
2745 goto fail;
Sage Weil2f2dc052009-10-06 11:31:09 -07002746 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002747
Sage Weil20cb34a2010-05-12 15:21:32 -07002748 if (recon_state.flock)
2749 reply->hdr.version = cpu_to_le16(2);
Yan, Zheng44c99752013-09-22 10:28:10 +08002750
2751 /* raced with cap release? */
2752 if (s_nr_caps != recon_state.nr_caps) {
2753 struct page *page = list_first_entry(&pagelist->head,
2754 struct page, lru);
2755 __le32 *addr = kmap_atomic(page);
2756 *addr = cpu_to_le32(recon_state.nr_caps);
2757 kunmap_atomic(addr);
Alex Elderebf18f42013-03-04 22:29:57 -06002758 }
Yan, Zheng44c99752013-09-22 10:28:10 +08002759
2760 reply->hdr.data_len = cpu_to_le32(pagelist->length);
2761 ceph_msg_data_add_pagelist(reply, pagelist);
Sage Weil2f2dc052009-10-06 11:31:09 -07002762 ceph_con_send(&session->s_con, reply);
2763
Sage Weil9abf82b2010-05-10 21:58:38 -07002764 mutex_unlock(&session->s_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002765
Sage Weil9abf82b2010-05-10 21:58:38 -07002766 mutex_lock(&mdsc->mutex);
2767 __wake_requests(mdsc, &session->s_waiting);
2768 mutex_unlock(&mdsc->mutex);
2769
Sage Weil2f2dc052009-10-06 11:31:09 -07002770 up_read(&mdsc->snap_rwsem);
Sage Weil2f2dc052009-10-06 11:31:09 -07002771 return;
2772
Sage Weil93cea5b2009-12-23 12:21:51 -08002773fail:
Sage Weil2f2dc052009-10-06 11:31:09 -07002774 ceph_msg_put(reply);
Sage Weil9abf82b2010-05-10 21:58:38 -07002775 up_read(&mdsc->snap_rwsem);
2776 mutex_unlock(&session->s_mutex);
Sage Weil93cea5b2009-12-23 12:21:51 -08002777fail_nomsg:
2778 ceph_pagelist_release(pagelist);
2779 kfree(pagelist);
2780fail_nopagelist:
Sage Weil9abf82b2010-05-10 21:58:38 -07002781 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
Sage Weil9abf82b2010-05-10 21:58:38 -07002782 return;
Sage Weil2f2dc052009-10-06 11:31:09 -07002783}
2784
2785
2786/*
2787 * compare old and new mdsmaps, kicking requests
2788 * and closing out old connections as necessary
2789 *
2790 * called under mdsc->mutex.
2791 */
2792static void check_new_map(struct ceph_mds_client *mdsc,
2793 struct ceph_mdsmap *newmap,
2794 struct ceph_mdsmap *oldmap)
2795{
2796 int i;
2797 int oldstate, newstate;
2798 struct ceph_mds_session *s;
2799
2800 dout("check_new_map new %u old %u\n",
2801 newmap->m_epoch, oldmap->m_epoch);
2802
2803 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2804 if (mdsc->sessions[i] == NULL)
2805 continue;
2806 s = mdsc->sessions[i];
2807 oldstate = ceph_mdsmap_get_state(oldmap, i);
2808 newstate = ceph_mdsmap_get_state(newmap, i);
2809
Sage Weil0deb01c2010-06-17 14:19:01 -07002810 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07002811 i, ceph_mds_state_name(oldstate),
Sage Weil0deb01c2010-06-17 14:19:01 -07002812 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
Sage Weil2f2dc052009-10-06 11:31:09 -07002813 ceph_mds_state_name(newstate),
Sage Weil0deb01c2010-06-17 14:19:01 -07002814 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
Sage Weil2f2dc052009-10-06 11:31:09 -07002815 session_state_name(s->s_state));
2816
Yan, Zheng3e8f43a2012-09-20 17:42:25 +08002817 if (i >= newmap->m_max_mds ||
2818 memcmp(ceph_mdsmap_get_addr(oldmap, i),
Sage Weil2f2dc052009-10-06 11:31:09 -07002819 ceph_mdsmap_get_addr(newmap, i),
2820 sizeof(struct ceph_entity_addr))) {
2821 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2822 /* the session never opened, just close it
2823 * out now */
2824 __wake_requests(mdsc, &s->s_waiting);
Sage Weil2600d2d2010-02-22 15:12:16 -08002825 __unregister_session(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07002826 } else {
2827 /* just close it */
2828 mutex_unlock(&mdsc->mutex);
2829 mutex_lock(&s->s_mutex);
2830 mutex_lock(&mdsc->mutex);
2831 ceph_con_close(&s->s_con);
2832 mutex_unlock(&s->s_mutex);
2833 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2834 }
2835
2836 /* kick any requests waiting on the recovering mds */
Sage Weil29790f22010-03-18 14:45:05 -07002837 kick_requests(mdsc, i);
Sage Weil2f2dc052009-10-06 11:31:09 -07002838 } else if (oldstate == newstate) {
2839 continue; /* nothing new with this mds */
2840 }
2841
2842 /*
2843 * send reconnect?
2844 */
2845 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
Sage Weil34b6c852010-05-10 16:31:25 -07002846 newstate >= CEPH_MDS_STATE_RECONNECT) {
2847 mutex_unlock(&mdsc->mutex);
2848 send_mds_reconnect(mdsc, s);
2849 mutex_lock(&mdsc->mutex);
2850 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002851
2852 /*
Sage Weil29790f22010-03-18 14:45:05 -07002853 * kick request on any mds that has gone active.
Sage Weil2f2dc052009-10-06 11:31:09 -07002854 */
2855 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2856 newstate >= CEPH_MDS_STATE_ACTIVE) {
Sage Weil29790f22010-03-18 14:45:05 -07002857 if (oldstate != CEPH_MDS_STATE_CREATING &&
2858 oldstate != CEPH_MDS_STATE_STARTING)
2859 pr_info("mds%d recovery completed\n", s->s_mds);
2860 kick_requests(mdsc, i);
Sage Weil2f2dc052009-10-06 11:31:09 -07002861 ceph_kick_flushing_caps(mdsc, s);
Sage Weil0dc25702009-11-20 13:43:45 -08002862 wake_up_session_caps(s, 1);
Sage Weil2f2dc052009-10-06 11:31:09 -07002863 }
2864 }
Sage Weilcb170a22010-06-21 13:38:35 -07002865
2866 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
2867 s = mdsc->sessions[i];
2868 if (!s)
2869 continue;
2870 if (!ceph_mdsmap_is_laggy(newmap, i))
2871 continue;
2872 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2873 s->s_state == CEPH_MDS_SESSION_HUNG ||
2874 s->s_state == CEPH_MDS_SESSION_CLOSING) {
2875 dout(" connecting to export targets of laggy mds%d\n",
2876 i);
2877 __open_export_target_sessions(mdsc, s);
2878 }
2879 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002880}
2881
2882
2883
2884/*
2885 * leases
2886 */
2887
2888/*
2889 * caller must hold session s_mutex, dentry->d_lock
2890 */
2891void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
2892{
2893 struct ceph_dentry_info *di = ceph_dentry(dentry);
2894
2895 ceph_put_mds_session(di->lease_session);
2896 di->lease_session = NULL;
2897}
2898
Sage Weil2600d2d2010-02-22 15:12:16 -08002899static void handle_lease(struct ceph_mds_client *mdsc,
2900 struct ceph_mds_session *session,
2901 struct ceph_msg *msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07002902{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002903 struct super_block *sb = mdsc->fsc->sb;
Sage Weil2f2dc052009-10-06 11:31:09 -07002904 struct inode *inode;
Sage Weil2f2dc052009-10-06 11:31:09 -07002905 struct dentry *parent, *dentry;
2906 struct ceph_dentry_info *di;
Sage Weil2600d2d2010-02-22 15:12:16 -08002907 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07002908 struct ceph_mds_lease *h = msg->front.iov_base;
Sage Weil1e5ea232010-06-04 10:05:40 -07002909 u32 seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002910 struct ceph_vino vino;
Sage Weil2f2dc052009-10-06 11:31:09 -07002911 struct qstr dname;
2912 int release = 0;
2913
Sage Weil2f2dc052009-10-06 11:31:09 -07002914 dout("handle_lease from mds%d\n", mds);
2915
2916 /* decode */
2917 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
2918 goto bad;
2919 vino.ino = le64_to_cpu(h->ino);
2920 vino.snap = CEPH_NOSNAP;
Sage Weil1e5ea232010-06-04 10:05:40 -07002921 seq = le32_to_cpu(h->seq);
Sage Weil2f2dc052009-10-06 11:31:09 -07002922 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
2923 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
2924 if (dname.len != get_unaligned_le32(h+1))
2925 goto bad;
2926
Sage Weil2f2dc052009-10-06 11:31:09 -07002927 mutex_lock(&session->s_mutex);
2928 session->s_seq++;
2929
2930 /* lookup inode */
2931 inode = ceph_find_inode(sb, vino);
Sage Weil2f90b852011-07-26 11:28:25 -07002932 dout("handle_lease %s, ino %llx %p %.*s\n",
2933 ceph_lease_op_name(h->action), vino.ino, inode,
Sage Weil1e5ea232010-06-04 10:05:40 -07002934 dname.len, dname.name);
Sage Weil2f2dc052009-10-06 11:31:09 -07002935 if (inode == NULL) {
2936 dout("handle_lease no inode %llx\n", vino.ino);
2937 goto release;
2938 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002939
2940 /* dentry */
2941 parent = d_find_alias(inode);
2942 if (!parent) {
2943 dout("no parent dentry on inode %p\n", inode);
2944 WARN_ON(1);
2945 goto release; /* hrm... */
2946 }
2947 dname.hash = full_name_hash(dname.name, dname.len);
2948 dentry = d_lookup(parent, &dname);
2949 dput(parent);
2950 if (!dentry)
2951 goto release;
2952
2953 spin_lock(&dentry->d_lock);
2954 di = ceph_dentry(dentry);
2955 switch (h->action) {
2956 case CEPH_MDS_LEASE_REVOKE:
Sage Weil3d8eb7a2011-11-11 09:48:53 -08002957 if (di->lease_session == session) {
Sage Weil1e5ea232010-06-04 10:05:40 -07002958 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
2959 h->seq = cpu_to_le32(di->lease_seq);
Sage Weil2f2dc052009-10-06 11:31:09 -07002960 __ceph_mdsc_drop_dentry_lease(dentry);
2961 }
2962 release = 1;
2963 break;
2964
2965 case CEPH_MDS_LEASE_RENEW:
Sage Weil3d8eb7a2011-11-11 09:48:53 -08002966 if (di->lease_session == session &&
Sage Weil2f2dc052009-10-06 11:31:09 -07002967 di->lease_gen == session->s_cap_gen &&
2968 di->lease_renew_from &&
2969 di->lease_renew_after == 0) {
2970 unsigned long duration =
2971 le32_to_cpu(h->duration_ms) * HZ / 1000;
2972
Sage Weil1e5ea232010-06-04 10:05:40 -07002973 di->lease_seq = seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002974 dentry->d_time = di->lease_renew_from + duration;
2975 di->lease_renew_after = di->lease_renew_from +
2976 (duration >> 1);
2977 di->lease_renew_from = 0;
2978 }
2979 break;
2980 }
2981 spin_unlock(&dentry->d_lock);
2982 dput(dentry);
2983
2984 if (!release)
2985 goto out;
2986
2987release:
2988 /* let's just reuse the same message */
2989 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
2990 ceph_msg_get(msg);
2991 ceph_con_send(&session->s_con, msg);
2992
2993out:
2994 iput(inode);
2995 mutex_unlock(&session->s_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002996 return;
2997
2998bad:
2999 pr_err("corrupt lease message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08003000 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003001}
3002
3003void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3004 struct inode *inode,
3005 struct dentry *dentry, char action,
3006 u32 seq)
3007{
3008 struct ceph_msg *msg;
3009 struct ceph_mds_lease *lease;
3010 int len = sizeof(*lease) + sizeof(u32);
3011 int dnamelen = 0;
3012
3013 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3014 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3015 dnamelen = dentry->d_name.len;
3016 len += dnamelen;
3017
Sage Weilb61c2762011-08-09 15:03:46 -07003018 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07003019 if (!msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07003020 return;
3021 lease = msg->front.iov_base;
3022 lease->action = action;
Sage Weil2f2dc052009-10-06 11:31:09 -07003023 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3024 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3025 lease->seq = cpu_to_le32(seq);
3026 put_unaligned_le32(dnamelen, lease + 1);
3027 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3028
3029 /*
3030 * if this is a preemptive lease RELEASE, no need to
3031 * flush request stream, since the actual request will
3032 * soon follow.
3033 */
3034 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3035
3036 ceph_con_send(&session->s_con, msg);
3037}
3038
3039/*
3040 * Preemptively release a lease we expect to invalidate anyway.
3041 * Pass @inode always, @dentry is optional.
3042 */
3043void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
Sage Weil2f90b852011-07-26 11:28:25 -07003044 struct dentry *dentry)
Sage Weil2f2dc052009-10-06 11:31:09 -07003045{
3046 struct ceph_dentry_info *di;
3047 struct ceph_mds_session *session;
3048 u32 seq;
3049
3050 BUG_ON(inode == NULL);
3051 BUG_ON(dentry == NULL);
Sage Weil2f2dc052009-10-06 11:31:09 -07003052
3053 /* is dentry lease valid? */
3054 spin_lock(&dentry->d_lock);
3055 di = ceph_dentry(dentry);
3056 if (!di || !di->lease_session ||
3057 di->lease_session->s_mds < 0 ||
3058 di->lease_gen != di->lease_session->s_cap_gen ||
3059 !time_before(jiffies, dentry->d_time)) {
3060 dout("lease_release inode %p dentry %p -- "
Sage Weil2f90b852011-07-26 11:28:25 -07003061 "no lease\n",
3062 inode, dentry);
Sage Weil2f2dc052009-10-06 11:31:09 -07003063 spin_unlock(&dentry->d_lock);
3064 return;
3065 }
3066
3067 /* we do have a lease on this dentry; note mds and seq */
3068 session = ceph_get_mds_session(di->lease_session);
3069 seq = di->lease_seq;
3070 __ceph_mdsc_drop_dentry_lease(dentry);
3071 spin_unlock(&dentry->d_lock);
3072
Sage Weil2f90b852011-07-26 11:28:25 -07003073 dout("lease_release inode %p dentry %p to mds%d\n",
3074 inode, dentry, session->s_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07003075 ceph_mdsc_lease_send_msg(session, inode, dentry,
3076 CEPH_MDS_LEASE_RELEASE, seq);
3077 ceph_put_mds_session(session);
3078}
3079
3080/*
3081 * drop all leases (and dentry refs) in preparation for umount
3082 */
3083static void drop_leases(struct ceph_mds_client *mdsc)
3084{
3085 int i;
3086
3087 dout("drop_leases\n");
3088 mutex_lock(&mdsc->mutex);
3089 for (i = 0; i < mdsc->max_sessions; i++) {
3090 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3091 if (!s)
3092 continue;
3093 mutex_unlock(&mdsc->mutex);
3094 mutex_lock(&s->s_mutex);
3095 mutex_unlock(&s->s_mutex);
3096 ceph_put_mds_session(s);
3097 mutex_lock(&mdsc->mutex);
3098 }
3099 mutex_unlock(&mdsc->mutex);
3100}
3101
3102
3103
3104/*
3105 * delayed work -- periodically trim expired leases, renew caps with mds
3106 */
3107static void schedule_delayed(struct ceph_mds_client *mdsc)
3108{
3109 int delay = 5;
3110 unsigned hz = round_jiffies_relative(HZ * delay);
3111 schedule_delayed_work(&mdsc->delayed_work, hz);
3112}
3113
3114static void delayed_work(struct work_struct *work)
3115{
3116 int i;
3117 struct ceph_mds_client *mdsc =
3118 container_of(work, struct ceph_mds_client, delayed_work.work);
3119 int renew_interval;
3120 int renew_caps;
3121
3122 dout("mdsc delayed_work\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07003123 ceph_check_delayed_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07003124
3125 mutex_lock(&mdsc->mutex);
3126 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3127 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3128 mdsc->last_renew_caps);
3129 if (renew_caps)
3130 mdsc->last_renew_caps = jiffies;
3131
3132 for (i = 0; i < mdsc->max_sessions; i++) {
3133 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3134 if (s == NULL)
3135 continue;
3136 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3137 dout("resending session close request for mds%d\n",
3138 s->s_mds);
3139 request_close_session(mdsc, s);
3140 ceph_put_mds_session(s);
3141 continue;
3142 }
3143 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3144 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3145 s->s_state = CEPH_MDS_SESSION_HUNG;
3146 pr_info("mds%d hung\n", s->s_mds);
3147 }
3148 }
3149 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3150 /* this mds is failed or recovering, just wait */
3151 ceph_put_mds_session(s);
3152 continue;
3153 }
3154 mutex_unlock(&mdsc->mutex);
3155
3156 mutex_lock(&s->s_mutex);
3157 if (renew_caps)
3158 send_renew_caps(mdsc, s);
3159 else
3160 ceph_con_keepalive(&s->s_con);
Sage Weilee6b2722010-06-10 12:55:52 -07003161 ceph_add_cap_releases(mdsc, s);
Sage Weilaab53dd2010-03-17 16:30:21 -07003162 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3163 s->s_state == CEPH_MDS_SESSION_HUNG)
Sage Weil3d7ded42010-06-09 16:47:10 -07003164 ceph_send_cap_releases(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07003165 mutex_unlock(&s->s_mutex);
3166 ceph_put_mds_session(s);
3167
3168 mutex_lock(&mdsc->mutex);
3169 }
3170 mutex_unlock(&mdsc->mutex);
3171
3172 schedule_delayed(mdsc);
3173}
3174
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003175int ceph_mdsc_init(struct ceph_fs_client *fsc)
Sage Weil2f2dc052009-10-06 11:31:09 -07003176
Sage Weil2f2dc052009-10-06 11:31:09 -07003177{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003178 struct ceph_mds_client *mdsc;
3179
3180 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3181 if (!mdsc)
3182 return -ENOMEM;
3183 mdsc->fsc = fsc;
3184 fsc->mdsc = mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07003185 mutex_init(&mdsc->mutex);
3186 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
majianpengfb3101b2013-06-25 14:48:19 +08003187 if (mdsc->mdsmap == NULL) {
3188 kfree(mdsc);
Cheng Renquan2d06eeb2010-03-26 18:04:40 +08003189 return -ENOMEM;
majianpengfb3101b2013-06-25 14:48:19 +08003190 }
Cheng Renquan2d06eeb2010-03-26 18:04:40 +08003191
Sage Weil2f2dc052009-10-06 11:31:09 -07003192 init_completion(&mdsc->safe_umount_waiters);
Sage Weilf3c60c52010-08-11 14:51:23 -07003193 init_waitqueue_head(&mdsc->session_close_wq);
Sage Weil2f2dc052009-10-06 11:31:09 -07003194 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3195 mdsc->sessions = NULL;
3196 mdsc->max_sessions = 0;
3197 mdsc->stopping = 0;
3198 init_rwsem(&mdsc->snap_rwsem);
Sage Weila105f002010-02-15 14:37:55 -08003199 mdsc->snap_realms = RB_ROOT;
Sage Weil2f2dc052009-10-06 11:31:09 -07003200 INIT_LIST_HEAD(&mdsc->snap_empty);
3201 spin_lock_init(&mdsc->snap_empty_lock);
3202 mdsc->last_tid = 0;
Sage Weil44ca18f2010-02-15 12:08:46 -08003203 mdsc->request_tree = RB_ROOT;
Sage Weil2f2dc052009-10-06 11:31:09 -07003204 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3205 mdsc->last_renew_caps = jiffies;
3206 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3207 spin_lock_init(&mdsc->cap_delay_lock);
3208 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3209 spin_lock_init(&mdsc->snap_flush_lock);
3210 mdsc->cap_flush_seq = 0;
3211 INIT_LIST_HEAD(&mdsc->cap_dirty);
Sage Weildb354052011-05-24 11:46:31 -07003212 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
Sage Weil2f2dc052009-10-06 11:31:09 -07003213 mdsc->num_cap_flushing = 0;
3214 spin_lock_init(&mdsc->cap_dirty_lock);
3215 init_waitqueue_head(&mdsc->cap_flushing_wq);
3216 spin_lock_init(&mdsc->dentry_lru_lock);
3217 INIT_LIST_HEAD(&mdsc->dentry_lru);
Cheng Renquan2d06eeb2010-03-26 18:04:40 +08003218
Yehuda Sadeh37151662010-06-17 16:16:12 -07003219 ceph_caps_init(mdsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003220 ceph_adjust_min_caps(mdsc, fsc->min_caps);
Yehuda Sadeh37151662010-06-17 16:16:12 -07003221
Sage Weil5f44f142009-11-18 14:52:18 -08003222 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07003223}
3224
3225/*
3226 * Wait for safe replies on open mds requests. If we time out, drop
3227 * all requests from the tree to avoid dangling dentry refs.
3228 */
3229static void wait_requests(struct ceph_mds_client *mdsc)
3230{
3231 struct ceph_mds_request *req;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003232 struct ceph_fs_client *fsc = mdsc->fsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07003233
3234 mutex_lock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08003235 if (__get_oldest_req(mdsc)) {
Sage Weil2f2dc052009-10-06 11:31:09 -07003236 mutex_unlock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08003237
Sage Weil2f2dc052009-10-06 11:31:09 -07003238 dout("wait_requests waiting for requests\n");
3239 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003240 fsc->client->options->mount_timeout * HZ);
Sage Weil2f2dc052009-10-06 11:31:09 -07003241
3242 /* tear down remaining requests */
Sage Weil44ca18f2010-02-15 12:08:46 -08003243 mutex_lock(&mdsc->mutex);
3244 while ((req = __get_oldest_req(mdsc))) {
Sage Weil2f2dc052009-10-06 11:31:09 -07003245 dout("wait_requests timed out on tid %llu\n",
3246 req->r_tid);
Sage Weil44ca18f2010-02-15 12:08:46 -08003247 __unregister_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -07003248 }
3249 }
3250 mutex_unlock(&mdsc->mutex);
3251 dout("wait_requests done\n");
3252}
3253
3254/*
3255 * called before mount is ro, and before dentries are torn down.
3256 * (hmm, does this still race with new lookups?)
3257 */
3258void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3259{
3260 dout("pre_umount\n");
3261 mdsc->stopping = 1;
3262
3263 drop_leases(mdsc);
Sage Weilafcdaea2009-10-14 14:27:38 -07003264 ceph_flush_dirty_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07003265 wait_requests(mdsc);
Sage Weil17c688c2010-06-21 16:12:26 -07003266
3267 /*
3268 * wait for reply handlers to drop their request refs and
3269 * their inode/dcache refs
3270 */
3271 ceph_msgr_flush();
Sage Weil2f2dc052009-10-06 11:31:09 -07003272}
3273
3274/*
3275 * wait for all write mds requests to flush.
3276 */
3277static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3278{
Sage Weil80fc7312010-03-16 15:28:54 -07003279 struct ceph_mds_request *req = NULL, *nextreq;
Sage Weil44ca18f2010-02-15 12:08:46 -08003280 struct rb_node *n;
Sage Weil2f2dc052009-10-06 11:31:09 -07003281
3282 mutex_lock(&mdsc->mutex);
3283 dout("wait_unsafe_requests want %lld\n", want_tid);
Sage Weil80fc7312010-03-16 15:28:54 -07003284restart:
Sage Weil44ca18f2010-02-15 12:08:46 -08003285 req = __get_oldest_req(mdsc);
3286 while (req && req->r_tid <= want_tid) {
Sage Weil80fc7312010-03-16 15:28:54 -07003287 /* find next request */
3288 n = rb_next(&req->r_node);
3289 if (n)
3290 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3291 else
3292 nextreq = NULL;
Sage Weil44ca18f2010-02-15 12:08:46 -08003293 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
3294 /* write op */
3295 ceph_mdsc_get_request(req);
Sage Weil80fc7312010-03-16 15:28:54 -07003296 if (nextreq)
3297 ceph_mdsc_get_request(nextreq);
Sage Weil44ca18f2010-02-15 12:08:46 -08003298 mutex_unlock(&mdsc->mutex);
3299 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3300 req->r_tid, want_tid);
3301 wait_for_completion(&req->r_safe_completion);
3302 mutex_lock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08003303 ceph_mdsc_put_request(req);
Sage Weil80fc7312010-03-16 15:28:54 -07003304 if (!nextreq)
3305 break; /* next dne before, so we're done! */
3306 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3307 /* next request was removed from tree */
3308 ceph_mdsc_put_request(nextreq);
3309 goto restart;
3310 }
3311 ceph_mdsc_put_request(nextreq); /* won't go away */
Sage Weil44ca18f2010-02-15 12:08:46 -08003312 }
Sage Weil80fc7312010-03-16 15:28:54 -07003313 req = nextreq;
Sage Weil2f2dc052009-10-06 11:31:09 -07003314 }
3315 mutex_unlock(&mdsc->mutex);
3316 dout("wait_unsafe_requests done\n");
3317}
3318
3319void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3320{
3321 u64 want_tid, want_flush;
3322
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003323 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
Sage Weil56b7cf92010-05-03 15:22:00 -07003324 return;
3325
Sage Weil2f2dc052009-10-06 11:31:09 -07003326 dout("sync\n");
3327 mutex_lock(&mdsc->mutex);
3328 want_tid = mdsc->last_tid;
3329 want_flush = mdsc->cap_flush_seq;
3330 mutex_unlock(&mdsc->mutex);
3331 dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
3332
Sage Weilafcdaea2009-10-14 14:27:38 -07003333 ceph_flush_dirty_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07003334
3335 wait_unsafe_requests(mdsc, want_tid);
3336 wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush));
3337}
3338
Sage Weilf3c60c52010-08-11 14:51:23 -07003339/*
3340 * true if all sessions are closed, or we force unmount
3341 */
H Hartley Sweeten7fd7d102011-09-23 13:22:11 -07003342static bool done_closing_sessions(struct ceph_mds_client *mdsc)
Sage Weilf3c60c52010-08-11 14:51:23 -07003343{
3344 int i, n = 0;
3345
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003346 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
Sage Weilf3c60c52010-08-11 14:51:23 -07003347 return true;
3348
3349 mutex_lock(&mdsc->mutex);
3350 for (i = 0; i < mdsc->max_sessions; i++)
3351 if (mdsc->sessions[i])
3352 n++;
3353 mutex_unlock(&mdsc->mutex);
3354 return n == 0;
3355}
Sage Weil2f2dc052009-10-06 11:31:09 -07003356
3357/*
3358 * called after sb is ro.
3359 */
3360void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3361{
3362 struct ceph_mds_session *session;
3363 int i;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003364 struct ceph_fs_client *fsc = mdsc->fsc;
3365 unsigned long timeout = fsc->client->options->mount_timeout * HZ;
Sage Weil2f2dc052009-10-06 11:31:09 -07003366
3367 dout("close_sessions\n");
3368
Sage Weil2f2dc052009-10-06 11:31:09 -07003369 /* close sessions */
Sage Weilf3c60c52010-08-11 14:51:23 -07003370 mutex_lock(&mdsc->mutex);
3371 for (i = 0; i < mdsc->max_sessions; i++) {
3372 session = __ceph_lookup_mds_session(mdsc, i);
3373 if (!session)
3374 continue;
Sage Weil2f2dc052009-10-06 11:31:09 -07003375 mutex_unlock(&mdsc->mutex);
Sage Weilf3c60c52010-08-11 14:51:23 -07003376 mutex_lock(&session->s_mutex);
3377 __close_session(mdsc, session);
3378 mutex_unlock(&session->s_mutex);
3379 ceph_put_mds_session(session);
Sage Weil2f2dc052009-10-06 11:31:09 -07003380 mutex_lock(&mdsc->mutex);
3381 }
Sage Weilf3c60c52010-08-11 14:51:23 -07003382 mutex_unlock(&mdsc->mutex);
3383
3384 dout("waiting for sessions to close\n");
3385 wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3386 timeout);
Sage Weil2f2dc052009-10-06 11:31:09 -07003387
3388 /* tear down remaining sessions */
Sage Weilf3c60c52010-08-11 14:51:23 -07003389 mutex_lock(&mdsc->mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07003390 for (i = 0; i < mdsc->max_sessions; i++) {
3391 if (mdsc->sessions[i]) {
3392 session = get_session(mdsc->sessions[i]);
Sage Weil2600d2d2010-02-22 15:12:16 -08003393 __unregister_session(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07003394 mutex_unlock(&mdsc->mutex);
3395 mutex_lock(&session->s_mutex);
3396 remove_session_caps(session);
3397 mutex_unlock(&session->s_mutex);
3398 ceph_put_mds_session(session);
3399 mutex_lock(&mdsc->mutex);
3400 }
3401 }
Sage Weil2f2dc052009-10-06 11:31:09 -07003402 WARN_ON(!list_empty(&mdsc->cap_delay_list));
Sage Weil2f2dc052009-10-06 11:31:09 -07003403 mutex_unlock(&mdsc->mutex);
3404
3405 ceph_cleanup_empty_realms(mdsc);
3406
3407 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3408
3409 dout("stopped\n");
3410}
3411
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003412static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
Sage Weil2f2dc052009-10-06 11:31:09 -07003413{
3414 dout("stop\n");
3415 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3416 if (mdsc->mdsmap)
3417 ceph_mdsmap_destroy(mdsc->mdsmap);
3418 kfree(mdsc->sessions);
Yehuda Sadeh37151662010-06-17 16:16:12 -07003419 ceph_caps_finalize(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07003420}
3421
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003422void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3423{
3424 struct ceph_mds_client *mdsc = fsc->mdsc;
3425
Sage Weilef550f62011-03-25 13:27:48 -07003426 dout("mdsc_destroy %p\n", mdsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003427 ceph_mdsc_stop(mdsc);
Sage Weilef550f62011-03-25 13:27:48 -07003428
3429 /* flush out any connection work with references to us */
3430 ceph_msgr_flush();
3431
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003432 fsc->mdsc = NULL;
3433 kfree(mdsc);
Sage Weilef550f62011-03-25 13:27:48 -07003434 dout("mdsc_destroy %p done\n", mdsc);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003435}
3436
Sage Weil2f2dc052009-10-06 11:31:09 -07003437
3438/*
3439 * handle mds map update.
3440 */
3441void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3442{
3443 u32 epoch;
3444 u32 maplen;
3445 void *p = msg->front.iov_base;
3446 void *end = p + msg->front.iov_len;
3447 struct ceph_mdsmap *newmap, *oldmap;
3448 struct ceph_fsid fsid;
3449 int err = -EINVAL;
3450
3451 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3452 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003453 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
Sage Weil07433042009-11-18 16:50:41 -08003454 return;
Sage Weilc89136e2009-10-14 09:59:09 -07003455 epoch = ceph_decode_32(&p);
3456 maplen = ceph_decode_32(&p);
Sage Weil2f2dc052009-10-06 11:31:09 -07003457 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3458
3459 /* do we need it? */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003460 ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
Sage Weil2f2dc052009-10-06 11:31:09 -07003461 mutex_lock(&mdsc->mutex);
3462 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3463 dout("handle_map epoch %u <= our %u\n",
3464 epoch, mdsc->mdsmap->m_epoch);
3465 mutex_unlock(&mdsc->mutex);
3466 return;
3467 }
3468
3469 newmap = ceph_mdsmap_decode(&p, end);
3470 if (IS_ERR(newmap)) {
3471 err = PTR_ERR(newmap);
3472 goto bad_unlock;
3473 }
3474
3475 /* swap into place */
3476 if (mdsc->mdsmap) {
3477 oldmap = mdsc->mdsmap;
3478 mdsc->mdsmap = newmap;
3479 check_new_map(mdsc, newmap, oldmap);
3480 ceph_mdsmap_destroy(oldmap);
3481 } else {
3482 mdsc->mdsmap = newmap; /* first mds map */
3483 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003484 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
Sage Weil2f2dc052009-10-06 11:31:09 -07003485
3486 __wake_requests(mdsc, &mdsc->waiting_for_map);
3487
3488 mutex_unlock(&mdsc->mutex);
3489 schedule_delayed(mdsc);
3490 return;
3491
3492bad_unlock:
3493 mutex_unlock(&mdsc->mutex);
3494bad:
3495 pr_err("error decoding mdsmap %d\n", err);
3496 return;
3497}
3498
3499static struct ceph_connection *con_get(struct ceph_connection *con)
3500{
3501 struct ceph_mds_session *s = con->private;
3502
3503 if (get_session(s)) {
Sage Weil2600d2d2010-02-22 15:12:16 -08003504 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
Sage Weil2f2dc052009-10-06 11:31:09 -07003505 return con;
3506 }
3507 dout("mdsc con_get %p FAIL\n", s);
3508 return NULL;
3509}
3510
3511static void con_put(struct ceph_connection *con)
3512{
3513 struct ceph_mds_session *s = con->private;
3514
Henry C Chang7d8e18a2011-05-11 10:29:52 +00003515 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
Sage Weil2f2dc052009-10-06 11:31:09 -07003516 ceph_put_mds_session(s);
3517}
3518
3519/*
3520 * if the client is unresponsive for long enough, the mds will kill
3521 * the session entirely.
3522 */
3523static void peer_reset(struct ceph_connection *con)
3524{
3525 struct ceph_mds_session *s = con->private;
Sage Weil7e70f0e2010-03-18 13:59:12 -07003526 struct ceph_mds_client *mdsc = s->s_mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07003527
Sage Weil7e70f0e2010-03-18 13:59:12 -07003528 pr_warning("mds%d closed our session\n", s->s_mds);
3529 send_mds_reconnect(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07003530}
3531
3532static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3533{
3534 struct ceph_mds_session *s = con->private;
3535 struct ceph_mds_client *mdsc = s->s_mdsc;
3536 int type = le16_to_cpu(msg->hdr.type);
3537
Sage Weil2600d2d2010-02-22 15:12:16 -08003538 mutex_lock(&mdsc->mutex);
3539 if (__verify_registered_session(mdsc, s) < 0) {
3540 mutex_unlock(&mdsc->mutex);
3541 goto out;
3542 }
3543 mutex_unlock(&mdsc->mutex);
3544
Sage Weil2f2dc052009-10-06 11:31:09 -07003545 switch (type) {
3546 case CEPH_MSG_MDS_MAP:
3547 ceph_mdsc_handle_map(mdsc, msg);
3548 break;
3549 case CEPH_MSG_CLIENT_SESSION:
3550 handle_session(s, msg);
3551 break;
3552 case CEPH_MSG_CLIENT_REPLY:
3553 handle_reply(s, msg);
3554 break;
3555 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
Sage Weil2600d2d2010-02-22 15:12:16 -08003556 handle_forward(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003557 break;
3558 case CEPH_MSG_CLIENT_CAPS:
3559 ceph_handle_caps(s, msg);
3560 break;
3561 case CEPH_MSG_CLIENT_SNAP:
Sage Weil2600d2d2010-02-22 15:12:16 -08003562 ceph_handle_snap(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003563 break;
3564 case CEPH_MSG_CLIENT_LEASE:
Sage Weil2600d2d2010-02-22 15:12:16 -08003565 handle_lease(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003566 break;
3567
3568 default:
3569 pr_err("received unknown message type %d %s\n", type,
3570 ceph_msg_type_name(type));
3571 }
Sage Weil2600d2d2010-02-22 15:12:16 -08003572out:
Sage Weil2f2dc052009-10-06 11:31:09 -07003573 ceph_msg_put(msg);
3574}
3575
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003576/*
3577 * authentication
3578 */
Alex Eldera3530df2012-05-16 15:16:39 -05003579
3580/*
3581 * Note: returned pointer is the address of a structure that's
3582 * managed separately. Caller must *not* attempt to free it.
3583 */
3584static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
Alex Elder8f43fb52012-05-16 15:16:39 -05003585 int *proto, int force_new)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003586{
3587 struct ceph_mds_session *s = con->private;
3588 struct ceph_mds_client *mdsc = s->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003589 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
Alex Elder74f18692012-05-16 15:16:39 -05003590 struct ceph_auth_handshake *auth = &s->s_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003591
Alex Elder74f18692012-05-16 15:16:39 -05003592 if (force_new && auth->authorizer) {
Sage Weil27859f92013-03-25 10:26:14 -07003593 ceph_auth_destroy_authorizer(ac, auth->authorizer);
Alex Elder74f18692012-05-16 15:16:39 -05003594 auth->authorizer = NULL;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003595 }
Sage Weil27859f92013-03-25 10:26:14 -07003596 if (!auth->authorizer) {
3597 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3598 auth);
Sage Weil0bed9b52013-03-25 10:26:01 -07003599 if (ret)
3600 return ERR_PTR(ret);
Sage Weil27859f92013-03-25 10:26:14 -07003601 } else {
3602 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3603 auth);
Alex Eldera2556512012-05-16 15:16:39 -05003604 if (ret)
Alex Eldera3530df2012-05-16 15:16:39 -05003605 return ERR_PTR(ret);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003606 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003607 *proto = ac->protocol;
Alex Elder74f18692012-05-16 15:16:39 -05003608
Alex Eldera3530df2012-05-16 15:16:39 -05003609 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003610}
3611
3612
3613static int verify_authorizer_reply(struct ceph_connection *con, int len)
3614{
3615 struct ceph_mds_session *s = con->private;
3616 struct ceph_mds_client *mdsc = s->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003617 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003618
Sage Weil27859f92013-03-25 10:26:14 -07003619 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003620}
3621
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003622static int invalidate_authorizer(struct ceph_connection *con)
3623{
3624 struct ceph_mds_session *s = con->private;
3625 struct ceph_mds_client *mdsc = s->s_mdsc;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003626 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003627
Sage Weil27859f92013-03-25 10:26:14 -07003628 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003629
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003630 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003631}
3632
Alex Elder53ded492013-03-01 18:00:14 -06003633static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3634 struct ceph_msg_header *hdr, int *skip)
3635{
3636 struct ceph_msg *msg;
3637 int type = (int) le16_to_cpu(hdr->type);
3638 int front_len = (int) le32_to_cpu(hdr->front_len);
3639
3640 if (con->in_msg)
3641 return con->in_msg;
3642
3643 *skip = 0;
3644 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3645 if (!msg) {
3646 pr_err("unable to allocate msg type %d len %d\n",
3647 type, front_len);
3648 return NULL;
3649 }
Alex Elder53ded492013-03-01 18:00:14 -06003650
3651 return msg;
3652}
3653
Tobias Klauser9e327892010-05-20 10:40:19 +02003654static const struct ceph_connection_operations mds_con_ops = {
Sage Weil2f2dc052009-10-06 11:31:09 -07003655 .get = con_get,
3656 .put = con_put,
3657 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003658 .get_authorizer = get_authorizer,
3659 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003660 .invalidate_authorizer = invalidate_authorizer,
Sage Weil2f2dc052009-10-06 11:31:09 -07003661 .peer_reset = peer_reset,
Alex Elder53ded492013-03-01 18:00:14 -06003662 .alloc_msg = mds_alloc_msg,
Sage Weil2f2dc052009-10-06 11:31:09 -07003663};
3664
Sage Weil2f2dc052009-10-06 11:31:09 -07003665/* eof */