projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: define struct ceph_sparse_extent and add some helpers
[linux-2.6-block.git] / net / ceph / osd_client.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
a4ce40a9 2
3d14c5d2 3#include <linux/ceph/ceph_debug.h>
f24e9980 4
3d14c5d2 5#include <linux/module.h>
f24e9980
SW		 6#include <linux/err.h>
7#include <linux/highmem.h>
8#include <linux/mm.h>
9#include <linux/pagemap.h>
10#include <linux/slab.h>
11#include <linux/uaccess.h>
68b4476b
YS		 12#ifdef CONFIG_BLOCK
13#include <linux/bio.h>
14#endif
f24e9980 15
8cb441c0 16#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 17#include <linux/ceph/libceph.h>
18#include <linux/ceph/osd_client.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/auth.h>
22#include <linux/ceph/pagelist.h>
08c1ac50 23#include <linux/ceph/striper.h>
f24e9980 24
c16e7869 25#define OSD_OPREPLY_FRONT_LEN 512
0d59ab81 26
5522ae0b
AE		 27static struct kmem_cache *ceph_osd_request_cache;
28
9e32789f 29static const struct ceph_connection_operations osd_con_ops;
f24e9980 30
f24e9980
SW		 31/*
32 * Implement client access to distributed object storage cluster.
33 *
34 * All data objects are stored within a cluster/cloud of OSDs, or
35 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
36 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
37 * remote daemons serving up and coordinating consistent and safe
38 * access to storage.
39 *
40 * Cluster membership and the mapping of data objects onto storage devices
41 * are described by the osd map.
42 *
43 * We keep track of pending OSD requests (read, write), resubmit
44 * requests to different OSDs when the cluster topology/data layout
45 * change, or retry the affected requests when the communications
46 * channel with an OSD is reset.
47 */
48
5aea3dcd
ID		 49static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req);
50static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req);
922dab61
ID		 51static void link_linger(struct ceph_osd *osd,
52 struct ceph_osd_linger_request *lreq);
53static void unlink_linger(struct ceph_osd *osd,
54 struct ceph_osd_linger_request *lreq);
a02a946d 55static void clear_backoffs(struct ceph_osd *osd);
5aea3dcd
ID		 56
57#if 1
58static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
59{
60 bool wrlocked = true;
61
62 if (unlikely(down_read_trylock(sem))) {
63 wrlocked = false;
64 up_read(sem);
65 }
66
67 return wrlocked;
68}
69static inline void verify_osdc_locked(struct ceph_osd_client *osdc)
70{
71 WARN_ON(!rwsem_is_locked(&osdc->lock));
72}
73static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc)
74{
75 WARN_ON(!rwsem_is_wrlocked(&osdc->lock));
76}
77static inline void verify_osd_locked(struct ceph_osd *osd)
78{
79 struct ceph_osd_client *osdc = osd->o_osdc;
80
81 WARN_ON(!(mutex_is_locked(&osd->lock) &&
82 rwsem_is_locked(&osdc->lock)) &&
83 !rwsem_is_wrlocked(&osdc->lock));
84}
922dab61
ID		 85static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq)
86{
87 WARN_ON(!mutex_is_locked(&lreq->lock));
88}
5aea3dcd
ID		 89#else
90static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { }
91static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { }
92static inline void verify_osd_locked(struct ceph_osd *osd) { }
922dab61 93static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { }
5aea3dcd
ID		 94#endif
95
f24e9980
SW		 96/*
97 * calculate the mapping of a file extent onto an object, and fill out the
98 * request accordingly. shorten extent as necessary if it crosses an
99 * object boundary.
100 *
101 * fill osd op in request message.
102 */
dbe0fc41 103static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
a19dadfb 104 u64 *objnum, u64 *objoff, u64 *objlen)
f24e9980 105{
60e56f13 106 u64 orig_len = *plen;
dccbf080 107 u32 xlen;
f24e9980 108
60e56f13 109 /* object extent? */
dccbf080
ID		 110 ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
111 objoff, &xlen);
112 *objlen = xlen;
75d1c941
AE		 113 if (*objlen < orig_len) {
114 *plen = *objlen;
60e56f13
AE		 115 dout(" skipping last %llu, final file extent %llu~%llu\n",
116 orig_len - *plen, off, *plen);
117 }
118
75d1c941 119 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
3ff5f385 120 return 0;
f24e9980
SW		 121}
122
c54d47bf
AE		 123static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
124{
125 memset(osd_data, 0, sizeof (*osd_data));
126 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
127}
128
89486833
ID		 129/*
130 * Consumes @pages if @own_pages is true.
131 */
a4ce40a9 132static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
43bfe5de
AE		 133 struct page **pages, u64 length, u32 alignment,
134 bool pages_from_pool, bool own_pages)
135{
136 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
137 osd_data->pages = pages;
138 osd_data->length = length;
139 osd_data->alignment = alignment;
140 osd_data->pages_from_pool = pages_from_pool;
141 osd_data->own_pages = own_pages;
142}
43bfe5de 143
89486833
ID		 144/*
145 * Consumes a ref on @pagelist.
146 */
a4ce40a9 147static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
43bfe5de
AE		 148 struct ceph_pagelist *pagelist)
149{
150 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
151 osd_data->pagelist = pagelist;
152}
43bfe5de
AE		 153
154#ifdef CONFIG_BLOCK
a4ce40a9 155static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
5359a17d
ID		 156 struct ceph_bio_iter *bio_pos,
157 u32 bio_length)
43bfe5de
AE		 158{
159 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
5359a17d 160 osd_data->bio_pos = *bio_pos;
43bfe5de
AE		 161 osd_data->bio_length = bio_length;
162}
43bfe5de
AE		 163#endif /* CONFIG_BLOCK */
164
b9e281c2 165static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data,
0010f705
ID		 166 struct ceph_bvec_iter *bvec_pos,
167 u32 num_bvecs)
b9e281c2
ID		 168{
169 osd_data->type = CEPH_OSD_DATA_TYPE_BVECS;
170 osd_data->bvec_pos = *bvec_pos;
0010f705 171 osd_data->num_bvecs = num_bvecs;
b9e281c2
ID		 172}
173
49719778
AE		 174static struct ceph_osd_data *
175osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
176{
177 BUG_ON(which >= osd_req->r_num_ops);
178
179 return &osd_req->r_ops[which].raw_data_in;
180}
181
a4ce40a9
AE		 182struct ceph_osd_data *
183osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
406e2c9f 184 unsigned int which)
a4ce40a9 185{
863c7eb5 186 return osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9
AE		 187}
188EXPORT_SYMBOL(osd_req_op_extent_osd_data);
189
49719778
AE		 190void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
191 unsigned int which, struct page **pages,
192 u64 length, u32 alignment,
193 bool pages_from_pool, bool own_pages)
194{
195 struct ceph_osd_data *osd_data;
196
197 osd_data = osd_req_op_raw_data_in(osd_req, which);
198 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
199 pages_from_pool, own_pages);
200}
201EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
202
a4ce40a9 203void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
406e2c9f
AE		 204 unsigned int which, struct page **pages,
205 u64 length, u32 alignment,
a4ce40a9
AE		 206 bool pages_from_pool, bool own_pages)
207{
208 struct ceph_osd_data *osd_data;
209
863c7eb5 210 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9
AE		 211 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
212 pages_from_pool, own_pages);
a4ce40a9
AE		 213}
214EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
215
216void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
406e2c9f 217 unsigned int which, struct ceph_pagelist *pagelist)
a4ce40a9
AE		 218{
219 struct ceph_osd_data *osd_data;
220
863c7eb5 221 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
a4ce40a9 222 ceph_osd_data_pagelist_init(osd_data, pagelist);
a4ce40a9
AE		 223}
224EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
225
226#ifdef CONFIG_BLOCK
227void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
5359a17d
ID		 228 unsigned int which,
229 struct ceph_bio_iter *bio_pos,
230 u32 bio_length)
a4ce40a9
AE		 231{
232 struct ceph_osd_data *osd_data;
863c7eb5
AE		 233
234 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
5359a17d 235 ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
a4ce40a9
AE		 236}
237EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
238#endif /* CONFIG_BLOCK */
239
0010f705
ID		 240void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
241 unsigned int which,
242 struct bio_vec *bvecs, u32 num_bvecs,
243 u32 bytes)
244{
245 struct ceph_osd_data *osd_data;
246 struct ceph_bvec_iter it = {
247 .bvecs = bvecs,
248 .iter = { .bi_size = bytes },
249 };
250
251 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
252 ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
253}
254EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
255
b9e281c2
ID		 256void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
257 unsigned int which,
258 struct ceph_bvec_iter *bvec_pos)
259{
260 struct ceph_osd_data *osd_data;
261
262 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
0010f705 263 ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
b9e281c2
ID		 264}
265EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
266
a4ce40a9
AE		 267static void osd_req_op_cls_request_info_pagelist(
268 struct ceph_osd_request *osd_req,
269 unsigned int which, struct ceph_pagelist *pagelist)
270{
271 struct ceph_osd_data *osd_data;
272
863c7eb5 273 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
a4ce40a9 274 ceph_osd_data_pagelist_init(osd_data, pagelist);
a4ce40a9
AE		 275}
276
04017e29
AE		 277void osd_req_op_cls_request_data_pagelist(
278 struct ceph_osd_request *osd_req,
279 unsigned int which, struct ceph_pagelist *pagelist)
280{
281 struct ceph_osd_data *osd_data;
282
863c7eb5 283 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
04017e29 284 ceph_osd_data_pagelist_init(osd_data, pagelist);
bb873b53
ID		 285 osd_req->r_ops[which].cls.indata_len += pagelist->length;
286 osd_req->r_ops[which].indata_len += pagelist->length;
04017e29
AE		 287}
288EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
289
6c57b554
AE		 290void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
291 unsigned int which, struct page **pages, u64 length,
292 u32 alignment, bool pages_from_pool, bool own_pages)
293{
294 struct ceph_osd_data *osd_data;
295
296 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
297 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
298 pages_from_pool, own_pages);
bb873b53
ID		 299 osd_req->r_ops[which].cls.indata_len += length;
300 osd_req->r_ops[which].indata_len += length;
6c57b554
AE		 301}
302EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
303
b9e281c2
ID		 304void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req,
305 unsigned int which,
0010f705
ID		 306 struct bio_vec *bvecs, u32 num_bvecs,
307 u32 bytes)
b9e281c2
ID		 308{
309 struct ceph_osd_data *osd_data;
310 struct ceph_bvec_iter it = {
311 .bvecs = bvecs,
312 .iter = { .bi_size = bytes },
313 };
314
315 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
0010f705 316 ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
b9e281c2
ID		 317 osd_req->r_ops[which].cls.indata_len += bytes;
318 osd_req->r_ops[which].indata_len += bytes;
319}
320EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs);
321
a4ce40a9
AE		 322void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
323 unsigned int which, struct page **pages, u64 length,
324 u32 alignment, bool pages_from_pool, bool own_pages)
325{
326 struct ceph_osd_data *osd_data;
327
863c7eb5 328 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
a4ce40a9
AE		 329 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
330 pages_from_pool, own_pages);
a4ce40a9
AE		 331}
332EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
333
23c08a9c
AE		 334static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
335{
336 switch (osd_data->type) {
337 case CEPH_OSD_DATA_TYPE_NONE:
338 return 0;
339 case CEPH_OSD_DATA_TYPE_PAGES:
340 return osd_data->length;
341 case CEPH_OSD_DATA_TYPE_PAGELIST:
342 return (u64)osd_data->pagelist->length;
343#ifdef CONFIG_BLOCK
344 case CEPH_OSD_DATA_TYPE_BIO:
345 return (u64)osd_data->bio_length;
346#endif /* CONFIG_BLOCK */
b9e281c2
ID		 347 case CEPH_OSD_DATA_TYPE_BVECS:
348 return osd_data->bvec_pos.iter.bi_size;
23c08a9c
AE		 349 default:
350 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
351 return 0;
352 }
353}
354
c54d47bf
AE		 355static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
356{
5476492f 357 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
c54d47bf
AE		 358 int num_pages;
359
360 num_pages = calc_pages_for((u64)osd_data->alignment,
361 (u64)osd_data->length);
362 ceph_release_page_vector(osd_data->pages, num_pages);
89486833
ID		 363 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
364 ceph_pagelist_release(osd_data->pagelist);
c54d47bf 365 }
5476492f
AE		 366 ceph_osd_data_init(osd_data);
367}
368
369static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
370 unsigned int which)
371{
372 struct ceph_osd_req_op *op;
373
374 BUG_ON(which >= osd_req->r_num_ops);
375 op = &osd_req->r_ops[which];
376
377 switch (op->op) {
378 case CEPH_OSD_OP_READ:
379 case CEPH_OSD_OP_WRITE:
e30b7577 380 case CEPH_OSD_OP_WRITEFULL:
a679e50f 381 kfree(op->extent.sparse_ext);
5476492f
AE		 382 ceph_osd_data_release(&op->extent.osd_data);
383 break;
384 case CEPH_OSD_OP_CALL:
385 ceph_osd_data_release(&op->cls.request_info);
04017e29 386 ceph_osd_data_release(&op->cls.request_data);
5476492f
AE		 387 ceph_osd_data_release(&op->cls.response_data);
388 break;
d74b50be
YZ		 389 case CEPH_OSD_OP_SETXATTR:
390 case CEPH_OSD_OP_CMPXATTR:
391 ceph_osd_data_release(&op->xattr.osd_data);
392 break;
66ba609f
YZ		 393 case CEPH_OSD_OP_STAT:
394 ceph_osd_data_release(&op->raw_data_in);
395 break;
922dab61
ID		 396 case CEPH_OSD_OP_NOTIFY_ACK:
397 ceph_osd_data_release(&op->notify_ack.request_data);
398 break;
19079203
ID		 399 case CEPH_OSD_OP_NOTIFY:
400 ceph_osd_data_release(&op->notify.request_data);
401 ceph_osd_data_release(&op->notify.response_data);
402 break;
a4ed38d7
DF		 403 case CEPH_OSD_OP_LIST_WATCHERS:
404 ceph_osd_data_release(&op->list_watchers.response_data);
405 break;
78beb0ff 406 case CEPH_OSD_OP_COPY_FROM2:
23ddf9be
LH		 407 ceph_osd_data_release(&op->copy_from.osd_data);
408 break;
5476492f
AE		 409 default:
410 break;
411 }
c54d47bf
AE		 412}
413
63244fa1
ID		 414/*
415 * Assumes @t is zero-initialized.
416 */
417static void target_init(struct ceph_osd_request_target *t)
418{
419 ceph_oid_init(&t->base_oid);
420 ceph_oloc_init(&t->base_oloc);
421 ceph_oid_init(&t->target_oid);
422 ceph_oloc_init(&t->target_oloc);
423
424 ceph_osds_init(&t->acting);
425 ceph_osds_init(&t->up);
426 t->size = -1;
427 t->min_size = -1;
428
429 t->osd = CEPH_HOMELESS_OSD;
430}
431
922dab61
ID		 432static void target_copy(struct ceph_osd_request_target *dest,
433 const struct ceph_osd_request_target *src)
434{
435 ceph_oid_copy(&dest->base_oid, &src->base_oid);
436 ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
437 ceph_oid_copy(&dest->target_oid, &src->target_oid);
438 ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
439
440 dest->pgid = src->pgid; /* struct */
dc98ff72 441 dest->spgid = src->spgid; /* struct */
922dab61
ID		 442 dest->pg_num = src->pg_num;
443 dest->pg_num_mask = src->pg_num_mask;
444 ceph_osds_copy(&dest->acting, &src->acting);
445 ceph_osds_copy(&dest->up, &src->up);
446 dest->size = src->size;
447 dest->min_size = src->min_size;
448 dest->sort_bitwise = src->sort_bitwise;
2f3fead6 449 dest->recovery_deletes = src->recovery_deletes;
922dab61
ID		 450
451 dest->flags = src->flags;
7ed286f3 452 dest->used_replica = src->used_replica;
922dab61
ID		 453 dest->paused = src->paused;
454
04c7d789 455 dest->epoch = src->epoch;
dc93e0e2
ID		 456 dest->last_force_resend = src->last_force_resend;
457
922dab61
ID		 458 dest->osd = src->osd;
459}
460
63244fa1
ID		 461static void target_destroy(struct ceph_osd_request_target *t)
462{
463 ceph_oid_destroy(&t->base_oid);
30c156d9 464 ceph_oloc_destroy(&t->base_oloc);
63244fa1 465 ceph_oid_destroy(&t->target_oid);
30c156d9 466 ceph_oloc_destroy(&t->target_oloc);
63244fa1
ID		 467}
468
f24e9980
SW		 469/*
470 * requests
471 */
3540bfdb
ID		 472static void request_release_checks(struct ceph_osd_request *req)
473{
474 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
4609245e 475 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
94e85771 476 WARN_ON(!list_empty(&req->r_private_item));
3540bfdb
ID		 477 WARN_ON(req->r_osd);
478}
479
9e94af20 480static void ceph_osdc_release_request(struct kref *kref)
f24e9980 481{
9e94af20
ID		 482 struct ceph_osd_request *req = container_of(kref,
483 struct ceph_osd_request, r_kref);
5476492f 484 unsigned int which;
415e49a9 485
9e94af20
ID		 486 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
487 req->r_request, req->r_reply);
3540bfdb 488 request_release_checks(req);
9e94af20 489
415e49a9
SW		 490 if (req->r_request)
491 ceph_msg_put(req->r_request);
5aea3dcd 492 if (req->r_reply)
ab8cb34a 493 ceph_msg_put(req->r_reply);
0fff87ec 494
5476492f
AE		 495 for (which = 0; which < req->r_num_ops; which++)
496 osd_req_op_data_release(req, which);
0fff87ec 497
a66dd383 498 target_destroy(&req->r_t);
415e49a9 499 ceph_put_snap_context(req->r_snapc);
d30291b9 500
415e49a9
SW		 501 if (req->r_mempool)
502 mempool_free(req, req->r_osdc->req_mempool);
3f1af42a 503 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
5522ae0b 504 kmem_cache_free(ceph_osd_request_cache, req);
3f1af42a
ID		 505 else
506 kfree(req);
f24e9980 507}
9e94af20
ID		 508
509void ceph_osdc_get_request(struct ceph_osd_request *req)
510{
511 dout("%s %p (was %d)\n", __func__, req,
2c935bc5 512 kref_read(&req->r_kref));
9e94af20
ID		 513 kref_get(&req->r_kref);
514}
515EXPORT_SYMBOL(ceph_osdc_get_request);
516
517void ceph_osdc_put_request(struct ceph_osd_request *req)
518{
3ed97d63
ID		 519 if (req) {
520 dout("%s %p (was %d)\n", __func__, req,
2c935bc5 521 kref_read(&req->r_kref));
3ed97d63
ID		 522 kref_put(&req->r_kref, ceph_osdc_release_request);
523 }
9e94af20
ID		 524}
525EXPORT_SYMBOL(ceph_osdc_put_request);
68b4476b 526
3540bfdb
ID		 527static void request_init(struct ceph_osd_request *req)
528{
042f6498 529 /* req only, each op is zeroed in osd_req_op_init() */
3540bfdb
ID		 530 memset(req, 0, sizeof(*req));
531
532 kref_init(&req->r_kref);
533 init_completion(&req->r_completion);
3540bfdb 534 RB_CLEAR_NODE(&req->r_node);
4609245e 535 RB_CLEAR_NODE(&req->r_mc_node);
94e85771 536 INIT_LIST_HEAD(&req->r_private_item);
3540bfdb
ID		 537
538 target_init(&req->r_t);
539}
540
3499e8a5 541struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
f24e9980 542 struct ceph_snap_context *snapc,
1b83bef2 543 unsigned int num_ops,
3499e8a5 544 bool use_mempool,
54a54007 545 gfp_t gfp_flags)
f24e9980
SW		 546{
547 struct ceph_osd_request *req;
1b83bef2 548
f24e9980 549 if (use_mempool) {
3f1af42a 550 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
3499e8a5 551 req = mempool_alloc(osdc->req_mempool, gfp_flags);
3f1af42a
ID		 552 } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
553 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
f24e9980 554 } else {
3f1af42a 555 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
acafe7e3 556 req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
f24e9980 557 }
3f1af42a 558 if (unlikely(!req))
a79832f2 559 return NULL;
f24e9980 560
3540bfdb 561 request_init(req);
f24e9980
SW		 562 req->r_osdc = osdc;
563 req->r_mempool = use_mempool;
79528734 564 req->r_num_ops = num_ops;
84127282
ID		 565 req->r_snapid = CEPH_NOSNAP;
566 req->r_snapc = ceph_get_snap_context(snapc);
68b4476b 567
13d1ad16
ID		 568 dout("%s req %p\n", __func__, req);
569 return req;
570}
571EXPORT_SYMBOL(ceph_osdc_alloc_request);
3f1af42a 572
2e59ffd1 573static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
30c156d9
YZ		 574{
575 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
576}
577
0d9c1ab3
ID		 578static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
579 int num_request_data_items,
580 int num_reply_data_items)
13d1ad16
ID		 581{
582 struct ceph_osd_client *osdc = req->r_osdc;
583 struct ceph_msg *msg;
584 int msg_size;
c16e7869 585
0d9c1ab3 586 WARN_ON(req->r_request || req->r_reply);
d30291b9 587 WARN_ON(ceph_oid_empty(&req->r_base_oid));
30c156d9 588 WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
d30291b9 589
13d1ad16 590 /* create request message */
8cb441c0
ID		 591 msg_size = CEPH_ENCODING_START_BLK_LEN +
592 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
593 msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
594 msg_size += CEPH_ENCODING_START_BLK_LEN +
595 sizeof(struct ceph_osd_reqid); /* reqid */
596 msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
597 msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
30c156d9
YZ		 598 msg_size += CEPH_ENCODING_START_BLK_LEN +
599 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
13d1ad16
ID		 600 msg_size += 4 + req->r_base_oid.name_len; /* oid */
601 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
ae458f5a
ID		 602 msg_size += 8; /* snapid */
603 msg_size += 8; /* snap_seq */
13d1ad16 604 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
8cb441c0 605 msg_size += 4 + 8; /* retry_attempt, features */
ae458f5a 606
13d1ad16 607 if (req->r_mempool)
0d9c1ab3
ID		 608 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
609 num_request_data_items);
f24e9980 610 else
0d9c1ab3
ID		 611 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
612 num_request_data_items, gfp, true);
13d1ad16
ID		 613 if (!msg)
614 return -ENOMEM;
68b4476b 615
f24e9980 616 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5 617 req->r_request = msg;
3499e8a5 618
13d1ad16
ID		 619 /* create reply message */
620 msg_size = OSD_OPREPLY_FRONT_LEN;
711da55d
ID		 621 msg_size += req->r_base_oid.name_len;
622 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
13d1ad16
ID		 623
624 if (req->r_mempool)
0d9c1ab3
ID		 625 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
626 num_reply_data_items);
13d1ad16 627 else
0d9c1ab3
ID		 628 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
629 num_reply_data_items, gfp, true);
13d1ad16
ID		 630 if (!msg)
631 return -ENOMEM;
632
633 req->r_reply = msg;
634
635 return 0;
3499e8a5
YS		 636}
637
a8dd0a37 638static bool osd_req_opcode_valid(u16 opcode)
68b4476b 639{
a8dd0a37 640 switch (opcode) {
70b5bfa3
ID		 641#define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
642__CEPH_FORALL_OSD_OPS(GENERATE_CASE)
643#undef GENERATE_CASE
a8dd0a37
AE		 644 default:
645 return false;
646 }
647}
648
0d9c1ab3
ID		 649static void get_num_data_items(struct ceph_osd_request *req,
650 int *num_request_data_items,
651 int *num_reply_data_items)
652{
653 struct ceph_osd_req_op *op;
654
655 *num_request_data_items = 0;
656 *num_reply_data_items = 0;
657
658 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
659 switch (op->op) {
660 /* request */
661 case CEPH_OSD_OP_WRITE:
662 case CEPH_OSD_OP_WRITEFULL:
663 case CEPH_OSD_OP_SETXATTR:
664 case CEPH_OSD_OP_CMPXATTR:
665 case CEPH_OSD_OP_NOTIFY_ACK:
78beb0ff 666 case CEPH_OSD_OP_COPY_FROM2:
0d9c1ab3
ID		 667 *num_request_data_items += 1;
668 break;
669
670 /* reply */
671 case CEPH_OSD_OP_STAT:
672 case CEPH_OSD_OP_READ:
673 case CEPH_OSD_OP_LIST_WATCHERS:
674 *num_reply_data_items += 1;
675 break;
676
677 /* both */
678 case CEPH_OSD_OP_NOTIFY:
679 *num_request_data_items += 1;
680 *num_reply_data_items += 1;
681 break;
682 case CEPH_OSD_OP_CALL:
683 *num_request_data_items += 2;
684 *num_reply_data_items += 1;
685 break;
686
687 default:
688 WARN_ON(!osd_req_opcode_valid(op->op));
689 break;
690 }
691 }
692}
693
694/*
695 * oid, oloc and OSD op opcode(s) must be filled in before this function
696 * is called.
697 */
698int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
699{
700 int num_request_data_items, num_reply_data_items;
701
702 get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
703 return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
704 num_reply_data_items);
705}
706EXPORT_SYMBOL(ceph_osdc_alloc_messages);
707
33803f33
AE		 708/*
709 * This is an osd op init function for opcodes that have no data or
710 * other information associated with them. It also serves as a
711 * common init routine for all the other init functions, below.
712 */
042f6498
JL		 713struct ceph_osd_req_op *
714osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
144cba14 715 u16 opcode, u32 flags)
33803f33 716{
c99d2d4a
AE		 717 struct ceph_osd_req_op *op;
718
719 BUG_ON(which >= osd_req->r_num_ops);
33803f33
AE		 720 BUG_ON(!osd_req_opcode_valid(opcode));
721
c99d2d4a 722 op = &osd_req->r_ops[which];
33803f33 723 memset(op, 0, sizeof (*op));
33803f33 724 op->op = opcode;
144cba14 725 op->flags = flags;
c99d2d4a
AE		 726
727 return op;
33803f33 728}
49719778
AE		 729EXPORT_SYMBOL(osd_req_op_init);
730
c99d2d4a
AE		 731void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
732 unsigned int which, u16 opcode,
33803f33
AE		 733 u64 offset, u64 length,
734 u64 truncate_size, u32 truncate_seq)
735{
042f6498
JL		 736 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
737 opcode, 0);
33803f33
AE		 738 size_t payload_len = 0;
739
ad7a60de 740 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
e30b7577
ID		 741 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
742 opcode != CEPH_OSD_OP_TRUNCATE);
33803f33 743
33803f33
AE		 744 op->extent.offset = offset;
745 op->extent.length = length;
746 op->extent.truncate_size = truncate_size;
747 op->extent.truncate_seq = truncate_seq;
e30b7577 748 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
33803f33
AE		 749 payload_len += length;
750
de2aa102 751 op->indata_len = payload_len;
33803f33
AE		 752}
753EXPORT_SYMBOL(osd_req_op_extent_init);
754
c99d2d4a
AE		 755void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
756 unsigned int which, u64 length)
e5975c7c 757{
c99d2d4a
AE		 758 struct ceph_osd_req_op *op;
759 u64 previous;
760
761 BUG_ON(which >= osd_req->r_num_ops);
762 op = &osd_req->r_ops[which];
763 previous = op->extent.length;
e5975c7c
AE		 764
765 if (length == previous)
766 return; /* Nothing to do */
767 BUG_ON(length > previous);
768
769 op->extent.length = length;
d641df81
YZ		 770 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
771 op->indata_len -= previous - length;
e5975c7c
AE		 772}
773EXPORT_SYMBOL(osd_req_op_extent_update);
774
2c63f49a
YZ		 775void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
776 unsigned int which, u64 offset_inc)
777{
778 struct ceph_osd_req_op *op, *prev_op;
779
780 BUG_ON(which + 1 >= osd_req->r_num_ops);
781
782 prev_op = &osd_req->r_ops[which];
042f6498 783 op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
2c63f49a
YZ		 784 /* dup previous one */
785 op->indata_len = prev_op->indata_len;
786 op->outdata_len = prev_op->outdata_len;
787 op->extent = prev_op->extent;
788 /* adjust offset */
789 op->extent.offset += offset_inc;
790 op->extent.length -= offset_inc;
791
792 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
793 op->indata_len -= offset_inc;
794}
795EXPORT_SYMBOL(osd_req_op_extent_dup_last);
796
fe943d50 797int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
24639ce5 798 const char *class, const char *method)
33803f33 799{
24639ce5 800 struct ceph_osd_req_op *op;
5f562df5 801 struct ceph_pagelist *pagelist;
33803f33
AE		 802 size_t payload_len = 0;
803 size_t size;
4766815b 804 int ret;
33803f33 805
042f6498 806 op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
33803f33 807
33165d47 808 pagelist = ceph_pagelist_alloc(GFP_NOFS);
fe943d50
CX		 809 if (!pagelist)
810 return -ENOMEM;
811
33803f33
AE		 812 op->cls.class_name = class;
813 size = strlen(class);
814 BUG_ON(size > (size_t) U8_MAX);
815 op->cls.class_len = size;
4766815b
DD		 816 ret = ceph_pagelist_append(pagelist, class, size);
817 if (ret)
818 goto err_pagelist_free;
33803f33
AE		 819 payload_len += size;
820
821 op->cls.method_name = method;
822 size = strlen(method);
823 BUG_ON(size > (size_t) U8_MAX);
824 op->cls.method_len = size;
4766815b
DD		 825 ret = ceph_pagelist_append(pagelist, method, size);
826 if (ret)
827 goto err_pagelist_free;
33803f33
AE		 828 payload_len += size;
829
a4ce40a9 830 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
de2aa102 831 op->indata_len = payload_len;
fe943d50 832 return 0;
4766815b
DD		 833
834err_pagelist_free:
835 ceph_pagelist_release(pagelist);
836 return ret;
33803f33
AE		 837}
838EXPORT_SYMBOL(osd_req_op_cls_init);
8c042b0d 839
d74b50be
YZ		 840int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
841 u16 opcode, const char *name, const void *value,
842 size_t size, u8 cmp_op, u8 cmp_mode)
843{
042f6498
JL		 844 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
845 opcode, 0);
d74b50be
YZ		 846 struct ceph_pagelist *pagelist;
847 size_t payload_len;
4766815b 848 int ret;
d74b50be
YZ		 849
850 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
851
33165d47 852 pagelist = ceph_pagelist_alloc(GFP_NOFS);
d74b50be
YZ		 853 if (!pagelist)
854 return -ENOMEM;
855
d74b50be
YZ		 856 payload_len = strlen(name);
857 op->xattr.name_len = payload_len;
4766815b
DD		 858 ret = ceph_pagelist_append(pagelist, name, payload_len);
859 if (ret)
860 goto err_pagelist_free;
d74b50be
YZ		 861
862 op->xattr.value_len = size;
4766815b
DD		 863 ret = ceph_pagelist_append(pagelist, value, size);
864 if (ret)
865 goto err_pagelist_free;
d74b50be
YZ		 866 payload_len += size;
867
868 op->xattr.cmp_op = cmp_op;
869 op->xattr.cmp_mode = cmp_mode;
870
871 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
de2aa102 872 op->indata_len = payload_len;
d74b50be 873 return 0;
4766815b
DD		 874
875err_pagelist_free:
876 ceph_pagelist_release(pagelist);
877 return ret;
d74b50be
YZ		 878}
879EXPORT_SYMBOL(osd_req_op_xattr_init);
880
922dab61
ID		 881/*
882 * @watch_opcode: CEPH_OSD_WATCH_OP_*
883 */
884static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
75dbb685 885 u8 watch_opcode, u64 cookie, u32 gen)
33803f33 886{
922dab61 887 struct ceph_osd_req_op *op;
33803f33 888
042f6498 889 op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
33803f33 890 op->watch.cookie = cookie;
922dab61 891 op->watch.op = watch_opcode;
75dbb685
ID		 892 op->watch.gen = gen;
893}
894
895/*
896 * prot_ver, timeout and notify payload (may be empty) should already be
897 * encoded in @request_pl
898 */
899static void osd_req_op_notify_init(struct ceph_osd_request *req, int which,
900 u64 cookie, struct ceph_pagelist *request_pl)
901{
902 struct ceph_osd_req_op *op;
903
904 op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
905 op->notify.cookie = cookie;
906
907 ceph_osd_data_pagelist_init(&op->notify.request_data, request_pl);
908 op->indata_len = request_pl->length;
33803f33 909}
33803f33 910
d3798acc
ID		 911/*
912 * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
913 */
c647b8a8
ID		 914void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
915 unsigned int which,
916 u64 expected_object_size,
d3798acc
ID		 917 u64 expected_write_size,
918 u32 flags)
c647b8a8 919{
042f6498 920 struct ceph_osd_req_op *op;
c647b8a8 921
042f6498 922 op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0);
c647b8a8
ID		 923 op->alloc_hint.expected_object_size = expected_object_size;
924 op->alloc_hint.expected_write_size = expected_write_size;
d3798acc 925 op->alloc_hint.flags = flags;
c647b8a8
ID		 926
927 /*
928 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
929 * not worth a feature bit. Set FAILOK per-op flag to make
930 * sure older osds don't trip over an unsupported opcode.
931 */
932 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
933}
934EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
935
90af3602 936static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
ec9123c5
AE		 937 struct ceph_osd_data *osd_data)
938{
939 u64 length = ceph_osd_data_length(osd_data);
940
941 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
942 BUG_ON(length > (u64) SIZE_MAX);
943 if (length)
90af3602 944 ceph_msg_data_add_pages(msg, osd_data->pages,
e8862740 945 length, osd_data->alignment, false);
ec9123c5
AE		 946 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
947 BUG_ON(!length);
90af3602 948 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
ec9123c5
AE		 949#ifdef CONFIG_BLOCK
950 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
5359a17d 951 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
ec9123c5 952#endif
b9e281c2
ID		 953 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
954 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
ec9123c5
AE		 955 } else {
956 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
957 }
958}
959
bb873b53
ID		 960static u32 osd_req_encode_op(struct ceph_osd_op *dst,
961 const struct ceph_osd_req_op *src)
a8dd0a37 962{
a8dd0a37
AE		 963 switch (src->op) {
964 case CEPH_OSD_OP_STAT:
965 break;
966 case CEPH_OSD_OP_READ:
967 case CEPH_OSD_OP_WRITE:
e30b7577 968 case CEPH_OSD_OP_WRITEFULL:
ad7a60de 969 case CEPH_OSD_OP_ZERO:
ad7a60de 970 case CEPH_OSD_OP_TRUNCATE:
a8dd0a37
AE		 971 dst->extent.offset = cpu_to_le64(src->extent.offset);
972 dst->extent.length = cpu_to_le64(src->extent.length);
973 dst->extent.truncate_size =
974 cpu_to_le64(src->extent.truncate_size);
975 dst->extent.truncate_seq =
976 cpu_to_le32(src->extent.truncate_seq);
977 break;
978 case CEPH_OSD_OP_CALL:
a8dd0a37
AE		 979 dst->cls.class_len = src->cls.class_len;
980 dst->cls.method_len = src->cls.method_len;
bb873b53 981 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
a8dd0a37 982 break;
a8dd0a37
AE		 983 case CEPH_OSD_OP_WATCH:
984 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
922dab61
ID		 985 dst->watch.ver = cpu_to_le64(0);
986 dst->watch.op = src->watch.op;
987 dst->watch.gen = cpu_to_le32(src->watch.gen);
988 break;
989 case CEPH_OSD_OP_NOTIFY_ACK:
a8dd0a37 990 break;
19079203
ID		 991 case CEPH_OSD_OP_NOTIFY:
992 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
993 break;
a4ed38d7
DF		 994 case CEPH_OSD_OP_LIST_WATCHERS:
995 break;
c647b8a8
ID		 996 case CEPH_OSD_OP_SETALLOCHINT:
997 dst->alloc_hint.expected_object_size =
998 cpu_to_le64(src->alloc_hint.expected_object_size);
999 dst->alloc_hint.expected_write_size =
1000 cpu_to_le64(src->alloc_hint.expected_write_size);
d3798acc 1001 dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
c647b8a8 1002 break;
d74b50be
YZ		 1003 case CEPH_OSD_OP_SETXATTR:
1004 case CEPH_OSD_OP_CMPXATTR:
1005 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1006 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1007 dst->xattr.cmp_op = src->xattr.cmp_op;
1008 dst->xattr.cmp_mode = src->xattr.cmp_mode;
d74b50be 1009 break;
864e9197
YZ		 1010 case CEPH_OSD_OP_CREATE:
1011 case CEPH_OSD_OP_DELETE:
1012 break;
78beb0ff 1013 case CEPH_OSD_OP_COPY_FROM2:
23ddf9be
LH		 1014 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1015 dst->copy_from.src_version =
1016 cpu_to_le64(src->copy_from.src_version);
1017 dst->copy_from.flags = src->copy_from.flags;
1018 dst->copy_from.src_fadvise_flags =
1019 cpu_to_le32(src->copy_from.src_fadvise_flags);
1020 break;
a8dd0a37 1021 default:
4c46459c 1022 pr_err("unsupported osd opcode %s\n",
8f63ca2d 1023 ceph_osd_op_name(src->op));
4c46459c 1024 WARN_ON(1);
a8dd0a37
AE		 1025
1026 return 0;
68b4476b 1027 }
7b25bf5f 1028
a8dd0a37 1029 dst->op = cpu_to_le16(src->op);
7b25bf5f 1030 dst->flags = cpu_to_le32(src->flags);
de2aa102 1031 dst->payload_len = cpu_to_le32(src->indata_len);
175face2 1032
bb873b53 1033 return src->indata_len;
68b4476b
YS		 1034}
1035
3499e8a5
YS		 1036/*
1037 * build new request AND message, calculate layout, and adjust file
1038 * extent as needed.
1039 *
1040 * if the file was recently truncated, we include information about its
1041 * old and new size so that the object can be updated appropriately. (we
1042 * avoid synchronously deleting truncated objects because it's slow.)
3499e8a5
YS		 1043 */
1044struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1045 struct ceph_file_layout *layout,
1046 struct ceph_vino vino,
715e4cd4
YZ		 1047 u64 off, u64 *plen,
1048 unsigned int which, int num_ops,
3499e8a5
YS		 1049 int opcode, int flags,
1050 struct ceph_snap_context *snapc,
3499e8a5
YS		 1051 u32 truncate_seq,
1052 u64 truncate_size,
153e5167 1053 bool use_mempool)
3499e8a5 1054{
68b4476b 1055 struct ceph_osd_request *req;
75d1c941
AE		 1056 u64 objnum = 0;
1057 u64 objoff = 0;
1058 u64 objlen = 0;
6816282d 1059 int r;
68b4476b 1060
ad7a60de 1061 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
864e9197
YZ		 1062 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1063 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
68b4476b 1064
acead002 1065 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
ae7ca4a3 1066 GFP_NOFS);
13d1ad16
ID		 1067 if (!req) {
1068 r = -ENOMEM;
1069 goto fail;
1070 }
79528734 1071
3499e8a5 1072 /* calculate max write size */
a19dadfb 1073 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
13d1ad16
ID		 1074 if (r)
1075 goto fail;
a19dadfb 1076
864e9197 1077 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
144cba14 1078 osd_req_op_init(req, which, opcode, 0);
864e9197 1079 } else {
7627151e 1080 u32 object_size = layout->object_size;
864e9197
YZ		 1081 u32 object_base = off - objoff;
1082 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1083 if (truncate_size <= object_base) {
1084 truncate_size = 0;
1085 } else {
1086 truncate_size -= object_base;
1087 if (truncate_size > object_size)
1088 truncate_size = object_size;
1089 }
ccca4e37 1090 }
715e4cd4 1091 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
864e9197
YZ		 1092 truncate_size, truncate_seq);
1093 }
d18d1e28 1094
7627151e 1095 req->r_base_oloc.pool = layout->pool_id;
30c156d9 1096 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
d30291b9 1097 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
22d2cfdf 1098 req->r_flags = flags | osdc->client->options->read_from_replica;
dbe0fc41 1099
bb873b53
ID		 1100 req->r_snapid = vino.snap;
1101 if (flags & CEPH_OSD_FLAG_WRITE)
1102 req->r_data_offset = off;
1103
0d9c1ab3
ID		 1104 if (num_ops > 1)
1105 /*
1106 * This is a special case for ceph_writepages_start(), but it
1107 * also covers ceph_uninline_data(). If more multi-op request
1108 * use cases emerge, we will need a separate helper.
1109 */
1110 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0);
1111 else
1112 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
13d1ad16
ID		 1113 if (r)
1114 goto fail;
1115
f24e9980 1116 return req;
13d1ad16
ID		 1117
1118fail:
1119 ceph_osdc_put_request(req);
1120 return ERR_PTR(r);
f24e9980 1121}
3d14c5d2 1122EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980 1123
a679e50f
JL		 1124int __ceph_alloc_sparse_ext_map(struct ceph_osd_req_op *op, int cnt)
1125{
1126 op->extent.sparse_ext_cnt = cnt;
1127 op->extent.sparse_ext = kmalloc_array(cnt,
1128 sizeof(*op->extent.sparse_ext),
1129 GFP_NOFS);
1130 if (!op->extent.sparse_ext)
1131 return -ENOMEM;
1132 return 0;
1133}
1134EXPORT_SYMBOL(__ceph_alloc_sparse_ext_map);
1135
f24e9980
SW		 1136/*
1137 * We keep osd requests in an rbtree, sorted by ->r_tid.
1138 */
fcd00b68 1139DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
4609245e 1140DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
f24e9980 1141
66850df5
ID		 1142/*
1143 * Call @fn on each OSD request as long as @fn returns 0.
1144 */
1145static void for_each_request(struct ceph_osd_client *osdc,
1146 int (*fn)(struct ceph_osd_request *req, void *arg),
1147 void *arg)
1148{
1149 struct rb_node *n, *p;
1150
1151 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1152 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1153
1154 for (p = rb_first(&osd->o_requests); p; ) {
1155 struct ceph_osd_request *req =
1156 rb_entry(p, struct ceph_osd_request, r_node);
1157
1158 p = rb_next(p);
1159 if (fn(req, arg))
1160 return;
1161 }
1162 }
1163
1164 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1165 struct ceph_osd_request *req =
1166 rb_entry(p, struct ceph_osd_request, r_node);
1167
1168 p = rb_next(p);
1169 if (fn(req, arg))
1170 return;
1171 }
1172}
1173
0247a0cf
ID		 1174static bool osd_homeless(struct ceph_osd *osd)
1175{
1176 return osd->o_osd == CEPH_HOMELESS_OSD;
1177}
1178
5aea3dcd 1179static bool osd_registered(struct ceph_osd *osd)
f24e9980 1180{
5aea3dcd 1181 verify_osdc_locked(osd->o_osdc);
f24e9980 1182
5aea3dcd 1183 return !RB_EMPTY_NODE(&osd->o_node);
f24e9980
SW		 1184}
1185
0247a0cf
ID		 1186/*
1187 * Assumes @osd is zero-initialized.
1188 */
1189static void osd_init(struct ceph_osd *osd)
1190{
02113a0f 1191 refcount_set(&osd->o_ref, 1);
0247a0cf 1192 RB_CLEAR_NODE(&osd->o_node);
08b8a044 1193 spin_lock_init(&osd->o_requests_lock);
5aea3dcd 1194 osd->o_requests = RB_ROOT;
922dab61 1195 osd->o_linger_requests = RB_ROOT;
a02a946d
ID		 1196 osd->o_backoff_mappings = RB_ROOT;
1197 osd->o_backoffs_by_id = RB_ROOT;
0247a0cf
ID		 1198 INIT_LIST_HEAD(&osd->o_osd_lru);
1199 INIT_LIST_HEAD(&osd->o_keepalive_item);
1200 osd->o_incarnation = 1;
5aea3dcd 1201 mutex_init(&osd->lock);
0247a0cf
ID		 1202}
1203
1204static void osd_cleanup(struct ceph_osd *osd)
1205{
1206 WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
5aea3dcd 1207 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
922dab61 1208 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
a02a946d
ID		 1209 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1210 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
0247a0cf
ID		 1211 WARN_ON(!list_empty(&osd->o_osd_lru));
1212 WARN_ON(!list_empty(&osd->o_keepalive_item));
1213
1214 if (osd->o_auth.authorizer) {
1215 WARN_ON(osd_homeless(osd));
1216 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1217 }
1218}
1219
f24e9980
SW		 1220/*
1221 * Track open sessions with osds.
1222 */
e10006f8 1223static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 1224{
1225 struct ceph_osd *osd;
1226
0247a0cf
ID		 1227 WARN_ON(onum == CEPH_HOMELESS_OSD);
1228
7a28f59b 1229 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
0247a0cf 1230 osd_init(osd);
f24e9980 1231 osd->o_osdc = osdc;
e10006f8 1232 osd->o_osd = onum;
f24e9980 1233
b7a9e5dd 1234 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 1235
f24e9980
SW		 1236 return osd;
1237}
1238
1239static struct ceph_osd *get_osd(struct ceph_osd *osd)
1240{
02113a0f
ER		 1241 if (refcount_inc_not_zero(&osd->o_ref)) {
1242 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1243 refcount_read(&osd->o_ref));
f24e9980
SW		 1244 return osd;
1245 } else {
1246 dout("get_osd %p FAIL\n", osd);
1247 return NULL;
1248 }
1249}
1250
1251static void put_osd(struct ceph_osd *osd)
1252{
02113a0f
ER		 1253 dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1254 refcount_read(&osd->o_ref) - 1);
1255 if (refcount_dec_and_test(&osd->o_ref)) {
0247a0cf 1256 osd_cleanup(osd);
f24e9980 1257 kfree(osd);
79494d1b 1258 }
f24e9980
SW		 1259}
1260
fcd00b68
ID		 1261DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1262
9dd2845c 1263static void __move_osd_to_lru(struct ceph_osd *osd)
f5a2041b 1264{
9dd2845c
ID		 1265 struct ceph_osd_client *osdc = osd->o_osdc;
1266
1267 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
f5a2041b 1268 BUG_ON(!list_empty(&osd->o_osd_lru));
bbf37ec3 1269
9dd2845c 1270 spin_lock(&osdc->osd_lru_lock);
f5a2041b 1271 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
9dd2845c
ID		 1272 spin_unlock(&osdc->osd_lru_lock);
1273
a319bf56 1274 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
f5a2041b
YS		 1275}
1276
9dd2845c 1277static void maybe_move_osd_to_lru(struct ceph_osd *osd)
bbf37ec3 1278{
5aea3dcd 1279 if (RB_EMPTY_ROOT(&osd->o_requests) &&
922dab61 1280 RB_EMPTY_ROOT(&osd->o_linger_requests))
9dd2845c 1281 __move_osd_to_lru(osd);
bbf37ec3
ID		 1282}
1283
f5a2041b
YS		 1284static void __remove_osd_from_lru(struct ceph_osd *osd)
1285{
9dd2845c
ID		 1286 struct ceph_osd_client *osdc = osd->o_osdc;
1287
1288 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1289
1290 spin_lock(&osdc->osd_lru_lock);
f5a2041b
YS		 1291 if (!list_empty(&osd->o_osd_lru))
1292 list_del_init(&osd->o_osd_lru);
9dd2845c 1293 spin_unlock(&osdc->osd_lru_lock);
f5a2041b
YS		 1294}
1295
5aea3dcd
ID		 1296/*
1297 * Close the connection and assign any leftover requests to the
1298 * homeless session.
1299 */
1300static void close_osd(struct ceph_osd *osd)
1301{
1302 struct ceph_osd_client *osdc = osd->o_osdc;
1303 struct rb_node *n;
1304
1305 verify_osdc_wrlocked(osdc);
1306 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1307
1308 ceph_con_close(&osd->o_con);
1309
1310 for (n = rb_first(&osd->o_requests); n; ) {
1311 struct ceph_osd_request *req =
1312 rb_entry(n, struct ceph_osd_request, r_node);
1313
1314 n = rb_next(n); /* unlink_request() */
1315
1316 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1317 unlink_request(osd, req);
1318 link_request(&osdc->homeless_osd, req);
1319 }
922dab61
ID		 1320 for (n = rb_first(&osd->o_linger_requests); n; ) {
1321 struct ceph_osd_linger_request *lreq =
1322 rb_entry(n, struct ceph_osd_linger_request, node);
1323
1324 n = rb_next(n); /* unlink_linger() */
1325
1326 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1327 lreq->linger_id);
1328 unlink_linger(osd, lreq);
1329 link_linger(&osdc->homeless_osd, lreq);
1330 }
a02a946d 1331 clear_backoffs(osd);
5aea3dcd
ID		 1332
1333 __remove_osd_from_lru(osd);
1334 erase_osd(&osdc->osds, osd);
1335 put_osd(osd);
1336}
1337
f24e9980
SW		 1338/*
1339 * reset osd connect
1340 */
5aea3dcd 1341static int reopen_osd(struct ceph_osd *osd)
f24e9980 1342{
c3acb181 1343 struct ceph_entity_addr *peer_addr;
f24e9980 1344
5aea3dcd
ID		 1345 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1346
1347 if (RB_EMPTY_ROOT(&osd->o_requests) &&
922dab61 1348 RB_EMPTY_ROOT(&osd->o_linger_requests)) {
5aea3dcd 1349 close_osd(osd);
c3acb181
AE		 1350 return -ENODEV;
1351 }
1352
5aea3dcd 1353 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
c3acb181
AE		 1354 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1355 !ceph_con_opened(&osd->o_con)) {
5aea3dcd 1356 struct rb_node *n;
c3acb181 1357
0b4af2e8
ID		 1358 dout("osd addr hasn't changed and connection never opened, "
1359 "letting msgr retry\n");
87b315a5 1360 /* touch each r_stamp for handle_timeout()'s benfit */
5aea3dcd
ID		 1361 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1362 struct ceph_osd_request *req =
1363 rb_entry(n, struct ceph_osd_request, r_node);
87b315a5 1364 req->r_stamp = jiffies;
5aea3dcd 1365 }
c3acb181
AE		 1366
1367 return -EAGAIN;
f24e9980 1368 }
c3acb181
AE		 1369
1370 ceph_con_close(&osd->o_con);
1371 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1372 osd->o_incarnation++;
1373
1374 return 0;
f24e9980
SW		 1375}
1376
5aea3dcd
ID		 1377static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1378 bool wrlocked)
f24e9980 1379{
5aea3dcd 1380 struct ceph_osd *osd;
35f9f8a0 1381
5aea3dcd
ID		 1382 if (wrlocked)
1383 verify_osdc_wrlocked(osdc);
1384 else
1385 verify_osdc_locked(osdc);
f24e9980 1386
5aea3dcd
ID		 1387 if (o != CEPH_HOMELESS_OSD)
1388 osd = lookup_osd(&osdc->osds, o);
1389 else
1390 osd = &osdc->homeless_osd;
1391 if (!osd) {
1392 if (!wrlocked)
1393 return ERR_PTR(-EAGAIN);
0ba6478d 1394
5aea3dcd
ID		 1395 osd = create_osd(osdc, o);
1396 insert_osd(&osdc->osds, osd);
1397 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1398 &osdc->osdmap->osd_addr[osd->o_osd]);
0ba6478d 1399 }
f24e9980 1400
5aea3dcd
ID		 1401 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1402 return osd;
f24e9980
SW		 1403}
1404
1405/*
5aea3dcd
ID		 1406 * Create request <-> OSD session relation.
1407 *
1408 * @req has to be assigned a tid, @osd may be homeless.
f24e9980 1409 */
5aea3dcd 1410static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
f24e9980 1411{
5aea3dcd
ID		 1412 verify_osd_locked(osd);
1413 WARN_ON(!req->r_tid || req->r_osd);
1414 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1415 req, req->r_tid);
1416
1417 if (!osd_homeless(osd))
1418 __remove_osd_from_lru(osd);
1419 else
1420 atomic_inc(&osd->o_osdc->num_homeless);
1421
1422 get_osd(osd);
08b8a044 1423 spin_lock(&osd->o_requests_lock);
5aea3dcd 1424 insert_request(&osd->o_requests, req);
08b8a044 1425 spin_unlock(&osd->o_requests_lock);
5aea3dcd 1426 req->r_osd = osd;
f24e9980
SW		 1427}
1428
5aea3dcd
ID		 1429static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1430{
1431 verify_osd_locked(osd);
1432 WARN_ON(req->r_osd != osd);
1433 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1434 req, req->r_tid);
1435
1436 req->r_osd = NULL;
08b8a044 1437 spin_lock(&osd->o_requests_lock);
5aea3dcd 1438 erase_request(&osd->o_requests, req);
08b8a044 1439 spin_unlock(&osd->o_requests_lock);
5aea3dcd
ID		 1440 put_osd(osd);
1441
1442 if (!osd_homeless(osd))
1443 maybe_move_osd_to_lru(osd);
1444 else
1445 atomic_dec(&osd->o_osdc->num_homeless);
1446}
1447
63244fa1
ID		 1448static bool __pool_full(struct ceph_pg_pool_info *pi)
1449{
1450 return pi->flags & CEPH_POOL_FLAG_FULL;
1451}
1452
42c1b124
ID		 1453static bool have_pool_full(struct ceph_osd_client *osdc)
1454{
1455 struct rb_node *n;
1456
1457 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1458 struct ceph_pg_pool_info *pi =
1459 rb_entry(n, struct ceph_pg_pool_info, node);
1460
1461 if (__pool_full(pi))
1462 return true;
1463 }
1464
1465 return false;
1466}
1467
5aea3dcd
ID		 1468static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1469{
1470 struct ceph_pg_pool_info *pi;
1471
1472 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1473 if (!pi)
1474 return false;
1475
1476 return __pool_full(pi);
1477}
1478
d29adb34
JD		 1479/*
1480 * Returns whether a request should be blocked from being sent
1481 * based on the current osdmap and osd_client settings.
d29adb34 1482 */
63244fa1
ID		 1483static bool target_should_be_paused(struct ceph_osd_client *osdc,
1484 const struct ceph_osd_request_target *t,
1485 struct ceph_pg_pool_info *pi)
1486{
b7ec35b3
ID		 1487 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1488 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1489 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
63244fa1
ID		 1490 __pool_full(pi);
1491
6d637a54 1492 WARN_ON(pi->id != t->target_oloc.pool);
58eb7932
JL		 1493 return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1494 ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1495 (osdc->osdmap->epoch < osdc->epoch_barrier);
63244fa1
ID		 1496}
1497
117d96a0
ID		 1498static int pick_random_replica(const struct ceph_osds *acting)
1499{
8032bf12 1500 int i = get_random_u32_below(acting->size);
117d96a0
ID		 1501
1502 dout("%s picked osd%d, primary osd%d\n", __func__,
1503 acting->osds[i], acting->primary);
1504 return i;
1505}
1506
1507/*
1508 * Picks the closest replica based on client's location given by
1509 * crush_location option. Prefers the primary if the locality is
1510 * the same.
1511 */
1512static int pick_closest_replica(struct ceph_osd_client *osdc,
1513 const struct ceph_osds *acting)
1514{
1515 struct ceph_options *opt = osdc->client->options;
1516 int best_i, best_locality;
1517 int i = 0, locality;
1518
1519 do {
1520 locality = ceph_get_crush_locality(osdc->osdmap,
1521 acting->osds[i],
1522 &opt->crush_locs);
1523 if (i == 0 ||
1524 (locality >= 0 && best_locality < 0) ||
1525 (locality >= 0 && best_locality >= 0 &&
1526 locality < best_locality)) {
1527 best_i = i;
1528 best_locality = locality;
1529 }
1530 } while (++i < acting->size);
1531
1532 dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1533 acting->osds[best_i], best_locality, acting->primary);
1534 return best_i;
1535}
1536
63244fa1
ID		 1537enum calc_target_result {
1538 CALC_TARGET_NO_ACTION = 0,
1539 CALC_TARGET_NEED_RESEND,
1540 CALC_TARGET_POOL_DNE,
1541};
1542
1543static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1544 struct ceph_osd_request_target *t,
63244fa1
ID		 1545 bool any_change)
1546{
1547 struct ceph_pg_pool_info *pi;
1548 struct ceph_pg pgid, last_pgid;
1549 struct ceph_osds up, acting;
117d96a0
ID		 1550 bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1551 bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
63244fa1 1552 bool force_resend = false;
84ed45df 1553 bool unpaused = false;
a5613724 1554 bool legacy_change = false;
7de030d6 1555 bool split = false;
b7ec35b3 1556 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
ae78dd81
ID		 1557 bool recovery_deletes = ceph_osdmap_flag(osdc,
1558 CEPH_OSDMAP_RECOVERY_DELETES);
63244fa1 1559 enum calc_target_result ct_res;
63244fa1 1560
04c7d789 1561 t->epoch = osdc->osdmap->epoch;
63244fa1
ID		 1562 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1563 if (!pi) {
1564 t->osd = CEPH_HOMELESS_OSD;
1565 ct_res = CALC_TARGET_POOL_DNE;
1566 goto out;
1567 }
1568
1569 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
dc93e0e2
ID		 1570 if (t->last_force_resend < pi->last_force_request_resend) {
1571 t->last_force_resend = pi->last_force_request_resend;
63244fa1 1572 force_resend = true;
dc93e0e2 1573 } else if (t->last_force_resend == 0) {
63244fa1
ID		 1574 force_resend = true;
1575 }
1576 }
63244fa1 1577
db098ec4
ID		 1578 /* apply tiering */
1579 ceph_oid_copy(&t->target_oid, &t->base_oid);
1580 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1581 if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
117d96a0 1582 if (is_read && pi->read_tier >= 0)
63244fa1 1583 t->target_oloc.pool = pi->read_tier;
117d96a0 1584 if (is_write && pi->write_tier >= 0)
63244fa1 1585 t->target_oloc.pool = pi->write_tier;
6d637a54
ID		 1586
1587 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1588 if (!pi) {
1589 t->osd = CEPH_HOMELESS_OSD;
1590 ct_res = CALC_TARGET_POOL_DNE;
1591 goto out;
1592 }
63244fa1
ID		 1593 }
1594
a86f009f 1595 __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
63244fa1
ID		 1596 last_pgid.pool = pgid.pool;
1597 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1598
df28152d 1599 ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
63244fa1
ID		 1600 if (any_change &&
1601 ceph_is_new_interval(&t->acting,
1602 &acting,
1603 &t->up,
1604 &up,
1605 t->size,
1606 pi->size,
1607 t->min_size,
1608 pi->min_size,
1609 t->pg_num,
1610 pi->pg_num,
1611 t->sort_bitwise,
1612 sort_bitwise,
ae78dd81
ID		 1613 t->recovery_deletes,
1614 recovery_deletes,
63244fa1
ID		 1615 &last_pgid))
1616 force_resend = true;
1617
1618 if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1619 t->paused = false;
84ed45df 1620 unpaused = true;
63244fa1 1621 }
84ed45df 1622 legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
117d96a0
ID		 1623 ceph_osds_changed(&t->acting, &acting,
1624 t->used_replica || any_change);
7de030d6
ID		 1625 if (t->pg_num)
1626 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
63244fa1 1627
7de030d6 1628 if (legacy_change || force_resend || split) {
63244fa1 1629 t->pgid = pgid; /* struct */
df28152d 1630 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
63244fa1
ID		 1631 ceph_osds_copy(&t->acting, &acting);
1632 ceph_osds_copy(&t->up, &up);
1633 t->size = pi->size;
1634 t->min_size = pi->min_size;
1635 t->pg_num = pi->pg_num;
1636 t->pg_num_mask = pi->pg_num_mask;
1637 t->sort_bitwise = sort_bitwise;
ae78dd81 1638 t->recovery_deletes = recovery_deletes;
63244fa1 1639
117d96a0
ID		 1640 if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1641 CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1642 !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1643 acting.size > 1) {
1644 int pos;
1645
1646 WARN_ON(!is_read || acting.osds[0] != acting.primary);
1647 if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1648 pos = pick_random_replica(&acting);
1649 } else {
1650 pos = pick_closest_replica(osdc, &acting);
1651 }
1652 t->osd = acting.osds[pos];
1653 t->used_replica = pos > 0;
1654 } else {
1655 t->osd = acting.primary;
1656 t->used_replica = false;
1657 }
63244fa1
ID		 1658 }
1659
a5613724 1660 if (unpaused || legacy_change || force_resend || split)
84ed45df
ID		 1661 ct_res = CALC_TARGET_NEED_RESEND;
1662 else
1663 ct_res = CALC_TARGET_NO_ACTION;
1664
63244fa1 1665out:
a5613724
ID		 1666 dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1667 legacy_change, force_resend, split, ct_res, t->osd);
63244fa1
ID		 1668 return ct_res;
1669}
1670
a02a946d
ID		 1671static struct ceph_spg_mapping *alloc_spg_mapping(void)
1672{
1673 struct ceph_spg_mapping *spg;
1674
1675 spg = kmalloc(sizeof(*spg), GFP_NOIO);
1676 if (!spg)
1677 return NULL;
1678
1679 RB_CLEAR_NODE(&spg->node);
1680 spg->backoffs = RB_ROOT;
1681 return spg;
1682}
1683
1684static void free_spg_mapping(struct ceph_spg_mapping *spg)
1685{
1686 WARN_ON(!RB_EMPTY_NODE(&spg->node));
1687 WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1688
1689 kfree(spg);
1690}
1691
1692/*
1693 * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1694 * ceph_pg_mapping. Used to track OSD backoffs -- a backoff [range] is
1695 * defined only within a specific spgid; it does not pass anything to
1696 * children on split, or to another primary.
1697 */
1698DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1699 RB_BYPTR, const struct ceph_spg *, node)
1700
1701static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1702{
1703 return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1704}
1705
1706static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1707 void **pkey, size_t *pkey_len)
1708{
1709 if (hoid->key_len) {
1710 *pkey = hoid->key;
1711 *pkey_len = hoid->key_len;
1712 } else {
1713 *pkey = hoid->oid;
1714 *pkey_len = hoid->oid_len;
1715 }
1716}
1717
1718static int compare_names(const void *name1, size_t name1_len,
1719 const void *name2, size_t name2_len)
1720{
1721 int ret;
1722
1723 ret = memcmp(name1, name2, min(name1_len, name2_len));
1724 if (!ret) {
1725 if (name1_len < name2_len)
1726 ret = -1;
1727 else if (name1_len > name2_len)
1728 ret = 1;
1729 }
1730 return ret;
1731}
1732
1733static int hoid_compare(const struct ceph_hobject_id *lhs,
1734 const struct ceph_hobject_id *rhs)
1735{
1736 void *effective_key1, *effective_key2;
1737 size_t effective_key1_len, effective_key2_len;
1738 int ret;
1739
1740 if (lhs->is_max < rhs->is_max)
1741 return -1;
1742 if (lhs->is_max > rhs->is_max)
1743 return 1;
1744
1745 if (lhs->pool < rhs->pool)
1746 return -1;
1747 if (lhs->pool > rhs->pool)
1748 return 1;
1749
1750 if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1751 return -1;
1752 if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1753 return 1;
1754
1755 ret = compare_names(lhs->nspace, lhs->nspace_len,
1756 rhs->nspace, rhs->nspace_len);
1757 if (ret)
1758 return ret;
1759
1760 hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1761 hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1762 ret = compare_names(effective_key1, effective_key1_len,
1763 effective_key2, effective_key2_len);
1764 if (ret)
1765 return ret;
1766
1767 ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1768 if (ret)
1769 return ret;
1770
1771 if (lhs->snapid < rhs->snapid)
1772 return -1;
1773 if (lhs->snapid > rhs->snapid)
1774 return 1;
1775
1776 return 0;
1777}
1778
1779/*
1780 * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1781 * compat stuff here.
1782 *
1783 * Assumes @hoid is zero-initialized.
1784 */
1785static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1786{
1787 u8 struct_v;
1788 u32 struct_len;
1789 int ret;
1790
1791 ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1792 &struct_len);
1793 if (ret)
1794 return ret;
1795
1796 if (struct_v < 4) {
1797 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1798 goto e_inval;
1799 }
1800
1801 hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1802 GFP_NOIO);
1803 if (IS_ERR(hoid->key)) {
1804 ret = PTR_ERR(hoid->key);
1805 hoid->key = NULL;
1806 return ret;
1807 }
1808
1809 hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1810 GFP_NOIO);
1811 if (IS_ERR(hoid->oid)) {
1812 ret = PTR_ERR(hoid->oid);
1813 hoid->oid = NULL;
1814 return ret;
1815 }
1816
1817 ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1818 ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1819 ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1820
1821 hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1822 GFP_NOIO);
1823 if (IS_ERR(hoid->nspace)) {
1824 ret = PTR_ERR(hoid->nspace);
1825 hoid->nspace = NULL;
1826 return ret;
1827 }
1828
1829 ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1830
1831 ceph_hoid_build_hash_cache(hoid);
1832 return 0;
1833
1834e_inval:
1835 return -EINVAL;
1836}
1837
1838static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1839{
1840 return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1841 4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1842}
1843
1844static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1845{
1846 ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1847 ceph_encode_string(p, end, hoid->key, hoid->key_len);
1848 ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1849 ceph_encode_64(p, hoid->snapid);
1850 ceph_encode_32(p, hoid->hash);
1851 ceph_encode_8(p, hoid->is_max);
1852 ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1853 ceph_encode_64(p, hoid->pool);
1854}
1855
1856static void free_hoid(struct ceph_hobject_id *hoid)
1857{
1858 if (hoid) {
1859 kfree(hoid->key);
1860 kfree(hoid->oid);
1861 kfree(hoid->nspace);
1862 kfree(hoid);
1863 }
1864}
1865
1866static struct ceph_osd_backoff *alloc_backoff(void)
1867{
1868 struct ceph_osd_backoff *backoff;
1869
1870 backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1871 if (!backoff)
1872 return NULL;
1873
1874 RB_CLEAR_NODE(&backoff->spg_node);
1875 RB_CLEAR_NODE(&backoff->id_node);
1876 return backoff;
1877}
1878
1879static void free_backoff(struct ceph_osd_backoff *backoff)
1880{
1881 WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1882 WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1883
1884 free_hoid(backoff->begin);
1885 free_hoid(backoff->end);
1886 kfree(backoff);
1887}
1888
1889/*
1890 * Within a specific spgid, backoffs are managed by ->begin hoid.
1891 */
1892DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1893 RB_BYVAL, spg_node);
1894
1895static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1896 const struct ceph_hobject_id *hoid)
1897{
1898 struct rb_node *n = root->rb_node;
1899
1900 while (n) {
1901 struct ceph_osd_backoff *cur =
1902 rb_entry(n, struct ceph_osd_backoff, spg_node);
1903 int cmp;
1904
1905 cmp = hoid_compare(hoid, cur->begin);
1906 if (cmp < 0) {
1907 n = n->rb_left;
1908 } else if (cmp > 0) {
1909 if (hoid_compare(hoid, cur->end) < 0)
1910 return cur;
1911
1912 n = n->rb_right;
1913 } else {
1914 return cur;
1915 }
1916 }
1917
1918 return NULL;
1919}
1920
1921/*
1922 * Each backoff has a unique id within its OSD session.
1923 */
1924DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1925
1926static void clear_backoffs(struct ceph_osd *osd)
1927{
1928 while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1929 struct ceph_spg_mapping *spg =
1930 rb_entry(rb_first(&osd->o_backoff_mappings),
1931 struct ceph_spg_mapping, node);
1932
1933 while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1934 struct ceph_osd_backoff *backoff =
1935 rb_entry(rb_first(&spg->backoffs),
1936 struct ceph_osd_backoff, spg_node);
1937
1938 erase_backoff(&spg->backoffs, backoff);
1939 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1940 free_backoff(backoff);
1941 }
1942 erase_spg_mapping(&osd->o_backoff_mappings, spg);
1943 free_spg_mapping(spg);
1944 }
1945}
1946
1947/*
1948 * Set up a temporary, non-owning view into @t.
1949 */
1950static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1951 const struct ceph_osd_request_target *t)
1952{
1953 hoid->key = NULL;
1954 hoid->key_len = 0;
1955 hoid->oid = t->target_oid.name;
1956 hoid->oid_len = t->target_oid.name_len;
1957 hoid->snapid = CEPH_NOSNAP;
1958 hoid->hash = t->pgid.seed;
1959 hoid->is_max = false;
1960 if (t->target_oloc.pool_ns) {
1961 hoid->nspace = t->target_oloc.pool_ns->str;
1962 hoid->nspace_len = t->target_oloc.pool_ns->len;
1963 } else {
1964 hoid->nspace = NULL;
1965 hoid->nspace_len = 0;
1966 }
1967 hoid->pool = t->target_oloc.pool;
1968 ceph_hoid_build_hash_cache(hoid);
1969}
1970
1971static bool should_plug_request(struct ceph_osd_request *req)
1972{
1973 struct ceph_osd *osd = req->r_osd;
1974 struct ceph_spg_mapping *spg;
1975 struct ceph_osd_backoff *backoff;
1976 struct ceph_hobject_id hoid;
1977
1978 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
1979 if (!spg)
1980 return false;
1981
1982 hoid_fill_from_target(&hoid, &req->r_t);
1983 backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
1984 if (!backoff)
1985 return false;
1986
1987 dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
1988 __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
1989 backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
1990 return true;
1991}
1992
0d9c1ab3
ID		 1993/*
1994 * Keep get_num_data_items() in sync with this function.
1995 */
98c4bfe9 1996static void setup_request_data(struct ceph_osd_request *req)
f24e9980 1997{
98c4bfe9
ID		 1998 struct ceph_msg *request_msg = req->r_request;
1999 struct ceph_msg *reply_msg = req->r_reply;
2000 struct ceph_osd_req_op *op;
bb873b53 2001
98c4bfe9 2002 if (req->r_request->num_data_items || req->r_reply->num_data_items)
bb873b53 2003 return;
f24e9980 2004
98c4bfe9
ID		 2005 WARN_ON(request_msg->data_length || reply_msg->data_length);
2006 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
bb873b53
ID		 2007 switch (op->op) {
2008 /* request */
2009 case CEPH_OSD_OP_WRITE:
2010 case CEPH_OSD_OP_WRITEFULL:
2011 WARN_ON(op->indata_len != op->extent.length);
98c4bfe9
ID		 2012 ceph_osdc_msg_data_add(request_msg,
2013 &op->extent.osd_data);
bb873b53
ID		 2014 break;
2015 case CEPH_OSD_OP_SETXATTR:
2016 case CEPH_OSD_OP_CMPXATTR:
2017 WARN_ON(op->indata_len != op->xattr.name_len +
2018 op->xattr.value_len);
98c4bfe9
ID		 2019 ceph_osdc_msg_data_add(request_msg,
2020 &op->xattr.osd_data);
bb873b53 2021 break;
922dab61 2022 case CEPH_OSD_OP_NOTIFY_ACK:
98c4bfe9 2023 ceph_osdc_msg_data_add(request_msg,
922dab61
ID		 2024 &op->notify_ack.request_data);
2025 break;
78beb0ff 2026 case CEPH_OSD_OP_COPY_FROM2:
23ddf9be
LH		 2027 ceph_osdc_msg_data_add(request_msg,
2028 &op->copy_from.osd_data);
2029 break;
bb873b53
ID		 2030
2031 /* reply */
2032 case CEPH_OSD_OP_STAT:
98c4bfe9 2033 ceph_osdc_msg_data_add(reply_msg,
bb873b53
ID		 2034 &op->raw_data_in);
2035 break;
2036 case CEPH_OSD_OP_READ:
98c4bfe9 2037 ceph_osdc_msg_data_add(reply_msg,
bb873b53
ID		 2038 &op->extent.osd_data);
2039 break;
a4ed38d7 2040 case CEPH_OSD_OP_LIST_WATCHERS:
98c4bfe9 2041 ceph_osdc_msg_data_add(reply_msg,
a4ed38d7
DF		 2042 &op->list_watchers.response_data);
2043 break;
bb873b53
ID		 2044
2045 /* both */
2046 case CEPH_OSD_OP_CALL:
2047 WARN_ON(op->indata_len != op->cls.class_len +
2048 op->cls.method_len +
2049 op->cls.indata_len);
98c4bfe9
ID		 2050 ceph_osdc_msg_data_add(request_msg,
2051 &op->cls.request_info);
bb873b53 2052 /* optional, can be NONE */
98c4bfe9
ID		 2053 ceph_osdc_msg_data_add(request_msg,
2054 &op->cls.request_data);
bb873b53 2055 /* optional, can be NONE */
98c4bfe9 2056 ceph_osdc_msg_data_add(reply_msg,
bb873b53
ID		 2057 &op->cls.response_data);
2058 break;
19079203 2059 case CEPH_OSD_OP_NOTIFY:
98c4bfe9 2060 ceph_osdc_msg_data_add(request_msg,
19079203 2061 &op->notify.request_data);
98c4bfe9 2062 ceph_osdc_msg_data_add(reply_msg,
19079203
ID		 2063 &op->notify.response_data);
2064 break;
bb873b53 2065 }
bb873b53 2066 }
bb873b53
ID		 2067}
2068
2e59ffd1
ID		 2069static void encode_pgid(void **p, const struct ceph_pg *pgid)
2070{
2071 ceph_encode_8(p, 1);
2072 ceph_encode_64(p, pgid->pool);
2073 ceph_encode_32(p, pgid->seed);
2074 ceph_encode_32(p, -1); /* preferred */
2075}
2076
8cb441c0
ID		 2077static void encode_spgid(void **p, const struct ceph_spg *spgid)
2078{
2079 ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2080 encode_pgid(p, &spgid->pgid);
2081 ceph_encode_8(p, spgid->shard);
2082}
2083
2e59ffd1
ID		 2084static void encode_oloc(void **p, void *end,
2085 const struct ceph_object_locator *oloc)
2086{
2087 ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2088 ceph_encode_64(p, oloc->pool);
2089 ceph_encode_32(p, -1); /* preferred */
2090 ceph_encode_32(p, 0); /* key len */
2091 if (oloc->pool_ns)
2092 ceph_encode_string(p, end, oloc->pool_ns->str,
2093 oloc->pool_ns->len);
2094 else
2095 ceph_encode_32(p, 0);
2096}
2097
8cb441c0
ID		 2098static void encode_request_partial(struct ceph_osd_request *req,
2099 struct ceph_msg *msg)
bb873b53
ID		 2100{
2101 void *p = msg->front.iov_base;
2102 void *const end = p + msg->front_alloc_len;
2103 u32 data_len = 0;
2104 int i;
2105
2106 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2107 /* snapshots aren't writeable */
2108 WARN_ON(req->r_snapid != CEPH_NOSNAP);
2109 } else {
2110 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2111 req->r_data_offset || req->r_snapc);
2112 }
2113
98c4bfe9 2114 setup_request_data(req);
bb873b53 2115
8cb441c0
ID		 2116 encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2117 ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
bb873b53
ID		 2118 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2119 ceph_encode_32(&p, req->r_flags);
8cb441c0
ID		 2120
2121 /* reqid */
2122 ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2123 memset(p, 0, sizeof(struct ceph_osd_reqid));
2124 p += sizeof(struct ceph_osd_reqid);
2125
2126 /* trace */
2127 memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2128 p += sizeof(struct ceph_blkin_trace_info);
2129
2130 ceph_encode_32(&p, 0); /* client_inc, always 0 */
fac02ddf 2131 ceph_encode_timespec64(p, &req->r_mtime);
bb873b53 2132 p += sizeof(struct ceph_timespec);
aa26d662 2133
2e59ffd1 2134 encode_oloc(&p, end, &req->r_t.target_oloc);
2e59ffd1
ID		 2135 ceph_encode_string(&p, end, req->r_t.target_oid.name,
2136 req->r_t.target_oid.name_len);
f24e9980 2137
bb873b53
ID		 2138 /* ops, can imply data */
2139 ceph_encode_16(&p, req->r_num_ops);
2140 for (i = 0; i < req->r_num_ops; i++) {
2141 data_len += osd_req_encode_op(p, &req->r_ops[i]);
2142 p += sizeof(struct ceph_osd_op);
2143 }
26be8808 2144
bb873b53
ID		 2145 ceph_encode_64(&p, req->r_snapid); /* snapid */
2146 if (req->r_snapc) {
2147 ceph_encode_64(&p, req->r_snapc->seq);
2148 ceph_encode_32(&p, req->r_snapc->num_snaps);
2149 for (i = 0; i < req->r_snapc->num_snaps; i++)
2150 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2151 } else {
2152 ceph_encode_64(&p, 0); /* snap_seq */
2153 ceph_encode_32(&p, 0); /* snaps len */
2154 }
2155
2156 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
986e8989 2157 BUG_ON(p > end - 8); /* space for features */
bb873b53 2158
8cb441c0
ID		 2159 msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2160 /* front_len is finalized in encode_request_finish() */
986e8989
ID		 2161 msg->front.iov_len = p - msg->front.iov_base;
2162 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
bb873b53
ID		 2163 msg->hdr.data_len = cpu_to_le32(data_len);
2164 /*
2165 * The header "data_off" is a hint to the receiver allowing it
2166 * to align received data into its buffers such that there's no
2167 * need to re-copy it before writing it to disk (direct I/O).
2168 */
2169 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
26be8808 2170
8cb441c0
ID		 2171 dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2172 req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2173}
2174
2175static void encode_request_finish(struct ceph_msg *msg)
2176{
2177 void *p = msg->front.iov_base;
986e8989 2178 void *const partial_end = p + msg->front.iov_len;
8cb441c0
ID		 2179 void *const end = p + msg->front_alloc_len;
2180
2181 if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2182 /* luminous OSD -- encode features and be done */
986e8989 2183 p = partial_end;
8cb441c0
ID		 2184 ceph_encode_64(&p, msg->con->peer_features);
2185 } else {
2186 struct {
2187 char spgid[CEPH_ENCODING_START_BLK_LEN +
2188 CEPH_PGID_ENCODING_LEN + 1];
2189 __le32 hash;
2190 __le32 epoch;
2191 __le32 flags;
2192 char reqid[CEPH_ENCODING_START_BLK_LEN +
2193 sizeof(struct ceph_osd_reqid)];
2194 char trace[sizeof(struct ceph_blkin_trace_info)];
2195 __le32 client_inc;
2196 struct ceph_timespec mtime;
2197 } __packed head;
2198 struct ceph_pg pgid;
2199 void *oloc, *oid, *tail;
2200 int oloc_len, oid_len, tail_len;
2201 int len;
2202
2203 /*
2204 * Pre-luminous OSD -- reencode v8 into v4 using @head
2205 * as a temporary buffer. Encode the raw PG; the rest
2206 * is just a matter of moving oloc, oid and tail blobs
2207 * around.
2208 */
2209 memcpy(&head, p, sizeof(head));
2210 p += sizeof(head);
2211
2212 oloc = p;
2213 p += CEPH_ENCODING_START_BLK_LEN;
2214 pgid.pool = ceph_decode_64(&p);
2215 p += 4 + 4; /* preferred, key len */
2216 len = ceph_decode_32(&p);
2217 p += len; /* nspace */
2218 oloc_len = p - oloc;
2219
2220 oid = p;
2221 len = ceph_decode_32(&p);
2222 p += len;
2223 oid_len = p - oid;
2224
2225 tail = p;
986e8989 2226 tail_len = partial_end - p;
8cb441c0
ID		 2227
2228 p = msg->front.iov_base;
2229 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2230 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2231 ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2232 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2233
2234 /* reassert_version */
2235 memset(p, 0, sizeof(struct ceph_eversion));
2236 p += sizeof(struct ceph_eversion);
2237
2238 BUG_ON(p >= oloc);
2239 memmove(p, oloc, oloc_len);
2240 p += oloc_len;
2241
2242 pgid.seed = le32_to_cpu(head.hash);
2243 encode_pgid(&p, &pgid); /* raw pg */
2244
2245 BUG_ON(p >= oid);
2246 memmove(p, oid, oid_len);
2247 p += oid_len;
2248
2249 /* tail -- ops, snapid, snapc, retry_attempt */
2250 BUG_ON(p >= tail);
2251 memmove(p, tail, tail_len);
2252 p += tail_len;
2253
2254 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2255 }
2256
2257 BUG_ON(p > end);
2258 msg->front.iov_len = p - msg->front.iov_base;
2259 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2260
2261 dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2262 le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2263 le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2264 le16_to_cpu(msg->hdr.version));
bb873b53
ID		 2265}
2266
2267/*
2268 * @req has to be assigned a tid and registered.
2269 */
2270static void send_request(struct ceph_osd_request *req)
2271{
2272 struct ceph_osd *osd = req->r_osd;
2273
5aea3dcd 2274 verify_osd_locked(osd);
bb873b53
ID		 2275 WARN_ON(osd->o_osd != req->r_t.osd);
2276
a02a946d
ID		 2277 /* backoff? */
2278 if (should_plug_request(req))
2279 return;
2280
5aea3dcd
ID		 2281 /*
2282 * We may have a previously queued request message hanging
2283 * around. Cancel it to avoid corrupting the msgr.
2284 */
2285 if (req->r_sent)
2286 ceph_msg_revoke(req->r_request);
2287
bb873b53
ID		 2288 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2289 if (req->r_attempts)
2290 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2291 else
2292 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2293
8cb441c0 2294 encode_request_partial(req, req->r_request);
bb873b53 2295
04c7d789 2296 dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
bb873b53 2297 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
dc98ff72 2298 req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
04c7d789
ID		 2299 req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2300 req->r_attempts);
bb873b53
ID		 2301
2302 req->r_t.paused = false;
2303 req->r_stamp = jiffies;
2304 req->r_attempts++;
2305
2306 req->r_sent = osd->o_incarnation;
2307 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2308 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
f24e9980
SW		 2309}
2310
42c1b124
ID		 2311static void maybe_request_map(struct ceph_osd_client *osdc)
2312{
2313 bool continuous = false;
2314
5aea3dcd 2315 verify_osdc_locked(osdc);
42c1b124
ID		 2316 WARN_ON(!osdc->osdmap->epoch);
2317
b7ec35b3
ID		 2318 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2319 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2320 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
42c1b124
ID		 2321 dout("%s osdc %p continuous\n", __func__, osdc);
2322 continuous = true;
2323 } else {
2324 dout("%s osdc %p onetime\n", __func__, osdc);
2325 }
2326
2327 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2328 osdc->osdmap->epoch + 1, continuous))
2329 ceph_monc_renew_subs(&osdc->client->monc);
2330}
2331
a1f4020a 2332static void complete_request(struct ceph_osd_request *req, int err);
4609245e
ID		 2333static void send_map_check(struct ceph_osd_request *req);
2334
5aea3dcd 2335static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
0bbfdfe8 2336{
5aea3dcd
ID		 2337 struct ceph_osd_client *osdc = req->r_osdc;
2338 struct ceph_osd *osd;
4609245e 2339 enum calc_target_result ct_res;
66850df5 2340 int err = 0;
5aea3dcd
ID		 2341 bool need_send = false;
2342 bool promoted = false;
0bbfdfe8 2343
b18b9550 2344 WARN_ON(req->r_tid);
5aea3dcd
ID		 2345 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2346
2347again:
8edf84ba 2348 ct_res = calc_target(osdc, &req->r_t, false);
4609245e
ID		 2349 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2350 goto promote;
2351
5aea3dcd
ID		 2352 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2353 if (IS_ERR(osd)) {
2354 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2355 goto promote;
0bbfdfe8
ID		 2356 }
2357
66850df5
ID		 2358 if (osdc->abort_err) {
2359 dout("req %p abort_err %d\n", req, osdc->abort_err);
2360 err = osdc->abort_err;
2361 } else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
58eb7932
JL		 2362 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2363 osdc->epoch_barrier);
2364 req->r_t.paused = true;
2365 maybe_request_map(osdc);
2366 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2367 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
5aea3dcd
ID		 2368 dout("req %p pausewr\n", req);
2369 req->r_t.paused = true;
2370 maybe_request_map(osdc);
2371 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
b7ec35b3 2372 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
5aea3dcd
ID		 2373 dout("req %p pauserd\n", req);
2374 req->r_t.paused = true;
2375 maybe_request_map(osdc);
2376 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2377 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2378 CEPH_OSD_FLAG_FULL_FORCE)) &&
b7ec35b3 2379 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
5aea3dcd
ID		 2380 pool_full(osdc, req->r_t.base_oloc.pool))) {
2381 dout("req %p full/pool_full\n", req);
02b2f549 2382 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
66850df5 2383 err = -ENOSPC;
29e87820 2384 } else {
dc9b0dc4
ID		 2385 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL))
2386 pr_warn_ratelimited("cluster is full (osdmap FULL)\n");
2387 else
2388 pr_warn_ratelimited("pool %lld is full or reached quota\n",
2389 req->r_t.base_oloc.pool);
29e87820
ID		 2390 req->r_t.paused = true;
2391 maybe_request_map(osdc);
2392 }
5aea3dcd
ID		 2393 } else if (!osd_homeless(osd)) {
2394 need_send = true;
0bbfdfe8 2395 } else {
5aea3dcd 2396 maybe_request_map(osdc);
0bbfdfe8
ID		 2397 }
2398
5aea3dcd
ID		 2399 mutex_lock(&osd->lock);
2400 /*
2401 * Assign the tid atomically with send_request() to protect
2402 * multiple writes to the same object from racing with each
2403 * other, resulting in out of order ops on the OSDs.
2404 */
2405 req->r_tid = atomic64_inc_return(&osdc->last_tid);
2406 link_request(osd, req);
2407 if (need_send)
2408 send_request(req);
66850df5
ID		 2409 else if (err)
2410 complete_request(req, err);
5aea3dcd
ID		 2411 mutex_unlock(&osd->lock);
2412
6001567c 2413 if (!err && ct_res == CALC_TARGET_POOL_DNE)
4609245e
ID		 2414 send_map_check(req);
2415
5aea3dcd
ID		 2416 if (promoted)
2417 downgrade_write(&osdc->lock);
2418 return;
2419
2420promote:
2421 up_read(&osdc->lock);
2422 down_write(&osdc->lock);
2423 wrlocked = true;
2424 promoted = true;
2425 goto again;
2426}
2427
2428static void account_request(struct ceph_osd_request *req)
2429{
54ea0046 2430 WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
b18b9550 2431 WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
5aea3dcd 2432
b18b9550 2433 req->r_flags |= CEPH_OSD_FLAG_ONDISK;
22d2cfdf 2434 atomic_inc(&req->r_osdc->num_requests);
7cc5e38f
ID		 2435
2436 req->r_start_stamp = jiffies;
97e27aaa 2437 req->r_start_latency = ktime_get();
5aea3dcd
ID		 2438}
2439
2440static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2441{
2442 ceph_osdc_get_request(req);
2443 account_request(req);
2444 __submit_request(req, wrlocked);
2445}
2446
45ee2c1d 2447static void finish_request(struct ceph_osd_request *req)
5aea3dcd
ID		 2448{
2449 struct ceph_osd_client *osdc = req->r_osdc;
5aea3dcd 2450
04c7d789 2451 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
5aea3dcd
ID		 2452 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2453
97e27aaa
XL		 2454 req->r_end_latency = ktime_get();
2455
04c7d789
ID		 2456 if (req->r_osd)
2457 unlink_request(req->r_osd, req);
5aea3dcd
ID		 2458 atomic_dec(&osdc->num_requests);
2459
2460 /*
2461 * If an OSD has failed or returned and a request has been sent
2462 * twice, it's possible to get a reply and end up here while the
2463 * request message is queued for delivery. We will ignore the
2464 * reply, so not a big deal, but better to try and catch it.
2465 */
2466 ceph_msg_revoke(req->r_request);
2467 ceph_msg_revoke_incoming(req->r_reply);
2468}
2469
fe5da05e
ID		 2470static void __complete_request(struct ceph_osd_request *req)
2471{
d75f773c 2472 dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
26df726b
ID		 2473 req->r_tid, req->r_callback, req->r_result);
2474
2475 if (req->r_callback)
fe5da05e 2476 req->r_callback(req);
26df726b
ID		 2477 complete_all(&req->r_completion);
2478 ceph_osdc_put_request(req);
fe5da05e
ID		 2479}
2480
88bc1922
ID		 2481static void complete_request_workfn(struct work_struct *work)
2482{
2483 struct ceph_osd_request *req =
2484 container_of(work, struct ceph_osd_request, r_complete_work);
2485
2486 __complete_request(req);
fe5da05e
ID		 2487}
2488
4609245e 2489/*
b18b9550 2490 * This is open-coded in handle_reply().
4609245e
ID		 2491 */
2492static void complete_request(struct ceph_osd_request *req, int err)
2493{
2494 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2495
2496 req->r_result = err;
45ee2c1d 2497 finish_request(req);
88bc1922
ID		 2498
2499 INIT_WORK(&req->r_complete_work, complete_request_workfn);
2500 queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
4609245e
ID		 2501}
2502
2503static void cancel_map_check(struct ceph_osd_request *req)
2504{
2505 struct ceph_osd_client *osdc = req->r_osdc;
2506 struct ceph_osd_request *lookup_req;
2507
2508 verify_osdc_wrlocked(osdc);
2509
2510 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2511 if (!lookup_req)
2512 return;
2513
2514 WARN_ON(lookup_req != req);
2515 erase_request_mc(&osdc->map_checks, req);
2516 ceph_osdc_put_request(req);
2517}
2518
5aea3dcd
ID		 2519static void cancel_request(struct ceph_osd_request *req)
2520{
2521 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2522
4609245e 2523 cancel_map_check(req);
45ee2c1d 2524 finish_request(req);
b18b9550 2525 complete_all(&req->r_completion);
c297eb42 2526 ceph_osdc_put_request(req);
5aea3dcd
ID		 2527}
2528
7cc5e38f
ID		 2529static void abort_request(struct ceph_osd_request *req, int err)
2530{
2531 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2532
2533 cancel_map_check(req);
2534 complete_request(req, err);
2535}
2536
66850df5
ID		 2537static int abort_fn(struct ceph_osd_request *req, void *arg)
2538{
2539 int err = *(int *)arg;
2540
2541 abort_request(req, err);
2542 return 0; /* continue iteration */
2543}
2544
2545/*
2546 * Abort all in-flight requests with @err and arrange for all future
2547 * requests to be failed immediately.
2548 */
2549void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2550{
2551 dout("%s osdc %p err %d\n", __func__, osdc, err);
2552 down_write(&osdc->lock);
2553 for_each_request(osdc, abort_fn, &err);
2554 osdc->abort_err = err;
2555 up_write(&osdc->lock);
2556}
2557EXPORT_SYMBOL(ceph_osdc_abort_requests);
2558
2cef0ba8
YZ		 2559void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2560{
2561 down_write(&osdc->lock);
2562 osdc->abort_err = 0;
2563 up_write(&osdc->lock);
2564}
2565EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2566
58eb7932
JL		 2567static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2568{
2569 if (likely(eb > osdc->epoch_barrier)) {
2570 dout("updating epoch_barrier from %u to %u\n",
2571 osdc->epoch_barrier, eb);
2572 osdc->epoch_barrier = eb;
2573 /* Request map if we're not to the barrier yet */
2574 if (eb > osdc->osdmap->epoch)
2575 maybe_request_map(osdc);
2576 }
2577}
2578
2579void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2580{
2581 down_read(&osdc->lock);
2582 if (unlikely(eb > osdc->epoch_barrier)) {
2583 up_read(&osdc->lock);
2584 down_write(&osdc->lock);
2585 update_epoch_barrier(osdc, eb);
2586 up_write(&osdc->lock);
2587 } else {
2588 up_read(&osdc->lock);
2589 }
2590}
2591EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2592
4eea0fef
ID		 2593/*
2594 * We can end up releasing caps as a result of abort_request().
2595 * In that case, we probably want to ensure that the cap release message
2596 * has an updated epoch barrier in it, so set the epoch barrier prior to
2597 * aborting the first request.
2598 */
2599static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2600{
2601 struct ceph_osd_client *osdc = req->r_osdc;
2602 bool *victims = arg;
2603
c843d13c 2604 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
4eea0fef 2605 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
690f951d 2606 pool_full(osdc, req->r_t.base_oloc.pool))) {
4eea0fef
ID		 2607 if (!*victims) {
2608 update_epoch_barrier(osdc, osdc->osdmap->epoch);
2609 *victims = true;
2610 }
2611 abort_request(req, -ENOSPC);
2612 }
2613
2614 return 0; /* continue iteration */
2615}
2616
fc36d0a4
JL		 2617/*
2618 * Drop all pending requests that are stalled waiting on a full condition to
58eb7932
JL		 2619 * clear, and complete them with ENOSPC as the return code. Set the
2620 * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2621 * cancelled.
fc36d0a4
JL		 2622 */
2623static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2624{
58eb7932 2625 bool victims = false;
fc36d0a4 2626
02b2f549 2627 if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
c843d13c 2628 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
4eea0fef 2629 for_each_request(osdc, abort_on_full_fn, &victims);
fc36d0a4
JL		 2630}
2631
4609245e
ID		 2632static void check_pool_dne(struct ceph_osd_request *req)
2633{
2634 struct ceph_osd_client *osdc = req->r_osdc;
2635 struct ceph_osdmap *map = osdc->osdmap;
2636
2637 verify_osdc_wrlocked(osdc);
2638 WARN_ON(!map->epoch);
2639
2640 if (req->r_attempts) {
2641 /*
2642 * We sent a request earlier, which means that
2643 * previously the pool existed, and now it does not
2644 * (i.e., it was deleted).
2645 */
2646 req->r_map_dne_bound = map->epoch;
2647 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2648 req->r_tid);
2649 } else {
2650 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2651 req, req->r_tid, req->r_map_dne_bound, map->epoch);
2652 }
2653
2654 if (req->r_map_dne_bound) {
2655 if (map->epoch >= req->r_map_dne_bound) {
2656 /* we had a new enough map */
2657 pr_info_ratelimited("tid %llu pool does not exist\n",
2658 req->r_tid);
2659 complete_request(req, -ENOENT);
2660 }
2661 } else {
2662 send_map_check(req);
2663 }
2664}
2665
2666static void map_check_cb(struct ceph_mon_generic_request *greq)
2667{
2668 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2669 struct ceph_osd_request *req;
2670 u64 tid = greq->private_data;
2671
2672 WARN_ON(greq->result || !greq->u.newest);
2673
2674 down_write(&osdc->lock);
2675 req = lookup_request_mc(&osdc->map_checks, tid);
2676 if (!req) {
2677 dout("%s tid %llu dne\n", __func__, tid);
2678 goto out_unlock;
2679 }
2680
2681 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2682 req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2683 if (!req->r_map_dne_bound)
2684 req->r_map_dne_bound = greq->u.newest;
2685 erase_request_mc(&osdc->map_checks, req);
2686 check_pool_dne(req);
2687
2688 ceph_osdc_put_request(req);
2689out_unlock:
2690 up_write(&osdc->lock);
2691}
2692
2693static void send_map_check(struct ceph_osd_request *req)
2694{
2695 struct ceph_osd_client *osdc = req->r_osdc;
2696 struct ceph_osd_request *lookup_req;
2697 int ret;
2698
2699 verify_osdc_wrlocked(osdc);
2700
2701 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2702 if (lookup_req) {
2703 WARN_ON(lookup_req != req);
2704 return;
2705 }
2706
2707 ceph_osdc_get_request(req);
2708 insert_request_mc(&osdc->map_checks, req);
2709 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2710 map_check_cb, req->r_tid);
2711 WARN_ON(ret);
2712}
2713
922dab61
ID		 2714/*
2715 * lingering requests, watch/notify v2 infrastructure
2716 */
2717static void linger_release(struct kref *kref)
2718{
2719 struct ceph_osd_linger_request *lreq =
2720 container_of(kref, struct ceph_osd_linger_request, kref);
2721
2722 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2723 lreq->reg_req, lreq->ping_req);
2724 WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2725 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
4609245e 2726 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
922dab61 2727 WARN_ON(!list_empty(&lreq->scan_item));
b07d3c4b 2728 WARN_ON(!list_empty(&lreq->pending_lworks));
922dab61
ID		 2729 WARN_ON(lreq->osd);
2730
75dbb685
ID		 2731 if (lreq->request_pl)
2732 ceph_pagelist_release(lreq->request_pl);
2733 if (lreq->notify_id_pages)
2734 ceph_release_page_vector(lreq->notify_id_pages, 1);
2735
2736 ceph_osdc_put_request(lreq->reg_req);
2737 ceph_osdc_put_request(lreq->ping_req);
922dab61
ID		 2738 target_destroy(&lreq->t);
2739 kfree(lreq);
2740}
2741
2742static void linger_put(struct ceph_osd_linger_request *lreq)
2743{
2744 if (lreq)
2745 kref_put(&lreq->kref, linger_release);
2746}
2747
2748static struct ceph_osd_linger_request *
2749linger_get(struct ceph_osd_linger_request *lreq)
2750{
2751 kref_get(&lreq->kref);
2752 return lreq;
2753}
2754
2755static struct ceph_osd_linger_request *
2756linger_alloc(struct ceph_osd_client *osdc)
2757{
2758 struct ceph_osd_linger_request *lreq;
2759
2760 lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2761 if (!lreq)
2762 return NULL;
2763
2764 kref_init(&lreq->kref);
2765 mutex_init(&lreq->lock);
2766 RB_CLEAR_NODE(&lreq->node);
2767 RB_CLEAR_NODE(&lreq->osdc_node);
4609245e 2768 RB_CLEAR_NODE(&lreq->mc_node);
922dab61 2769 INIT_LIST_HEAD(&lreq->scan_item);
b07d3c4b 2770 INIT_LIST_HEAD(&lreq->pending_lworks);
922dab61 2771 init_completion(&lreq->reg_commit_wait);
19079203 2772 init_completion(&lreq->notify_finish_wait);
922dab61
ID		 2773
2774 lreq->osdc = osdc;
2775 target_init(&lreq->t);
2776
2777 dout("%s lreq %p\n", __func__, lreq);
2778 return lreq;
2779}
2780
2781DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2782DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
4609245e 2783DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
922dab61
ID		 2784
2785/*
2786 * Create linger request <-> OSD session relation.
2787 *
2788 * @lreq has to be registered, @osd may be homeless.
2789 */
2790static void link_linger(struct ceph_osd *osd,
2791 struct ceph_osd_linger_request *lreq)
2792{
2793 verify_osd_locked(osd);
2794 WARN_ON(!lreq->linger_id || lreq->osd);
2795 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2796 osd->o_osd, lreq, lreq->linger_id);
2797
2798 if (!osd_homeless(osd))
2799 __remove_osd_from_lru(osd);
2800 else
2801 atomic_inc(&osd->o_osdc->num_homeless);
2802
2803 get_osd(osd);
2804 insert_linger(&osd->o_linger_requests, lreq);
2805 lreq->osd = osd;
2806}
2807
2808static void unlink_linger(struct ceph_osd *osd,
2809 struct ceph_osd_linger_request *lreq)
2810{
2811 verify_osd_locked(osd);
2812 WARN_ON(lreq->osd != osd);
2813 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2814 osd->o_osd, lreq, lreq->linger_id);
2815
2816 lreq->osd = NULL;
2817 erase_linger(&osd->o_linger_requests, lreq);
2818 put_osd(osd);
2819
2820 if (!osd_homeless(osd))
2821 maybe_move_osd_to_lru(osd);
2822 else
2823 atomic_dec(&osd->o_osdc->num_homeless);
2824}
2825
2826static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2827{
2828 verify_osdc_locked(lreq->osdc);
2829
2830 return !RB_EMPTY_NODE(&lreq->osdc_node);
2831}
2832
2833static bool linger_registered(struct ceph_osd_linger_request *lreq)
2834{
2835 struct ceph_osd_client *osdc = lreq->osdc;
2836 bool registered;
2837
2838 down_read(&osdc->lock);
2839 registered = __linger_registered(lreq);
2840 up_read(&osdc->lock);
2841
2842 return registered;
2843}
2844
2845static void linger_register(struct ceph_osd_linger_request *lreq)
2846{
2847 struct ceph_osd_client *osdc = lreq->osdc;
2848
2849 verify_osdc_wrlocked(osdc);
2850 WARN_ON(lreq->linger_id);
2851
2852 linger_get(lreq);
2853 lreq->linger_id = ++osdc->last_linger_id;
2854 insert_linger_osdc(&osdc->linger_requests, lreq);
2855}
2856
2857static void linger_unregister(struct ceph_osd_linger_request *lreq)
2858{
2859 struct ceph_osd_client *osdc = lreq->osdc;
2860
2861 verify_osdc_wrlocked(osdc);
2862
2863 erase_linger_osdc(&osdc->linger_requests, lreq);
2864 linger_put(lreq);
2865}
2866
2867static void cancel_linger_request(struct ceph_osd_request *req)
2868{
2869 struct ceph_osd_linger_request *lreq = req->r_priv;
2870
2871 WARN_ON(!req->r_linger);
2872 cancel_request(req);
2873 linger_put(lreq);
2874}
2875
2876struct linger_work {
2877 struct work_struct work;
2878 struct ceph_osd_linger_request *lreq;
b07d3c4b
ID		 2879 struct list_head pending_item;
2880 unsigned long queued_stamp;
922dab61
ID		 2881
2882 union {
2883 struct {
2884 u64 notify_id;
2885 u64 notifier_id;
2886 void *payload; /* points into @msg front */
2887 size_t payload_len;
2888
2889 struct ceph_msg *msg; /* for ceph_msg_put() */
2890 } notify;
2891 struct {
2892 int err;
2893 } error;
2894 };
2895};
2896
2897static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2898 work_func_t workfn)
2899{
2900 struct linger_work *lwork;
2901
2902 lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2903 if (!lwork)
2904 return NULL;
2905
2906 INIT_WORK(&lwork->work, workfn);
b07d3c4b 2907 INIT_LIST_HEAD(&lwork->pending_item);
922dab61
ID		 2908 lwork->lreq = linger_get(lreq);
2909
2910 return lwork;
2911}
2912
2913static void lwork_free(struct linger_work *lwork)
2914{
2915 struct ceph_osd_linger_request *lreq = lwork->lreq;
2916
b07d3c4b
ID		 2917 mutex_lock(&lreq->lock);
2918 list_del(&lwork->pending_item);
2919 mutex_unlock(&lreq->lock);
2920
922dab61
ID		 2921 linger_put(lreq);
2922 kfree(lwork);
2923}
2924
2925static void lwork_queue(struct linger_work *lwork)
2926{
2927 struct ceph_osd_linger_request *lreq = lwork->lreq;
2928 struct ceph_osd_client *osdc = lreq->osdc;
2929
2930 verify_lreq_locked(lreq);
b07d3c4b
ID		 2931 WARN_ON(!list_empty(&lwork->pending_item));
2932
2933 lwork->queued_stamp = jiffies;
2934 list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
922dab61
ID		 2935 queue_work(osdc->notify_wq, &lwork->work);
2936}
2937
2938static void do_watch_notify(struct work_struct *w)
2939{
2940 struct linger_work *lwork = container_of(w, struct linger_work, work);
2941 struct ceph_osd_linger_request *lreq = lwork->lreq;
2942
2943 if (!linger_registered(lreq)) {
2944 dout("%s lreq %p not registered\n", __func__, lreq);
2945 goto out;
2946 }
2947
19079203 2948 WARN_ON(!lreq->is_watch);
922dab61
ID		 2949 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2950 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2951 lwork->notify.payload_len);
2952 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2953 lwork->notify.notifier_id, lwork->notify.payload,
2954 lwork->notify.payload_len);
2955
2956out:
2957 ceph_msg_put(lwork->notify.msg);
2958 lwork_free(lwork);
2959}
2960
2961static void do_watch_error(struct work_struct *w)
2962{
2963 struct linger_work *lwork = container_of(w, struct linger_work, work);
2964 struct ceph_osd_linger_request *lreq = lwork->lreq;
2965
2966 if (!linger_registered(lreq)) {
2967 dout("%s lreq %p not registered\n", __func__, lreq);
2968 goto out;
2969 }
2970
2971 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2972 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2973
2974out:
2975 lwork_free(lwork);
2976}
2977
2978static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2979{
2980 struct linger_work *lwork;
2981
2982 lwork = lwork_alloc(lreq, do_watch_error);
2983 if (!lwork) {
2984 pr_err("failed to allocate error-lwork\n");
2985 return;
2986 }
2987
2988 lwork->error.err = lreq->last_error;
2989 lwork_queue(lwork);
2990}
2991
2992static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
2993 int result)
2994{
2995 if (!completion_done(&lreq->reg_commit_wait)) {
2996 lreq->reg_commit_error = (result <= 0 ? result : 0);
2997 complete_all(&lreq->reg_commit_wait);
2998 }
2999}
3000
3001static void linger_commit_cb(struct ceph_osd_request *req)
3002{
3003 struct ceph_osd_linger_request *lreq = req->r_priv;
3004
3005 mutex_lock(&lreq->lock);
75dbb685
ID		 3006 if (req != lreq->reg_req) {
3007 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3008 __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3009 goto out;
3010 }
3011
922dab61
ID		 3012 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3013 lreq->linger_id, req->r_result);
922dab61
ID		 3014 linger_reg_commit_complete(lreq, req->r_result);
3015 lreq->committed = true;
3016
19079203
ID		 3017 if (!lreq->is_watch) {
3018 struct ceph_osd_data *osd_data =
3019 osd_req_op_data(req, 0, notify, response_data);
3020 void *p = page_address(osd_data->pages[0]);
3021
3022 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3023 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3024
3025 /* make note of the notify_id */
3026 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3027 lreq->notify_id = ceph_decode_64(&p);
3028 dout("lreq %p notify_id %llu\n", lreq,
3029 lreq->notify_id);
3030 } else {
3031 dout("lreq %p no notify_id\n", lreq);
3032 }
3033 }
3034
75dbb685 3035out:
922dab61
ID		 3036 mutex_unlock(&lreq->lock);
3037 linger_put(lreq);
3038}
3039
3040static int normalize_watch_error(int err)
3041{
3042 /*
3043 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3044 * notification and a failure to reconnect because we raced with
3045 * the delete appear the same to the user.
3046 */
3047 if (err == -ENOENT)
3048 err = -ENOTCONN;
3049
3050 return err;
3051}
3052
3053static void linger_reconnect_cb(struct ceph_osd_request *req)
3054{
3055 struct ceph_osd_linger_request *lreq = req->r_priv;
3056
3057 mutex_lock(&lreq->lock);
75dbb685
ID		 3058 if (req != lreq->reg_req) {
3059 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3060 __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3061 goto out;
3062 }
3063
922dab61
ID		 3064 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3065 lreq, lreq->linger_id, req->r_result, lreq->last_error);
3066 if (req->r_result < 0) {
3067 if (!lreq->last_error) {
3068 lreq->last_error = normalize_watch_error(req->r_result);
3069 queue_watch_error(lreq);
3070 }
3071 }
3072
75dbb685 3073out:
922dab61
ID		 3074 mutex_unlock(&lreq->lock);
3075 linger_put(lreq);
3076}
3077
3078static void send_linger(struct ceph_osd_linger_request *lreq)
3079{
75dbb685
ID		 3080 struct ceph_osd_client *osdc = lreq->osdc;
3081 struct ceph_osd_request *req;
3082 int ret;
922dab61 3083
75dbb685
ID		 3084 verify_osdc_wrlocked(osdc);
3085 mutex_lock(&lreq->lock);
922dab61
ID		 3086 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3087
75dbb685
ID		 3088 if (lreq->reg_req) {
3089 if (lreq->reg_req->r_osd)
3090 cancel_linger_request(lreq->reg_req);
3091 ceph_osdc_put_request(lreq->reg_req);
3092 }
3093
3094 req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3095 BUG_ON(!req);
922dab61 3096
5133ba8f 3097 target_copy(&req->r_t, &lreq->t);
922dab61
ID		 3098 req->r_mtime = lreq->mtime;
3099
19079203 3100 if (lreq->is_watch && lreq->committed) {
75dbb685
ID		 3101 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_RECONNECT,
3102 lreq->linger_id, ++lreq->register_gen);
922dab61 3103 dout("lreq %p reconnect register_gen %u\n", lreq,
75dbb685 3104 req->r_ops[0].watch.gen);
922dab61
ID		 3105 req->r_callback = linger_reconnect_cb;
3106 } else {
75dbb685
ID		 3107 if (lreq->is_watch) {
3108 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_WATCH,
3109 lreq->linger_id, 0);
3110 } else {
19079203 3111 lreq->notify_id = 0;
75dbb685
ID		 3112
3113 refcount_inc(&lreq->request_pl->refcnt);
3114 osd_req_op_notify_init(req, 0, lreq->linger_id,
3115 lreq->request_pl);
3116 ceph_osd_data_pages_init(
3117 osd_req_op_data(req, 0, notify, response_data),
3118 lreq->notify_id_pages, PAGE_SIZE, 0, false, false);
3119 }
922dab61
ID		 3120 dout("lreq %p register\n", lreq);
3121 req->r_callback = linger_commit_cb;
3122 }
75dbb685
ID		 3123
3124 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3125 BUG_ON(ret);
922dab61
ID		 3126
3127 req->r_priv = linger_get(lreq);
3128 req->r_linger = true;
75dbb685
ID		 3129 lreq->reg_req = req;
3130 mutex_unlock(&lreq->lock);
922dab61
ID		 3131
3132 submit_request(req, true);
3133}
3134
3135static void linger_ping_cb(struct ceph_osd_request *req)
3136{
3137 struct ceph_osd_linger_request *lreq = req->r_priv;
3138
3139 mutex_lock(&lreq->lock);
75dbb685
ID		 3140 if (req != lreq->ping_req) {
3141 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3142 __func__, lreq, lreq->linger_id, req, lreq->ping_req);
3143 goto out;
3144 }
3145
922dab61
ID		 3146 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3147 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3148 lreq->last_error);
3149 if (lreq->register_gen == req->r_ops[0].watch.gen) {
b07d3c4b
ID		 3150 if (!req->r_result) {
3151 lreq->watch_valid_thru = lreq->ping_sent;
3152 } else if (!lreq->last_error) {
922dab61
ID		 3153 lreq->last_error = normalize_watch_error(req->r_result);
3154 queue_watch_error(lreq);
3155 }
3156 } else {
3157 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3158 lreq->register_gen, req->r_ops[0].watch.gen);
3159 }
3160
75dbb685 3161out:
922dab61
ID		 3162 mutex_unlock(&lreq->lock);
3163 linger_put(lreq);
3164}
3165
3166static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3167{
3168 struct ceph_osd_client *osdc = lreq->osdc;
75dbb685
ID		 3169 struct ceph_osd_request *req;
3170 int ret;
922dab61 3171
b7ec35b3 3172 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
922dab61
ID		 3173 dout("%s PAUSERD\n", __func__);
3174 return;
3175 }
3176
3177 lreq->ping_sent = jiffies;
3178 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3179 __func__, lreq, lreq->linger_id, lreq->ping_sent,
3180 lreq->register_gen);
3181
75dbb685
ID		 3182 if (lreq->ping_req) {
3183 if (lreq->ping_req->r_osd)
3184 cancel_linger_request(lreq->ping_req);
3185 ceph_osdc_put_request(lreq->ping_req);
3186 }
922dab61 3187
75dbb685
ID		 3188 req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3189 BUG_ON(!req);
922dab61 3190
75dbb685
ID		 3191 target_copy(&req->r_t, &lreq->t);
3192 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_PING, lreq->linger_id,
3193 lreq->register_gen);
922dab61 3194 req->r_callback = linger_ping_cb;
75dbb685
ID		 3195
3196 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3197 BUG_ON(ret);
3198
922dab61
ID		 3199 req->r_priv = linger_get(lreq);
3200 req->r_linger = true;
75dbb685 3201 lreq->ping_req = req;
922dab61
ID		 3202
3203 ceph_osdc_get_request(req);
3204 account_request(req);
3205 req->r_tid = atomic64_inc_return(&osdc->last_tid);
3206 link_request(lreq->osd, req);
3207 send_request(req);
3208}
3209
3210static void linger_submit(struct ceph_osd_linger_request *lreq)
3211{
3212 struct ceph_osd_client *osdc = lreq->osdc;
3213 struct ceph_osd *osd;
3214
81c65213
ID		 3215 down_write(&osdc->lock);
3216 linger_register(lreq);
81c65213 3217
8edf84ba 3218 calc_target(osdc, &lreq->t, false);
922dab61
ID		 3219 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3220 link_linger(osd, lreq);
3221
3222 send_linger(lreq);
81c65213 3223 up_write(&osdc->lock);
922dab61
ID		 3224}
3225
4609245e
ID		 3226static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3227{
3228 struct ceph_osd_client *osdc = lreq->osdc;
3229 struct ceph_osd_linger_request *lookup_lreq;
3230
3231 verify_osdc_wrlocked(osdc);
3232
3233 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3234 lreq->linger_id);
3235 if (!lookup_lreq)
3236 return;
3237
3238 WARN_ON(lookup_lreq != lreq);
3239 erase_linger_mc(&osdc->linger_map_checks, lreq);
3240 linger_put(lreq);
3241}
3242
922dab61
ID		 3243/*
3244 * @lreq has to be both registered and linked.
3245 */
3246static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3247{
75dbb685 3248 if (lreq->ping_req && lreq->ping_req->r_osd)
922dab61 3249 cancel_linger_request(lreq->ping_req);
75dbb685 3250 if (lreq->reg_req && lreq->reg_req->r_osd)
922dab61 3251 cancel_linger_request(lreq->reg_req);
4609245e 3252 cancel_linger_map_check(lreq);
922dab61
ID		 3253 unlink_linger(lreq->osd, lreq);
3254 linger_unregister(lreq);
3255}
3256
3257static void linger_cancel(struct ceph_osd_linger_request *lreq)
3258{
3259 struct ceph_osd_client *osdc = lreq->osdc;
3260
3261 down_write(&osdc->lock);
3262 if (__linger_registered(lreq))
3263 __linger_cancel(lreq);
3264 up_write(&osdc->lock);
3265}
3266
4609245e
ID		 3267static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3268
3269static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3270{
3271 struct ceph_osd_client *osdc = lreq->osdc;
3272 struct ceph_osdmap *map = osdc->osdmap;
3273
3274 verify_osdc_wrlocked(osdc);
3275 WARN_ON(!map->epoch);
3276
3277 if (lreq->register_gen) {
3278 lreq->map_dne_bound = map->epoch;
3279 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3280 lreq, lreq->linger_id);
3281 } else {
3282 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3283 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3284 map->epoch);
3285 }
3286
3287 if (lreq->map_dne_bound) {
3288 if (map->epoch >= lreq->map_dne_bound) {
3289 /* we had a new enough map */
3290 pr_info("linger_id %llu pool does not exist\n",
3291 lreq->linger_id);
3292 linger_reg_commit_complete(lreq, -ENOENT);
3293 __linger_cancel(lreq);
3294 }
3295 } else {
3296 send_linger_map_check(lreq);
3297 }
3298}
3299
3300static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3301{
3302 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3303 struct ceph_osd_linger_request *lreq;
3304 u64 linger_id = greq->private_data;
3305
3306 WARN_ON(greq->result || !greq->u.newest);
3307
3308 down_write(&osdc->lock);
3309 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3310 if (!lreq) {
3311 dout("%s linger_id %llu dne\n", __func__, linger_id);
3312 goto out_unlock;
3313 }
3314
3315 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3316 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3317 greq->u.newest);
3318 if (!lreq->map_dne_bound)
3319 lreq->map_dne_bound = greq->u.newest;
3320 erase_linger_mc(&osdc->linger_map_checks, lreq);
3321 check_linger_pool_dne(lreq);
3322
3323 linger_put(lreq);
3324out_unlock:
3325 up_write(&osdc->lock);
3326}
3327
3328static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3329{
3330 struct ceph_osd_client *osdc = lreq->osdc;
3331 struct ceph_osd_linger_request *lookup_lreq;
3332 int ret;
3333
3334 verify_osdc_wrlocked(osdc);
3335
3336 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3337 lreq->linger_id);
3338 if (lookup_lreq) {
3339 WARN_ON(lookup_lreq != lreq);
3340 return;
3341 }
3342
3343 linger_get(lreq);
3344 insert_linger_mc(&osdc->linger_map_checks, lreq);
3345 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3346 linger_map_check_cb, lreq->linger_id);
3347 WARN_ON(ret);
3348}
3349
922dab61
ID		 3350static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3351{
3352 int ret;
3353
3354 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
e6e28432 3355 ret = wait_for_completion_killable(&lreq->reg_commit_wait);
922dab61
ID		 3356 return ret ?: lreq->reg_commit_error;
3357}
3358
e6e28432
ID		 3359static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq,
3360 unsigned long timeout)
19079203 3361{
e6e28432 3362 long left;
19079203
ID		 3363
3364 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
e6e28432
ID		 3365 left = wait_for_completion_killable_timeout(&lreq->notify_finish_wait,
3366 ceph_timeout_jiffies(timeout));
3367 if (left <= 0)
3368 left = left ?: -ETIMEDOUT;
3369 else
3370 left = lreq->notify_finish_error; /* completed */
3371
3372 return left;
19079203
ID		 3373}
3374
f24e9980 3375/*
fbca9635
ID		 3376 * Timeout callback, called every N seconds. When 1 or more OSD
3377 * requests has been active for more than N seconds, we send a keepalive
3378 * (tag + timestamp) to its OSD to ensure any communications channel
3379 * reset is detected.
f24e9980
SW		 3380 */
3381static void handle_timeout(struct work_struct *work)
3382{
3383 struct ceph_osd_client *osdc =
3384 container_of(work, struct ceph_osd_client, timeout_work.work);
a319bf56 3385 struct ceph_options *opts = osdc->client->options;
5aea3dcd 3386 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
7cc5e38f 3387 unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
5aea3dcd
ID		 3388 LIST_HEAD(slow_osds);
3389 struct rb_node *n, *p;
f24e9980 3390
5aea3dcd
ID		 3391 dout("%s osdc %p\n", __func__, osdc);
3392 down_write(&osdc->lock);
f24e9980 3393
422d2cb8
YS		 3394 /*
3395 * ping osds that are a bit slow. this ensures that if there
3396 * is a break in the TCP connection we will notice, and reopen
3397 * a connection with that osd (from the fault callback).
3398 */
5aea3dcd
ID		 3399 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3400 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3401 bool found = false;
3402
7cc5e38f 3403 for (p = rb_first(&osd->o_requests); p; ) {
5aea3dcd
ID		 3404 struct ceph_osd_request *req =
3405 rb_entry(p, struct ceph_osd_request, r_node);
3406
7cc5e38f
ID		 3407 p = rb_next(p); /* abort_request() */
3408
5aea3dcd
ID		 3409 if (time_before(req->r_stamp, cutoff)) {
3410 dout(" req %p tid %llu on osd%d is laggy\n",
3411 req, req->r_tid, osd->o_osd);
3412 found = true;
3413 }
7cc5e38f
ID		 3414 if (opts->osd_request_timeout &&
3415 time_before(req->r_start_stamp, expiry_cutoff)) {
3416 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3417 req->r_tid, osd->o_osd);
3418 abort_request(req, -ETIMEDOUT);
3419 }
5aea3dcd 3420 }
922dab61
ID		 3421 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3422 struct ceph_osd_linger_request *lreq =
3423 rb_entry(p, struct ceph_osd_linger_request, node);
3424
3425 dout(" lreq %p linger_id %llu is served by osd%d\n",
3426 lreq, lreq->linger_id, osd->o_osd);
3427 found = true;
3428
3429 mutex_lock(&lreq->lock);
19079203 3430 if (lreq->is_watch && lreq->committed && !lreq->last_error)
922dab61
ID		 3431 send_linger_ping(lreq);
3432 mutex_unlock(&lreq->lock);
3433 }
422d2cb8 3434
5aea3dcd
ID		 3435 if (found)
3436 list_move_tail(&osd->o_keepalive_item, &slow_osds);
422d2cb8 3437 }
5aea3dcd 3438
7cc5e38f
ID		 3439 if (opts->osd_request_timeout) {
3440 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3441 struct ceph_osd_request *req =
3442 rb_entry(p, struct ceph_osd_request, r_node);
3443
3444 p = rb_next(p); /* abort_request() */
3445
3446 if (time_before(req->r_start_stamp, expiry_cutoff)) {
3447 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3448 req->r_tid, osdc->homeless_osd.o_osd);
3449 abort_request(req, -ETIMEDOUT);
3450 }
3451 }
3452 }
3453
5aea3dcd
ID		 3454 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3455 maybe_request_map(osdc);
3456
422d2cb8 3457 while (!list_empty(&slow_osds)) {
5aea3dcd
ID		 3458 struct ceph_osd *osd = list_first_entry(&slow_osds,
3459 struct ceph_osd,
3460 o_keepalive_item);
422d2cb8 3461 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 3462 ceph_con_keepalive(&osd->o_con);
3463 }
3464
5aea3dcd 3465 up_write(&osdc->lock);
fbca9635
ID		 3466 schedule_delayed_work(&osdc->timeout_work,
3467 osdc->client->options->osd_keepalive_timeout);
f24e9980
SW		 3468}
3469
f5a2041b
YS		 3470static void handle_osds_timeout(struct work_struct *work)
3471{
3472 struct ceph_osd_client *osdc =
3473 container_of(work, struct ceph_osd_client,
3474 osds_timeout_work.work);
a319bf56 3475 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
42a2c09f 3476 struct ceph_osd *osd, *nosd;
f5a2041b 3477
42a2c09f 3478 dout("%s osdc %p\n", __func__, osdc);
5aea3dcd 3479 down_write(&osdc->lock);
42a2c09f
ID		 3480 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3481 if (time_before(jiffies, osd->lru_ttl))
3482 break;
3483
5aea3dcd 3484 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
922dab61 3485 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
5aea3dcd 3486 close_osd(osd);
42a2c09f 3487 }
f5a2041b 3488
5aea3dcd 3489 up_write(&osdc->lock);
f5a2041b
YS		 3490 schedule_delayed_work(&osdc->osds_timeout_work,
3491 round_jiffies_relative(delay));
3492}
3493
205ee118
ID		 3494static int ceph_oloc_decode(void **p, void *end,
3495 struct ceph_object_locator *oloc)
3496{
3497 u8 struct_v, struct_cv;
3498 u32 len;
3499 void *struct_end;
3500 int ret = 0;
3501
3502 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3503 struct_v = ceph_decode_8(p);
3504 struct_cv = ceph_decode_8(p);
3505 if (struct_v < 3) {
3506 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3507 struct_v, struct_cv);
3508 goto e_inval;
3509 }
3510 if (struct_cv > 6) {
3511 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3512 struct_v, struct_cv);
3513 goto e_inval;
3514 }
3515 len = ceph_decode_32(p);
3516 ceph_decode_need(p, end, len, e_inval);
3517 struct_end = *p + len;
3518
3519 oloc->pool = ceph_decode_64(p);
3520 *p += 4; /* skip preferred */
3521
3522 len = ceph_decode_32(p);
3523 if (len > 0) {
3524 pr_warn("ceph_object_locator::key is set\n");
3525 goto e_inval;
3526 }
3527
3528 if (struct_v >= 5) {
cd08e0a2
YZ		 3529 bool changed = false;
3530
205ee118
ID		 3531 len = ceph_decode_32(p);
3532 if (len > 0) {
30c156d9 3533 ceph_decode_need(p, end, len, e_inval);
cd08e0a2
YZ		 3534 if (!oloc->pool_ns ||
3535 ceph_compare_string(oloc->pool_ns, *p, len))
3536 changed = true;
30c156d9 3537 *p += len;
cd08e0a2
YZ		 3538 } else {
3539 if (oloc->pool_ns)
3540 changed = true;
3541 }
3542 if (changed) {
3543 /* redirect changes namespace */
3544 pr_warn("ceph_object_locator::nspace is changed\n");
3545 goto e_inval;
205ee118
ID		 3546 }
3547 }
3548
3549 if (struct_v >= 6) {
3550 s64 hash = ceph_decode_64(p);
3551 if (hash != -1) {
3552 pr_warn("ceph_object_locator::hash is set\n");
3553 goto e_inval;
3554 }
3555 }
3556
3557 /* skip the rest */
3558 *p = struct_end;
3559out:
3560 return ret;
3561
3562e_inval:
3563 ret = -EINVAL;
3564 goto out;
3565}
3566
3567static int ceph_redirect_decode(void **p, void *end,
3568 struct ceph_request_redirect *redir)
3569{
3570 u8 struct_v, struct_cv;
3571 u32 len;
3572 void *struct_end;
3573 int ret;
3574
3575 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3576 struct_v = ceph_decode_8(p);
3577 struct_cv = ceph_decode_8(p);
3578 if (struct_cv > 1) {
3579 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3580 struct_v, struct_cv);
3581 goto e_inval;
3582 }
3583 len = ceph_decode_32(p);
3584 ceph_decode_need(p, end, len, e_inval);
3585 struct_end = *p + len;
3586
3587 ret = ceph_oloc_decode(p, end, &redir->oloc);
3588 if (ret)
3589 goto out;
3590
3591 len = ceph_decode_32(p);
3592 if (len > 0) {
3593 pr_warn("ceph_request_redirect::object_name is set\n");
3594 goto e_inval;
3595 }
3596
205ee118
ID		 3597 /* skip the rest */
3598 *p = struct_end;
3599out:
3600 return ret;
3601
3602e_inval:
3603 ret = -EINVAL;
3604 goto out;
3605}
3606
fe5da05e
ID		 3607struct MOSDOpReply {
3608 struct ceph_pg pgid;
3609 u64 flags;
3610 int result;
3611 u32 epoch;
3612 int num_ops;
3613 u32 outdata_len[CEPH_OSD_MAX_OPS];
3614 s32 rval[CEPH_OSD_MAX_OPS];
3615 int retry_attempt;
3616 struct ceph_eversion replay_version;
3617 u64 user_version;
3618 struct ceph_request_redirect redirect;
3619};
25845472 3620
fe5da05e 3621static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
f24e9980 3622{
fe5da05e
ID		 3623 void *p = msg->front.iov_base;
3624 void *const end = p + msg->front.iov_len;
3625 u16 version = le16_to_cpu(msg->hdr.version);
3626 struct ceph_eversion bad_replay_version;
b0b31a8f 3627 u8 decode_redir;
fe5da05e
ID		 3628 u32 len;
3629 int ret;
3630 int i;
1b83bef2 3631
fe5da05e
ID		 3632 ceph_decode_32_safe(&p, end, len, e_inval);
3633 ceph_decode_need(&p, end, len, e_inval);
3634 p += len; /* skip oid */
1b83bef2 3635
fe5da05e
ID		 3636 ret = ceph_decode_pgid(&p, end, &m->pgid);
3637 if (ret)
3638 return ret;
1b83bef2 3639
fe5da05e
ID		 3640 ceph_decode_64_safe(&p, end, m->flags, e_inval);
3641 ceph_decode_32_safe(&p, end, m->result, e_inval);
3642 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3643 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3644 p += sizeof(bad_replay_version);
3645 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
1b83bef2 3646
fe5da05e
ID		 3647 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3648 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3649 goto e_inval;
1b83bef2 3650
fe5da05e
ID		 3651 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3652 e_inval);
3653 for (i = 0; i < m->num_ops; i++) {
1b83bef2 3654 struct ceph_osd_op *op = p;
1b83bef2 3655
fe5da05e 3656 m->outdata_len[i] = le32_to_cpu(op->payload_len);
1b83bef2
SW		 3657 p += sizeof(*op);
3658 }
1b83bef2 3659
fe5da05e
ID		 3660 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3661 for (i = 0; i < m->num_ops; i++)
3662 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
f24e9980 3663
fe5da05e
ID		 3664 if (version >= 5) {
3665 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3666 memcpy(&m->replay_version, p, sizeof(m->replay_version));
3667 p += sizeof(m->replay_version);
3668 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3669 } else {
3670 m->replay_version = bad_replay_version; /* struct */
3671 m->user_version = le64_to_cpu(m->replay_version.version);
3672 }
eb845ff1 3673
fe5da05e
ID		 3674 if (version >= 6) {
3675 if (version >= 7)
3676 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
b0b31a8f
ID		 3677 else
3678 decode_redir = 1;
3679 } else {
3680 decode_redir = 0;
3681 }
3682
3683 if (decode_redir) {
fe5da05e
ID		 3684 ret = ceph_redirect_decode(&p, end, &m->redirect);
3685 if (ret)
3686 return ret;
205ee118 3687 } else {
fe5da05e 3688 ceph_oloc_init(&m->redirect.oloc);
205ee118 3689 }
f24e9980 3690
fe5da05e
ID		 3691 return 0;
3692
3693e_inval:
3694 return -EINVAL;
3695}
3696
3697/*
b18b9550
ID		 3698 * Handle MOSDOpReply. Set ->r_result and call the callback if it is
3699 * specified.
fe5da05e 3700 */
5aea3dcd 3701static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
fe5da05e 3702{
5aea3dcd 3703 struct ceph_osd_client *osdc = osd->o_osdc;
fe5da05e
ID		 3704 struct ceph_osd_request *req;
3705 struct MOSDOpReply m;
3706 u64 tid = le64_to_cpu(msg->hdr.tid);
3707 u32 data_len = 0;
fe5da05e
ID		 3708 int ret;
3709 int i;
3710
3711 dout("%s msg %p tid %llu\n", __func__, msg, tid);
3712
5aea3dcd
ID		 3713 down_read(&osdc->lock);
3714 if (!osd_registered(osd)) {
3715 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3716 goto out_unlock_osdc;
3717 }
3718 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3719
3720 mutex_lock(&osd->lock);
3721 req = lookup_request(&osd->o_requests, tid);
fe5da05e 3722 if (!req) {
5aea3dcd
ID		 3723 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3724 goto out_unlock_session;
fe5da05e 3725 }
fe5da05e 3726
cd08e0a2 3727 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
fe5da05e 3728 ret = decode_MOSDOpReply(msg, &m);
cd08e0a2 3729 m.redirect.oloc.pool_ns = NULL;
fe5da05e
ID		 3730 if (ret) {
3731 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3732 req->r_tid, ret);
3733 ceph_msg_dump(msg);
3734 goto fail_request;
3735 }
3736 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3737 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3738 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3739 le64_to_cpu(m.replay_version.version), m.user_version);
3740
3741 if (m.retry_attempt >= 0) {
3742 if (m.retry_attempt != req->r_attempts - 1) {
3743 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3744 req, req->r_tid, m.retry_attempt,
3745 req->r_attempts - 1);
5aea3dcd 3746 goto out_unlock_session;
fe5da05e
ID		 3747 }
3748 } else {
3749 WARN_ON(1); /* MOSDOpReply v4 is assumed */
3750 }
3751
3752 if (!ceph_oloc_empty(&m.redirect.oloc)) {
3753 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3754 m.redirect.oloc.pool);
5aea3dcd
ID		 3755 unlink_request(osd, req);
3756 mutex_unlock(&osd->lock);
205ee118 3757
30c156d9
YZ		 3758 /*
3759 * Not ceph_oloc_copy() - changing pool_ns is not
3760 * supported.
3761 */
3762 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
890bd0f8
JL		 3763 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3764 CEPH_OSD_FLAG_IGNORE_OVERLAY |
3765 CEPH_OSD_FLAG_IGNORE_CACHE;
5aea3dcd
ID		 3766 req->r_tid = 0;
3767 __submit_request(req, false);
3768 goto out_unlock_osdc;
205ee118
ID		 3769 }
3770
117d96a0
ID		 3771 if (m.result == -EAGAIN) {
3772 dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3773 unlink_request(osd, req);
3774 mutex_unlock(&osd->lock);
3775
3776 /*
3777 * The object is missing on the replica or not (yet)
3778 * readable. Clear pgid to force a resend to the primary
3779 * via legacy_change.
3780 */
3781 req->r_t.pgid.pool = 0;
3782 req->r_t.pgid.seed = 0;
3783 WARN_ON(!req->r_t.used_replica);
3784 req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3785 CEPH_OSD_FLAG_LOCALIZE_READS);
3786 req->r_tid = 0;
3787 __submit_request(req, false);
3788 goto out_unlock_osdc;
3789 }
3790
fe5da05e
ID		 3791 if (m.num_ops != req->r_num_ops) {
3792 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3793 req->r_num_ops, req->r_tid);
3794 goto fail_request;
f24e9980 3795 }
fe5da05e
ID		 3796 for (i = 0; i < req->r_num_ops; i++) {
3797 dout(" req %p tid %llu op %d rval %d len %u\n", req,
3798 req->r_tid, i, m.rval[i], m.outdata_len[i]);
3799 req->r_ops[i].rval = m.rval[i];
3800 req->r_ops[i].outdata_len = m.outdata_len[i];
3801 data_len += m.outdata_len[i];
3802 }
3803 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3804 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3805 le32_to_cpu(msg->hdr.data_len), req->r_tid);
3806 goto fail_request;
3807 }
b18b9550
ID		 3808 dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3809 req, req->r_tid, m.result, data_len);
f24e9980 3810
b18b9550
ID		 3811 /*
3812 * Since we only ever request ONDISK, we should only ever get
3813 * one (type of) reply back.
3814 */
3815 WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3816 req->r_result = m.result ?: data_len;
3817 finish_request(req);
5aea3dcd
ID		 3818 mutex_unlock(&osd->lock);
3819 up_read(&osdc->lock);
f24e9980 3820
b18b9550 3821 __complete_request(req);
f24e9980
SW		 3822 return;
3823
fe5da05e 3824fail_request:
4609245e 3825 complete_request(req, -EIO);
5aea3dcd
ID		 3826out_unlock_session:
3827 mutex_unlock(&osd->lock);
3828out_unlock_osdc:
3829 up_read(&osdc->lock);
f24e9980
SW		 3830}
3831
42c1b124
ID		 3832static void set_pool_was_full(struct ceph_osd_client *osdc)
3833{
3834 struct rb_node *n;
3835
3836 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3837 struct ceph_pg_pool_info *pi =
3838 rb_entry(n, struct ceph_pg_pool_info, node);
3839
3840 pi->was_full = __pool_full(pi);
3841 }
3842}
3843
5aea3dcd 3844static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
f24e9980 3845{
5aea3dcd 3846 struct ceph_pg_pool_info *pi;
f24e9980 3847
5aea3dcd
ID		 3848 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3849 if (!pi)
3850 return false;
f24e9980 3851
5aea3dcd 3852 return pi->was_full && !__pool_full(pi);
422d2cb8
YS		 3853}
3854
922dab61
ID		 3855static enum calc_target_result
3856recalc_linger_target(struct ceph_osd_linger_request *lreq)
3857{
3858 struct ceph_osd_client *osdc = lreq->osdc;
3859 enum calc_target_result ct_res;
3860
8edf84ba 3861 ct_res = calc_target(osdc, &lreq->t, true);
922dab61
ID		 3862 if (ct_res == CALC_TARGET_NEED_RESEND) {
3863 struct ceph_osd *osd;
3864
3865 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3866 if (osd != lreq->osd) {
3867 unlink_linger(lreq->osd, lreq);
3868 link_linger(osd, lreq);
3869 }
3870 }
3871
3872 return ct_res;
3873}
3874
422d2cb8 3875/*
5aea3dcd 3876 * Requeue requests whose mapping to an OSD has changed.
422d2cb8 3877 */
5aea3dcd
ID		 3878static void scan_requests(struct ceph_osd *osd,
3879 bool force_resend,
3880 bool cleared_full,
3881 bool check_pool_cleared_full,
3882 struct rb_root *need_resend,
3883 struct list_head *need_resend_linger)
422d2cb8 3884{
5aea3dcd
ID		 3885 struct ceph_osd_client *osdc = osd->o_osdc;
3886 struct rb_node *n;
3887 bool force_resend_writes;
3888
922dab61
ID		 3889 for (n = rb_first(&osd->o_linger_requests); n; ) {
3890 struct ceph_osd_linger_request *lreq =
3891 rb_entry(n, struct ceph_osd_linger_request, node);
3892 enum calc_target_result ct_res;
3893
3894 n = rb_next(n); /* recalc_linger_target() */
3895
3896 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3897 lreq->linger_id);
3898 ct_res = recalc_linger_target(lreq);
3899 switch (ct_res) {
3900 case CALC_TARGET_NO_ACTION:
3901 force_resend_writes = cleared_full ||
3902 (check_pool_cleared_full &&
3903 pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3904 if (!force_resend && !force_resend_writes)
3905 break;
3906
df561f66 3907 fallthrough;
922dab61 3908 case CALC_TARGET_NEED_RESEND:
4609245e 3909 cancel_linger_map_check(lreq);
922dab61
ID		 3910 /*
3911 * scan_requests() for the previous epoch(s)
3912 * may have already added it to the list, since
3913 * it's not unlinked here.
3914 */
3915 if (list_empty(&lreq->scan_item))
3916 list_add_tail(&lreq->scan_item, need_resend_linger);
3917 break;
3918 case CALC_TARGET_POOL_DNE:
a10bcb19 3919 list_del_init(&lreq->scan_item);
4609245e 3920 check_linger_pool_dne(lreq);
922dab61
ID		 3921 break;
3922 }
3923 }
3924
5aea3dcd
ID		 3925 for (n = rb_first(&osd->o_requests); n; ) {
3926 struct ceph_osd_request *req =
3927 rb_entry(n, struct ceph_osd_request, r_node);
3928 enum calc_target_result ct_res;
3929
4609245e 3930 n = rb_next(n); /* unlink_request(), check_pool_dne() */
5aea3dcd
ID		 3931
3932 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
8edf84ba 3933 ct_res = calc_target(osdc, &req->r_t, false);
5aea3dcd
ID		 3934 switch (ct_res) {
3935 case CALC_TARGET_NO_ACTION:
3936 force_resend_writes = cleared_full ||
3937 (check_pool_cleared_full &&
3938 pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3939 if (!force_resend &&
3940 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3941 !force_resend_writes))
3942 break;
3943
df561f66 3944 fallthrough;
5aea3dcd 3945 case CALC_TARGET_NEED_RESEND:
4609245e 3946 cancel_map_check(req);
5aea3dcd
ID		 3947 unlink_request(osd, req);
3948 insert_request(need_resend, req);
3949 break;
3950 case CALC_TARGET_POOL_DNE:
4609245e 3951 check_pool_dne(req);
5aea3dcd 3952 break;
b0494532 3953 }
6f6c7006 3954 }
422d2cb8 3955}
6f6c7006 3956
42c1b124 3957static int handle_one_map(struct ceph_osd_client *osdc,
5aea3dcd
ID		 3958 void *p, void *end, bool incremental,
3959 struct rb_root *need_resend,
3960 struct list_head *need_resend_linger)
42c1b124
ID		 3961{
3962 struct ceph_osdmap *newmap;
3963 struct rb_node *n;
3964 bool skipped_map = false;
3965 bool was_full;
3966
b7ec35b3 3967 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
42c1b124
ID		 3968 set_pool_was_full(osdc);
3969
3970 if (incremental)
cd1a677c
ID		 3971 newmap = osdmap_apply_incremental(&p, end,
3972 ceph_msgr2(osdc->client),
3973 osdc->osdmap);
42c1b124 3974 else
cd1a677c 3975 newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client));
42c1b124
ID		 3976 if (IS_ERR(newmap))
3977 return PTR_ERR(newmap);
3978
3979 if (newmap != osdc->osdmap) {
3980 /*
3981 * Preserve ->was_full before destroying the old map.
3982 * For pools that weren't in the old map, ->was_full
3983 * should be false.
3984 */
3985 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
3986 struct ceph_pg_pool_info *pi =
3987 rb_entry(n, struct ceph_pg_pool_info, node);
3988 struct ceph_pg_pool_info *old_pi;
3989
3990 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
3991 if (old_pi)
3992 pi->was_full = old_pi->was_full;
3993 else
3994 WARN_ON(pi->was_full);
3995 }
3996
3997 if (osdc->osdmap->epoch &&
3998 osdc->osdmap->epoch + 1 < newmap->epoch) {
3999 WARN_ON(incremental);
4000 skipped_map = true;
4001 }
4002
4003 ceph_osdmap_destroy(osdc->osdmap);
4004 osdc->osdmap = newmap;
4005 }
4006
b7ec35b3 4007 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
5aea3dcd
ID		 4008 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
4009 need_resend, need_resend_linger);
4010
4011 for (n = rb_first(&osdc->osds); n; ) {
4012 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4013
4014 n = rb_next(n); /* close_osd() */
4015
4016 scan_requests(osd, skipped_map, was_full, true, need_resend,
4017 need_resend_linger);
4018 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
4019 memcmp(&osd->o_con.peer_addr,
4020 ceph_osd_addr(osdc->osdmap, osd->o_osd),
4021 sizeof(struct ceph_entity_addr)))
4022 close_osd(osd);
4023 }
42c1b124
ID		 4024
4025 return 0;
4026}
6f6c7006 4027
5aea3dcd
ID		 4028static void kick_requests(struct ceph_osd_client *osdc,
4029 struct rb_root *need_resend,
4030 struct list_head *need_resend_linger)
4031{
922dab61 4032 struct ceph_osd_linger_request *lreq, *nlreq;
04c7d789 4033 enum calc_target_result ct_res;
5aea3dcd
ID		 4034 struct rb_node *n;
4035
04c7d789
ID		 4036 /* make sure need_resend targets reflect latest map */
4037 for (n = rb_first(need_resend); n; ) {
4038 struct ceph_osd_request *req =
4039 rb_entry(n, struct ceph_osd_request, r_node);
4040
4041 n = rb_next(n);
4042
4043 if (req->r_t.epoch < osdc->osdmap->epoch) {
8edf84ba 4044 ct_res = calc_target(osdc, &req->r_t, false);
04c7d789
ID		 4045 if (ct_res == CALC_TARGET_POOL_DNE) {
4046 erase_request(need_resend, req);
4047 check_pool_dne(req);
4048 }
4049 }
4050 }
4051
5aea3dcd
ID		 4052 for (n = rb_first(need_resend); n; ) {
4053 struct ceph_osd_request *req =
4054 rb_entry(n, struct ceph_osd_request, r_node);
4055 struct ceph_osd *osd;
4056
4057 n = rb_next(n);
4058 erase_request(need_resend, req); /* before link_request() */
4059
5aea3dcd
ID		 4060 osd = lookup_create_osd(osdc, req->r_t.osd, true);
4061 link_request(osd, req);
4062 if (!req->r_linger) {
4063 if (!osd_homeless(osd) && !req->r_t.paused)
4064 send_request(req);
922dab61
ID		 4065 } else {
4066 cancel_linger_request(req);
5aea3dcd
ID		 4067 }
4068 }
922dab61
ID		 4069
4070 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4071 if (!osd_homeless(lreq->osd))
4072 send_linger(lreq);
4073
4074 list_del_init(&lreq->scan_item);
4075 }
5aea3dcd
ID		 4076}
4077
f24e9980
SW		 4078/*
4079 * Process updated osd map.
4080 *
4081 * The message contains any number of incremental and full maps, normally
4082 * indicating some sort of topology change in the cluster. Kick requests
4083 * off to different OSDs as needed.
4084 */
4085void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4086{
42c1b124
ID		 4087 void *p = msg->front.iov_base;
4088 void *const end = p + msg->front.iov_len;
f24e9980
SW		 4089 u32 nr_maps, maplen;
4090 u32 epoch;
f24e9980 4091 struct ceph_fsid fsid;
5aea3dcd
ID		 4092 struct rb_root need_resend = RB_ROOT;
4093 LIST_HEAD(need_resend_linger);
42c1b124
ID		 4094 bool handled_incremental = false;
4095 bool was_pauserd, was_pausewr;
4096 bool pauserd, pausewr;
4097 int err;
f24e9980 4098
42c1b124 4099 dout("%s have %u\n", __func__, osdc->osdmap->epoch);
5aea3dcd 4100 down_write(&osdc->lock);
f24e9980
SW		 4101
4102 /* verify fsid */
4103 ceph_decode_need(&p, end, sizeof(fsid), bad);
4104 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d 4105 if (ceph_check_fsid(osdc->client, &fsid) < 0)
42c1b124 4106 goto bad;
f24e9980 4107
b7ec35b3
ID		 4108 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4109 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4110 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
42c1b124 4111 have_pool_full(osdc);
9a1ea2db 4112
f24e9980
SW		 4113 /* incremental maps */
4114 ceph_decode_32_safe(&p, end, nr_maps, bad);
4115 dout(" %d inc maps\n", nr_maps);
4116 while (nr_maps > 0) {
4117 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 4118 epoch = ceph_decode_32(&p);
4119 maplen = ceph_decode_32(&p);
f24e9980 4120 ceph_decode_need(&p, end, maplen, bad);
42c1b124
ID		 4121 if (osdc->osdmap->epoch &&
4122 osdc->osdmap->epoch + 1 == epoch) {
f24e9980
SW		 4123 dout("applying incremental map %u len %d\n",
4124 epoch, maplen);
5aea3dcd
ID		 4125 err = handle_one_map(osdc, p, p + maplen, true,
4126 &need_resend, &need_resend_linger);
42c1b124 4127 if (err)
f24e9980 4128 goto bad;
42c1b124 4129 handled_incremental = true;
f24e9980
SW		 4130 } else {
4131 dout("ignoring incremental map %u len %d\n",
4132 epoch, maplen);
4133 }
42c1b124 4134 p += maplen;
f24e9980
SW		 4135 nr_maps--;
4136 }
42c1b124 4137 if (handled_incremental)
f24e9980
SW		 4138 goto done;
4139
4140 /* full maps */
4141 ceph_decode_32_safe(&p, end, nr_maps, bad);
4142 dout(" %d full maps\n", nr_maps);
4143 while (nr_maps) {
4144 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 4145 epoch = ceph_decode_32(&p);
4146 maplen = ceph_decode_32(&p);
f24e9980
SW		 4147 ceph_decode_need(&p, end, maplen, bad);
4148 if (nr_maps > 1) {
4149 dout("skipping non-latest full map %u len %d\n",
4150 epoch, maplen);
e5253a7b 4151 } else if (osdc->osdmap->epoch >= epoch) {
f24e9980
SW		 4152 dout("skipping full map %u len %d, "
4153 "older than our %u\n", epoch, maplen,
4154 osdc->osdmap->epoch);
4155 } else {
4156 dout("taking full map %u len %d\n", epoch, maplen);
5aea3dcd
ID		 4157 err = handle_one_map(osdc, p, p + maplen, false,
4158 &need_resend, &need_resend_linger);
42c1b124 4159 if (err)
f24e9980 4160 goto bad;
f24e9980
SW		 4161 }
4162 p += maplen;
4163 nr_maps--;
4164 }
4165
4166done:
cd634fb6
SW		 4167 /*
4168 * subscribe to subsequent osdmap updates if full to ensure
4169 * we find out when we are no longer full and stop returning
4170 * ENOSPC.
4171 */
b7ec35b3
ID		 4172 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4173 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4174 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
42c1b124 4175 have_pool_full(osdc);
58eb7932
JL		 4176 if (was_pauserd || was_pausewr || pauserd || pausewr ||
4177 osdc->osdmap->epoch < osdc->epoch_barrier)
42c1b124 4178 maybe_request_map(osdc);
cd634fb6 4179
5aea3dcd 4180 kick_requests(osdc, &need_resend, &need_resend_linger);
42c1b124 4181
fc36d0a4 4182 ceph_osdc_abort_on_full(osdc);
42c1b124
ID		 4183 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4184 osdc->osdmap->epoch);
5aea3dcd 4185 up_write(&osdc->lock);
03066f23 4186 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 4187 return;
4188
4189bad:
4190 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 4191 ceph_msg_dump(msg);
5aea3dcd
ID		 4192 up_write(&osdc->lock);
4193}
4194
4195/*
4196 * Resubmit requests pending on the given osd.
4197 */
4198static void kick_osd_requests(struct ceph_osd *osd)
4199{
4200 struct rb_node *n;
4201
a02a946d
ID		 4202 clear_backoffs(osd);
4203
922dab61 4204 for (n = rb_first(&osd->o_requests); n; ) {
5aea3dcd
ID		 4205 struct ceph_osd_request *req =
4206 rb_entry(n, struct ceph_osd_request, r_node);
4207
922dab61
ID		 4208 n = rb_next(n); /* cancel_linger_request() */
4209
5aea3dcd
ID		 4210 if (!req->r_linger) {
4211 if (!req->r_t.paused)
4212 send_request(req);
922dab61
ID		 4213 } else {
4214 cancel_linger_request(req);
5aea3dcd
ID		 4215 }
4216 }
922dab61
ID		 4217 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4218 struct ceph_osd_linger_request *lreq =
4219 rb_entry(n, struct ceph_osd_linger_request, node);
4220
4221 send_linger(lreq);
4222 }
5aea3dcd
ID		 4223}
4224
4225/*
4226 * If the osd connection drops, we need to resubmit all requests.
4227 */
4228static void osd_fault(struct ceph_connection *con)
4229{
4230 struct ceph_osd *osd = con->private;
4231 struct ceph_osd_client *osdc = osd->o_osdc;
4232
4233 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4234
4235 down_write(&osdc->lock);
4236 if (!osd_registered(osd)) {
4237 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4238 goto out_unlock;
4239 }
4240
4241 if (!reopen_osd(osd))
4242 kick_osd_requests(osd);
4243 maybe_request_map(osdc);
4244
4245out_unlock:
4246 up_write(&osdc->lock);
f24e9980
SW		 4247}
4248
a02a946d
ID		 4249struct MOSDBackoff {
4250 struct ceph_spg spgid;
4251 u32 map_epoch;
4252 u8 op;
4253 u64 id;
4254 struct ceph_hobject_id *begin;
4255 struct ceph_hobject_id *end;
4256};
4257
4258static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4259{
4260 void *p = msg->front.iov_base;
4261 void *const end = p + msg->front.iov_len;
4262 u8 struct_v;
4263 u32 struct_len;
4264 int ret;
4265
4266 ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4267 if (ret)
4268 return ret;
4269
4270 ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4271 if (ret)
4272 return ret;
4273
4274 ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4275 ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4276 ceph_decode_8_safe(&p, end, m->op, e_inval);
4277 ceph_decode_64_safe(&p, end, m->id, e_inval);
4278
4279 m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4280 if (!m->begin)
4281 return -ENOMEM;
4282
4283 ret = decode_hoid(&p, end, m->begin);
4284 if (ret) {
4285 free_hoid(m->begin);
4286 return ret;
4287 }
4288
4289 m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4290 if (!m->end) {
4291 free_hoid(m->begin);
4292 return -ENOMEM;
4293 }
4294
4295 ret = decode_hoid(&p, end, m->end);
4296 if (ret) {
4297 free_hoid(m->begin);
4298 free_hoid(m->end);
4299 return ret;
4300 }
4301
4302 return 0;
4303
4304e_inval:
4305 return -EINVAL;
4306}
4307
4308static struct ceph_msg *create_backoff_message(
4309 const struct ceph_osd_backoff *backoff,
4310 u32 map_epoch)
4311{
4312 struct ceph_msg *msg;
4313 void *p, *end;
4314 int msg_size;
4315
4316 msg_size = CEPH_ENCODING_START_BLK_LEN +
4317 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4318 msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4319 msg_size += CEPH_ENCODING_START_BLK_LEN +
4320 hoid_encoding_size(backoff->begin);
4321 msg_size += CEPH_ENCODING_START_BLK_LEN +
4322 hoid_encoding_size(backoff->end);
4323
4324 msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4325 if (!msg)
4326 return NULL;
4327
4328 p = msg->front.iov_base;
4329 end = p + msg->front_alloc_len;
4330
4331 encode_spgid(&p, &backoff->spgid);
4332 ceph_encode_32(&p, map_epoch);
4333 ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4334 ceph_encode_64(&p, backoff->id);
4335 encode_hoid(&p, end, backoff->begin);
4336 encode_hoid(&p, end, backoff->end);
4337 BUG_ON(p != end);
4338
4339 msg->front.iov_len = p - msg->front.iov_base;
4340 msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4341 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4342
4343 return msg;
4344}
4345
4346static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4347{
4348 struct ceph_spg_mapping *spg;
4349 struct ceph_osd_backoff *backoff;
4350 struct ceph_msg *msg;
4351
4352 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4353 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4354
4355 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4356 if (!spg) {
4357 spg = alloc_spg_mapping();
4358 if (!spg) {
4359 pr_err("%s failed to allocate spg\n", __func__);
4360 return;
4361 }
4362 spg->spgid = m->spgid; /* struct */
4363 insert_spg_mapping(&osd->o_backoff_mappings, spg);
4364 }
4365
4366 backoff = alloc_backoff();
4367 if (!backoff) {
4368 pr_err("%s failed to allocate backoff\n", __func__);
4369 return;
4370 }
4371 backoff->spgid = m->spgid; /* struct */
4372 backoff->id = m->id;
4373 backoff->begin = m->begin;
4374 m->begin = NULL; /* backoff now owns this */
4375 backoff->end = m->end;
4376 m->end = NULL; /* ditto */
4377
4378 insert_backoff(&spg->backoffs, backoff);
4379 insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4380
4381 /*
4382 * Ack with original backoff's epoch so that the OSD can
4383 * discard this if there was a PG split.
4384 */
4385 msg = create_backoff_message(backoff, m->map_epoch);
4386 if (!msg) {
4387 pr_err("%s failed to allocate msg\n", __func__);
4388 return;
4389 }
4390 ceph_con_send(&osd->o_con, msg);
4391}
4392
4393static bool target_contained_by(const struct ceph_osd_request_target *t,
4394 const struct ceph_hobject_id *begin,
4395 const struct ceph_hobject_id *end)
4396{
4397 struct ceph_hobject_id hoid;
4398 int cmp;
4399
4400 hoid_fill_from_target(&hoid, t);
4401 cmp = hoid_compare(&hoid, begin);
4402 return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4403}
4404
4405static void handle_backoff_unblock(struct ceph_osd *osd,
4406 const struct MOSDBackoff *m)
4407{
4408 struct ceph_spg_mapping *spg;
4409 struct ceph_osd_backoff *backoff;
4410 struct rb_node *n;
4411
4412 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4413 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4414
4415 backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4416 if (!backoff) {
4417 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4418 __func__, osd->o_osd, m->spgid.pgid.pool,
4419 m->spgid.pgid.seed, m->spgid.shard, m->id);
4420 return;
4421 }
4422
4423 if (hoid_compare(backoff->begin, m->begin) &&
4424 hoid_compare(backoff->end, m->end)) {
4425 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4426 __func__, osd->o_osd, m->spgid.pgid.pool,
4427 m->spgid.pgid.seed, m->spgid.shard, m->id);
4428 /* unblock it anyway... */
4429 }
4430
4431 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4432 BUG_ON(!spg);
4433
4434 erase_backoff(&spg->backoffs, backoff);
4435 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4436 free_backoff(backoff);
4437
4438 if (RB_EMPTY_ROOT(&spg->backoffs)) {
4439 erase_spg_mapping(&osd->o_backoff_mappings, spg);
4440 free_spg_mapping(spg);
4441 }
4442
4443 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4444 struct ceph_osd_request *req =
4445 rb_entry(n, struct ceph_osd_request, r_node);
4446
4447 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4448 /*
4449 * Match against @m, not @backoff -- the PG may
4450 * have split on the OSD.
4451 */
4452 if (target_contained_by(&req->r_t, m->begin, m->end)) {
4453 /*
4454 * If no other installed backoff applies,
4455 * resend.
4456 */
4457 send_request(req);
4458 }
4459 }
4460 }
4461}
4462
4463static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4464{
4465 struct ceph_osd_client *osdc = osd->o_osdc;
4466 struct MOSDBackoff m;
4467 int ret;
4468
4469 down_read(&osdc->lock);
4470 if (!osd_registered(osd)) {
4471 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4472 up_read(&osdc->lock);
4473 return;
4474 }
4475 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4476
4477 mutex_lock(&osd->lock);
4478 ret = decode_MOSDBackoff(msg, &m);
4479 if (ret) {
4480 pr_err("failed to decode MOSDBackoff: %d\n", ret);
4481 ceph_msg_dump(msg);
4482 goto out_unlock;
4483 }
4484
4485 switch (m.op) {
4486 case CEPH_OSD_BACKOFF_OP_BLOCK:
4487 handle_backoff_block(osd, &m);
4488 break;
4489 case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4490 handle_backoff_unblock(osd, &m);
4491 break;
4492 default:
4493 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4494 }
4495
4496 free_hoid(m.begin);
4497 free_hoid(m.end);
4498
4499out_unlock:
4500 mutex_unlock(&osd->lock);
4501 up_read(&osdc->lock);
4502}
4503
a40c4f10
YS		 4504/*
4505 * Process osd watch notifications
4506 */
3c663bbd
AE		 4507static void handle_watch_notify(struct ceph_osd_client *osdc,
4508 struct ceph_msg *msg)
a40c4f10 4509{
922dab61
ID		 4510 void *p = msg->front.iov_base;
4511 void *const end = p + msg->front.iov_len;
4512 struct ceph_osd_linger_request *lreq;
4513 struct linger_work *lwork;
4514 u8 proto_ver, opcode;
4515 u64 cookie, notify_id;
4516 u64 notifier_id = 0;
19079203 4517 s32 return_code = 0;
922dab61
ID		 4518 void *payload = NULL;
4519 u32 payload_len = 0;
a40c4f10
YS		 4520
4521 ceph_decode_8_safe(&p, end, proto_ver, bad);
4522 ceph_decode_8_safe(&p, end, opcode, bad);
4523 ceph_decode_64_safe(&p, end, cookie, bad);
922dab61 4524 p += 8; /* skip ver */
a40c4f10
YS		 4525 ceph_decode_64_safe(&p, end, notify_id, bad);
4526
922dab61
ID		 4527 if (proto_ver >= 1) {
4528 ceph_decode_32_safe(&p, end, payload_len, bad);
4529 ceph_decode_need(&p, end, payload_len, bad);
4530 payload = p;
4531 p += payload_len;
4532 }
4533
4534 if (le16_to_cpu(msg->hdr.version) >= 2)
19079203 4535 ceph_decode_32_safe(&p, end, return_code, bad);
922dab61
ID		 4536
4537 if (le16_to_cpu(msg->hdr.version) >= 3)
4538 ceph_decode_64_safe(&p, end, notifier_id, bad);
4539
4540 down_read(&osdc->lock);
4541 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4542 if (!lreq) {
4543 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4544 cookie);
4545 goto out_unlock_osdc;
4546 }
4547
4548 mutex_lock(&lreq->lock);
19079203
ID		 4549 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4550 opcode, cookie, lreq, lreq->is_watch);
922dab61
ID		 4551 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4552 if (!lreq->last_error) {
4553 lreq->last_error = -ENOTCONN;
4554 queue_watch_error(lreq);
4555 }
19079203
ID		 4556 } else if (!lreq->is_watch) {
4557 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4558 if (lreq->notify_id && lreq->notify_id != notify_id) {
4559 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4560 lreq->notify_id, notify_id);
4561 } else if (!completion_done(&lreq->notify_finish_wait)) {
4562 struct ceph_msg_data *data =
0d9c1ab3 4563 msg->num_data_items ? &msg->data[0] : NULL;
19079203
ID		 4564
4565 if (data) {
4566 if (lreq->preply_pages) {
4567 WARN_ON(data->type !=
4568 CEPH_MSG_DATA_PAGES);
4569 *lreq->preply_pages = data->pages;
4570 *lreq->preply_len = data->length;
e8862740 4571 data->own_pages = false;
19079203
ID		 4572 }
4573 }
4574 lreq->notify_finish_error = return_code;
4575 complete_all(&lreq->notify_finish_wait);
4576 }
922dab61
ID		 4577 } else {
4578 /* CEPH_WATCH_EVENT_NOTIFY */
4579 lwork = lwork_alloc(lreq, do_watch_notify);
4580 if (!lwork) {
4581 pr_err("failed to allocate notify-lwork\n");
4582 goto out_unlock_lreq;
a40c4f10 4583 }
a40c4f10 4584
922dab61
ID		 4585 lwork->notify.notify_id = notify_id;
4586 lwork->notify.notifier_id = notifier_id;
4587 lwork->notify.payload = payload;
4588 lwork->notify.payload_len = payload_len;
4589 lwork->notify.msg = ceph_msg_get(msg);
4590 lwork_queue(lwork);
91883cd2 4591 }
a40c4f10 4592
922dab61
ID		 4593out_unlock_lreq:
4594 mutex_unlock(&lreq->lock);
4595out_unlock_osdc:
4596 up_read(&osdc->lock);
a40c4f10
YS		 4597 return;
4598
4599bad:
4600 pr_err("osdc handle_watch_notify corrupt msg\n");
a40c4f10
YS		 4601}
4602
70636773
AE		 4603/*
4604 * Register request, send initial attempt.
4605 */
a8af0d68
JL		 4606void ceph_osdc_start_request(struct ceph_osd_client *osdc,
4607 struct ceph_osd_request *req)
70636773 4608{
5aea3dcd
ID		 4609 down_read(&osdc->lock);
4610 submit_request(req, false);
4611 up_read(&osdc->lock);
f24e9980 4612}
3d14c5d2 4613EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980 4614
c9f9b93d 4615/*
d0bb883c
ID		 4616 * Unregister request. If @req was registered, it isn't completed:
4617 * r_result isn't set and __complete_request() isn't invoked.
4618 *
4619 * If @req wasn't registered, this call may have raced with
4620 * handle_reply(), in which case r_result would already be set and
4621 * __complete_request() would be getting invoked, possibly even
4622 * concurrently with this call.
c9f9b93d
ID		 4623 */
4624void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4625{
4626 struct ceph_osd_client *osdc = req->r_osdc;
4627
5aea3dcd 4628 down_write(&osdc->lock);
5aea3dcd
ID		 4629 if (req->r_osd)
4630 cancel_request(req);
4631 up_write(&osdc->lock);
c9f9b93d
ID		 4632}
4633EXPORT_SYMBOL(ceph_osdc_cancel_request);
4634
f24e9980 4635/*
42b06965 4636 * @timeout: in jiffies, 0 means "wait forever"
f24e9980 4637 */
42b06965
ID		 4638static int wait_request_timeout(struct ceph_osd_request *req,
4639 unsigned long timeout)
f24e9980 4640{
42b06965 4641 long left;
c9f9b93d 4642
42b06965 4643 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
0e76abf2 4644 left = wait_for_completion_killable_timeout(&req->r_completion,
42b06965
ID		 4645 ceph_timeout_jiffies(timeout));
4646 if (left <= 0) {
4647 left = left ?: -ETIMEDOUT;
c9f9b93d 4648 ceph_osdc_cancel_request(req);
42b06965
ID		 4649 } else {
4650 left = req->r_result; /* completed */
f24e9980
SW		 4651 }
4652
42b06965
ID		 4653 return left;
4654}
4655
4656/*
4657 * wait for a request to complete
4658 */
4659int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4660 struct ceph_osd_request *req)
4661{
4662 return wait_request_timeout(req, 0);
f24e9980 4663}
3d14c5d2 4664EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 4665
4666/*
4667 * sync - wait for all in-flight requests to flush. avoid starvation.
4668 */
4669void ceph_osdc_sync(struct ceph_osd_client *osdc)
4670{
5aea3dcd
ID		 4671 struct rb_node *n, *p;
4672 u64 last_tid = atomic64_read(&osdc->last_tid);
f24e9980 4673
5aea3dcd
ID		 4674again:
4675 down_read(&osdc->lock);
4676 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4677 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4678
4679 mutex_lock(&osd->lock);
4680 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4681 struct ceph_osd_request *req =
4682 rb_entry(p, struct ceph_osd_request, r_node);
4683
4684 if (req->r_tid > last_tid)
4685 break;
4686
4687 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4688 continue;
f24e9980 4689
5aea3dcd
ID		 4690 ceph_osdc_get_request(req);
4691 mutex_unlock(&osd->lock);
4692 up_read(&osdc->lock);
4693 dout("%s waiting on req %p tid %llu last_tid %llu\n",
4694 __func__, req, req->r_tid, last_tid);
b18b9550 4695 wait_for_completion(&req->r_completion);
5aea3dcd
ID		 4696 ceph_osdc_put_request(req);
4697 goto again;
4698 }
f24e9980 4699
5aea3dcd 4700 mutex_unlock(&osd->lock);
f24e9980 4701 }
5aea3dcd
ID		 4702
4703 up_read(&osdc->lock);
4704 dout("%s done last_tid %llu\n", __func__, last_tid);
f24e9980 4705}
3d14c5d2 4706EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 4707
922dab61
ID		 4708/*
4709 * Returns a handle, caller owns a ref.
4710 */
4711struct ceph_osd_linger_request *
4712ceph_osdc_watch(struct ceph_osd_client *osdc,
4713 struct ceph_object_id *oid,
4714 struct ceph_object_locator *oloc,
4715 rados_watchcb2_t wcb,
4716 rados_watcherrcb_t errcb,
4717 void *data)
4718{
4719 struct ceph_osd_linger_request *lreq;
4720 int ret;
4721
4722 lreq = linger_alloc(osdc);
4723 if (!lreq)
4724 return ERR_PTR(-ENOMEM);
4725
19079203 4726 lreq->is_watch = true;
922dab61
ID		 4727 lreq->wcb = wcb;
4728 lreq->errcb = errcb;
4729 lreq->data = data;
b07d3c4b 4730 lreq->watch_valid_thru = jiffies;
922dab61
ID		 4731
4732 ceph_oid_copy(&lreq->t.base_oid, oid);
4733 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
54ea0046 4734 lreq->t.flags = CEPH_OSD_FLAG_WRITE;
fac02ddf 4735 ktime_get_real_ts64(&lreq->mtime);
922dab61 4736
81c65213 4737 linger_submit(lreq);
922dab61
ID		 4738 ret = linger_reg_commit_wait(lreq);
4739 if (ret) {
4740 linger_cancel(lreq);
4741 goto err_put_lreq;
4742 }
4743
4744 return lreq;
4745
4746err_put_lreq:
4747 linger_put(lreq);
4748 return ERR_PTR(ret);
4749}
4750EXPORT_SYMBOL(ceph_osdc_watch);
4751
4752/*
4753 * Releases a ref.
4754 *
4755 * Times out after mount_timeout to preserve rbd unmap behaviour
4756 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4757 * with mount_timeout").
4758 */
4759int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4760 struct ceph_osd_linger_request *lreq)
4761{
4762 struct ceph_options *opts = osdc->client->options;
4763 struct ceph_osd_request *req;
4764 int ret;
4765
4766 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4767 if (!req)
4768 return -ENOMEM;
4769
4770 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4771 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
54ea0046 4772 req->r_flags = CEPH_OSD_FLAG_WRITE;
fac02ddf 4773 ktime_get_real_ts64(&req->r_mtime);
75dbb685
ID		 4774 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_UNWATCH,
4775 lreq->linger_id, 0);
922dab61
ID		 4776
4777 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4778 if (ret)
4779 goto out_put_req;
4780
a8af0d68 4781 ceph_osdc_start_request(osdc, req);
922dab61
ID		 4782 linger_cancel(lreq);
4783 linger_put(lreq);
4784 ret = wait_request_timeout(req, opts->mount_timeout);
4785
4786out_put_req:
4787 ceph_osdc_put_request(req);
4788 return ret;
4789}
4790EXPORT_SYMBOL(ceph_osdc_unwatch);
4791
4792static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4793 u64 notify_id, u64 cookie, void *payload,
6d54228f 4794 u32 payload_len)
922dab61
ID		 4795{
4796 struct ceph_osd_req_op *op;
4797 struct ceph_pagelist *pl;
4798 int ret;
4799
042f6498 4800 op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
922dab61 4801
33165d47 4802 pl = ceph_pagelist_alloc(GFP_NOIO);
922dab61
ID		 4803 if (!pl)
4804 return -ENOMEM;
4805
922dab61
ID		 4806 ret = ceph_pagelist_encode_64(pl, notify_id);
4807 ret |= ceph_pagelist_encode_64(pl, cookie);
4808 if (payload) {
4809 ret |= ceph_pagelist_encode_32(pl, payload_len);
4810 ret |= ceph_pagelist_append(pl, payload, payload_len);
4811 } else {
4812 ret |= ceph_pagelist_encode_32(pl, 0);
4813 }
4814 if (ret) {
4815 ceph_pagelist_release(pl);
4816 return -ENOMEM;
4817 }
4818
4819 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4820 op->indata_len = pl->length;
4821 return 0;
4822}
4823
4824int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4825 struct ceph_object_id *oid,
4826 struct ceph_object_locator *oloc,
4827 u64 notify_id,
4828 u64 cookie,
4829 void *payload,
6d54228f 4830 u32 payload_len)
922dab61
ID		 4831{
4832 struct ceph_osd_request *req;
4833 int ret;
4834
4835 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4836 if (!req)
4837 return -ENOMEM;
4838
4839 ceph_oid_copy(&req->r_base_oid, oid);
4840 ceph_oloc_copy(&req->r_base_oloc, oloc);
4841 req->r_flags = CEPH_OSD_FLAG_READ;
4842
26f887e0
ID		 4843 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4844 payload_len);
922dab61
ID		 4845 if (ret)
4846 goto out_put_req;
4847
26f887e0 4848 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
922dab61
ID		 4849 if (ret)
4850 goto out_put_req;
4851
a8af0d68 4852 ceph_osdc_start_request(osdc, req);
922dab61
ID		 4853 ret = ceph_osdc_wait_request(osdc, req);
4854
4855out_put_req:
4856 ceph_osdc_put_request(req);
4857 return ret;
4858}
4859EXPORT_SYMBOL(ceph_osdc_notify_ack);
4860
19079203
ID		 4861/*
4862 * @timeout: in seconds
4863 *
4864 * @preply_{pages,len} are initialized both on success and error.
4865 * The caller is responsible for:
4866 *
4867 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4868 */
4869int ceph_osdc_notify(struct ceph_osd_client *osdc,
4870 struct ceph_object_id *oid,
4871 struct ceph_object_locator *oloc,
4872 void *payload,
6d54228f 4873 u32 payload_len,
19079203
ID		 4874 u32 timeout,
4875 struct page ***preply_pages,
4876 size_t *preply_len)
4877{
4878 struct ceph_osd_linger_request *lreq;
19079203
ID		 4879 int ret;
4880
4881 WARN_ON(!timeout);
4882 if (preply_pages) {
4883 *preply_pages = NULL;
4884 *preply_len = 0;
4885 }
4886
4887 lreq = linger_alloc(osdc);
4888 if (!lreq)
4889 return -ENOMEM;
4890
75dbb685
ID		 4891 lreq->request_pl = ceph_pagelist_alloc(GFP_NOIO);
4892 if (!lreq->request_pl) {
19079203
ID		 4893 ret = -ENOMEM;
4894 goto out_put_lreq;
4895 }
4896
75dbb685
ID		 4897 ret = ceph_pagelist_encode_32(lreq->request_pl, 1); /* prot_ver */
4898 ret |= ceph_pagelist_encode_32(lreq->request_pl, timeout);
4899 ret |= ceph_pagelist_encode_32(lreq->request_pl, payload_len);
4900 ret |= ceph_pagelist_append(lreq->request_pl, payload, payload_len);
4901 if (ret) {
4902 ret = -ENOMEM;
81c65213 4903 goto out_put_lreq;
75dbb685 4904 }
81c65213 4905
19079203 4906 /* for notify_id */
75dbb685
ID		 4907 lreq->notify_id_pages = ceph_alloc_page_vector(1, GFP_NOIO);
4908 if (IS_ERR(lreq->notify_id_pages)) {
4909 ret = PTR_ERR(lreq->notify_id_pages);
4910 lreq->notify_id_pages = NULL;
19079203
ID		 4911 goto out_put_lreq;
4912 }
19079203 4913
75dbb685
ID		 4914 lreq->preply_pages = preply_pages;
4915 lreq->preply_len = preply_len;
4916
4917 ceph_oid_copy(&lreq->t.base_oid, oid);
4918 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4919 lreq->t.flags = CEPH_OSD_FLAG_READ;
26f887e0 4920
81c65213 4921 linger_submit(lreq);
19079203
ID		 4922 ret = linger_reg_commit_wait(lreq);
4923 if (!ret)
e6e28432
ID		 4924 ret = linger_notify_finish_wait(lreq,
4925 msecs_to_jiffies(2 * timeout * MSEC_PER_SEC));
19079203
ID		 4926 else
4927 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4928
4929 linger_cancel(lreq);
4930out_put_lreq:
4931 linger_put(lreq);
4932 return ret;
4933}
4934EXPORT_SYMBOL(ceph_osdc_notify);
4935
b07d3c4b
ID		 4936/*
4937 * Return the number of milliseconds since the watch was last
4938 * confirmed, or an error. If there is an error, the watch is no
4939 * longer valid, and should be destroyed with ceph_osdc_unwatch().
4940 */
4941int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
4942 struct ceph_osd_linger_request *lreq)
4943{
4944 unsigned long stamp, age;
4945 int ret;
4946
4947 down_read(&osdc->lock);
4948 mutex_lock(&lreq->lock);
4949 stamp = lreq->watch_valid_thru;
4950 if (!list_empty(&lreq->pending_lworks)) {
4951 struct linger_work *lwork =
4952 list_first_entry(&lreq->pending_lworks,
4953 struct linger_work,
4954 pending_item);
4955
4956 if (time_before(lwork->queued_stamp, stamp))
4957 stamp = lwork->queued_stamp;
4958 }
4959 age = jiffies - stamp;
4960 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
4961 lreq, lreq->linger_id, age, lreq->last_error);
4962 /* we are truncating to msecs, so return a safe upper bound */
4963 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
4964
4965 mutex_unlock(&lreq->lock);
4966 up_read(&osdc->lock);
4967 return ret;
4968}
4969
a4ed38d7
DF		 4970static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
4971{
4972 u8 struct_v;
4973 u32 struct_len;
4974 int ret;
4975
4976 ret = ceph_start_decoding(p, end, 2, "watch_item_t",
4977 &struct_v, &struct_len);
4978 if (ret)
51fc7ab4
JL		 4979 goto bad;
4980
4981 ret = -EINVAL;
4982 ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
4983 ceph_decode_64_safe(p, end, item->cookie, bad);
4984 ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
a4ed38d7 4985
a4ed38d7 4986 if (struct_v >= 2) {
51fc7ab4
JL		 4987 ret = ceph_decode_entity_addr(p, end, &item->addr);
4988 if (ret)
4989 goto bad;
4990 } else {
4991 ret = 0;
a4ed38d7
DF		 4992 }
4993
4994 dout("%s %s%llu cookie %llu addr %s\n", __func__,
4995 ENTITY_NAME(item->name), item->cookie,
b726ec97 4996 ceph_pr_addr(&item->addr));
51fc7ab4
JL		 4997bad:
4998 return ret;
a4ed38d7
DF		 4999}
5000
5001static int decode_watchers(void **p, void *end,
5002 struct ceph_watch_item **watchers,
5003 u32 *num_watchers)
5004{
5005 u8 struct_v;
5006 u32 struct_len;
5007 int i;
5008 int ret;
5009
5010 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
5011 &struct_v, &struct_len);
5012 if (ret)
5013 return ret;
5014
5015 *num_watchers = ceph_decode_32(p);
5016 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
5017 if (!*watchers)
5018 return -ENOMEM;
5019
5020 for (i = 0; i < *num_watchers; i++) {
5021 ret = decode_watcher(p, end, *watchers + i);
5022 if (ret) {
5023 kfree(*watchers);
5024 return ret;
5025 }
5026 }
5027
5028 return 0;
5029}
5030
5031/*
5032 * On success, the caller is responsible for:
5033 *
5034 * kfree(watchers);
5035 */
5036int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5037 struct ceph_object_id *oid,
5038 struct ceph_object_locator *oloc,
5039 struct ceph_watch_item **watchers,
5040 u32 *num_watchers)
5041{
5042 struct ceph_osd_request *req;
5043 struct page **pages;
5044 int ret;
5045
5046 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5047 if (!req)
5048 return -ENOMEM;
5049
5050 ceph_oid_copy(&req->r_base_oid, oid);
5051 ceph_oloc_copy(&req->r_base_oloc, oloc);
5052 req->r_flags = CEPH_OSD_FLAG_READ;
5053
a4ed38d7
DF		 5054 pages = ceph_alloc_page_vector(1, GFP_NOIO);
5055 if (IS_ERR(pages)) {
5056 ret = PTR_ERR(pages);
5057 goto out_put_req;
5058 }
5059
5060 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5061 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5062 response_data),
5063 pages, PAGE_SIZE, 0, false, true);
5064
26f887e0
ID		 5065 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5066 if (ret)
5067 goto out_put_req;
5068
a8af0d68 5069 ceph_osdc_start_request(osdc, req);
a4ed38d7
DF		 5070 ret = ceph_osdc_wait_request(osdc, req);
5071 if (ret >= 0) {
5072 void *p = page_address(pages[0]);
5073 void *const end = p + req->r_ops[0].outdata_len;
5074
5075 ret = decode_watchers(&p, end, watchers, num_watchers);
5076 }
5077
5078out_put_req:
5079 ceph_osdc_put_request(req);
5080 return ret;
5081}
5082EXPORT_SYMBOL(ceph_osdc_list_watchers);
5083
dd935f44
JD		 5084/*
5085 * Call all pending notify callbacks - for use after a watch is
5086 * unregistered, to make sure no more callbacks for it will be invoked
5087 */
f6479449 5088void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
dd935f44 5089{
99d16943 5090 dout("%s osdc %p\n", __func__, osdc);
dd935f44
JD		 5091 flush_workqueue(osdc->notify_wq);
5092}
5093EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5094
7cca78c9
ID		 5095void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5096{
5097 down_read(&osdc->lock);
5098 maybe_request_map(osdc);
5099 up_read(&osdc->lock);
5100}
5101EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
dd935f44 5102
428a7158
DF		 5103/*
5104 * Execute an OSD class method on an object.
5105 *
5106 * @flags: CEPH_OSD_FLAG_*
2544a020 5107 * @resp_len: in/out param for reply length
428a7158
DF		 5108 */
5109int ceph_osdc_call(struct ceph_osd_client *osdc,
5110 struct ceph_object_id *oid,
5111 struct ceph_object_locator *oloc,
5112 const char *class, const char *method,
5113 unsigned int flags,
5114 struct page *req_page, size_t req_len,
68ada915 5115 struct page **resp_pages, size_t *resp_len)
428a7158
DF		 5116{
5117 struct ceph_osd_request *req;
5118 int ret;
5119
68ada915 5120 if (req_len > PAGE_SIZE)
2544a020
ID		 5121 return -E2BIG;
5122
428a7158
DF		 5123 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5124 if (!req)
5125 return -ENOMEM;
5126
5127 ceph_oid_copy(&req->r_base_oid, oid);
5128 ceph_oloc_copy(&req->r_base_oloc, oloc);
5129 req->r_flags = flags;
5130
24639ce5 5131 ret = osd_req_op_cls_init(req, 0, class, method);
fe943d50
CX		 5132 if (ret)
5133 goto out_put_req;
5134
428a7158
DF		 5135 if (req_page)
5136 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5137 0, false, false);
68ada915
ID		 5138 if (resp_pages)
5139 osd_req_op_cls_response_data_pages(req, 0, resp_pages,
2544a020 5140 *resp_len, 0, false, false);
428a7158 5141
26f887e0
ID		 5142 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5143 if (ret)
5144 goto out_put_req;
5145
a8af0d68 5146 ceph_osdc_start_request(osdc, req);
428a7158
DF		 5147 ret = ceph_osdc_wait_request(osdc, req);
5148 if (ret >= 0) {
5149 ret = req->r_ops[0].rval;
68ada915 5150 if (resp_pages)
428a7158
DF		 5151 *resp_len = req->r_ops[0].outdata_len;
5152 }
5153
5154out_put_req:
5155 ceph_osdc_put_request(req);
5156 return ret;
5157}
5158EXPORT_SYMBOL(ceph_osdc_call);
5159
120a75ea
YZ		 5160/*
5161 * reset all osd connections
5162 */
5163void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5164{
5165 struct rb_node *n;
5166
5167 down_write(&osdc->lock);
5168 for (n = rb_first(&osdc->osds); n; ) {
5169 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5170
5171 n = rb_next(n);
5172 if (!reopen_osd(osd))
5173 kick_osd_requests(osd);
5174 }
5175 up_write(&osdc->lock);
5176}
5177
f24e9980
SW		 5178/*
5179 * init, shutdown
5180 */
5181int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5182{
5183 int err;
5184
5185 dout("init\n");
5186 osdc->client = client;
5aea3dcd 5187 init_rwsem(&osdc->lock);
f24e9980 5188 osdc->osds = RB_ROOT;
f5a2041b 5189 INIT_LIST_HEAD(&osdc->osd_lru);
9dd2845c 5190 spin_lock_init(&osdc->osd_lru_lock);
5aea3dcd
ID		 5191 osd_init(&osdc->homeless_osd);
5192 osdc->homeless_osd.o_osdc = osdc;
5193 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
264048af 5194 osdc->last_linger_id = CEPH_LINGER_ID_START;
922dab61 5195 osdc->linger_requests = RB_ROOT;
4609245e
ID		 5196 osdc->map_checks = RB_ROOT;
5197 osdc->linger_map_checks = RB_ROOT;
f24e9980 5198 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b
YS		 5199 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5200
5f44f142 5201 err = -ENOMEM;
e5253a7b
ID		 5202 osdc->osdmap = ceph_osdmap_alloc();
5203 if (!osdc->osdmap)
5204 goto out;
5205
9e767adb
ID		 5206 osdc->req_mempool = mempool_create_slab_pool(10,
5207 ceph_osd_request_cache);
f24e9980 5208 if (!osdc->req_mempool)
e5253a7b 5209 goto out_map;
f24e9980 5210
d50b409f 5211 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
0d9c1ab3 5212 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
f24e9980 5213 if (err < 0)
5f44f142 5214 goto out_mempool;
d50b409f 5215 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
0d9c1ab3
ID		 5216 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5217 "osd_op_reply");
c16e7869
SW		 5218 if (err < 0)
5219 goto out_msgpool;
a40c4f10 5220
dbcae088 5221 err = -ENOMEM;
a40c4f10 5222 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
dbcae088 5223 if (!osdc->notify_wq)
c172ec5c
ID		 5224 goto out_msgpool_reply;
5225
88bc1922
ID		 5226 osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5227 if (!osdc->completion_wq)
5228 goto out_notify_wq;
5229
fbca9635
ID		 5230 schedule_delayed_work(&osdc->timeout_work,
5231 osdc->client->options->osd_keepalive_timeout);
b37ee1b9
ID		 5232 schedule_delayed_work(&osdc->osds_timeout_work,
5233 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5234
f24e9980 5235 return 0;
5f44f142 5236
88bc1922
ID		 5237out_notify_wq:
5238 destroy_workqueue(osdc->notify_wq);
c172ec5c
ID		 5239out_msgpool_reply:
5240 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
c16e7869
SW		 5241out_msgpool:
5242 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 5243out_mempool:
5244 mempool_destroy(osdc->req_mempool);
e5253a7b
ID		 5245out_map:
5246 ceph_osdmap_destroy(osdc->osdmap);
5f44f142
SW		 5247out:
5248 return err;
f24e9980
SW		 5249}
5250
5251void ceph_osdc_stop(struct ceph_osd_client *osdc)
5252{
88bc1922 5253 destroy_workqueue(osdc->completion_wq);
a40c4f10 5254 destroy_workqueue(osdc->notify_wq);
f24e9980 5255 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 5256 cancel_delayed_work_sync(&osdc->osds_timeout_work);
42a2c09f 5257
5aea3dcd 5258 down_write(&osdc->lock);
42a2c09f
ID		 5259 while (!RB_EMPTY_ROOT(&osdc->osds)) {
5260 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5261 struct ceph_osd, o_node);
5aea3dcd 5262 close_osd(osd);
42a2c09f 5263 }
5aea3dcd 5264 up_write(&osdc->lock);
02113a0f 5265 WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5aea3dcd
ID		 5266 osd_cleanup(&osdc->homeless_osd);
5267
5268 WARN_ON(!list_empty(&osdc->osd_lru));
922dab61 5269 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
4609245e
ID		 5270 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5271 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5aea3dcd
ID		 5272 WARN_ON(atomic_read(&osdc->num_requests));
5273 WARN_ON(atomic_read(&osdc->num_homeless));
42a2c09f 5274
e5253a7b 5275 ceph_osdmap_destroy(osdc->osdmap);
f24e9980
SW		 5276 mempool_destroy(osdc->req_mempool);
5277 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 5278 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 5279}
5280
aca39d9e
LH		 5281int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5282 u64 src_snapid, u64 src_version,
5283 struct ceph_object_id *src_oid,
5284 struct ceph_object_locator *src_oloc,
5285 u32 src_fadvise_flags,
5286 u32 dst_fadvise_flags,
5287 u32 truncate_seq, u64 truncate_size,
5288 u8 copy_from_flags)
23ddf9be
LH		 5289{
5290 struct ceph_osd_req_op *op;
5291 struct page **pages;
5292 void *p, *end;
5293
5294 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5295 if (IS_ERR(pages))
5296 return PTR_ERR(pages);
5297
042f6498
JL		 5298 op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5299 dst_fadvise_flags);
23ddf9be
LH		 5300 op->copy_from.snapid = src_snapid;
5301 op->copy_from.src_version = src_version;
5302 op->copy_from.flags = copy_from_flags;
5303 op->copy_from.src_fadvise_flags = src_fadvise_flags;
5304
5305 p = page_address(pages[0]);
5306 end = p + PAGE_SIZE;
5307 ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5308 encode_oloc(&p, end, src_oloc);
78beb0ff
LH		 5309 ceph_encode_32(&p, truncate_seq);
5310 ceph_encode_64(&p, truncate_size);
23ddf9be
LH		 5311 op->indata_len = PAGE_SIZE - (end - p);
5312
5313 ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5314 op->indata_len, 0, false, true);
5315 return 0;
5316}
aca39d9e 5317EXPORT_SYMBOL(osd_req_op_copy_from_init);
23ddf9be 5318
57a35dfb 5319int __init ceph_osdc_setup(void)
5522ae0b 5320{
3f1af42a
ID		 5321 size_t size = sizeof(struct ceph_osd_request) +
5322 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5323
5522ae0b 5324 BUG_ON(ceph_osd_request_cache);
3f1af42a
ID		 5325 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5326 0, 0, NULL);
5522ae0b
AE		 5327
5328 return ceph_osd_request_cache ? 0 : -ENOMEM;
5329}
5522ae0b
AE		 5330
5331void ceph_osdc_cleanup(void)
5332{
5333 BUG_ON(!ceph_osd_request_cache);
5334 kmem_cache_destroy(ceph_osd_request_cache);
5335 ceph_osd_request_cache = NULL;
5336}
5522ae0b 5337
f24e9980
SW		 5338/*
5339 * handle incoming message
5340 */
4972cf60 5341static void osd_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
f24e9980
SW		 5342{
5343 struct ceph_osd *osd = con->private;
5aea3dcd 5344 struct ceph_osd_client *osdc = osd->o_osdc;
f24e9980
SW		 5345 int type = le16_to_cpu(msg->hdr.type);
5346
f24e9980
SW		 5347 switch (type) {
5348 case CEPH_MSG_OSD_MAP:
5349 ceph_osdc_handle_map(osdc, msg);
5350 break;
5351 case CEPH_MSG_OSD_OPREPLY:
5aea3dcd 5352 handle_reply(osd, msg);
f24e9980 5353 break;
a02a946d
ID		 5354 case CEPH_MSG_OSD_BACKOFF:
5355 handle_backoff(osd, msg);
5356 break;
a40c4f10
YS		 5357 case CEPH_MSG_WATCH_NOTIFY:
5358 handle_watch_notify(osdc, msg);
5359 break;
f24e9980
SW		 5360
5361 default:
5362 pr_err("received unknown message type %d %s\n", type,
5363 ceph_msg_type_name(type));
5364 }
5aea3dcd 5365
f24e9980
SW		 5366 ceph_msg_put(msg);
5367}
5368
5b3a4db3 5369/*
d15f9d69
ID		 5370 * Lookup and return message for incoming reply. Don't try to do
5371 * anything about a larger than preallocated data portion of the
5372 * message at the moment - for now, just skip the message.
5b3a4db3
SW		 5373 */
5374static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 5375 struct ceph_msg_header *hdr,
5376 int *skip)
f24e9980
SW		 5377{
5378 struct ceph_osd *osd = con->private;
5379 struct ceph_osd_client *osdc = osd->o_osdc;
5aea3dcd 5380 struct ceph_msg *m = NULL;
0547a9b3 5381 struct ceph_osd_request *req;
3f0a4ac5 5382 int front_len = le32_to_cpu(hdr->front_len);
5b3a4db3 5383 int data_len = le32_to_cpu(hdr->data_len);
5aea3dcd 5384 u64 tid = le64_to_cpu(hdr->tid);
f24e9980 5385
5aea3dcd
ID		 5386 down_read(&osdc->lock);
5387 if (!osd_registered(osd)) {
5388 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5389 *skip = 1;
5390 goto out_unlock_osdc;
5391 }
5392 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5393
5394 mutex_lock(&osd->lock);
5395 req = lookup_request(&osd->o_requests, tid);
0547a9b3 5396 if (!req) {
cd8140c6
ID		 5397 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5398 osd->o_osd, tid);
d15f9d69 5399 *skip = 1;
5aea3dcd 5400 goto out_unlock_session;
0547a9b3 5401 }
c16e7869 5402
ace6d3a9 5403 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 5404
f2be82b0 5405 if (front_len > req->r_reply->front_alloc_len) {
d15f9d69
ID		 5406 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5407 __func__, osd->o_osd, req->r_tid, front_len,
5408 req->r_reply->front_alloc_len);
3f0a4ac5
ID		 5409 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5410 false);
a79832f2 5411 if (!m)
5aea3dcd 5412 goto out_unlock_session;
c16e7869
SW		 5413 ceph_msg_put(req->r_reply);
5414 req->r_reply = m;
5415 }
0fff87ec 5416
d15f9d69
ID		 5417 if (data_len > req->r_reply->data_length) {
5418 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5419 __func__, osd->o_osd, req->r_tid, data_len,
5420 req->r_reply->data_length);
5421 m = NULL;
5422 *skip = 1;
5aea3dcd 5423 goto out_unlock_session;
0547a9b3 5424 }
d15f9d69
ID		 5425
5426 m = ceph_msg_get(req->r_reply);
c16e7869 5427 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3 5428
5aea3dcd
ID		 5429out_unlock_session:
5430 mutex_unlock(&osd->lock);
5431out_unlock_osdc:
5432 up_read(&osdc->lock);
2450418c 5433 return m;
5b3a4db3
SW		 5434}
5435
19079203
ID		 5436static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5437{
5438 struct ceph_msg *m;
5439 int type = le16_to_cpu(hdr->type);
5440 u32 front_len = le32_to_cpu(hdr->front_len);
5441 u32 data_len = le32_to_cpu(hdr->data_len);
5442
0d9c1ab3 5443 m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
19079203
ID		 5444 if (!m)
5445 return NULL;
5446
5447 if (data_len) {
5448 struct page **pages;
19079203
ID		 5449
5450 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5451 GFP_NOIO);
c22e853a 5452 if (IS_ERR(pages)) {
19079203
ID		 5453 ceph_msg_put(m);
5454 return NULL;
5455 }
5456
e8862740 5457 ceph_msg_data_add_pages(m, pages, data_len, 0, true);
19079203
ID		 5458 }
5459
5460 return m;
5461}
5462
4972cf60
ID		 5463static struct ceph_msg *osd_alloc_msg(struct ceph_connection *con,
5464 struct ceph_msg_header *hdr,
5465 int *skip)
5b3a4db3
SW		 5466{
5467 struct ceph_osd *osd = con->private;
5468 int type = le16_to_cpu(hdr->type);
5b3a4db3 5469
1c20f2d2 5470 *skip = 0;
5b3a4db3
SW		 5471 switch (type) {
5472 case CEPH_MSG_OSD_MAP:
a02a946d 5473 case CEPH_MSG_OSD_BACKOFF:
a40c4f10 5474 case CEPH_MSG_WATCH_NOTIFY:
19079203 5475 return alloc_msg_with_page_vector(hdr);
5b3a4db3
SW		 5476 case CEPH_MSG_OSD_OPREPLY:
5477 return get_reply(con, hdr, skip);
5478 default:
5aea3dcd
ID		 5479 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5480 osd->o_osd, type);
5b3a4db3
SW		 5481 *skip = 1;
5482 return NULL;
5483 }
f24e9980
SW		 5484}
5485
5486/*
5487 * Wrappers to refcount containing ceph_osd struct
5488 */
4972cf60 5489static struct ceph_connection *osd_get_con(struct ceph_connection *con)
f24e9980
SW		 5490{
5491 struct ceph_osd *osd = con->private;
5492 if (get_osd(osd))
5493 return con;
5494 return NULL;
5495}
5496
4972cf60 5497static void osd_put_con(struct ceph_connection *con)
f24e9980
SW		 5498{
5499 struct ceph_osd *osd = con->private;
5500 put_osd(osd);
5501}
5502
4e7a5dcd
SW		 5503/*
5504 * authentication
5505 */
cd1a677c 5506
a3530df3
AE		 5507/*
5508 * Note: returned pointer is the address of a structure that's
5509 * managed separately. Caller must *not* attempt to free it.
5510 */
4972cf60
ID		 5511static struct ceph_auth_handshake *
5512osd_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
4e7a5dcd
SW		 5513{
5514 struct ceph_osd *o = con->private;
5515 struct ceph_osd_client *osdc = o->o_osdc;
5516 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 5517 struct ceph_auth_handshake *auth = &o->o_auth;
ce287162 5518 int ret;
4e7a5dcd 5519
ce287162
ID		 5520 ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5521 force_new, proto, NULL, NULL);
5522 if (ret)
5523 return ERR_PTR(ret);
74f1869f 5524
a3530df3 5525 return auth;
4e7a5dcd
SW		 5526}
5527
4972cf60 5528static int osd_add_authorizer_challenge(struct ceph_connection *con,
6daca13d
ID		 5529 void *challenge_buf, int challenge_buf_len)
5530{
5531 struct ceph_osd *o = con->private;
5532 struct ceph_osd_client *osdc = o->o_osdc;
5533 struct ceph_auth_client *ac = osdc->client->monc.auth;
5534
5535 return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5536 challenge_buf, challenge_buf_len);
5537}
4e7a5dcd 5538
4972cf60 5539static int osd_verify_authorizer_reply(struct ceph_connection *con)
4e7a5dcd
SW		 5540{
5541 struct ceph_osd *o = con->private;
5542 struct ceph_osd_client *osdc = o->o_osdc;
5543 struct ceph_auth_client *ac = osdc->client->monc.auth;
285ea34f 5544 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 5545
285ea34f
ID		 5546 return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5547 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5548 NULL, NULL, NULL, NULL);
4e7a5dcd
SW		 5549}
5550
4972cf60 5551static int osd_invalidate_authorizer(struct ceph_connection *con)
9bd2e6f8
SW		 5552{
5553 struct ceph_osd *o = con->private;
5554 struct ceph_osd_client *osdc = o->o_osdc;
5555 struct ceph_auth_client *ac = osdc->client->monc.auth;
5556
27859f97 5557 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 5558 return ceph_monc_validate_auth(&osdc->client->monc);
5559}
4e7a5dcd 5560
cd1a677c
ID		 5561static int osd_get_auth_request(struct ceph_connection *con,
5562 void *buf, int *buf_len,
5563 void **authorizer, int *authorizer_len)
5564{
5565 struct ceph_osd *o = con->private;
5566 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5567 struct ceph_auth_handshake *auth = &o->o_auth;
5568 int ret;
5569
5570 ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5571 buf, buf_len);
5572 if (ret)
5573 return ret;
5574
5575 *authorizer = auth->authorizer_buf;
5576 *authorizer_len = auth->authorizer_buf_len;
5577 return 0;
5578}
5579
5580static int osd_handle_auth_reply_more(struct ceph_connection *con,
5581 void *reply, int reply_len,
5582 void *buf, int *buf_len,
5583 void **authorizer, int *authorizer_len)
5584{
5585 struct ceph_osd *o = con->private;
5586 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5587 struct ceph_auth_handshake *auth = &o->o_auth;
5588 int ret;
5589
5590 ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5591 buf, buf_len);
5592 if (ret)
5593 return ret;
5594
5595 *authorizer = auth->authorizer_buf;
5596 *authorizer_len = auth->authorizer_buf_len;
5597 return 0;
5598}
5599
5600static int osd_handle_auth_done(struct ceph_connection *con,
5601 u64 global_id, void *reply, int reply_len,
5602 u8 *session_key, int *session_key_len,
5603 u8 *con_secret, int *con_secret_len)
5604{
5605 struct ceph_osd *o = con->private;
5606 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5607 struct ceph_auth_handshake *auth = &o->o_auth;
5608
5609 return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5610 session_key, session_key_len,
5611 con_secret, con_secret_len);
5612}
5613
5614static int osd_handle_auth_bad_method(struct ceph_connection *con,
5615 int used_proto, int result,
5616 const int *allowed_protos, int proto_cnt,
5617 const int *allowed_modes, int mode_cnt)
5618{
5619 struct ceph_osd *o = con->private;
5620 struct ceph_mon_client *monc = &o->o_osdc->client->monc;
5621 int ret;
5622
5623 if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD,
5624 used_proto, result,
5625 allowed_protos, proto_cnt,
5626 allowed_modes, mode_cnt)) {
5627 ret = ceph_monc_validate_auth(monc);
5628 if (ret)
5629 return ret;
5630 }
5631
5632 return -EACCES;
5633}
5634
8cb441c0
ID		 5635static void osd_reencode_message(struct ceph_msg *msg)
5636{
914902af
ID		 5637 int type = le16_to_cpu(msg->hdr.type);
5638
5639 if (type == CEPH_MSG_OSD_OP)
5640 encode_request_finish(msg);
8cb441c0
ID		 5641}
5642
79dbd1ba 5643static int osd_sign_message(struct ceph_msg *msg)
33d07337 5644{
79dbd1ba 5645 struct ceph_osd *o = msg->con->private;
33d07337 5646 struct ceph_auth_handshake *auth = &o->o_auth;
79dbd1ba 5647
33d07337
YZ		 5648 return ceph_auth_sign_message(auth, msg);
5649}
5650
79dbd1ba 5651static int osd_check_message_signature(struct ceph_msg *msg)
33d07337 5652{
79dbd1ba 5653 struct ceph_osd *o = msg->con->private;
33d07337 5654 struct ceph_auth_handshake *auth = &o->o_auth;
79dbd1ba 5655
33d07337
YZ		 5656 return ceph_auth_check_message_signature(auth, msg);
5657}
5658
9e32789f 5659static const struct ceph_connection_operations osd_con_ops = {
4972cf60
ID		 5660 .get = osd_get_con,
5661 .put = osd_put_con,
5662 .alloc_msg = osd_alloc_msg,
5663 .dispatch = osd_dispatch,
5664 .fault = osd_fault,
8cb441c0 5665 .reencode_message = osd_reencode_message,
4972cf60
ID		 5666 .get_authorizer = osd_get_authorizer,
5667 .add_authorizer_challenge = osd_add_authorizer_challenge,
5668 .verify_authorizer_reply = osd_verify_authorizer_reply,
5669 .invalidate_authorizer = osd_invalidate_authorizer,
79dbd1ba
ID		 5670 .sign_message = osd_sign_message,
5671 .check_message_signature = osd_check_message_signature,
cd1a677c
ID		 5672 .get_auth_request = osd_get_auth_request,
5673 .handle_auth_reply_more = osd_handle_auth_reply_more,
5674 .handle_auth_done = osd_handle_auth_done,
5675 .handle_auth_bad_method = osd_handle_auth_bad_method,
f24e9980 5676};
Linux 6.x block layer and io_uring tree(s)