projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: add sparse read support to OSD client
[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:
f628d799 379 case CEPH_OSD_OP_SPARSE_READ:
5476492f 380 case CEPH_OSD_OP_WRITE:
e30b7577 381 case CEPH_OSD_OP_WRITEFULL:
a679e50f 382 kfree(op->extent.sparse_ext);
5476492f
AE		 383 ceph_osd_data_release(&op->extent.osd_data);
384 break;
385 case CEPH_OSD_OP_CALL:
386 ceph_osd_data_release(&op->cls.request_info);
04017e29 387 ceph_osd_data_release(&op->cls.request_data);
5476492f
AE		 388 ceph_osd_data_release(&op->cls.response_data);
389 break;
d74b50be
YZ		 390 case CEPH_OSD_OP_SETXATTR:
391 case CEPH_OSD_OP_CMPXATTR:
392 ceph_osd_data_release(&op->xattr.osd_data);
393 break;
66ba609f
YZ		 394 case CEPH_OSD_OP_STAT:
395 ceph_osd_data_release(&op->raw_data_in);
396 break;
922dab61
ID		 397 case CEPH_OSD_OP_NOTIFY_ACK:
398 ceph_osd_data_release(&op->notify_ack.request_data);
399 break;
19079203
ID		 400 case CEPH_OSD_OP_NOTIFY:
401 ceph_osd_data_release(&op->notify.request_data);
402 ceph_osd_data_release(&op->notify.response_data);
403 break;
a4ed38d7
DF		 404 case CEPH_OSD_OP_LIST_WATCHERS:
405 ceph_osd_data_release(&op->list_watchers.response_data);
406 break;
78beb0ff 407 case CEPH_OSD_OP_COPY_FROM2:
23ddf9be
LH		 408 ceph_osd_data_release(&op->copy_from.osd_data);
409 break;
5476492f
AE		 410 default:
411 break;
412 }
c54d47bf
AE		 413}
414
63244fa1
ID		 415/*
416 * Assumes @t is zero-initialized.
417 */
418static void target_init(struct ceph_osd_request_target *t)
419{
420 ceph_oid_init(&t->base_oid);
421 ceph_oloc_init(&t->base_oloc);
422 ceph_oid_init(&t->target_oid);
423 ceph_oloc_init(&t->target_oloc);
424
425 ceph_osds_init(&t->acting);
426 ceph_osds_init(&t->up);
427 t->size = -1;
428 t->min_size = -1;
429
430 t->osd = CEPH_HOMELESS_OSD;
431}
432
922dab61
ID		 433static void target_copy(struct ceph_osd_request_target *dest,
434 const struct ceph_osd_request_target *src)
435{
436 ceph_oid_copy(&dest->base_oid, &src->base_oid);
437 ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
438 ceph_oid_copy(&dest->target_oid, &src->target_oid);
439 ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
440
441 dest->pgid = src->pgid; /* struct */
dc98ff72 442 dest->spgid = src->spgid; /* struct */
922dab61
ID		 443 dest->pg_num = src->pg_num;
444 dest->pg_num_mask = src->pg_num_mask;
445 ceph_osds_copy(&dest->acting, &src->acting);
446 ceph_osds_copy(&dest->up, &src->up);
447 dest->size = src->size;
448 dest->min_size = src->min_size;
449 dest->sort_bitwise = src->sort_bitwise;
2f3fead6 450 dest->recovery_deletes = src->recovery_deletes;
922dab61
ID		 451
452 dest->flags = src->flags;
7ed286f3 453 dest->used_replica = src->used_replica;
922dab61
ID		 454 dest->paused = src->paused;
455
04c7d789 456 dest->epoch = src->epoch;
dc93e0e2
ID		 457 dest->last_force_resend = src->last_force_resend;
458
922dab61
ID		 459 dest->osd = src->osd;
460}
461
63244fa1
ID		 462static void target_destroy(struct ceph_osd_request_target *t)
463{
464 ceph_oid_destroy(&t->base_oid);
30c156d9 465 ceph_oloc_destroy(&t->base_oloc);
63244fa1 466 ceph_oid_destroy(&t->target_oid);
30c156d9 467 ceph_oloc_destroy(&t->target_oloc);
63244fa1
ID		 468}
469
f24e9980
SW		 470/*
471 * requests
472 */
3540bfdb
ID		 473static void request_release_checks(struct ceph_osd_request *req)
474{
475 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
4609245e 476 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
94e85771 477 WARN_ON(!list_empty(&req->r_private_item));
3540bfdb
ID		 478 WARN_ON(req->r_osd);
479}
480
9e94af20 481static void ceph_osdc_release_request(struct kref *kref)
f24e9980 482{
9e94af20
ID		 483 struct ceph_osd_request *req = container_of(kref,
484 struct ceph_osd_request, r_kref);
5476492f 485 unsigned int which;
415e49a9 486
9e94af20
ID		 487 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
488 req->r_request, req->r_reply);
3540bfdb 489 request_release_checks(req);
9e94af20 490
415e49a9
SW		 491 if (req->r_request)
492 ceph_msg_put(req->r_request);
5aea3dcd 493 if (req->r_reply)
ab8cb34a 494 ceph_msg_put(req->r_reply);
0fff87ec 495
5476492f
AE		 496 for (which = 0; which < req->r_num_ops; which++)
497 osd_req_op_data_release(req, which);
0fff87ec 498
a66dd383 499 target_destroy(&req->r_t);
415e49a9 500 ceph_put_snap_context(req->r_snapc);
d30291b9 501
415e49a9
SW		 502 if (req->r_mempool)
503 mempool_free(req, req->r_osdc->req_mempool);
3f1af42a 504 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
5522ae0b 505 kmem_cache_free(ceph_osd_request_cache, req);
3f1af42a
ID		 506 else
507 kfree(req);
f24e9980 508}
9e94af20
ID		 509
510void ceph_osdc_get_request(struct ceph_osd_request *req)
511{
512 dout("%s %p (was %d)\n", __func__, req,
2c935bc5 513 kref_read(&req->r_kref));
9e94af20
ID		 514 kref_get(&req->r_kref);
515}
516EXPORT_SYMBOL(ceph_osdc_get_request);
517
518void ceph_osdc_put_request(struct ceph_osd_request *req)
519{
3ed97d63
ID		 520 if (req) {
521 dout("%s %p (was %d)\n", __func__, req,
2c935bc5 522 kref_read(&req->r_kref));
3ed97d63
ID		 523 kref_put(&req->r_kref, ceph_osdc_release_request);
524 }
9e94af20
ID		 525}
526EXPORT_SYMBOL(ceph_osdc_put_request);
68b4476b 527
3540bfdb
ID		 528static void request_init(struct ceph_osd_request *req)
529{
042f6498 530 /* req only, each op is zeroed in osd_req_op_init() */
3540bfdb
ID		 531 memset(req, 0, sizeof(*req));
532
533 kref_init(&req->r_kref);
534 init_completion(&req->r_completion);
3540bfdb 535 RB_CLEAR_NODE(&req->r_node);
4609245e 536 RB_CLEAR_NODE(&req->r_mc_node);
94e85771 537 INIT_LIST_HEAD(&req->r_private_item);
3540bfdb
ID		 538
539 target_init(&req->r_t);
540}
541
3499e8a5 542struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
f24e9980 543 struct ceph_snap_context *snapc,
1b83bef2 544 unsigned int num_ops,
3499e8a5 545 bool use_mempool,
54a54007 546 gfp_t gfp_flags)
f24e9980
SW		 547{
548 struct ceph_osd_request *req;
1b83bef2 549
f24e9980 550 if (use_mempool) {
3f1af42a 551 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
3499e8a5 552 req = mempool_alloc(osdc->req_mempool, gfp_flags);
3f1af42a
ID		 553 } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
554 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
f24e9980 555 } else {
3f1af42a 556 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
acafe7e3 557 req = kmalloc(struct_size(req, r_ops, num_ops), gfp_flags);
f24e9980 558 }
3f1af42a 559 if (unlikely(!req))
a79832f2 560 return NULL;
f24e9980 561
3540bfdb 562 request_init(req);
f24e9980
SW		 563 req->r_osdc = osdc;
564 req->r_mempool = use_mempool;
79528734 565 req->r_num_ops = num_ops;
84127282
ID		 566 req->r_snapid = CEPH_NOSNAP;
567 req->r_snapc = ceph_get_snap_context(snapc);
68b4476b 568
13d1ad16
ID		 569 dout("%s req %p\n", __func__, req);
570 return req;
571}
572EXPORT_SYMBOL(ceph_osdc_alloc_request);
3f1af42a 573
2e59ffd1 574static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
30c156d9
YZ		 575{
576 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
577}
578
0d9c1ab3
ID		 579static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
580 int num_request_data_items,
581 int num_reply_data_items)
13d1ad16
ID		 582{
583 struct ceph_osd_client *osdc = req->r_osdc;
584 struct ceph_msg *msg;
585 int msg_size;
c16e7869 586
0d9c1ab3 587 WARN_ON(req->r_request || req->r_reply);
d30291b9 588 WARN_ON(ceph_oid_empty(&req->r_base_oid));
30c156d9 589 WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
d30291b9 590
13d1ad16 591 /* create request message */
8cb441c0
ID		 592 msg_size = CEPH_ENCODING_START_BLK_LEN +
593 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
594 msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
595 msg_size += CEPH_ENCODING_START_BLK_LEN +
596 sizeof(struct ceph_osd_reqid); /* reqid */
597 msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
598 msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
30c156d9
YZ		 599 msg_size += CEPH_ENCODING_START_BLK_LEN +
600 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
13d1ad16
ID		 601 msg_size += 4 + req->r_base_oid.name_len; /* oid */
602 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
ae458f5a
ID		 603 msg_size += 8; /* snapid */
604 msg_size += 8; /* snap_seq */
13d1ad16 605 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
8cb441c0 606 msg_size += 4 + 8; /* retry_attempt, features */
ae458f5a 607
13d1ad16 608 if (req->r_mempool)
0d9c1ab3
ID		 609 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
610 num_request_data_items);
f24e9980 611 else
0d9c1ab3
ID		 612 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
613 num_request_data_items, gfp, true);
13d1ad16
ID		 614 if (!msg)
615 return -ENOMEM;
68b4476b 616
f24e9980 617 memset(msg->front.iov_base, 0, msg->front.iov_len);
3499e8a5 618 req->r_request = msg;
3499e8a5 619
13d1ad16
ID		 620 /* create reply message */
621 msg_size = OSD_OPREPLY_FRONT_LEN;
711da55d
ID		 622 msg_size += req->r_base_oid.name_len;
623 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
13d1ad16
ID		 624
625 if (req->r_mempool)
0d9c1ab3
ID		 626 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
627 num_reply_data_items);
13d1ad16 628 else
0d9c1ab3
ID		 629 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
630 num_reply_data_items, gfp, true);
13d1ad16
ID		 631 if (!msg)
632 return -ENOMEM;
633
634 req->r_reply = msg;
635
636 return 0;
3499e8a5
YS		 637}
638
a8dd0a37 639static bool osd_req_opcode_valid(u16 opcode)
68b4476b 640{
a8dd0a37 641 switch (opcode) {
70b5bfa3
ID		 642#define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
643__CEPH_FORALL_OSD_OPS(GENERATE_CASE)
644#undef GENERATE_CASE
a8dd0a37
AE		 645 default:
646 return false;
647 }
648}
649
0d9c1ab3
ID		 650static void get_num_data_items(struct ceph_osd_request *req,
651 int *num_request_data_items,
652 int *num_reply_data_items)
653{
654 struct ceph_osd_req_op *op;
655
656 *num_request_data_items = 0;
657 *num_reply_data_items = 0;
658
659 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
660 switch (op->op) {
661 /* request */
662 case CEPH_OSD_OP_WRITE:
663 case CEPH_OSD_OP_WRITEFULL:
664 case CEPH_OSD_OP_SETXATTR:
665 case CEPH_OSD_OP_CMPXATTR:
666 case CEPH_OSD_OP_NOTIFY_ACK:
78beb0ff 667 case CEPH_OSD_OP_COPY_FROM2:
0d9c1ab3
ID		 668 *num_request_data_items += 1;
669 break;
670
671 /* reply */
672 case CEPH_OSD_OP_STAT:
673 case CEPH_OSD_OP_READ:
f628d799 674 case CEPH_OSD_OP_SPARSE_READ:
0d9c1ab3
ID		 675 case CEPH_OSD_OP_LIST_WATCHERS:
676 *num_reply_data_items += 1;
677 break;
678
679 /* both */
680 case CEPH_OSD_OP_NOTIFY:
681 *num_request_data_items += 1;
682 *num_reply_data_items += 1;
683 break;
684 case CEPH_OSD_OP_CALL:
685 *num_request_data_items += 2;
686 *num_reply_data_items += 1;
687 break;
688
689 default:
690 WARN_ON(!osd_req_opcode_valid(op->op));
691 break;
692 }
693 }
694}
695
696/*
697 * oid, oloc and OSD op opcode(s) must be filled in before this function
698 * is called.
699 */
700int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
701{
702 int num_request_data_items, num_reply_data_items;
703
704 get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
705 return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
706 num_reply_data_items);
707}
708EXPORT_SYMBOL(ceph_osdc_alloc_messages);
709
33803f33
AE		 710/*
711 * This is an osd op init function for opcodes that have no data or
712 * other information associated with them. It also serves as a
713 * common init routine for all the other init functions, below.
714 */
042f6498
JL		 715struct ceph_osd_req_op *
716osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
144cba14 717 u16 opcode, u32 flags)
33803f33 718{
c99d2d4a
AE		 719 struct ceph_osd_req_op *op;
720
721 BUG_ON(which >= osd_req->r_num_ops);
33803f33
AE		 722 BUG_ON(!osd_req_opcode_valid(opcode));
723
c99d2d4a 724 op = &osd_req->r_ops[which];
33803f33 725 memset(op, 0, sizeof (*op));
33803f33 726 op->op = opcode;
144cba14 727 op->flags = flags;
c99d2d4a
AE		 728
729 return op;
33803f33 730}
49719778
AE		 731EXPORT_SYMBOL(osd_req_op_init);
732
c99d2d4a
AE		 733void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
734 unsigned int which, u16 opcode,
33803f33
AE		 735 u64 offset, u64 length,
736 u64 truncate_size, u32 truncate_seq)
737{
042f6498
JL		 738 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
739 opcode, 0);
33803f33
AE		 740 size_t payload_len = 0;
741
ad7a60de 742 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
e30b7577 743 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
f628d799 744 opcode != CEPH_OSD_OP_TRUNCATE && opcode != CEPH_OSD_OP_SPARSE_READ);
33803f33 745
33803f33
AE		 746 op->extent.offset = offset;
747 op->extent.length = length;
748 op->extent.truncate_size = truncate_size;
749 op->extent.truncate_seq = truncate_seq;
e30b7577 750 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
33803f33
AE		 751 payload_len += length;
752
de2aa102 753 op->indata_len = payload_len;
33803f33
AE		 754}
755EXPORT_SYMBOL(osd_req_op_extent_init);
756
c99d2d4a
AE		 757void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
758 unsigned int which, u64 length)
e5975c7c 759{
c99d2d4a
AE		 760 struct ceph_osd_req_op *op;
761 u64 previous;
762
763 BUG_ON(which >= osd_req->r_num_ops);
764 op = &osd_req->r_ops[which];
765 previous = op->extent.length;
e5975c7c
AE		 766
767 if (length == previous)
768 return; /* Nothing to do */
769 BUG_ON(length > previous);
770
771 op->extent.length = length;
d641df81
YZ		 772 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
773 op->indata_len -= previous - length;
e5975c7c
AE		 774}
775EXPORT_SYMBOL(osd_req_op_extent_update);
776
2c63f49a
YZ		 777void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
778 unsigned int which, u64 offset_inc)
779{
780 struct ceph_osd_req_op *op, *prev_op;
781
782 BUG_ON(which + 1 >= osd_req->r_num_ops);
783
784 prev_op = &osd_req->r_ops[which];
042f6498 785 op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
2c63f49a
YZ		 786 /* dup previous one */
787 op->indata_len = prev_op->indata_len;
788 op->outdata_len = prev_op->outdata_len;
789 op->extent = prev_op->extent;
790 /* adjust offset */
791 op->extent.offset += offset_inc;
792 op->extent.length -= offset_inc;
793
794 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
795 op->indata_len -= offset_inc;
796}
797EXPORT_SYMBOL(osd_req_op_extent_dup_last);
798
fe943d50 799int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
24639ce5 800 const char *class, const char *method)
33803f33 801{
24639ce5 802 struct ceph_osd_req_op *op;
5f562df5 803 struct ceph_pagelist *pagelist;
33803f33
AE		 804 size_t payload_len = 0;
805 size_t size;
4766815b 806 int ret;
33803f33 807
042f6498 808 op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
33803f33 809
33165d47 810 pagelist = ceph_pagelist_alloc(GFP_NOFS);
fe943d50
CX		 811 if (!pagelist)
812 return -ENOMEM;
813
33803f33
AE		 814 op->cls.class_name = class;
815 size = strlen(class);
816 BUG_ON(size > (size_t) U8_MAX);
817 op->cls.class_len = size;
4766815b
DD		 818 ret = ceph_pagelist_append(pagelist, class, size);
819 if (ret)
820 goto err_pagelist_free;
33803f33
AE		 821 payload_len += size;
822
823 op->cls.method_name = method;
824 size = strlen(method);
825 BUG_ON(size > (size_t) U8_MAX);
826 op->cls.method_len = size;
4766815b
DD		 827 ret = ceph_pagelist_append(pagelist, method, size);
828 if (ret)
829 goto err_pagelist_free;
33803f33
AE		 830 payload_len += size;
831
a4ce40a9 832 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
de2aa102 833 op->indata_len = payload_len;
fe943d50 834 return 0;
4766815b
DD		 835
836err_pagelist_free:
837 ceph_pagelist_release(pagelist);
838 return ret;
33803f33
AE		 839}
840EXPORT_SYMBOL(osd_req_op_cls_init);
8c042b0d 841
d74b50be
YZ		 842int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
843 u16 opcode, const char *name, const void *value,
844 size_t size, u8 cmp_op, u8 cmp_mode)
845{
042f6498
JL		 846 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
847 opcode, 0);
d74b50be
YZ		 848 struct ceph_pagelist *pagelist;
849 size_t payload_len;
4766815b 850 int ret;
d74b50be
YZ		 851
852 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
853
33165d47 854 pagelist = ceph_pagelist_alloc(GFP_NOFS);
d74b50be
YZ		 855 if (!pagelist)
856 return -ENOMEM;
857
d74b50be
YZ		 858 payload_len = strlen(name);
859 op->xattr.name_len = payload_len;
4766815b
DD		 860 ret = ceph_pagelist_append(pagelist, name, payload_len);
861 if (ret)
862 goto err_pagelist_free;
d74b50be
YZ		 863
864 op->xattr.value_len = size;
4766815b
DD		 865 ret = ceph_pagelist_append(pagelist, value, size);
866 if (ret)
867 goto err_pagelist_free;
d74b50be
YZ		 868 payload_len += size;
869
870 op->xattr.cmp_op = cmp_op;
871 op->xattr.cmp_mode = cmp_mode;
872
873 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
de2aa102 874 op->indata_len = payload_len;
d74b50be 875 return 0;
4766815b
DD		 876
877err_pagelist_free:
878 ceph_pagelist_release(pagelist);
879 return ret;
d74b50be
YZ		 880}
881EXPORT_SYMBOL(osd_req_op_xattr_init);
882
922dab61
ID		 883/*
884 * @watch_opcode: CEPH_OSD_WATCH_OP_*
885 */
886static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
75dbb685 887 u8 watch_opcode, u64 cookie, u32 gen)
33803f33 888{
922dab61 889 struct ceph_osd_req_op *op;
33803f33 890
042f6498 891 op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
33803f33 892 op->watch.cookie = cookie;
922dab61 893 op->watch.op = watch_opcode;
75dbb685
ID		 894 op->watch.gen = gen;
895}
896
897/*
898 * prot_ver, timeout and notify payload (may be empty) should already be
899 * encoded in @request_pl
900 */
901static void osd_req_op_notify_init(struct ceph_osd_request *req, int which,
902 u64 cookie, struct ceph_pagelist *request_pl)
903{
904 struct ceph_osd_req_op *op;
905
906 op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
907 op->notify.cookie = cookie;
908
909 ceph_osd_data_pagelist_init(&op->notify.request_data, request_pl);
910 op->indata_len = request_pl->length;
33803f33 911}
33803f33 912
d3798acc
ID		 913/*
914 * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
915 */
c647b8a8
ID		 916void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
917 unsigned int which,
918 u64 expected_object_size,
d3798acc
ID		 919 u64 expected_write_size,
920 u32 flags)
c647b8a8 921{
042f6498 922 struct ceph_osd_req_op *op;
c647b8a8 923
042f6498 924 op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0);
c647b8a8
ID		 925 op->alloc_hint.expected_object_size = expected_object_size;
926 op->alloc_hint.expected_write_size = expected_write_size;
d3798acc 927 op->alloc_hint.flags = flags;
c647b8a8
ID		 928
929 /*
930 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
931 * not worth a feature bit. Set FAILOK per-op flag to make
932 * sure older osds don't trip over an unsupported opcode.
933 */
934 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
935}
936EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
937
90af3602 938static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
ec9123c5
AE		 939 struct ceph_osd_data *osd_data)
940{
941 u64 length = ceph_osd_data_length(osd_data);
942
943 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
944 BUG_ON(length > (u64) SIZE_MAX);
945 if (length)
90af3602 946 ceph_msg_data_add_pages(msg, osd_data->pages,
e8862740 947 length, osd_data->alignment, false);
ec9123c5
AE		 948 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
949 BUG_ON(!length);
90af3602 950 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
ec9123c5
AE		 951#ifdef CONFIG_BLOCK
952 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
5359a17d 953 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
ec9123c5 954#endif
b9e281c2
ID		 955 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
956 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
ec9123c5
AE		 957 } else {
958 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
959 }
960}
961
bb873b53
ID		 962static u32 osd_req_encode_op(struct ceph_osd_op *dst,
963 const struct ceph_osd_req_op *src)
a8dd0a37 964{
a8dd0a37
AE		 965 switch (src->op) {
966 case CEPH_OSD_OP_STAT:
967 break;
968 case CEPH_OSD_OP_READ:
f628d799 969 case CEPH_OSD_OP_SPARSE_READ:
a8dd0a37 970 case CEPH_OSD_OP_WRITE:
e30b7577 971 case CEPH_OSD_OP_WRITEFULL:
ad7a60de 972 case CEPH_OSD_OP_ZERO:
ad7a60de 973 case CEPH_OSD_OP_TRUNCATE:
a8dd0a37
AE		 974 dst->extent.offset = cpu_to_le64(src->extent.offset);
975 dst->extent.length = cpu_to_le64(src->extent.length);
976 dst->extent.truncate_size =
977 cpu_to_le64(src->extent.truncate_size);
978 dst->extent.truncate_seq =
979 cpu_to_le32(src->extent.truncate_seq);
980 break;
981 case CEPH_OSD_OP_CALL:
a8dd0a37
AE		 982 dst->cls.class_len = src->cls.class_len;
983 dst->cls.method_len = src->cls.method_len;
bb873b53 984 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
a8dd0a37 985 break;
a8dd0a37
AE		 986 case CEPH_OSD_OP_WATCH:
987 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
922dab61
ID		 988 dst->watch.ver = cpu_to_le64(0);
989 dst->watch.op = src->watch.op;
990 dst->watch.gen = cpu_to_le32(src->watch.gen);
991 break;
992 case CEPH_OSD_OP_NOTIFY_ACK:
a8dd0a37 993 break;
19079203
ID		 994 case CEPH_OSD_OP_NOTIFY:
995 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
996 break;
a4ed38d7
DF		 997 case CEPH_OSD_OP_LIST_WATCHERS:
998 break;
c647b8a8
ID		 999 case CEPH_OSD_OP_SETALLOCHINT:
1000 dst->alloc_hint.expected_object_size =
1001 cpu_to_le64(src->alloc_hint.expected_object_size);
1002 dst->alloc_hint.expected_write_size =
1003 cpu_to_le64(src->alloc_hint.expected_write_size);
d3798acc 1004 dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
c647b8a8 1005 break;
d74b50be
YZ		 1006 case CEPH_OSD_OP_SETXATTR:
1007 case CEPH_OSD_OP_CMPXATTR:
1008 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1009 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1010 dst->xattr.cmp_op = src->xattr.cmp_op;
1011 dst->xattr.cmp_mode = src->xattr.cmp_mode;
d74b50be 1012 break;
864e9197
YZ		 1013 case CEPH_OSD_OP_CREATE:
1014 case CEPH_OSD_OP_DELETE:
1015 break;
78beb0ff 1016 case CEPH_OSD_OP_COPY_FROM2:
23ddf9be
LH		 1017 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1018 dst->copy_from.src_version =
1019 cpu_to_le64(src->copy_from.src_version);
1020 dst->copy_from.flags = src->copy_from.flags;
1021 dst->copy_from.src_fadvise_flags =
1022 cpu_to_le32(src->copy_from.src_fadvise_flags);
1023 break;
a8dd0a37 1024 default:
4c46459c 1025 pr_err("unsupported osd opcode %s\n",
8f63ca2d 1026 ceph_osd_op_name(src->op));
4c46459c 1027 WARN_ON(1);
a8dd0a37
AE		 1028
1029 return 0;
68b4476b 1030 }
7b25bf5f 1031
a8dd0a37 1032 dst->op = cpu_to_le16(src->op);
7b25bf5f 1033 dst->flags = cpu_to_le32(src->flags);
de2aa102 1034 dst->payload_len = cpu_to_le32(src->indata_len);
175face2 1035
bb873b53 1036 return src->indata_len;
68b4476b
YS		 1037}
1038
3499e8a5
YS		 1039/*
1040 * build new request AND message, calculate layout, and adjust file
1041 * extent as needed.
1042 *
1043 * if the file was recently truncated, we include information about its
1044 * old and new size so that the object can be updated appropriately. (we
1045 * avoid synchronously deleting truncated objects because it's slow.)
3499e8a5
YS		 1046 */
1047struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1048 struct ceph_file_layout *layout,
1049 struct ceph_vino vino,
715e4cd4
YZ		 1050 u64 off, u64 *plen,
1051 unsigned int which, int num_ops,
3499e8a5
YS		 1052 int opcode, int flags,
1053 struct ceph_snap_context *snapc,
3499e8a5
YS		 1054 u32 truncate_seq,
1055 u64 truncate_size,
153e5167 1056 bool use_mempool)
3499e8a5 1057{
68b4476b 1058 struct ceph_osd_request *req;
75d1c941
AE		 1059 u64 objnum = 0;
1060 u64 objoff = 0;
1061 u64 objlen = 0;
6816282d 1062 int r;
68b4476b 1063
ad7a60de 1064 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
864e9197 1065 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
f628d799
JL		 1066 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE &&
1067 opcode != CEPH_OSD_OP_SPARSE_READ);
68b4476b 1068
acead002 1069 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
ae7ca4a3 1070 GFP_NOFS);
13d1ad16
ID		 1071 if (!req) {
1072 r = -ENOMEM;
1073 goto fail;
1074 }
79528734 1075
3499e8a5 1076 /* calculate max write size */
a19dadfb 1077 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
13d1ad16
ID		 1078 if (r)
1079 goto fail;
a19dadfb 1080
864e9197 1081 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
144cba14 1082 osd_req_op_init(req, which, opcode, 0);
864e9197 1083 } else {
7627151e 1084 u32 object_size = layout->object_size;
864e9197
YZ		 1085 u32 object_base = off - objoff;
1086 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1087 if (truncate_size <= object_base) {
1088 truncate_size = 0;
1089 } else {
1090 truncate_size -= object_base;
1091 if (truncate_size > object_size)
1092 truncate_size = object_size;
1093 }
ccca4e37 1094 }
715e4cd4 1095 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
864e9197
YZ		 1096 truncate_size, truncate_seq);
1097 }
d18d1e28 1098
7627151e 1099 req->r_base_oloc.pool = layout->pool_id;
30c156d9 1100 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
d30291b9 1101 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
22d2cfdf 1102 req->r_flags = flags | osdc->client->options->read_from_replica;
dbe0fc41 1103
bb873b53
ID		 1104 req->r_snapid = vino.snap;
1105 if (flags & CEPH_OSD_FLAG_WRITE)
1106 req->r_data_offset = off;
1107
0d9c1ab3
ID		 1108 if (num_ops > 1)
1109 /*
1110 * This is a special case for ceph_writepages_start(), but it
1111 * also covers ceph_uninline_data(). If more multi-op request
1112 * use cases emerge, we will need a separate helper.
1113 */
1114 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_ops, 0);
1115 else
1116 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
13d1ad16
ID		 1117 if (r)
1118 goto fail;
1119
f24e9980 1120 return req;
13d1ad16
ID		 1121
1122fail:
1123 ceph_osdc_put_request(req);
1124 return ERR_PTR(r);
f24e9980 1125}
3d14c5d2 1126EXPORT_SYMBOL(ceph_osdc_new_request);
f24e9980 1127
a679e50f
JL		 1128int __ceph_alloc_sparse_ext_map(struct ceph_osd_req_op *op, int cnt)
1129{
1130 op->extent.sparse_ext_cnt = cnt;
1131 op->extent.sparse_ext = kmalloc_array(cnt,
1132 sizeof(*op->extent.sparse_ext),
1133 GFP_NOFS);
1134 if (!op->extent.sparse_ext)
1135 return -ENOMEM;
1136 return 0;
1137}
1138EXPORT_SYMBOL(__ceph_alloc_sparse_ext_map);
1139
f24e9980
SW		 1140/*
1141 * We keep osd requests in an rbtree, sorted by ->r_tid.
1142 */
fcd00b68 1143DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
4609245e 1144DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
f24e9980 1145
66850df5
ID		 1146/*
1147 * Call @fn on each OSD request as long as @fn returns 0.
1148 */
1149static void for_each_request(struct ceph_osd_client *osdc,
1150 int (*fn)(struct ceph_osd_request *req, void *arg),
1151 void *arg)
1152{
1153 struct rb_node *n, *p;
1154
1155 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1156 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1157
1158 for (p = rb_first(&osd->o_requests); p; ) {
1159 struct ceph_osd_request *req =
1160 rb_entry(p, struct ceph_osd_request, r_node);
1161
1162 p = rb_next(p);
1163 if (fn(req, arg))
1164 return;
1165 }
1166 }
1167
1168 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1169 struct ceph_osd_request *req =
1170 rb_entry(p, struct ceph_osd_request, r_node);
1171
1172 p = rb_next(p);
1173 if (fn(req, arg))
1174 return;
1175 }
1176}
1177
0247a0cf
ID		 1178static bool osd_homeless(struct ceph_osd *osd)
1179{
1180 return osd->o_osd == CEPH_HOMELESS_OSD;
1181}
1182
5aea3dcd 1183static bool osd_registered(struct ceph_osd *osd)
f24e9980 1184{
5aea3dcd 1185 verify_osdc_locked(osd->o_osdc);
f24e9980 1186
5aea3dcd 1187 return !RB_EMPTY_NODE(&osd->o_node);
f24e9980
SW		 1188}
1189
0247a0cf
ID		 1190/*
1191 * Assumes @osd is zero-initialized.
1192 */
1193static void osd_init(struct ceph_osd *osd)
1194{
02113a0f 1195 refcount_set(&osd->o_ref, 1);
0247a0cf 1196 RB_CLEAR_NODE(&osd->o_node);
08b8a044 1197 spin_lock_init(&osd->o_requests_lock);
5aea3dcd 1198 osd->o_requests = RB_ROOT;
922dab61 1199 osd->o_linger_requests = RB_ROOT;
a02a946d
ID		 1200 osd->o_backoff_mappings = RB_ROOT;
1201 osd->o_backoffs_by_id = RB_ROOT;
0247a0cf
ID		 1202 INIT_LIST_HEAD(&osd->o_osd_lru);
1203 INIT_LIST_HEAD(&osd->o_keepalive_item);
1204 osd->o_incarnation = 1;
5aea3dcd 1205 mutex_init(&osd->lock);
0247a0cf
ID		 1206}
1207
f628d799
JL		 1208static void ceph_init_sparse_read(struct ceph_sparse_read *sr)
1209{
1210 kfree(sr->sr_extent);
1211 memset(sr, '\0', sizeof(*sr));
1212 sr->sr_state = CEPH_SPARSE_READ_HDR;
1213}
1214
0247a0cf
ID		 1215static void osd_cleanup(struct ceph_osd *osd)
1216{
1217 WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
5aea3dcd 1218 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
922dab61 1219 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
a02a946d
ID		 1220 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1221 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
0247a0cf
ID		 1222 WARN_ON(!list_empty(&osd->o_osd_lru));
1223 WARN_ON(!list_empty(&osd->o_keepalive_item));
1224
f628d799
JL		 1225 ceph_init_sparse_read(&osd->o_sparse_read);
1226
0247a0cf
ID		 1227 if (osd->o_auth.authorizer) {
1228 WARN_ON(osd_homeless(osd));
1229 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1230 }
1231}
1232
f24e9980
SW		 1233/*
1234 * Track open sessions with osds.
1235 */
e10006f8 1236static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
f24e9980
SW		 1237{
1238 struct ceph_osd *osd;
1239
0247a0cf
ID		 1240 WARN_ON(onum == CEPH_HOMELESS_OSD);
1241
7a28f59b 1242 osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL);
0247a0cf 1243 osd_init(osd);
f24e9980 1244 osd->o_osdc = osdc;
e10006f8 1245 osd->o_osd = onum;
f628d799
JL		 1246 osd->o_sparse_op_idx = -1;
1247
1248 ceph_init_sparse_read(&osd->o_sparse_read);
f24e9980 1249
b7a9e5dd 1250 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
4e7a5dcd 1251
f24e9980
SW		 1252 return osd;
1253}
1254
1255static struct ceph_osd *get_osd(struct ceph_osd *osd)
1256{
02113a0f
ER		 1257 if (refcount_inc_not_zero(&osd->o_ref)) {
1258 dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1,
1259 refcount_read(&osd->o_ref));
f24e9980
SW		 1260 return osd;
1261 } else {
1262 dout("get_osd %p FAIL\n", osd);
1263 return NULL;
1264 }
1265}
1266
1267static void put_osd(struct ceph_osd *osd)
1268{
02113a0f
ER		 1269 dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref),
1270 refcount_read(&osd->o_ref) - 1);
1271 if (refcount_dec_and_test(&osd->o_ref)) {
0247a0cf 1272 osd_cleanup(osd);
f24e9980 1273 kfree(osd);
79494d1b 1274 }
f24e9980
SW		 1275}
1276
fcd00b68
ID		 1277DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1278
9dd2845c 1279static void __move_osd_to_lru(struct ceph_osd *osd)
f5a2041b 1280{
9dd2845c
ID		 1281 struct ceph_osd_client *osdc = osd->o_osdc;
1282
1283 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
f5a2041b 1284 BUG_ON(!list_empty(&osd->o_osd_lru));
bbf37ec3 1285
9dd2845c 1286 spin_lock(&osdc->osd_lru_lock);
f5a2041b 1287 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
9dd2845c
ID		 1288 spin_unlock(&osdc->osd_lru_lock);
1289
a319bf56 1290 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
f5a2041b
YS		 1291}
1292
9dd2845c 1293static void maybe_move_osd_to_lru(struct ceph_osd *osd)
bbf37ec3 1294{
5aea3dcd 1295 if (RB_EMPTY_ROOT(&osd->o_requests) &&
922dab61 1296 RB_EMPTY_ROOT(&osd->o_linger_requests))
9dd2845c 1297 __move_osd_to_lru(osd);
bbf37ec3
ID		 1298}
1299
f5a2041b
YS		 1300static void __remove_osd_from_lru(struct ceph_osd *osd)
1301{
9dd2845c
ID		 1302 struct ceph_osd_client *osdc = osd->o_osdc;
1303
1304 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1305
1306 spin_lock(&osdc->osd_lru_lock);
f5a2041b
YS		 1307 if (!list_empty(&osd->o_osd_lru))
1308 list_del_init(&osd->o_osd_lru);
9dd2845c 1309 spin_unlock(&osdc->osd_lru_lock);
f5a2041b
YS		 1310}
1311
5aea3dcd
ID		 1312/*
1313 * Close the connection and assign any leftover requests to the
1314 * homeless session.
1315 */
1316static void close_osd(struct ceph_osd *osd)
1317{
1318 struct ceph_osd_client *osdc = osd->o_osdc;
1319 struct rb_node *n;
1320
1321 verify_osdc_wrlocked(osdc);
1322 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1323
1324 ceph_con_close(&osd->o_con);
1325
1326 for (n = rb_first(&osd->o_requests); n; ) {
1327 struct ceph_osd_request *req =
1328 rb_entry(n, struct ceph_osd_request, r_node);
1329
1330 n = rb_next(n); /* unlink_request() */
1331
1332 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1333 unlink_request(osd, req);
1334 link_request(&osdc->homeless_osd, req);
1335 }
922dab61
ID		 1336 for (n = rb_first(&osd->o_linger_requests); n; ) {
1337 struct ceph_osd_linger_request *lreq =
1338 rb_entry(n, struct ceph_osd_linger_request, node);
1339
1340 n = rb_next(n); /* unlink_linger() */
1341
1342 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1343 lreq->linger_id);
1344 unlink_linger(osd, lreq);
1345 link_linger(&osdc->homeless_osd, lreq);
1346 }
a02a946d 1347 clear_backoffs(osd);
5aea3dcd
ID		 1348
1349 __remove_osd_from_lru(osd);
1350 erase_osd(&osdc->osds, osd);
1351 put_osd(osd);
1352}
1353
f24e9980
SW		 1354/*
1355 * reset osd connect
1356 */
5aea3dcd 1357static int reopen_osd(struct ceph_osd *osd)
f24e9980 1358{
c3acb181 1359 struct ceph_entity_addr *peer_addr;
f24e9980 1360
5aea3dcd
ID		 1361 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1362
1363 if (RB_EMPTY_ROOT(&osd->o_requests) &&
922dab61 1364 RB_EMPTY_ROOT(&osd->o_linger_requests)) {
5aea3dcd 1365 close_osd(osd);
c3acb181
AE		 1366 return -ENODEV;
1367 }
1368
5aea3dcd 1369 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
c3acb181
AE		 1370 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1371 !ceph_con_opened(&osd->o_con)) {
5aea3dcd 1372 struct rb_node *n;
c3acb181 1373
0b4af2e8
ID		 1374 dout("osd addr hasn't changed and connection never opened, "
1375 "letting msgr retry\n");
87b315a5 1376 /* touch each r_stamp for handle_timeout()'s benfit */
5aea3dcd
ID		 1377 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1378 struct ceph_osd_request *req =
1379 rb_entry(n, struct ceph_osd_request, r_node);
87b315a5 1380 req->r_stamp = jiffies;
5aea3dcd 1381 }
c3acb181
AE		 1382
1383 return -EAGAIN;
f24e9980 1384 }
c3acb181
AE		 1385
1386 ceph_con_close(&osd->o_con);
1387 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1388 osd->o_incarnation++;
1389
1390 return 0;
f24e9980
SW		 1391}
1392
5aea3dcd
ID		 1393static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1394 bool wrlocked)
f24e9980 1395{
5aea3dcd 1396 struct ceph_osd *osd;
35f9f8a0 1397
5aea3dcd
ID		 1398 if (wrlocked)
1399 verify_osdc_wrlocked(osdc);
1400 else
1401 verify_osdc_locked(osdc);
f24e9980 1402
5aea3dcd
ID		 1403 if (o != CEPH_HOMELESS_OSD)
1404 osd = lookup_osd(&osdc->osds, o);
1405 else
1406 osd = &osdc->homeless_osd;
1407 if (!osd) {
1408 if (!wrlocked)
1409 return ERR_PTR(-EAGAIN);
0ba6478d 1410
5aea3dcd
ID		 1411 osd = create_osd(osdc, o);
1412 insert_osd(&osdc->osds, osd);
1413 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1414 &osdc->osdmap->osd_addr[osd->o_osd]);
0ba6478d 1415 }
f24e9980 1416
5aea3dcd
ID		 1417 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1418 return osd;
f24e9980
SW		 1419}
1420
1421/*
5aea3dcd
ID		 1422 * Create request <-> OSD session relation.
1423 *
1424 * @req has to be assigned a tid, @osd may be homeless.
f24e9980 1425 */
5aea3dcd 1426static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
f24e9980 1427{
5aea3dcd
ID		 1428 verify_osd_locked(osd);
1429 WARN_ON(!req->r_tid || req->r_osd);
1430 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1431 req, req->r_tid);
1432
1433 if (!osd_homeless(osd))
1434 __remove_osd_from_lru(osd);
1435 else
1436 atomic_inc(&osd->o_osdc->num_homeless);
1437
1438 get_osd(osd);
08b8a044 1439 spin_lock(&osd->o_requests_lock);
5aea3dcd 1440 insert_request(&osd->o_requests, req);
08b8a044 1441 spin_unlock(&osd->o_requests_lock);
5aea3dcd 1442 req->r_osd = osd;
f24e9980
SW		 1443}
1444
5aea3dcd
ID		 1445static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1446{
1447 verify_osd_locked(osd);
1448 WARN_ON(req->r_osd != osd);
1449 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1450 req, req->r_tid);
1451
1452 req->r_osd = NULL;
08b8a044 1453 spin_lock(&osd->o_requests_lock);
5aea3dcd 1454 erase_request(&osd->o_requests, req);
08b8a044 1455 spin_unlock(&osd->o_requests_lock);
5aea3dcd
ID		 1456 put_osd(osd);
1457
1458 if (!osd_homeless(osd))
1459 maybe_move_osd_to_lru(osd);
1460 else
1461 atomic_dec(&osd->o_osdc->num_homeless);
1462}
1463
63244fa1
ID		 1464static bool __pool_full(struct ceph_pg_pool_info *pi)
1465{
1466 return pi->flags & CEPH_POOL_FLAG_FULL;
1467}
1468
42c1b124
ID		 1469static bool have_pool_full(struct ceph_osd_client *osdc)
1470{
1471 struct rb_node *n;
1472
1473 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1474 struct ceph_pg_pool_info *pi =
1475 rb_entry(n, struct ceph_pg_pool_info, node);
1476
1477 if (__pool_full(pi))
1478 return true;
1479 }
1480
1481 return false;
1482}
1483
5aea3dcd
ID		 1484static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1485{
1486 struct ceph_pg_pool_info *pi;
1487
1488 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1489 if (!pi)
1490 return false;
1491
1492 return __pool_full(pi);
1493}
1494
d29adb34
JD		 1495/*
1496 * Returns whether a request should be blocked from being sent
1497 * based on the current osdmap and osd_client settings.
d29adb34 1498 */
63244fa1
ID		 1499static bool target_should_be_paused(struct ceph_osd_client *osdc,
1500 const struct ceph_osd_request_target *t,
1501 struct ceph_pg_pool_info *pi)
1502{
b7ec35b3
ID		 1503 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1504 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1505 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
63244fa1
ID		 1506 __pool_full(pi);
1507
6d637a54 1508 WARN_ON(pi->id != t->target_oloc.pool);
58eb7932
JL		 1509 return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1510 ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1511 (osdc->osdmap->epoch < osdc->epoch_barrier);
63244fa1
ID		 1512}
1513
117d96a0
ID		 1514static int pick_random_replica(const struct ceph_osds *acting)
1515{
8032bf12 1516 int i = get_random_u32_below(acting->size);
117d96a0
ID		 1517
1518 dout("%s picked osd%d, primary osd%d\n", __func__,
1519 acting->osds[i], acting->primary);
1520 return i;
1521}
1522
1523/*
1524 * Picks the closest replica based on client's location given by
1525 * crush_location option. Prefers the primary if the locality is
1526 * the same.
1527 */
1528static int pick_closest_replica(struct ceph_osd_client *osdc,
1529 const struct ceph_osds *acting)
1530{
1531 struct ceph_options *opt = osdc->client->options;
1532 int best_i, best_locality;
1533 int i = 0, locality;
1534
1535 do {
1536 locality = ceph_get_crush_locality(osdc->osdmap,
1537 acting->osds[i],
1538 &opt->crush_locs);
1539 if (i == 0 ||
1540 (locality >= 0 && best_locality < 0) ||
1541 (locality >= 0 && best_locality >= 0 &&
1542 locality < best_locality)) {
1543 best_i = i;
1544 best_locality = locality;
1545 }
1546 } while (++i < acting->size);
1547
1548 dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1549 acting->osds[best_i], best_locality, acting->primary);
1550 return best_i;
1551}
1552
63244fa1
ID		 1553enum calc_target_result {
1554 CALC_TARGET_NO_ACTION = 0,
1555 CALC_TARGET_NEED_RESEND,
1556 CALC_TARGET_POOL_DNE,
1557};
1558
1559static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1560 struct ceph_osd_request_target *t,
63244fa1
ID		 1561 bool any_change)
1562{
1563 struct ceph_pg_pool_info *pi;
1564 struct ceph_pg pgid, last_pgid;
1565 struct ceph_osds up, acting;
117d96a0
ID		 1566 bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1567 bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
63244fa1 1568 bool force_resend = false;
84ed45df 1569 bool unpaused = false;
a5613724 1570 bool legacy_change = false;
7de030d6 1571 bool split = false;
b7ec35b3 1572 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
ae78dd81
ID		 1573 bool recovery_deletes = ceph_osdmap_flag(osdc,
1574 CEPH_OSDMAP_RECOVERY_DELETES);
63244fa1 1575 enum calc_target_result ct_res;
63244fa1 1576
04c7d789 1577 t->epoch = osdc->osdmap->epoch;
63244fa1
ID		 1578 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1579 if (!pi) {
1580 t->osd = CEPH_HOMELESS_OSD;
1581 ct_res = CALC_TARGET_POOL_DNE;
1582 goto out;
1583 }
1584
1585 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
dc93e0e2
ID		 1586 if (t->last_force_resend < pi->last_force_request_resend) {
1587 t->last_force_resend = pi->last_force_request_resend;
63244fa1 1588 force_resend = true;
dc93e0e2 1589 } else if (t->last_force_resend == 0) {
63244fa1
ID		 1590 force_resend = true;
1591 }
1592 }
63244fa1 1593
db098ec4
ID		 1594 /* apply tiering */
1595 ceph_oid_copy(&t->target_oid, &t->base_oid);
1596 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1597 if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
117d96a0 1598 if (is_read && pi->read_tier >= 0)
63244fa1 1599 t->target_oloc.pool = pi->read_tier;
117d96a0 1600 if (is_write && pi->write_tier >= 0)
63244fa1 1601 t->target_oloc.pool = pi->write_tier;
6d637a54
ID		 1602
1603 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1604 if (!pi) {
1605 t->osd = CEPH_HOMELESS_OSD;
1606 ct_res = CALC_TARGET_POOL_DNE;
1607 goto out;
1608 }
63244fa1
ID		 1609 }
1610
a86f009f 1611 __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
63244fa1
ID		 1612 last_pgid.pool = pgid.pool;
1613 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1614
df28152d 1615 ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
63244fa1
ID		 1616 if (any_change &&
1617 ceph_is_new_interval(&t->acting,
1618 &acting,
1619 &t->up,
1620 &up,
1621 t->size,
1622 pi->size,
1623 t->min_size,
1624 pi->min_size,
1625 t->pg_num,
1626 pi->pg_num,
1627 t->sort_bitwise,
1628 sort_bitwise,
ae78dd81
ID		 1629 t->recovery_deletes,
1630 recovery_deletes,
63244fa1
ID		 1631 &last_pgid))
1632 force_resend = true;
1633
1634 if (t->paused && !target_should_be_paused(osdc, t, pi)) {
1635 t->paused = false;
84ed45df 1636 unpaused = true;
63244fa1 1637 }
84ed45df 1638 legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
117d96a0
ID		 1639 ceph_osds_changed(&t->acting, &acting,
1640 t->used_replica || any_change);
7de030d6
ID		 1641 if (t->pg_num)
1642 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
63244fa1 1643
7de030d6 1644 if (legacy_change || force_resend || split) {
63244fa1 1645 t->pgid = pgid; /* struct */
df28152d 1646 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
63244fa1
ID		 1647 ceph_osds_copy(&t->acting, &acting);
1648 ceph_osds_copy(&t->up, &up);
1649 t->size = pi->size;
1650 t->min_size = pi->min_size;
1651 t->pg_num = pi->pg_num;
1652 t->pg_num_mask = pi->pg_num_mask;
1653 t->sort_bitwise = sort_bitwise;
ae78dd81 1654 t->recovery_deletes = recovery_deletes;
63244fa1 1655
117d96a0
ID		 1656 if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1657 CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1658 !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1659 acting.size > 1) {
1660 int pos;
1661
1662 WARN_ON(!is_read || acting.osds[0] != acting.primary);
1663 if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1664 pos = pick_random_replica(&acting);
1665 } else {
1666 pos = pick_closest_replica(osdc, &acting);
1667 }
1668 t->osd = acting.osds[pos];
1669 t->used_replica = pos > 0;
1670 } else {
1671 t->osd = acting.primary;
1672 t->used_replica = false;
1673 }
63244fa1
ID		 1674 }
1675
a5613724 1676 if (unpaused || legacy_change || force_resend || split)
84ed45df
ID		 1677 ct_res = CALC_TARGET_NEED_RESEND;
1678 else
1679 ct_res = CALC_TARGET_NO_ACTION;
1680
63244fa1 1681out:
a5613724
ID		 1682 dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1683 legacy_change, force_resend, split, ct_res, t->osd);
63244fa1
ID		 1684 return ct_res;
1685}
1686
a02a946d
ID		 1687static struct ceph_spg_mapping *alloc_spg_mapping(void)
1688{
1689 struct ceph_spg_mapping *spg;
1690
1691 spg = kmalloc(sizeof(*spg), GFP_NOIO);
1692 if (!spg)
1693 return NULL;
1694
1695 RB_CLEAR_NODE(&spg->node);
1696 spg->backoffs = RB_ROOT;
1697 return spg;
1698}
1699
1700static void free_spg_mapping(struct ceph_spg_mapping *spg)
1701{
1702 WARN_ON(!RB_EMPTY_NODE(&spg->node));
1703 WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1704
1705 kfree(spg);
1706}
1707
1708/*
1709 * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1710 * ceph_pg_mapping. Used to track OSD backoffs -- a backoff [range] is
1711 * defined only within a specific spgid; it does not pass anything to
1712 * children on split, or to another primary.
1713 */
1714DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1715 RB_BYPTR, const struct ceph_spg *, node)
1716
1717static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1718{
1719 return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1720}
1721
1722static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1723 void **pkey, size_t *pkey_len)
1724{
1725 if (hoid->key_len) {
1726 *pkey = hoid->key;
1727 *pkey_len = hoid->key_len;
1728 } else {
1729 *pkey = hoid->oid;
1730 *pkey_len = hoid->oid_len;
1731 }
1732}
1733
1734static int compare_names(const void *name1, size_t name1_len,
1735 const void *name2, size_t name2_len)
1736{
1737 int ret;
1738
1739 ret = memcmp(name1, name2, min(name1_len, name2_len));
1740 if (!ret) {
1741 if (name1_len < name2_len)
1742 ret = -1;
1743 else if (name1_len > name2_len)
1744 ret = 1;
1745 }
1746 return ret;
1747}
1748
1749static int hoid_compare(const struct ceph_hobject_id *lhs,
1750 const struct ceph_hobject_id *rhs)
1751{
1752 void *effective_key1, *effective_key2;
1753 size_t effective_key1_len, effective_key2_len;
1754 int ret;
1755
1756 if (lhs->is_max < rhs->is_max)
1757 return -1;
1758 if (lhs->is_max > rhs->is_max)
1759 return 1;
1760
1761 if (lhs->pool < rhs->pool)
1762 return -1;
1763 if (lhs->pool > rhs->pool)
1764 return 1;
1765
1766 if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1767 return -1;
1768 if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1769 return 1;
1770
1771 ret = compare_names(lhs->nspace, lhs->nspace_len,
1772 rhs->nspace, rhs->nspace_len);
1773 if (ret)
1774 return ret;
1775
1776 hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1777 hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1778 ret = compare_names(effective_key1, effective_key1_len,
1779 effective_key2, effective_key2_len);
1780 if (ret)
1781 return ret;
1782
1783 ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1784 if (ret)
1785 return ret;
1786
1787 if (lhs->snapid < rhs->snapid)
1788 return -1;
1789 if (lhs->snapid > rhs->snapid)
1790 return 1;
1791
1792 return 0;
1793}
1794
1795/*
1796 * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1797 * compat stuff here.
1798 *
1799 * Assumes @hoid is zero-initialized.
1800 */
1801static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1802{
1803 u8 struct_v;
1804 u32 struct_len;
1805 int ret;
1806
1807 ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1808 &struct_len);
1809 if (ret)
1810 return ret;
1811
1812 if (struct_v < 4) {
1813 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1814 goto e_inval;
1815 }
1816
1817 hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1818 GFP_NOIO);
1819 if (IS_ERR(hoid->key)) {
1820 ret = PTR_ERR(hoid->key);
1821 hoid->key = NULL;
1822 return ret;
1823 }
1824
1825 hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1826 GFP_NOIO);
1827 if (IS_ERR(hoid->oid)) {
1828 ret = PTR_ERR(hoid->oid);
1829 hoid->oid = NULL;
1830 return ret;
1831 }
1832
1833 ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1834 ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1835 ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1836
1837 hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1838 GFP_NOIO);
1839 if (IS_ERR(hoid->nspace)) {
1840 ret = PTR_ERR(hoid->nspace);
1841 hoid->nspace = NULL;
1842 return ret;
1843 }
1844
1845 ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1846
1847 ceph_hoid_build_hash_cache(hoid);
1848 return 0;
1849
1850e_inval:
1851 return -EINVAL;
1852}
1853
1854static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1855{
1856 return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1857 4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1858}
1859
1860static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1861{
1862 ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1863 ceph_encode_string(p, end, hoid->key, hoid->key_len);
1864 ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1865 ceph_encode_64(p, hoid->snapid);
1866 ceph_encode_32(p, hoid->hash);
1867 ceph_encode_8(p, hoid->is_max);
1868 ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1869 ceph_encode_64(p, hoid->pool);
1870}
1871
1872static void free_hoid(struct ceph_hobject_id *hoid)
1873{
1874 if (hoid) {
1875 kfree(hoid->key);
1876 kfree(hoid->oid);
1877 kfree(hoid->nspace);
1878 kfree(hoid);
1879 }
1880}
1881
1882static struct ceph_osd_backoff *alloc_backoff(void)
1883{
1884 struct ceph_osd_backoff *backoff;
1885
1886 backoff = kzalloc(sizeof(*backoff), GFP_NOIO);
1887 if (!backoff)
1888 return NULL;
1889
1890 RB_CLEAR_NODE(&backoff->spg_node);
1891 RB_CLEAR_NODE(&backoff->id_node);
1892 return backoff;
1893}
1894
1895static void free_backoff(struct ceph_osd_backoff *backoff)
1896{
1897 WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1898 WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1899
1900 free_hoid(backoff->begin);
1901 free_hoid(backoff->end);
1902 kfree(backoff);
1903}
1904
1905/*
1906 * Within a specific spgid, backoffs are managed by ->begin hoid.
1907 */
1908DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1909 RB_BYVAL, spg_node);
1910
1911static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1912 const struct ceph_hobject_id *hoid)
1913{
1914 struct rb_node *n = root->rb_node;
1915
1916 while (n) {
1917 struct ceph_osd_backoff *cur =
1918 rb_entry(n, struct ceph_osd_backoff, spg_node);
1919 int cmp;
1920
1921 cmp = hoid_compare(hoid, cur->begin);
1922 if (cmp < 0) {
1923 n = n->rb_left;
1924 } else if (cmp > 0) {
1925 if (hoid_compare(hoid, cur->end) < 0)
1926 return cur;
1927
1928 n = n->rb_right;
1929 } else {
1930 return cur;
1931 }
1932 }
1933
1934 return NULL;
1935}
1936
1937/*
1938 * Each backoff has a unique id within its OSD session.
1939 */
1940DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1941
1942static void clear_backoffs(struct ceph_osd *osd)
1943{
1944 while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1945 struct ceph_spg_mapping *spg =
1946 rb_entry(rb_first(&osd->o_backoff_mappings),
1947 struct ceph_spg_mapping, node);
1948
1949 while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1950 struct ceph_osd_backoff *backoff =
1951 rb_entry(rb_first(&spg->backoffs),
1952 struct ceph_osd_backoff, spg_node);
1953
1954 erase_backoff(&spg->backoffs, backoff);
1955 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1956 free_backoff(backoff);
1957 }
1958 erase_spg_mapping(&osd->o_backoff_mappings, spg);
1959 free_spg_mapping(spg);
1960 }
1961}
1962
1963/*
1964 * Set up a temporary, non-owning view into @t.
1965 */
1966static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1967 const struct ceph_osd_request_target *t)
1968{
1969 hoid->key = NULL;
1970 hoid->key_len = 0;
1971 hoid->oid = t->target_oid.name;
1972 hoid->oid_len = t->target_oid.name_len;
1973 hoid->snapid = CEPH_NOSNAP;
1974 hoid->hash = t->pgid.seed;
1975 hoid->is_max = false;
1976 if (t->target_oloc.pool_ns) {
1977 hoid->nspace = t->target_oloc.pool_ns->str;
1978 hoid->nspace_len = t->target_oloc.pool_ns->len;
1979 } else {
1980 hoid->nspace = NULL;
1981 hoid->nspace_len = 0;
1982 }
1983 hoid->pool = t->target_oloc.pool;
1984 ceph_hoid_build_hash_cache(hoid);
1985}
1986
1987static bool should_plug_request(struct ceph_osd_request *req)
1988{
1989 struct ceph_osd *osd = req->r_osd;
1990 struct ceph_spg_mapping *spg;
1991 struct ceph_osd_backoff *backoff;
1992 struct ceph_hobject_id hoid;
1993
1994 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
1995 if (!spg)
1996 return false;
1997
1998 hoid_fill_from_target(&hoid, &req->r_t);
1999 backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
2000 if (!backoff)
2001 return false;
2002
2003 dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
2004 __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
2005 backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
2006 return true;
2007}
2008
0d9c1ab3
ID		 2009/*
2010 * Keep get_num_data_items() in sync with this function.
2011 */
98c4bfe9 2012static void setup_request_data(struct ceph_osd_request *req)
f24e9980 2013{
98c4bfe9
ID		 2014 struct ceph_msg *request_msg = req->r_request;
2015 struct ceph_msg *reply_msg = req->r_reply;
2016 struct ceph_osd_req_op *op;
bb873b53 2017
98c4bfe9 2018 if (req->r_request->num_data_items || req->r_reply->num_data_items)
bb873b53 2019 return;
f24e9980 2020
98c4bfe9
ID		 2021 WARN_ON(request_msg->data_length || reply_msg->data_length);
2022 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
bb873b53
ID		 2023 switch (op->op) {
2024 /* request */
2025 case CEPH_OSD_OP_WRITE:
2026 case CEPH_OSD_OP_WRITEFULL:
2027 WARN_ON(op->indata_len != op->extent.length);
98c4bfe9
ID		 2028 ceph_osdc_msg_data_add(request_msg,
2029 &op->extent.osd_data);
bb873b53
ID		 2030 break;
2031 case CEPH_OSD_OP_SETXATTR:
2032 case CEPH_OSD_OP_CMPXATTR:
2033 WARN_ON(op->indata_len != op->xattr.name_len +
2034 op->xattr.value_len);
98c4bfe9
ID		 2035 ceph_osdc_msg_data_add(request_msg,
2036 &op->xattr.osd_data);
bb873b53 2037 break;
922dab61 2038 case CEPH_OSD_OP_NOTIFY_ACK:
98c4bfe9 2039 ceph_osdc_msg_data_add(request_msg,
922dab61
ID		 2040 &op->notify_ack.request_data);
2041 break;
78beb0ff 2042 case CEPH_OSD_OP_COPY_FROM2:
23ddf9be
LH		 2043 ceph_osdc_msg_data_add(request_msg,
2044 &op->copy_from.osd_data);
2045 break;
bb873b53
ID		 2046
2047 /* reply */
2048 case CEPH_OSD_OP_STAT:
98c4bfe9 2049 ceph_osdc_msg_data_add(reply_msg,
bb873b53
ID		 2050 &op->raw_data_in);
2051 break;
2052 case CEPH_OSD_OP_READ:
f628d799 2053 case CEPH_OSD_OP_SPARSE_READ:
98c4bfe9 2054 ceph_osdc_msg_data_add(reply_msg,
bb873b53
ID		 2055 &op->extent.osd_data);
2056 break;
a4ed38d7 2057 case CEPH_OSD_OP_LIST_WATCHERS:
98c4bfe9 2058 ceph_osdc_msg_data_add(reply_msg,
a4ed38d7
DF		 2059 &op->list_watchers.response_data);
2060 break;
bb873b53
ID		 2061
2062 /* both */
2063 case CEPH_OSD_OP_CALL:
2064 WARN_ON(op->indata_len != op->cls.class_len +
2065 op->cls.method_len +
2066 op->cls.indata_len);
98c4bfe9
ID		 2067 ceph_osdc_msg_data_add(request_msg,
2068 &op->cls.request_info);
bb873b53 2069 /* optional, can be NONE */
98c4bfe9
ID		 2070 ceph_osdc_msg_data_add(request_msg,
2071 &op->cls.request_data);
bb873b53 2072 /* optional, can be NONE */
98c4bfe9 2073 ceph_osdc_msg_data_add(reply_msg,
bb873b53
ID		 2074 &op->cls.response_data);
2075 break;
19079203 2076 case CEPH_OSD_OP_NOTIFY:
98c4bfe9 2077 ceph_osdc_msg_data_add(request_msg,
19079203 2078 &op->notify.request_data);
98c4bfe9 2079 ceph_osdc_msg_data_add(reply_msg,
19079203
ID		 2080 &op->notify.response_data);
2081 break;
bb873b53 2082 }
bb873b53 2083 }
bb873b53
ID		 2084}
2085
2e59ffd1
ID		 2086static void encode_pgid(void **p, const struct ceph_pg *pgid)
2087{
2088 ceph_encode_8(p, 1);
2089 ceph_encode_64(p, pgid->pool);
2090 ceph_encode_32(p, pgid->seed);
2091 ceph_encode_32(p, -1); /* preferred */
2092}
2093
8cb441c0
ID		 2094static void encode_spgid(void **p, const struct ceph_spg *spgid)
2095{
2096 ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2097 encode_pgid(p, &spgid->pgid);
2098 ceph_encode_8(p, spgid->shard);
2099}
2100
2e59ffd1
ID		 2101static void encode_oloc(void **p, void *end,
2102 const struct ceph_object_locator *oloc)
2103{
2104 ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2105 ceph_encode_64(p, oloc->pool);
2106 ceph_encode_32(p, -1); /* preferred */
2107 ceph_encode_32(p, 0); /* key len */
2108 if (oloc->pool_ns)
2109 ceph_encode_string(p, end, oloc->pool_ns->str,
2110 oloc->pool_ns->len);
2111 else
2112 ceph_encode_32(p, 0);
2113}
2114
8cb441c0
ID		 2115static void encode_request_partial(struct ceph_osd_request *req,
2116 struct ceph_msg *msg)
bb873b53
ID		 2117{
2118 void *p = msg->front.iov_base;
2119 void *const end = p + msg->front_alloc_len;
2120 u32 data_len = 0;
2121 int i;
2122
2123 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2124 /* snapshots aren't writeable */
2125 WARN_ON(req->r_snapid != CEPH_NOSNAP);
2126 } else {
2127 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2128 req->r_data_offset || req->r_snapc);
2129 }
2130
98c4bfe9 2131 setup_request_data(req);
bb873b53 2132
8cb441c0
ID		 2133 encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2134 ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
bb873b53
ID		 2135 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2136 ceph_encode_32(&p, req->r_flags);
8cb441c0
ID		 2137
2138 /* reqid */
2139 ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2140 memset(p, 0, sizeof(struct ceph_osd_reqid));
2141 p += sizeof(struct ceph_osd_reqid);
2142
2143 /* trace */
2144 memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2145 p += sizeof(struct ceph_blkin_trace_info);
2146
2147 ceph_encode_32(&p, 0); /* client_inc, always 0 */
fac02ddf 2148 ceph_encode_timespec64(p, &req->r_mtime);
bb873b53 2149 p += sizeof(struct ceph_timespec);
aa26d662 2150
2e59ffd1 2151 encode_oloc(&p, end, &req->r_t.target_oloc);
2e59ffd1
ID		 2152 ceph_encode_string(&p, end, req->r_t.target_oid.name,
2153 req->r_t.target_oid.name_len);
f24e9980 2154
bb873b53
ID		 2155 /* ops, can imply data */
2156 ceph_encode_16(&p, req->r_num_ops);
2157 for (i = 0; i < req->r_num_ops; i++) {
2158 data_len += osd_req_encode_op(p, &req->r_ops[i]);
2159 p += sizeof(struct ceph_osd_op);
2160 }
26be8808 2161
bb873b53
ID		 2162 ceph_encode_64(&p, req->r_snapid); /* snapid */
2163 if (req->r_snapc) {
2164 ceph_encode_64(&p, req->r_snapc->seq);
2165 ceph_encode_32(&p, req->r_snapc->num_snaps);
2166 for (i = 0; i < req->r_snapc->num_snaps; i++)
2167 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2168 } else {
2169 ceph_encode_64(&p, 0); /* snap_seq */
2170 ceph_encode_32(&p, 0); /* snaps len */
2171 }
2172
2173 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
986e8989 2174 BUG_ON(p > end - 8); /* space for features */
bb873b53 2175
8cb441c0
ID		 2176 msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2177 /* front_len is finalized in encode_request_finish() */
986e8989
ID		 2178 msg->front.iov_len = p - msg->front.iov_base;
2179 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
bb873b53
ID		 2180 msg->hdr.data_len = cpu_to_le32(data_len);
2181 /*
2182 * The header "data_off" is a hint to the receiver allowing it
2183 * to align received data into its buffers such that there's no
2184 * need to re-copy it before writing it to disk (direct I/O).
2185 */
2186 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
26be8808 2187
8cb441c0
ID		 2188 dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2189 req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2190}
2191
2192static void encode_request_finish(struct ceph_msg *msg)
2193{
2194 void *p = msg->front.iov_base;
986e8989 2195 void *const partial_end = p + msg->front.iov_len;
8cb441c0
ID		 2196 void *const end = p + msg->front_alloc_len;
2197
2198 if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2199 /* luminous OSD -- encode features and be done */
986e8989 2200 p = partial_end;
8cb441c0
ID		 2201 ceph_encode_64(&p, msg->con->peer_features);
2202 } else {
2203 struct {
2204 char spgid[CEPH_ENCODING_START_BLK_LEN +
2205 CEPH_PGID_ENCODING_LEN + 1];
2206 __le32 hash;
2207 __le32 epoch;
2208 __le32 flags;
2209 char reqid[CEPH_ENCODING_START_BLK_LEN +
2210 sizeof(struct ceph_osd_reqid)];
2211 char trace[sizeof(struct ceph_blkin_trace_info)];
2212 __le32 client_inc;
2213 struct ceph_timespec mtime;
2214 } __packed head;
2215 struct ceph_pg pgid;
2216 void *oloc, *oid, *tail;
2217 int oloc_len, oid_len, tail_len;
2218 int len;
2219
2220 /*
2221 * Pre-luminous OSD -- reencode v8 into v4 using @head
2222 * as a temporary buffer. Encode the raw PG; the rest
2223 * is just a matter of moving oloc, oid and tail blobs
2224 * around.
2225 */
2226 memcpy(&head, p, sizeof(head));
2227 p += sizeof(head);
2228
2229 oloc = p;
2230 p += CEPH_ENCODING_START_BLK_LEN;
2231 pgid.pool = ceph_decode_64(&p);
2232 p += 4 + 4; /* preferred, key len */
2233 len = ceph_decode_32(&p);
2234 p += len; /* nspace */
2235 oloc_len = p - oloc;
2236
2237 oid = p;
2238 len = ceph_decode_32(&p);
2239 p += len;
2240 oid_len = p - oid;
2241
2242 tail = p;
986e8989 2243 tail_len = partial_end - p;
8cb441c0
ID		 2244
2245 p = msg->front.iov_base;
2246 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2247 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2248 ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2249 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2250
2251 /* reassert_version */
2252 memset(p, 0, sizeof(struct ceph_eversion));
2253 p += sizeof(struct ceph_eversion);
2254
2255 BUG_ON(p >= oloc);
2256 memmove(p, oloc, oloc_len);
2257 p += oloc_len;
2258
2259 pgid.seed = le32_to_cpu(head.hash);
2260 encode_pgid(&p, &pgid); /* raw pg */
2261
2262 BUG_ON(p >= oid);
2263 memmove(p, oid, oid_len);
2264 p += oid_len;
2265
2266 /* tail -- ops, snapid, snapc, retry_attempt */
2267 BUG_ON(p >= tail);
2268 memmove(p, tail, tail_len);
2269 p += tail_len;
2270
2271 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2272 }
2273
2274 BUG_ON(p > end);
2275 msg->front.iov_len = p - msg->front.iov_base;
2276 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2277
2278 dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2279 le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2280 le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2281 le16_to_cpu(msg->hdr.version));
bb873b53
ID		 2282}
2283
2284/*
2285 * @req has to be assigned a tid and registered.
2286 */
2287static void send_request(struct ceph_osd_request *req)
2288{
2289 struct ceph_osd *osd = req->r_osd;
2290
5aea3dcd 2291 verify_osd_locked(osd);
bb873b53
ID		 2292 WARN_ON(osd->o_osd != req->r_t.osd);
2293
a02a946d
ID		 2294 /* backoff? */
2295 if (should_plug_request(req))
2296 return;
2297
5aea3dcd
ID		 2298 /*
2299 * We may have a previously queued request message hanging
2300 * around. Cancel it to avoid corrupting the msgr.
2301 */
2302 if (req->r_sent)
2303 ceph_msg_revoke(req->r_request);
2304
bb873b53
ID		 2305 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2306 if (req->r_attempts)
2307 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2308 else
2309 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2310
8cb441c0 2311 encode_request_partial(req, req->r_request);
bb873b53 2312
04c7d789 2313 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 2314 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
dc98ff72 2315 req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
04c7d789
ID		 2316 req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2317 req->r_attempts);
bb873b53
ID		 2318
2319 req->r_t.paused = false;
2320 req->r_stamp = jiffies;
2321 req->r_attempts++;
2322
2323 req->r_sent = osd->o_incarnation;
2324 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2325 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
f24e9980
SW		 2326}
2327
42c1b124
ID		 2328static void maybe_request_map(struct ceph_osd_client *osdc)
2329{
2330 bool continuous = false;
2331
5aea3dcd 2332 verify_osdc_locked(osdc);
42c1b124
ID		 2333 WARN_ON(!osdc->osdmap->epoch);
2334
b7ec35b3
ID		 2335 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2336 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2337 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
42c1b124
ID		 2338 dout("%s osdc %p continuous\n", __func__, osdc);
2339 continuous = true;
2340 } else {
2341 dout("%s osdc %p onetime\n", __func__, osdc);
2342 }
2343
2344 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2345 osdc->osdmap->epoch + 1, continuous))
2346 ceph_monc_renew_subs(&osdc->client->monc);
2347}
2348
a1f4020a 2349static void complete_request(struct ceph_osd_request *req, int err);
4609245e
ID		 2350static void send_map_check(struct ceph_osd_request *req);
2351
5aea3dcd 2352static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
0bbfdfe8 2353{
5aea3dcd
ID		 2354 struct ceph_osd_client *osdc = req->r_osdc;
2355 struct ceph_osd *osd;
4609245e 2356 enum calc_target_result ct_res;
66850df5 2357 int err = 0;
5aea3dcd
ID		 2358 bool need_send = false;
2359 bool promoted = false;
0bbfdfe8 2360
b18b9550 2361 WARN_ON(req->r_tid);
5aea3dcd
ID		 2362 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2363
2364again:
8edf84ba 2365 ct_res = calc_target(osdc, &req->r_t, false);
4609245e
ID		 2366 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2367 goto promote;
2368
5aea3dcd
ID		 2369 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2370 if (IS_ERR(osd)) {
2371 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2372 goto promote;
0bbfdfe8
ID		 2373 }
2374
66850df5
ID		 2375 if (osdc->abort_err) {
2376 dout("req %p abort_err %d\n", req, osdc->abort_err);
2377 err = osdc->abort_err;
2378 } else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
58eb7932
JL		 2379 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2380 osdc->epoch_barrier);
2381 req->r_t.paused = true;
2382 maybe_request_map(osdc);
2383 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2384 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
5aea3dcd
ID		 2385 dout("req %p pausewr\n", req);
2386 req->r_t.paused = true;
2387 maybe_request_map(osdc);
2388 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
b7ec35b3 2389 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
5aea3dcd
ID		 2390 dout("req %p pauserd\n", req);
2391 req->r_t.paused = true;
2392 maybe_request_map(osdc);
2393 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2394 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2395 CEPH_OSD_FLAG_FULL_FORCE)) &&
b7ec35b3 2396 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
5aea3dcd
ID		 2397 pool_full(osdc, req->r_t.base_oloc.pool))) {
2398 dout("req %p full/pool_full\n", req);
02b2f549 2399 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
66850df5 2400 err = -ENOSPC;
29e87820 2401 } else {
dc9b0dc4
ID		 2402 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL))
2403 pr_warn_ratelimited("cluster is full (osdmap FULL)\n");
2404 else
2405 pr_warn_ratelimited("pool %lld is full or reached quota\n",
2406 req->r_t.base_oloc.pool);
29e87820
ID		 2407 req->r_t.paused = true;
2408 maybe_request_map(osdc);
2409 }
5aea3dcd
ID		 2410 } else if (!osd_homeless(osd)) {
2411 need_send = true;
0bbfdfe8 2412 } else {
5aea3dcd 2413 maybe_request_map(osdc);
0bbfdfe8
ID		 2414 }
2415
5aea3dcd
ID		 2416 mutex_lock(&osd->lock);
2417 /*
2418 * Assign the tid atomically with send_request() to protect
2419 * multiple writes to the same object from racing with each
2420 * other, resulting in out of order ops on the OSDs.
2421 */
2422 req->r_tid = atomic64_inc_return(&osdc->last_tid);
2423 link_request(osd, req);
2424 if (need_send)
2425 send_request(req);
66850df5
ID		 2426 else if (err)
2427 complete_request(req, err);
5aea3dcd
ID		 2428 mutex_unlock(&osd->lock);
2429
6001567c 2430 if (!err && ct_res == CALC_TARGET_POOL_DNE)
4609245e
ID		 2431 send_map_check(req);
2432
5aea3dcd
ID		 2433 if (promoted)
2434 downgrade_write(&osdc->lock);
2435 return;
2436
2437promote:
2438 up_read(&osdc->lock);
2439 down_write(&osdc->lock);
2440 wrlocked = true;
2441 promoted = true;
2442 goto again;
2443}
2444
2445static void account_request(struct ceph_osd_request *req)
2446{
54ea0046 2447 WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
b18b9550 2448 WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
5aea3dcd 2449
b18b9550 2450 req->r_flags |= CEPH_OSD_FLAG_ONDISK;
22d2cfdf 2451 atomic_inc(&req->r_osdc->num_requests);
7cc5e38f
ID		 2452
2453 req->r_start_stamp = jiffies;
97e27aaa 2454 req->r_start_latency = ktime_get();
5aea3dcd
ID		 2455}
2456
2457static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2458{
2459 ceph_osdc_get_request(req);
2460 account_request(req);
2461 __submit_request(req, wrlocked);
2462}
2463
45ee2c1d 2464static void finish_request(struct ceph_osd_request *req)
5aea3dcd
ID		 2465{
2466 struct ceph_osd_client *osdc = req->r_osdc;
5aea3dcd 2467
04c7d789 2468 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
5aea3dcd
ID		 2469 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2470
97e27aaa
XL		 2471 req->r_end_latency = ktime_get();
2472
f628d799
JL		 2473 if (req->r_osd) {
2474 ceph_init_sparse_read(&req->r_osd->o_sparse_read);
04c7d789 2475 unlink_request(req->r_osd, req);
f628d799 2476 }
5aea3dcd
ID		 2477 atomic_dec(&osdc->num_requests);
2478
2479 /*
2480 * If an OSD has failed or returned and a request has been sent
2481 * twice, it's possible to get a reply and end up here while the
2482 * request message is queued for delivery. We will ignore the
2483 * reply, so not a big deal, but better to try and catch it.
2484 */
2485 ceph_msg_revoke(req->r_request);
2486 ceph_msg_revoke_incoming(req->r_reply);
2487}
2488
fe5da05e
ID		 2489static void __complete_request(struct ceph_osd_request *req)
2490{
d75f773c 2491 dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
26df726b
ID		 2492 req->r_tid, req->r_callback, req->r_result);
2493
2494 if (req->r_callback)
fe5da05e 2495 req->r_callback(req);
26df726b
ID		 2496 complete_all(&req->r_completion);
2497 ceph_osdc_put_request(req);
fe5da05e
ID		 2498}
2499
88bc1922
ID		 2500static void complete_request_workfn(struct work_struct *work)
2501{
2502 struct ceph_osd_request *req =
2503 container_of(work, struct ceph_osd_request, r_complete_work);
2504
2505 __complete_request(req);
fe5da05e
ID		 2506}
2507
4609245e 2508/*
b18b9550 2509 * This is open-coded in handle_reply().
4609245e
ID		 2510 */
2511static void complete_request(struct ceph_osd_request *req, int err)
2512{
2513 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2514
2515 req->r_result = err;
45ee2c1d 2516 finish_request(req);
88bc1922
ID		 2517
2518 INIT_WORK(&req->r_complete_work, complete_request_workfn);
2519 queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
4609245e
ID		 2520}
2521
2522static void cancel_map_check(struct ceph_osd_request *req)
2523{
2524 struct ceph_osd_client *osdc = req->r_osdc;
2525 struct ceph_osd_request *lookup_req;
2526
2527 verify_osdc_wrlocked(osdc);
2528
2529 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2530 if (!lookup_req)
2531 return;
2532
2533 WARN_ON(lookup_req != req);
2534 erase_request_mc(&osdc->map_checks, req);
2535 ceph_osdc_put_request(req);
2536}
2537
5aea3dcd
ID		 2538static void cancel_request(struct ceph_osd_request *req)
2539{
2540 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2541
4609245e 2542 cancel_map_check(req);
45ee2c1d 2543 finish_request(req);
b18b9550 2544 complete_all(&req->r_completion);
c297eb42 2545 ceph_osdc_put_request(req);
5aea3dcd
ID		 2546}
2547
7cc5e38f
ID		 2548static void abort_request(struct ceph_osd_request *req, int err)
2549{
2550 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2551
2552 cancel_map_check(req);
2553 complete_request(req, err);
2554}
2555
66850df5
ID		 2556static int abort_fn(struct ceph_osd_request *req, void *arg)
2557{
2558 int err = *(int *)arg;
2559
2560 abort_request(req, err);
2561 return 0; /* continue iteration */
2562}
2563
2564/*
2565 * Abort all in-flight requests with @err and arrange for all future
2566 * requests to be failed immediately.
2567 */
2568void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2569{
2570 dout("%s osdc %p err %d\n", __func__, osdc, err);
2571 down_write(&osdc->lock);
2572 for_each_request(osdc, abort_fn, &err);
2573 osdc->abort_err = err;
2574 up_write(&osdc->lock);
2575}
2576EXPORT_SYMBOL(ceph_osdc_abort_requests);
2577
2cef0ba8
YZ		 2578void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2579{
2580 down_write(&osdc->lock);
2581 osdc->abort_err = 0;
2582 up_write(&osdc->lock);
2583}
2584EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2585
58eb7932
JL		 2586static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2587{
2588 if (likely(eb > osdc->epoch_barrier)) {
2589 dout("updating epoch_barrier from %u to %u\n",
2590 osdc->epoch_barrier, eb);
2591 osdc->epoch_barrier = eb;
2592 /* Request map if we're not to the barrier yet */
2593 if (eb > osdc->osdmap->epoch)
2594 maybe_request_map(osdc);
2595 }
2596}
2597
2598void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2599{
2600 down_read(&osdc->lock);
2601 if (unlikely(eb > osdc->epoch_barrier)) {
2602 up_read(&osdc->lock);
2603 down_write(&osdc->lock);
2604 update_epoch_barrier(osdc, eb);
2605 up_write(&osdc->lock);
2606 } else {
2607 up_read(&osdc->lock);
2608 }
2609}
2610EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2611
4eea0fef
ID		 2612/*
2613 * We can end up releasing caps as a result of abort_request().
2614 * In that case, we probably want to ensure that the cap release message
2615 * has an updated epoch barrier in it, so set the epoch barrier prior to
2616 * aborting the first request.
2617 */
2618static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2619{
2620 struct ceph_osd_client *osdc = req->r_osdc;
2621 bool *victims = arg;
2622
c843d13c 2623 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
4eea0fef 2624 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
690f951d 2625 pool_full(osdc, req->r_t.base_oloc.pool))) {
4eea0fef
ID		 2626 if (!*victims) {
2627 update_epoch_barrier(osdc, osdc->osdmap->epoch);
2628 *victims = true;
2629 }
2630 abort_request(req, -ENOSPC);
2631 }
2632
2633 return 0; /* continue iteration */
2634}
2635
fc36d0a4
JL		 2636/*
2637 * Drop all pending requests that are stalled waiting on a full condition to
58eb7932
JL		 2638 * clear, and complete them with ENOSPC as the return code. Set the
2639 * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2640 * cancelled.
fc36d0a4
JL		 2641 */
2642static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2643{
58eb7932 2644 bool victims = false;
fc36d0a4 2645
02b2f549 2646 if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
c843d13c 2647 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
4eea0fef 2648 for_each_request(osdc, abort_on_full_fn, &victims);
fc36d0a4
JL		 2649}
2650
4609245e
ID		 2651static void check_pool_dne(struct ceph_osd_request *req)
2652{
2653 struct ceph_osd_client *osdc = req->r_osdc;
2654 struct ceph_osdmap *map = osdc->osdmap;
2655
2656 verify_osdc_wrlocked(osdc);
2657 WARN_ON(!map->epoch);
2658
2659 if (req->r_attempts) {
2660 /*
2661 * We sent a request earlier, which means that
2662 * previously the pool existed, and now it does not
2663 * (i.e., it was deleted).
2664 */
2665 req->r_map_dne_bound = map->epoch;
2666 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2667 req->r_tid);
2668 } else {
2669 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2670 req, req->r_tid, req->r_map_dne_bound, map->epoch);
2671 }
2672
2673 if (req->r_map_dne_bound) {
2674 if (map->epoch >= req->r_map_dne_bound) {
2675 /* we had a new enough map */
2676 pr_info_ratelimited("tid %llu pool does not exist\n",
2677 req->r_tid);
2678 complete_request(req, -ENOENT);
2679 }
2680 } else {
2681 send_map_check(req);
2682 }
2683}
2684
2685static void map_check_cb(struct ceph_mon_generic_request *greq)
2686{
2687 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2688 struct ceph_osd_request *req;
2689 u64 tid = greq->private_data;
2690
2691 WARN_ON(greq->result || !greq->u.newest);
2692
2693 down_write(&osdc->lock);
2694 req = lookup_request_mc(&osdc->map_checks, tid);
2695 if (!req) {
2696 dout("%s tid %llu dne\n", __func__, tid);
2697 goto out_unlock;
2698 }
2699
2700 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2701 req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2702 if (!req->r_map_dne_bound)
2703 req->r_map_dne_bound = greq->u.newest;
2704 erase_request_mc(&osdc->map_checks, req);
2705 check_pool_dne(req);
2706
2707 ceph_osdc_put_request(req);
2708out_unlock:
2709 up_write(&osdc->lock);
2710}
2711
2712static void send_map_check(struct ceph_osd_request *req)
2713{
2714 struct ceph_osd_client *osdc = req->r_osdc;
2715 struct ceph_osd_request *lookup_req;
2716 int ret;
2717
2718 verify_osdc_wrlocked(osdc);
2719
2720 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2721 if (lookup_req) {
2722 WARN_ON(lookup_req != req);
2723 return;
2724 }
2725
2726 ceph_osdc_get_request(req);
2727 insert_request_mc(&osdc->map_checks, req);
2728 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2729 map_check_cb, req->r_tid);
2730 WARN_ON(ret);
2731}
2732
922dab61
ID		 2733/*
2734 * lingering requests, watch/notify v2 infrastructure
2735 */
2736static void linger_release(struct kref *kref)
2737{
2738 struct ceph_osd_linger_request *lreq =
2739 container_of(kref, struct ceph_osd_linger_request, kref);
2740
2741 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2742 lreq->reg_req, lreq->ping_req);
2743 WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2744 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
4609245e 2745 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
922dab61 2746 WARN_ON(!list_empty(&lreq->scan_item));
b07d3c4b 2747 WARN_ON(!list_empty(&lreq->pending_lworks));
922dab61
ID		 2748 WARN_ON(lreq->osd);
2749
75dbb685
ID		 2750 if (lreq->request_pl)
2751 ceph_pagelist_release(lreq->request_pl);
2752 if (lreq->notify_id_pages)
2753 ceph_release_page_vector(lreq->notify_id_pages, 1);
2754
2755 ceph_osdc_put_request(lreq->reg_req);
2756 ceph_osdc_put_request(lreq->ping_req);
922dab61
ID		 2757 target_destroy(&lreq->t);
2758 kfree(lreq);
2759}
2760
2761static void linger_put(struct ceph_osd_linger_request *lreq)
2762{
2763 if (lreq)
2764 kref_put(&lreq->kref, linger_release);
2765}
2766
2767static struct ceph_osd_linger_request *
2768linger_get(struct ceph_osd_linger_request *lreq)
2769{
2770 kref_get(&lreq->kref);
2771 return lreq;
2772}
2773
2774static struct ceph_osd_linger_request *
2775linger_alloc(struct ceph_osd_client *osdc)
2776{
2777 struct ceph_osd_linger_request *lreq;
2778
2779 lreq = kzalloc(sizeof(*lreq), GFP_NOIO);
2780 if (!lreq)
2781 return NULL;
2782
2783 kref_init(&lreq->kref);
2784 mutex_init(&lreq->lock);
2785 RB_CLEAR_NODE(&lreq->node);
2786 RB_CLEAR_NODE(&lreq->osdc_node);
4609245e 2787 RB_CLEAR_NODE(&lreq->mc_node);
922dab61 2788 INIT_LIST_HEAD(&lreq->scan_item);
b07d3c4b 2789 INIT_LIST_HEAD(&lreq->pending_lworks);
922dab61 2790 init_completion(&lreq->reg_commit_wait);
19079203 2791 init_completion(&lreq->notify_finish_wait);
922dab61
ID		 2792
2793 lreq->osdc = osdc;
2794 target_init(&lreq->t);
2795
2796 dout("%s lreq %p\n", __func__, lreq);
2797 return lreq;
2798}
2799
2800DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2801DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
4609245e 2802DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
922dab61
ID		 2803
2804/*
2805 * Create linger request <-> OSD session relation.
2806 *
2807 * @lreq has to be registered, @osd may be homeless.
2808 */
2809static void link_linger(struct ceph_osd *osd,
2810 struct ceph_osd_linger_request *lreq)
2811{
2812 verify_osd_locked(osd);
2813 WARN_ON(!lreq->linger_id || lreq->osd);
2814 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2815 osd->o_osd, lreq, lreq->linger_id);
2816
2817 if (!osd_homeless(osd))
2818 __remove_osd_from_lru(osd);
2819 else
2820 atomic_inc(&osd->o_osdc->num_homeless);
2821
2822 get_osd(osd);
2823 insert_linger(&osd->o_linger_requests, lreq);
2824 lreq->osd = osd;
2825}
2826
2827static void unlink_linger(struct ceph_osd *osd,
2828 struct ceph_osd_linger_request *lreq)
2829{
2830 verify_osd_locked(osd);
2831 WARN_ON(lreq->osd != osd);
2832 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2833 osd->o_osd, lreq, lreq->linger_id);
2834
2835 lreq->osd = NULL;
2836 erase_linger(&osd->o_linger_requests, lreq);
2837 put_osd(osd);
2838
2839 if (!osd_homeless(osd))
2840 maybe_move_osd_to_lru(osd);
2841 else
2842 atomic_dec(&osd->o_osdc->num_homeless);
2843}
2844
2845static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2846{
2847 verify_osdc_locked(lreq->osdc);
2848
2849 return !RB_EMPTY_NODE(&lreq->osdc_node);
2850}
2851
2852static bool linger_registered(struct ceph_osd_linger_request *lreq)
2853{
2854 struct ceph_osd_client *osdc = lreq->osdc;
2855 bool registered;
2856
2857 down_read(&osdc->lock);
2858 registered = __linger_registered(lreq);
2859 up_read(&osdc->lock);
2860
2861 return registered;
2862}
2863
2864static void linger_register(struct ceph_osd_linger_request *lreq)
2865{
2866 struct ceph_osd_client *osdc = lreq->osdc;
2867
2868 verify_osdc_wrlocked(osdc);
2869 WARN_ON(lreq->linger_id);
2870
2871 linger_get(lreq);
2872 lreq->linger_id = ++osdc->last_linger_id;
2873 insert_linger_osdc(&osdc->linger_requests, lreq);
2874}
2875
2876static void linger_unregister(struct ceph_osd_linger_request *lreq)
2877{
2878 struct ceph_osd_client *osdc = lreq->osdc;
2879
2880 verify_osdc_wrlocked(osdc);
2881
2882 erase_linger_osdc(&osdc->linger_requests, lreq);
2883 linger_put(lreq);
2884}
2885
2886static void cancel_linger_request(struct ceph_osd_request *req)
2887{
2888 struct ceph_osd_linger_request *lreq = req->r_priv;
2889
2890 WARN_ON(!req->r_linger);
2891 cancel_request(req);
2892 linger_put(lreq);
2893}
2894
2895struct linger_work {
2896 struct work_struct work;
2897 struct ceph_osd_linger_request *lreq;
b07d3c4b
ID		 2898 struct list_head pending_item;
2899 unsigned long queued_stamp;
922dab61
ID		 2900
2901 union {
2902 struct {
2903 u64 notify_id;
2904 u64 notifier_id;
2905 void *payload; /* points into @msg front */
2906 size_t payload_len;
2907
2908 struct ceph_msg *msg; /* for ceph_msg_put() */
2909 } notify;
2910 struct {
2911 int err;
2912 } error;
2913 };
2914};
2915
2916static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2917 work_func_t workfn)
2918{
2919 struct linger_work *lwork;
2920
2921 lwork = kzalloc(sizeof(*lwork), GFP_NOIO);
2922 if (!lwork)
2923 return NULL;
2924
2925 INIT_WORK(&lwork->work, workfn);
b07d3c4b 2926 INIT_LIST_HEAD(&lwork->pending_item);
922dab61
ID		 2927 lwork->lreq = linger_get(lreq);
2928
2929 return lwork;
2930}
2931
2932static void lwork_free(struct linger_work *lwork)
2933{
2934 struct ceph_osd_linger_request *lreq = lwork->lreq;
2935
b07d3c4b
ID		 2936 mutex_lock(&lreq->lock);
2937 list_del(&lwork->pending_item);
2938 mutex_unlock(&lreq->lock);
2939
922dab61
ID		 2940 linger_put(lreq);
2941 kfree(lwork);
2942}
2943
2944static void lwork_queue(struct linger_work *lwork)
2945{
2946 struct ceph_osd_linger_request *lreq = lwork->lreq;
2947 struct ceph_osd_client *osdc = lreq->osdc;
2948
2949 verify_lreq_locked(lreq);
b07d3c4b
ID		 2950 WARN_ON(!list_empty(&lwork->pending_item));
2951
2952 lwork->queued_stamp = jiffies;
2953 list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
922dab61
ID		 2954 queue_work(osdc->notify_wq, &lwork->work);
2955}
2956
2957static void do_watch_notify(struct work_struct *w)
2958{
2959 struct linger_work *lwork = container_of(w, struct linger_work, work);
2960 struct ceph_osd_linger_request *lreq = lwork->lreq;
2961
2962 if (!linger_registered(lreq)) {
2963 dout("%s lreq %p not registered\n", __func__, lreq);
2964 goto out;
2965 }
2966
19079203 2967 WARN_ON(!lreq->is_watch);
922dab61
ID		 2968 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2969 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
2970 lwork->notify.payload_len);
2971 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
2972 lwork->notify.notifier_id, lwork->notify.payload,
2973 lwork->notify.payload_len);
2974
2975out:
2976 ceph_msg_put(lwork->notify.msg);
2977 lwork_free(lwork);
2978}
2979
2980static void do_watch_error(struct work_struct *w)
2981{
2982 struct linger_work *lwork = container_of(w, struct linger_work, work);
2983 struct ceph_osd_linger_request *lreq = lwork->lreq;
2984
2985 if (!linger_registered(lreq)) {
2986 dout("%s lreq %p not registered\n", __func__, lreq);
2987 goto out;
2988 }
2989
2990 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
2991 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
2992
2993out:
2994 lwork_free(lwork);
2995}
2996
2997static void queue_watch_error(struct ceph_osd_linger_request *lreq)
2998{
2999 struct linger_work *lwork;
3000
3001 lwork = lwork_alloc(lreq, do_watch_error);
3002 if (!lwork) {
3003 pr_err("failed to allocate error-lwork\n");
3004 return;
3005 }
3006
3007 lwork->error.err = lreq->last_error;
3008 lwork_queue(lwork);
3009}
3010
3011static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
3012 int result)
3013{
3014 if (!completion_done(&lreq->reg_commit_wait)) {
3015 lreq->reg_commit_error = (result <= 0 ? result : 0);
3016 complete_all(&lreq->reg_commit_wait);
3017 }
3018}
3019
3020static void linger_commit_cb(struct ceph_osd_request *req)
3021{
3022 struct ceph_osd_linger_request *lreq = req->r_priv;
3023
3024 mutex_lock(&lreq->lock);
75dbb685
ID		 3025 if (req != lreq->reg_req) {
3026 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3027 __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3028 goto out;
3029 }
3030
922dab61
ID		 3031 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3032 lreq->linger_id, req->r_result);
922dab61
ID		 3033 linger_reg_commit_complete(lreq, req->r_result);
3034 lreq->committed = true;
3035
19079203
ID		 3036 if (!lreq->is_watch) {
3037 struct ceph_osd_data *osd_data =
3038 osd_req_op_data(req, 0, notify, response_data);
3039 void *p = page_address(osd_data->pages[0]);
3040
3041 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3042 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3043
3044 /* make note of the notify_id */
3045 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3046 lreq->notify_id = ceph_decode_64(&p);
3047 dout("lreq %p notify_id %llu\n", lreq,
3048 lreq->notify_id);
3049 } else {
3050 dout("lreq %p no notify_id\n", lreq);
3051 }
3052 }
3053
75dbb685 3054out:
922dab61
ID		 3055 mutex_unlock(&lreq->lock);
3056 linger_put(lreq);
3057}
3058
3059static int normalize_watch_error(int err)
3060{
3061 /*
3062 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3063 * notification and a failure to reconnect because we raced with
3064 * the delete appear the same to the user.
3065 */
3066 if (err == -ENOENT)
3067 err = -ENOTCONN;
3068
3069 return err;
3070}
3071
3072static void linger_reconnect_cb(struct ceph_osd_request *req)
3073{
3074 struct ceph_osd_linger_request *lreq = req->r_priv;
3075
3076 mutex_lock(&lreq->lock);
75dbb685
ID		 3077 if (req != lreq->reg_req) {
3078 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3079 __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3080 goto out;
3081 }
3082
922dab61
ID		 3083 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3084 lreq, lreq->linger_id, req->r_result, lreq->last_error);
3085 if (req->r_result < 0) {
3086 if (!lreq->last_error) {
3087 lreq->last_error = normalize_watch_error(req->r_result);
3088 queue_watch_error(lreq);
3089 }
3090 }
3091
75dbb685 3092out:
922dab61
ID		 3093 mutex_unlock(&lreq->lock);
3094 linger_put(lreq);
3095}
3096
3097static void send_linger(struct ceph_osd_linger_request *lreq)
3098{
75dbb685
ID		 3099 struct ceph_osd_client *osdc = lreq->osdc;
3100 struct ceph_osd_request *req;
3101 int ret;
922dab61 3102
75dbb685
ID		 3103 verify_osdc_wrlocked(osdc);
3104 mutex_lock(&lreq->lock);
922dab61
ID		 3105 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3106
75dbb685
ID		 3107 if (lreq->reg_req) {
3108 if (lreq->reg_req->r_osd)
3109 cancel_linger_request(lreq->reg_req);
3110 ceph_osdc_put_request(lreq->reg_req);
3111 }
3112
3113 req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3114 BUG_ON(!req);
922dab61 3115
5133ba8f 3116 target_copy(&req->r_t, &lreq->t);
922dab61
ID		 3117 req->r_mtime = lreq->mtime;
3118
19079203 3119 if (lreq->is_watch && lreq->committed) {
75dbb685
ID		 3120 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_RECONNECT,
3121 lreq->linger_id, ++lreq->register_gen);
922dab61 3122 dout("lreq %p reconnect register_gen %u\n", lreq,
75dbb685 3123 req->r_ops[0].watch.gen);
922dab61
ID		 3124 req->r_callback = linger_reconnect_cb;
3125 } else {
75dbb685
ID		 3126 if (lreq->is_watch) {
3127 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_WATCH,
3128 lreq->linger_id, 0);
3129 } else {
19079203 3130 lreq->notify_id = 0;
75dbb685
ID		 3131
3132 refcount_inc(&lreq->request_pl->refcnt);
3133 osd_req_op_notify_init(req, 0, lreq->linger_id,
3134 lreq->request_pl);
3135 ceph_osd_data_pages_init(
3136 osd_req_op_data(req, 0, notify, response_data),
3137 lreq->notify_id_pages, PAGE_SIZE, 0, false, false);
3138 }
922dab61
ID		 3139 dout("lreq %p register\n", lreq);
3140 req->r_callback = linger_commit_cb;
3141 }
75dbb685
ID		 3142
3143 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3144 BUG_ON(ret);
922dab61
ID		 3145
3146 req->r_priv = linger_get(lreq);
3147 req->r_linger = true;
75dbb685
ID		 3148 lreq->reg_req = req;
3149 mutex_unlock(&lreq->lock);
922dab61
ID		 3150
3151 submit_request(req, true);
3152}
3153
3154static void linger_ping_cb(struct ceph_osd_request *req)
3155{
3156 struct ceph_osd_linger_request *lreq = req->r_priv;
3157
3158 mutex_lock(&lreq->lock);
75dbb685
ID		 3159 if (req != lreq->ping_req) {
3160 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3161 __func__, lreq, lreq->linger_id, req, lreq->ping_req);
3162 goto out;
3163 }
3164
922dab61
ID		 3165 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3166 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3167 lreq->last_error);
3168 if (lreq->register_gen == req->r_ops[0].watch.gen) {
b07d3c4b
ID		 3169 if (!req->r_result) {
3170 lreq->watch_valid_thru = lreq->ping_sent;
3171 } else if (!lreq->last_error) {
922dab61
ID		 3172 lreq->last_error = normalize_watch_error(req->r_result);
3173 queue_watch_error(lreq);
3174 }
3175 } else {
3176 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3177 lreq->register_gen, req->r_ops[0].watch.gen);
3178 }
3179
75dbb685 3180out:
922dab61
ID		 3181 mutex_unlock(&lreq->lock);
3182 linger_put(lreq);
3183}
3184
3185static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3186{
3187 struct ceph_osd_client *osdc = lreq->osdc;
75dbb685
ID		 3188 struct ceph_osd_request *req;
3189 int ret;
922dab61 3190
b7ec35b3 3191 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
922dab61
ID		 3192 dout("%s PAUSERD\n", __func__);
3193 return;
3194 }
3195
3196 lreq->ping_sent = jiffies;
3197 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3198 __func__, lreq, lreq->linger_id, lreq->ping_sent,
3199 lreq->register_gen);
3200
75dbb685
ID		 3201 if (lreq->ping_req) {
3202 if (lreq->ping_req->r_osd)
3203 cancel_linger_request(lreq->ping_req);
3204 ceph_osdc_put_request(lreq->ping_req);
3205 }
922dab61 3206
75dbb685
ID		 3207 req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3208 BUG_ON(!req);
922dab61 3209
75dbb685
ID		 3210 target_copy(&req->r_t, &lreq->t);
3211 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_PING, lreq->linger_id,
3212 lreq->register_gen);
922dab61 3213 req->r_callback = linger_ping_cb;
75dbb685
ID		 3214
3215 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3216 BUG_ON(ret);
3217
922dab61
ID		 3218 req->r_priv = linger_get(lreq);
3219 req->r_linger = true;
75dbb685 3220 lreq->ping_req = req;
922dab61
ID		 3221
3222 ceph_osdc_get_request(req);
3223 account_request(req);
3224 req->r_tid = atomic64_inc_return(&osdc->last_tid);
3225 link_request(lreq->osd, req);
3226 send_request(req);
3227}
3228
3229static void linger_submit(struct ceph_osd_linger_request *lreq)
3230{
3231 struct ceph_osd_client *osdc = lreq->osdc;
3232 struct ceph_osd *osd;
3233
81c65213
ID		 3234 down_write(&osdc->lock);
3235 linger_register(lreq);
81c65213 3236
8edf84ba 3237 calc_target(osdc, &lreq->t, false);
922dab61
ID		 3238 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3239 link_linger(osd, lreq);
3240
3241 send_linger(lreq);
81c65213 3242 up_write(&osdc->lock);
922dab61
ID		 3243}
3244
4609245e
ID		 3245static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3246{
3247 struct ceph_osd_client *osdc = lreq->osdc;
3248 struct ceph_osd_linger_request *lookup_lreq;
3249
3250 verify_osdc_wrlocked(osdc);
3251
3252 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3253 lreq->linger_id);
3254 if (!lookup_lreq)
3255 return;
3256
3257 WARN_ON(lookup_lreq != lreq);
3258 erase_linger_mc(&osdc->linger_map_checks, lreq);
3259 linger_put(lreq);
3260}
3261
922dab61
ID		 3262/*
3263 * @lreq has to be both registered and linked.
3264 */
3265static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3266{
75dbb685 3267 if (lreq->ping_req && lreq->ping_req->r_osd)
922dab61 3268 cancel_linger_request(lreq->ping_req);
75dbb685 3269 if (lreq->reg_req && lreq->reg_req->r_osd)
922dab61 3270 cancel_linger_request(lreq->reg_req);
4609245e 3271 cancel_linger_map_check(lreq);
922dab61
ID		 3272 unlink_linger(lreq->osd, lreq);
3273 linger_unregister(lreq);
3274}
3275
3276static void linger_cancel(struct ceph_osd_linger_request *lreq)
3277{
3278 struct ceph_osd_client *osdc = lreq->osdc;
3279
3280 down_write(&osdc->lock);
3281 if (__linger_registered(lreq))
3282 __linger_cancel(lreq);
3283 up_write(&osdc->lock);
3284}
3285
4609245e
ID		 3286static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3287
3288static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3289{
3290 struct ceph_osd_client *osdc = lreq->osdc;
3291 struct ceph_osdmap *map = osdc->osdmap;
3292
3293 verify_osdc_wrlocked(osdc);
3294 WARN_ON(!map->epoch);
3295
3296 if (lreq->register_gen) {
3297 lreq->map_dne_bound = map->epoch;
3298 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3299 lreq, lreq->linger_id);
3300 } else {
3301 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3302 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3303 map->epoch);
3304 }
3305
3306 if (lreq->map_dne_bound) {
3307 if (map->epoch >= lreq->map_dne_bound) {
3308 /* we had a new enough map */
3309 pr_info("linger_id %llu pool does not exist\n",
3310 lreq->linger_id);
3311 linger_reg_commit_complete(lreq, -ENOENT);
3312 __linger_cancel(lreq);
3313 }
3314 } else {
3315 send_linger_map_check(lreq);
3316 }
3317}
3318
3319static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3320{
3321 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3322 struct ceph_osd_linger_request *lreq;
3323 u64 linger_id = greq->private_data;
3324
3325 WARN_ON(greq->result || !greq->u.newest);
3326
3327 down_write(&osdc->lock);
3328 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3329 if (!lreq) {
3330 dout("%s linger_id %llu dne\n", __func__, linger_id);
3331 goto out_unlock;
3332 }
3333
3334 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3335 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3336 greq->u.newest);
3337 if (!lreq->map_dne_bound)
3338 lreq->map_dne_bound = greq->u.newest;
3339 erase_linger_mc(&osdc->linger_map_checks, lreq);
3340 check_linger_pool_dne(lreq);
3341
3342 linger_put(lreq);
3343out_unlock:
3344 up_write(&osdc->lock);
3345}
3346
3347static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3348{
3349 struct ceph_osd_client *osdc = lreq->osdc;
3350 struct ceph_osd_linger_request *lookup_lreq;
3351 int ret;
3352
3353 verify_osdc_wrlocked(osdc);
3354
3355 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3356 lreq->linger_id);
3357 if (lookup_lreq) {
3358 WARN_ON(lookup_lreq != lreq);
3359 return;
3360 }
3361
3362 linger_get(lreq);
3363 insert_linger_mc(&osdc->linger_map_checks, lreq);
3364 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3365 linger_map_check_cb, lreq->linger_id);
3366 WARN_ON(ret);
3367}
3368
922dab61
ID		 3369static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3370{
3371 int ret;
3372
3373 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
e6e28432 3374 ret = wait_for_completion_killable(&lreq->reg_commit_wait);
922dab61
ID		 3375 return ret ?: lreq->reg_commit_error;
3376}
3377
e6e28432
ID		 3378static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq,
3379 unsigned long timeout)
19079203 3380{
e6e28432 3381 long left;
19079203
ID		 3382
3383 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
e6e28432
ID		 3384 left = wait_for_completion_killable_timeout(&lreq->notify_finish_wait,
3385 ceph_timeout_jiffies(timeout));
3386 if (left <= 0)
3387 left = left ?: -ETIMEDOUT;
3388 else
3389 left = lreq->notify_finish_error; /* completed */
3390
3391 return left;
19079203
ID		 3392}
3393
f24e9980 3394/*
fbca9635
ID		 3395 * Timeout callback, called every N seconds. When 1 or more OSD
3396 * requests has been active for more than N seconds, we send a keepalive
3397 * (tag + timestamp) to its OSD to ensure any communications channel
3398 * reset is detected.
f24e9980
SW		 3399 */
3400static void handle_timeout(struct work_struct *work)
3401{
3402 struct ceph_osd_client *osdc =
3403 container_of(work, struct ceph_osd_client, timeout_work.work);
a319bf56 3404 struct ceph_options *opts = osdc->client->options;
5aea3dcd 3405 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
7cc5e38f 3406 unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
5aea3dcd
ID		 3407 LIST_HEAD(slow_osds);
3408 struct rb_node *n, *p;
f24e9980 3409
5aea3dcd
ID		 3410 dout("%s osdc %p\n", __func__, osdc);
3411 down_write(&osdc->lock);
f24e9980 3412
422d2cb8
YS		 3413 /*
3414 * ping osds that are a bit slow. this ensures that if there
3415 * is a break in the TCP connection we will notice, and reopen
3416 * a connection with that osd (from the fault callback).
3417 */
5aea3dcd
ID		 3418 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3419 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3420 bool found = false;
3421
7cc5e38f 3422 for (p = rb_first(&osd->o_requests); p; ) {
5aea3dcd
ID		 3423 struct ceph_osd_request *req =
3424 rb_entry(p, struct ceph_osd_request, r_node);
3425
7cc5e38f
ID		 3426 p = rb_next(p); /* abort_request() */
3427
5aea3dcd
ID		 3428 if (time_before(req->r_stamp, cutoff)) {
3429 dout(" req %p tid %llu on osd%d is laggy\n",
3430 req, req->r_tid, osd->o_osd);
3431 found = true;
3432 }
7cc5e38f
ID		 3433 if (opts->osd_request_timeout &&
3434 time_before(req->r_start_stamp, expiry_cutoff)) {
3435 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3436 req->r_tid, osd->o_osd);
3437 abort_request(req, -ETIMEDOUT);
3438 }
5aea3dcd 3439 }
922dab61
ID		 3440 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3441 struct ceph_osd_linger_request *lreq =
3442 rb_entry(p, struct ceph_osd_linger_request, node);
3443
3444 dout(" lreq %p linger_id %llu is served by osd%d\n",
3445 lreq, lreq->linger_id, osd->o_osd);
3446 found = true;
3447
3448 mutex_lock(&lreq->lock);
19079203 3449 if (lreq->is_watch && lreq->committed && !lreq->last_error)
922dab61
ID		 3450 send_linger_ping(lreq);
3451 mutex_unlock(&lreq->lock);
3452 }
422d2cb8 3453
5aea3dcd
ID		 3454 if (found)
3455 list_move_tail(&osd->o_keepalive_item, &slow_osds);
422d2cb8 3456 }
5aea3dcd 3457
7cc5e38f
ID		 3458 if (opts->osd_request_timeout) {
3459 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3460 struct ceph_osd_request *req =
3461 rb_entry(p, struct ceph_osd_request, r_node);
3462
3463 p = rb_next(p); /* abort_request() */
3464
3465 if (time_before(req->r_start_stamp, expiry_cutoff)) {
3466 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3467 req->r_tid, osdc->homeless_osd.o_osd);
3468 abort_request(req, -ETIMEDOUT);
3469 }
3470 }
3471 }
3472
5aea3dcd
ID		 3473 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3474 maybe_request_map(osdc);
3475
422d2cb8 3476 while (!list_empty(&slow_osds)) {
5aea3dcd
ID		 3477 struct ceph_osd *osd = list_first_entry(&slow_osds,
3478 struct ceph_osd,
3479 o_keepalive_item);
422d2cb8 3480 list_del_init(&osd->o_keepalive_item);
f24e9980
SW		 3481 ceph_con_keepalive(&osd->o_con);
3482 }
3483
5aea3dcd 3484 up_write(&osdc->lock);
fbca9635
ID		 3485 schedule_delayed_work(&osdc->timeout_work,
3486 osdc->client->options->osd_keepalive_timeout);
f24e9980
SW		 3487}
3488
f5a2041b
YS		 3489static void handle_osds_timeout(struct work_struct *work)
3490{
3491 struct ceph_osd_client *osdc =
3492 container_of(work, struct ceph_osd_client,
3493 osds_timeout_work.work);
a319bf56 3494 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
42a2c09f 3495 struct ceph_osd *osd, *nosd;
f5a2041b 3496
42a2c09f 3497 dout("%s osdc %p\n", __func__, osdc);
5aea3dcd 3498 down_write(&osdc->lock);
42a2c09f
ID		 3499 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3500 if (time_before(jiffies, osd->lru_ttl))
3501 break;
3502
5aea3dcd 3503 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
922dab61 3504 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
5aea3dcd 3505 close_osd(osd);
42a2c09f 3506 }
f5a2041b 3507
5aea3dcd 3508 up_write(&osdc->lock);
f5a2041b
YS		 3509 schedule_delayed_work(&osdc->osds_timeout_work,
3510 round_jiffies_relative(delay));
3511}
3512
205ee118
ID		 3513static int ceph_oloc_decode(void **p, void *end,
3514 struct ceph_object_locator *oloc)
3515{
3516 u8 struct_v, struct_cv;
3517 u32 len;
3518 void *struct_end;
3519 int ret = 0;
3520
3521 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3522 struct_v = ceph_decode_8(p);
3523 struct_cv = ceph_decode_8(p);
3524 if (struct_v < 3) {
3525 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3526 struct_v, struct_cv);
3527 goto e_inval;
3528 }
3529 if (struct_cv > 6) {
3530 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3531 struct_v, struct_cv);
3532 goto e_inval;
3533 }
3534 len = ceph_decode_32(p);
3535 ceph_decode_need(p, end, len, e_inval);
3536 struct_end = *p + len;
3537
3538 oloc->pool = ceph_decode_64(p);
3539 *p += 4; /* skip preferred */
3540
3541 len = ceph_decode_32(p);
3542 if (len > 0) {
3543 pr_warn("ceph_object_locator::key is set\n");
3544 goto e_inval;
3545 }
3546
3547 if (struct_v >= 5) {
cd08e0a2
YZ		 3548 bool changed = false;
3549
205ee118
ID		 3550 len = ceph_decode_32(p);
3551 if (len > 0) {
30c156d9 3552 ceph_decode_need(p, end, len, e_inval);
cd08e0a2
YZ		 3553 if (!oloc->pool_ns ||
3554 ceph_compare_string(oloc->pool_ns, *p, len))
3555 changed = true;
30c156d9 3556 *p += len;
cd08e0a2
YZ		 3557 } else {
3558 if (oloc->pool_ns)
3559 changed = true;
3560 }
3561 if (changed) {
3562 /* redirect changes namespace */
3563 pr_warn("ceph_object_locator::nspace is changed\n");
3564 goto e_inval;
205ee118
ID		 3565 }
3566 }
3567
3568 if (struct_v >= 6) {
3569 s64 hash = ceph_decode_64(p);
3570 if (hash != -1) {
3571 pr_warn("ceph_object_locator::hash is set\n");
3572 goto e_inval;
3573 }
3574 }
3575
3576 /* skip the rest */
3577 *p = struct_end;
3578out:
3579 return ret;
3580
3581e_inval:
3582 ret = -EINVAL;
3583 goto out;
3584}
3585
3586static int ceph_redirect_decode(void **p, void *end,
3587 struct ceph_request_redirect *redir)
3588{
3589 u8 struct_v, struct_cv;
3590 u32 len;
3591 void *struct_end;
3592 int ret;
3593
3594 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3595 struct_v = ceph_decode_8(p);
3596 struct_cv = ceph_decode_8(p);
3597 if (struct_cv > 1) {
3598 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3599 struct_v, struct_cv);
3600 goto e_inval;
3601 }
3602 len = ceph_decode_32(p);
3603 ceph_decode_need(p, end, len, e_inval);
3604 struct_end = *p + len;
3605
3606 ret = ceph_oloc_decode(p, end, &redir->oloc);
3607 if (ret)
3608 goto out;
3609
3610 len = ceph_decode_32(p);
3611 if (len > 0) {
3612 pr_warn("ceph_request_redirect::object_name is set\n");
3613 goto e_inval;
3614 }
3615
205ee118
ID		 3616 /* skip the rest */
3617 *p = struct_end;
3618out:
3619 return ret;
3620
3621e_inval:
3622 ret = -EINVAL;
3623 goto out;
3624}
3625
fe5da05e
ID		 3626struct MOSDOpReply {
3627 struct ceph_pg pgid;
3628 u64 flags;
3629 int result;
3630 u32 epoch;
3631 int num_ops;
3632 u32 outdata_len[CEPH_OSD_MAX_OPS];
3633 s32 rval[CEPH_OSD_MAX_OPS];
3634 int retry_attempt;
3635 struct ceph_eversion replay_version;
3636 u64 user_version;
3637 struct ceph_request_redirect redirect;
3638};
25845472 3639
fe5da05e 3640static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
f24e9980 3641{
fe5da05e
ID		 3642 void *p = msg->front.iov_base;
3643 void *const end = p + msg->front.iov_len;
3644 u16 version = le16_to_cpu(msg->hdr.version);
3645 struct ceph_eversion bad_replay_version;
b0b31a8f 3646 u8 decode_redir;
fe5da05e
ID		 3647 u32 len;
3648 int ret;
3649 int i;
1b83bef2 3650
fe5da05e
ID		 3651 ceph_decode_32_safe(&p, end, len, e_inval);
3652 ceph_decode_need(&p, end, len, e_inval);
3653 p += len; /* skip oid */
1b83bef2 3654
fe5da05e
ID		 3655 ret = ceph_decode_pgid(&p, end, &m->pgid);
3656 if (ret)
3657 return ret;
1b83bef2 3658
fe5da05e
ID		 3659 ceph_decode_64_safe(&p, end, m->flags, e_inval);
3660 ceph_decode_32_safe(&p, end, m->result, e_inval);
3661 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3662 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3663 p += sizeof(bad_replay_version);
3664 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
1b83bef2 3665
fe5da05e
ID		 3666 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3667 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3668 goto e_inval;
1b83bef2 3669
fe5da05e
ID		 3670 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3671 e_inval);
3672 for (i = 0; i < m->num_ops; i++) {
1b83bef2 3673 struct ceph_osd_op *op = p;
1b83bef2 3674
fe5da05e 3675 m->outdata_len[i] = le32_to_cpu(op->payload_len);
1b83bef2
SW		 3676 p += sizeof(*op);
3677 }
1b83bef2 3678
fe5da05e
ID		 3679 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3680 for (i = 0; i < m->num_ops; i++)
3681 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
f24e9980 3682
fe5da05e
ID		 3683 if (version >= 5) {
3684 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3685 memcpy(&m->replay_version, p, sizeof(m->replay_version));
3686 p += sizeof(m->replay_version);
3687 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3688 } else {
3689 m->replay_version = bad_replay_version; /* struct */
3690 m->user_version = le64_to_cpu(m->replay_version.version);
3691 }
eb845ff1 3692
fe5da05e
ID		 3693 if (version >= 6) {
3694 if (version >= 7)
3695 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
b0b31a8f
ID		 3696 else
3697 decode_redir = 1;
3698 } else {
3699 decode_redir = 0;
3700 }
3701
3702 if (decode_redir) {
fe5da05e
ID		 3703 ret = ceph_redirect_decode(&p, end, &m->redirect);
3704 if (ret)
3705 return ret;
205ee118 3706 } else {
fe5da05e 3707 ceph_oloc_init(&m->redirect.oloc);
205ee118 3708 }
f24e9980 3709
fe5da05e
ID		 3710 return 0;
3711
3712e_inval:
3713 return -EINVAL;
3714}
3715
3716/*
b18b9550
ID		 3717 * Handle MOSDOpReply. Set ->r_result and call the callback if it is
3718 * specified.
fe5da05e 3719 */
5aea3dcd 3720static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
fe5da05e 3721{
5aea3dcd 3722 struct ceph_osd_client *osdc = osd->o_osdc;
fe5da05e
ID		 3723 struct ceph_osd_request *req;
3724 struct MOSDOpReply m;
3725 u64 tid = le64_to_cpu(msg->hdr.tid);
3726 u32 data_len = 0;
fe5da05e
ID		 3727 int ret;
3728 int i;
3729
3730 dout("%s msg %p tid %llu\n", __func__, msg, tid);
3731
5aea3dcd
ID		 3732 down_read(&osdc->lock);
3733 if (!osd_registered(osd)) {
3734 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3735 goto out_unlock_osdc;
3736 }
3737 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3738
3739 mutex_lock(&osd->lock);
3740 req = lookup_request(&osd->o_requests, tid);
fe5da05e 3741 if (!req) {
5aea3dcd
ID		 3742 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3743 goto out_unlock_session;
fe5da05e 3744 }
fe5da05e 3745
cd08e0a2 3746 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
fe5da05e 3747 ret = decode_MOSDOpReply(msg, &m);
cd08e0a2 3748 m.redirect.oloc.pool_ns = NULL;
fe5da05e
ID		 3749 if (ret) {
3750 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3751 req->r_tid, ret);
3752 ceph_msg_dump(msg);
3753 goto fail_request;
3754 }
3755 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3756 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3757 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3758 le64_to_cpu(m.replay_version.version), m.user_version);
3759
3760 if (m.retry_attempt >= 0) {
3761 if (m.retry_attempt != req->r_attempts - 1) {
3762 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3763 req, req->r_tid, m.retry_attempt,
3764 req->r_attempts - 1);
5aea3dcd 3765 goto out_unlock_session;
fe5da05e
ID		 3766 }
3767 } else {
3768 WARN_ON(1); /* MOSDOpReply v4 is assumed */
3769 }
3770
3771 if (!ceph_oloc_empty(&m.redirect.oloc)) {
3772 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3773 m.redirect.oloc.pool);
5aea3dcd
ID		 3774 unlink_request(osd, req);
3775 mutex_unlock(&osd->lock);
205ee118 3776
30c156d9
YZ		 3777 /*
3778 * Not ceph_oloc_copy() - changing pool_ns is not
3779 * supported.
3780 */
3781 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
890bd0f8
JL		 3782 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3783 CEPH_OSD_FLAG_IGNORE_OVERLAY |
3784 CEPH_OSD_FLAG_IGNORE_CACHE;
5aea3dcd
ID		 3785 req->r_tid = 0;
3786 __submit_request(req, false);
3787 goto out_unlock_osdc;
205ee118
ID		 3788 }
3789
117d96a0
ID		 3790 if (m.result == -EAGAIN) {
3791 dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3792 unlink_request(osd, req);
3793 mutex_unlock(&osd->lock);
3794
3795 /*
3796 * The object is missing on the replica or not (yet)
3797 * readable. Clear pgid to force a resend to the primary
3798 * via legacy_change.
3799 */
3800 req->r_t.pgid.pool = 0;
3801 req->r_t.pgid.seed = 0;
3802 WARN_ON(!req->r_t.used_replica);
3803 req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3804 CEPH_OSD_FLAG_LOCALIZE_READS);
3805 req->r_tid = 0;
3806 __submit_request(req, false);
3807 goto out_unlock_osdc;
3808 }
3809
fe5da05e
ID		 3810 if (m.num_ops != req->r_num_ops) {
3811 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3812 req->r_num_ops, req->r_tid);
3813 goto fail_request;
f24e9980 3814 }
fe5da05e
ID		 3815 for (i = 0; i < req->r_num_ops; i++) {
3816 dout(" req %p tid %llu op %d rval %d len %u\n", req,
3817 req->r_tid, i, m.rval[i], m.outdata_len[i]);
3818 req->r_ops[i].rval = m.rval[i];
3819 req->r_ops[i].outdata_len = m.outdata_len[i];
3820 data_len += m.outdata_len[i];
3821 }
3822 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3823 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3824 le32_to_cpu(msg->hdr.data_len), req->r_tid);
3825 goto fail_request;
3826 }
b18b9550
ID		 3827 dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3828 req, req->r_tid, m.result, data_len);
f24e9980 3829
b18b9550
ID		 3830 /*
3831 * Since we only ever request ONDISK, we should only ever get
3832 * one (type of) reply back.
3833 */
3834 WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3835 req->r_result = m.result ?: data_len;
3836 finish_request(req);
5aea3dcd
ID		 3837 mutex_unlock(&osd->lock);
3838 up_read(&osdc->lock);
f24e9980 3839
b18b9550 3840 __complete_request(req);
f24e9980
SW		 3841 return;
3842
fe5da05e 3843fail_request:
4609245e 3844 complete_request(req, -EIO);
5aea3dcd
ID		 3845out_unlock_session:
3846 mutex_unlock(&osd->lock);
3847out_unlock_osdc:
3848 up_read(&osdc->lock);
f24e9980
SW		 3849}
3850
42c1b124
ID		 3851static void set_pool_was_full(struct ceph_osd_client *osdc)
3852{
3853 struct rb_node *n;
3854
3855 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3856 struct ceph_pg_pool_info *pi =
3857 rb_entry(n, struct ceph_pg_pool_info, node);
3858
3859 pi->was_full = __pool_full(pi);
3860 }
3861}
3862
5aea3dcd 3863static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
f24e9980 3864{
5aea3dcd 3865 struct ceph_pg_pool_info *pi;
f24e9980 3866
5aea3dcd
ID		 3867 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3868 if (!pi)
3869 return false;
f24e9980 3870
5aea3dcd 3871 return pi->was_full && !__pool_full(pi);
422d2cb8
YS		 3872}
3873
922dab61
ID		 3874static enum calc_target_result
3875recalc_linger_target(struct ceph_osd_linger_request *lreq)
3876{
3877 struct ceph_osd_client *osdc = lreq->osdc;
3878 enum calc_target_result ct_res;
3879
8edf84ba 3880 ct_res = calc_target(osdc, &lreq->t, true);
922dab61
ID		 3881 if (ct_res == CALC_TARGET_NEED_RESEND) {
3882 struct ceph_osd *osd;
3883
3884 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3885 if (osd != lreq->osd) {
3886 unlink_linger(lreq->osd, lreq);
3887 link_linger(osd, lreq);
3888 }
3889 }
3890
3891 return ct_res;
3892}
3893
422d2cb8 3894/*
5aea3dcd 3895 * Requeue requests whose mapping to an OSD has changed.
422d2cb8 3896 */
5aea3dcd
ID		 3897static void scan_requests(struct ceph_osd *osd,
3898 bool force_resend,
3899 bool cleared_full,
3900 bool check_pool_cleared_full,
3901 struct rb_root *need_resend,
3902 struct list_head *need_resend_linger)
422d2cb8 3903{
5aea3dcd
ID		 3904 struct ceph_osd_client *osdc = osd->o_osdc;
3905 struct rb_node *n;
3906 bool force_resend_writes;
3907
922dab61
ID		 3908 for (n = rb_first(&osd->o_linger_requests); n; ) {
3909 struct ceph_osd_linger_request *lreq =
3910 rb_entry(n, struct ceph_osd_linger_request, node);
3911 enum calc_target_result ct_res;
3912
3913 n = rb_next(n); /* recalc_linger_target() */
3914
3915 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3916 lreq->linger_id);
3917 ct_res = recalc_linger_target(lreq);
3918 switch (ct_res) {
3919 case CALC_TARGET_NO_ACTION:
3920 force_resend_writes = cleared_full ||
3921 (check_pool_cleared_full &&
3922 pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3923 if (!force_resend && !force_resend_writes)
3924 break;
3925
df561f66 3926 fallthrough;
922dab61 3927 case CALC_TARGET_NEED_RESEND:
4609245e 3928 cancel_linger_map_check(lreq);
922dab61
ID		 3929 /*
3930 * scan_requests() for the previous epoch(s)
3931 * may have already added it to the list, since
3932 * it's not unlinked here.
3933 */
3934 if (list_empty(&lreq->scan_item))
3935 list_add_tail(&lreq->scan_item, need_resend_linger);
3936 break;
3937 case CALC_TARGET_POOL_DNE:
a10bcb19 3938 list_del_init(&lreq->scan_item);
4609245e 3939 check_linger_pool_dne(lreq);
922dab61
ID		 3940 break;
3941 }
3942 }
3943
5aea3dcd
ID		 3944 for (n = rb_first(&osd->o_requests); n; ) {
3945 struct ceph_osd_request *req =
3946 rb_entry(n, struct ceph_osd_request, r_node);
3947 enum calc_target_result ct_res;
3948
4609245e 3949 n = rb_next(n); /* unlink_request(), check_pool_dne() */
5aea3dcd
ID		 3950
3951 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
8edf84ba 3952 ct_res = calc_target(osdc, &req->r_t, false);
5aea3dcd
ID		 3953 switch (ct_res) {
3954 case CALC_TARGET_NO_ACTION:
3955 force_resend_writes = cleared_full ||
3956 (check_pool_cleared_full &&
3957 pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3958 if (!force_resend &&
3959 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3960 !force_resend_writes))
3961 break;
3962
df561f66 3963 fallthrough;
5aea3dcd 3964 case CALC_TARGET_NEED_RESEND:
4609245e 3965 cancel_map_check(req);
5aea3dcd
ID		 3966 unlink_request(osd, req);
3967 insert_request(need_resend, req);
3968 break;
3969 case CALC_TARGET_POOL_DNE:
4609245e 3970 check_pool_dne(req);
5aea3dcd 3971 break;
b0494532 3972 }
6f6c7006 3973 }
422d2cb8 3974}
6f6c7006 3975
42c1b124 3976static int handle_one_map(struct ceph_osd_client *osdc,
5aea3dcd
ID		 3977 void *p, void *end, bool incremental,
3978 struct rb_root *need_resend,
3979 struct list_head *need_resend_linger)
42c1b124
ID		 3980{
3981 struct ceph_osdmap *newmap;
3982 struct rb_node *n;
3983 bool skipped_map = false;
3984 bool was_full;
3985
b7ec35b3 3986 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
42c1b124
ID		 3987 set_pool_was_full(osdc);
3988
3989 if (incremental)
cd1a677c
ID		 3990 newmap = osdmap_apply_incremental(&p, end,
3991 ceph_msgr2(osdc->client),
3992 osdc->osdmap);
42c1b124 3993 else
cd1a677c 3994 newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client));
42c1b124
ID		 3995 if (IS_ERR(newmap))
3996 return PTR_ERR(newmap);
3997
3998 if (newmap != osdc->osdmap) {
3999 /*
4000 * Preserve ->was_full before destroying the old map.
4001 * For pools that weren't in the old map, ->was_full
4002 * should be false.
4003 */
4004 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
4005 struct ceph_pg_pool_info *pi =
4006 rb_entry(n, struct ceph_pg_pool_info, node);
4007 struct ceph_pg_pool_info *old_pi;
4008
4009 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
4010 if (old_pi)
4011 pi->was_full = old_pi->was_full;
4012 else
4013 WARN_ON(pi->was_full);
4014 }
4015
4016 if (osdc->osdmap->epoch &&
4017 osdc->osdmap->epoch + 1 < newmap->epoch) {
4018 WARN_ON(incremental);
4019 skipped_map = true;
4020 }
4021
4022 ceph_osdmap_destroy(osdc->osdmap);
4023 osdc->osdmap = newmap;
4024 }
4025
b7ec35b3 4026 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
5aea3dcd
ID		 4027 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
4028 need_resend, need_resend_linger);
4029
4030 for (n = rb_first(&osdc->osds); n; ) {
4031 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4032
4033 n = rb_next(n); /* close_osd() */
4034
4035 scan_requests(osd, skipped_map, was_full, true, need_resend,
4036 need_resend_linger);
4037 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
4038 memcmp(&osd->o_con.peer_addr,
4039 ceph_osd_addr(osdc->osdmap, osd->o_osd),
4040 sizeof(struct ceph_entity_addr)))
4041 close_osd(osd);
4042 }
42c1b124
ID		 4043
4044 return 0;
4045}
6f6c7006 4046
5aea3dcd
ID		 4047static void kick_requests(struct ceph_osd_client *osdc,
4048 struct rb_root *need_resend,
4049 struct list_head *need_resend_linger)
4050{
922dab61 4051 struct ceph_osd_linger_request *lreq, *nlreq;
04c7d789 4052 enum calc_target_result ct_res;
5aea3dcd
ID		 4053 struct rb_node *n;
4054
04c7d789
ID		 4055 /* make sure need_resend targets reflect latest map */
4056 for (n = rb_first(need_resend); n; ) {
4057 struct ceph_osd_request *req =
4058 rb_entry(n, struct ceph_osd_request, r_node);
4059
4060 n = rb_next(n);
4061
4062 if (req->r_t.epoch < osdc->osdmap->epoch) {
8edf84ba 4063 ct_res = calc_target(osdc, &req->r_t, false);
04c7d789
ID		 4064 if (ct_res == CALC_TARGET_POOL_DNE) {
4065 erase_request(need_resend, req);
4066 check_pool_dne(req);
4067 }
4068 }
4069 }
4070
5aea3dcd
ID		 4071 for (n = rb_first(need_resend); n; ) {
4072 struct ceph_osd_request *req =
4073 rb_entry(n, struct ceph_osd_request, r_node);
4074 struct ceph_osd *osd;
4075
4076 n = rb_next(n);
4077 erase_request(need_resend, req); /* before link_request() */
4078
5aea3dcd
ID		 4079 osd = lookup_create_osd(osdc, req->r_t.osd, true);
4080 link_request(osd, req);
4081 if (!req->r_linger) {
4082 if (!osd_homeless(osd) && !req->r_t.paused)
4083 send_request(req);
922dab61
ID		 4084 } else {
4085 cancel_linger_request(req);
5aea3dcd
ID		 4086 }
4087 }
922dab61
ID		 4088
4089 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4090 if (!osd_homeless(lreq->osd))
4091 send_linger(lreq);
4092
4093 list_del_init(&lreq->scan_item);
4094 }
5aea3dcd
ID		 4095}
4096
f24e9980
SW		 4097/*
4098 * Process updated osd map.
4099 *
4100 * The message contains any number of incremental and full maps, normally
4101 * indicating some sort of topology change in the cluster. Kick requests
4102 * off to different OSDs as needed.
4103 */
4104void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4105{
42c1b124
ID		 4106 void *p = msg->front.iov_base;
4107 void *const end = p + msg->front.iov_len;
f24e9980
SW		 4108 u32 nr_maps, maplen;
4109 u32 epoch;
f24e9980 4110 struct ceph_fsid fsid;
5aea3dcd
ID		 4111 struct rb_root need_resend = RB_ROOT;
4112 LIST_HEAD(need_resend_linger);
42c1b124
ID		 4113 bool handled_incremental = false;
4114 bool was_pauserd, was_pausewr;
4115 bool pauserd, pausewr;
4116 int err;
f24e9980 4117
42c1b124 4118 dout("%s have %u\n", __func__, osdc->osdmap->epoch);
5aea3dcd 4119 down_write(&osdc->lock);
f24e9980
SW		 4120
4121 /* verify fsid */
4122 ceph_decode_need(&p, end, sizeof(fsid), bad);
4123 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d 4124 if (ceph_check_fsid(osdc->client, &fsid) < 0)
42c1b124 4125 goto bad;
f24e9980 4126
b7ec35b3
ID		 4127 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4128 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4129 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
42c1b124 4130 have_pool_full(osdc);
9a1ea2db 4131
f24e9980
SW		 4132 /* incremental maps */
4133 ceph_decode_32_safe(&p, end, nr_maps, bad);
4134 dout(" %d inc maps\n", nr_maps);
4135 while (nr_maps > 0) {
4136 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 4137 epoch = ceph_decode_32(&p);
4138 maplen = ceph_decode_32(&p);
f24e9980 4139 ceph_decode_need(&p, end, maplen, bad);
42c1b124
ID		 4140 if (osdc->osdmap->epoch &&
4141 osdc->osdmap->epoch + 1 == epoch) {
f24e9980
SW		 4142 dout("applying incremental map %u len %d\n",
4143 epoch, maplen);
5aea3dcd
ID		 4144 err = handle_one_map(osdc, p, p + maplen, true,
4145 &need_resend, &need_resend_linger);
42c1b124 4146 if (err)
f24e9980 4147 goto bad;
42c1b124 4148 handled_incremental = true;
f24e9980
SW		 4149 } else {
4150 dout("ignoring incremental map %u len %d\n",
4151 epoch, maplen);
4152 }
42c1b124 4153 p += maplen;
f24e9980
SW		 4154 nr_maps--;
4155 }
42c1b124 4156 if (handled_incremental)
f24e9980
SW		 4157 goto done;
4158
4159 /* full maps */
4160 ceph_decode_32_safe(&p, end, nr_maps, bad);
4161 dout(" %d full maps\n", nr_maps);
4162 while (nr_maps) {
4163 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 4164 epoch = ceph_decode_32(&p);
4165 maplen = ceph_decode_32(&p);
f24e9980
SW		 4166 ceph_decode_need(&p, end, maplen, bad);
4167 if (nr_maps > 1) {
4168 dout("skipping non-latest full map %u len %d\n",
4169 epoch, maplen);
e5253a7b 4170 } else if (osdc->osdmap->epoch >= epoch) {
f24e9980
SW		 4171 dout("skipping full map %u len %d, "
4172 "older than our %u\n", epoch, maplen,
4173 osdc->osdmap->epoch);
4174 } else {
4175 dout("taking full map %u len %d\n", epoch, maplen);
5aea3dcd
ID		 4176 err = handle_one_map(osdc, p, p + maplen, false,
4177 &need_resend, &need_resend_linger);
42c1b124 4178 if (err)
f24e9980 4179 goto bad;
f24e9980
SW		 4180 }
4181 p += maplen;
4182 nr_maps--;
4183 }
4184
4185done:
cd634fb6
SW		 4186 /*
4187 * subscribe to subsequent osdmap updates if full to ensure
4188 * we find out when we are no longer full and stop returning
4189 * ENOSPC.
4190 */
b7ec35b3
ID		 4191 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4192 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4193 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
42c1b124 4194 have_pool_full(osdc);
58eb7932
JL		 4195 if (was_pauserd || was_pausewr || pauserd || pausewr ||
4196 osdc->osdmap->epoch < osdc->epoch_barrier)
42c1b124 4197 maybe_request_map(osdc);
cd634fb6 4198
5aea3dcd 4199 kick_requests(osdc, &need_resend, &need_resend_linger);
42c1b124 4200
fc36d0a4 4201 ceph_osdc_abort_on_full(osdc);
42c1b124
ID		 4202 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4203 osdc->osdmap->epoch);
5aea3dcd 4204 up_write(&osdc->lock);
03066f23 4205 wake_up_all(&osdc->client->auth_wq);
f24e9980
SW		 4206 return;
4207
4208bad:
4209 pr_err("osdc handle_map corrupt msg\n");
9ec7cab1 4210 ceph_msg_dump(msg);
5aea3dcd
ID		 4211 up_write(&osdc->lock);
4212}
4213
4214/*
4215 * Resubmit requests pending on the given osd.
4216 */
4217static void kick_osd_requests(struct ceph_osd *osd)
4218{
4219 struct rb_node *n;
4220
a02a946d
ID		 4221 clear_backoffs(osd);
4222
922dab61 4223 for (n = rb_first(&osd->o_requests); n; ) {
5aea3dcd
ID		 4224 struct ceph_osd_request *req =
4225 rb_entry(n, struct ceph_osd_request, r_node);
4226
922dab61
ID		 4227 n = rb_next(n); /* cancel_linger_request() */
4228
5aea3dcd
ID		 4229 if (!req->r_linger) {
4230 if (!req->r_t.paused)
4231 send_request(req);
922dab61
ID		 4232 } else {
4233 cancel_linger_request(req);
5aea3dcd
ID		 4234 }
4235 }
922dab61
ID		 4236 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4237 struct ceph_osd_linger_request *lreq =
4238 rb_entry(n, struct ceph_osd_linger_request, node);
4239
4240 send_linger(lreq);
4241 }
5aea3dcd
ID		 4242}
4243
4244/*
4245 * If the osd connection drops, we need to resubmit all requests.
4246 */
4247static void osd_fault(struct ceph_connection *con)
4248{
4249 struct ceph_osd *osd = con->private;
4250 struct ceph_osd_client *osdc = osd->o_osdc;
4251
4252 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4253
4254 down_write(&osdc->lock);
4255 if (!osd_registered(osd)) {
4256 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4257 goto out_unlock;
4258 }
4259
4260 if (!reopen_osd(osd))
4261 kick_osd_requests(osd);
4262 maybe_request_map(osdc);
4263
4264out_unlock:
4265 up_write(&osdc->lock);
f24e9980
SW		 4266}
4267
a02a946d
ID		 4268struct MOSDBackoff {
4269 struct ceph_spg spgid;
4270 u32 map_epoch;
4271 u8 op;
4272 u64 id;
4273 struct ceph_hobject_id *begin;
4274 struct ceph_hobject_id *end;
4275};
4276
4277static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4278{
4279 void *p = msg->front.iov_base;
4280 void *const end = p + msg->front.iov_len;
4281 u8 struct_v;
4282 u32 struct_len;
4283 int ret;
4284
4285 ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4286 if (ret)
4287 return ret;
4288
4289 ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4290 if (ret)
4291 return ret;
4292
4293 ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4294 ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4295 ceph_decode_8_safe(&p, end, m->op, e_inval);
4296 ceph_decode_64_safe(&p, end, m->id, e_inval);
4297
4298 m->begin = kzalloc(sizeof(*m->begin), GFP_NOIO);
4299 if (!m->begin)
4300 return -ENOMEM;
4301
4302 ret = decode_hoid(&p, end, m->begin);
4303 if (ret) {
4304 free_hoid(m->begin);
4305 return ret;
4306 }
4307
4308 m->end = kzalloc(sizeof(*m->end), GFP_NOIO);
4309 if (!m->end) {
4310 free_hoid(m->begin);
4311 return -ENOMEM;
4312 }
4313
4314 ret = decode_hoid(&p, end, m->end);
4315 if (ret) {
4316 free_hoid(m->begin);
4317 free_hoid(m->end);
4318 return ret;
4319 }
4320
4321 return 0;
4322
4323e_inval:
4324 return -EINVAL;
4325}
4326
4327static struct ceph_msg *create_backoff_message(
4328 const struct ceph_osd_backoff *backoff,
4329 u32 map_epoch)
4330{
4331 struct ceph_msg *msg;
4332 void *p, *end;
4333 int msg_size;
4334
4335 msg_size = CEPH_ENCODING_START_BLK_LEN +
4336 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4337 msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4338 msg_size += CEPH_ENCODING_START_BLK_LEN +
4339 hoid_encoding_size(backoff->begin);
4340 msg_size += CEPH_ENCODING_START_BLK_LEN +
4341 hoid_encoding_size(backoff->end);
4342
4343 msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4344 if (!msg)
4345 return NULL;
4346
4347 p = msg->front.iov_base;
4348 end = p + msg->front_alloc_len;
4349
4350 encode_spgid(&p, &backoff->spgid);
4351 ceph_encode_32(&p, map_epoch);
4352 ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4353 ceph_encode_64(&p, backoff->id);
4354 encode_hoid(&p, end, backoff->begin);
4355 encode_hoid(&p, end, backoff->end);
4356 BUG_ON(p != end);
4357
4358 msg->front.iov_len = p - msg->front.iov_base;
4359 msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4360 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4361
4362 return msg;
4363}
4364
4365static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4366{
4367 struct ceph_spg_mapping *spg;
4368 struct ceph_osd_backoff *backoff;
4369 struct ceph_msg *msg;
4370
4371 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4372 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4373
4374 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4375 if (!spg) {
4376 spg = alloc_spg_mapping();
4377 if (!spg) {
4378 pr_err("%s failed to allocate spg\n", __func__);
4379 return;
4380 }
4381 spg->spgid = m->spgid; /* struct */
4382 insert_spg_mapping(&osd->o_backoff_mappings, spg);
4383 }
4384
4385 backoff = alloc_backoff();
4386 if (!backoff) {
4387 pr_err("%s failed to allocate backoff\n", __func__);
4388 return;
4389 }
4390 backoff->spgid = m->spgid; /* struct */
4391 backoff->id = m->id;
4392 backoff->begin = m->begin;
4393 m->begin = NULL; /* backoff now owns this */
4394 backoff->end = m->end;
4395 m->end = NULL; /* ditto */
4396
4397 insert_backoff(&spg->backoffs, backoff);
4398 insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4399
4400 /*
4401 * Ack with original backoff's epoch so that the OSD can
4402 * discard this if there was a PG split.
4403 */
4404 msg = create_backoff_message(backoff, m->map_epoch);
4405 if (!msg) {
4406 pr_err("%s failed to allocate msg\n", __func__);
4407 return;
4408 }
4409 ceph_con_send(&osd->o_con, msg);
4410}
4411
4412static bool target_contained_by(const struct ceph_osd_request_target *t,
4413 const struct ceph_hobject_id *begin,
4414 const struct ceph_hobject_id *end)
4415{
4416 struct ceph_hobject_id hoid;
4417 int cmp;
4418
4419 hoid_fill_from_target(&hoid, t);
4420 cmp = hoid_compare(&hoid, begin);
4421 return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4422}
4423
4424static void handle_backoff_unblock(struct ceph_osd *osd,
4425 const struct MOSDBackoff *m)
4426{
4427 struct ceph_spg_mapping *spg;
4428 struct ceph_osd_backoff *backoff;
4429 struct rb_node *n;
4430
4431 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4432 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4433
4434 backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4435 if (!backoff) {
4436 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4437 __func__, osd->o_osd, m->spgid.pgid.pool,
4438 m->spgid.pgid.seed, m->spgid.shard, m->id);
4439 return;
4440 }
4441
4442 if (hoid_compare(backoff->begin, m->begin) &&
4443 hoid_compare(backoff->end, m->end)) {
4444 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4445 __func__, osd->o_osd, m->spgid.pgid.pool,
4446 m->spgid.pgid.seed, m->spgid.shard, m->id);
4447 /* unblock it anyway... */
4448 }
4449
4450 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4451 BUG_ON(!spg);
4452
4453 erase_backoff(&spg->backoffs, backoff);
4454 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4455 free_backoff(backoff);
4456
4457 if (RB_EMPTY_ROOT(&spg->backoffs)) {
4458 erase_spg_mapping(&osd->o_backoff_mappings, spg);
4459 free_spg_mapping(spg);
4460 }
4461
4462 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4463 struct ceph_osd_request *req =
4464 rb_entry(n, struct ceph_osd_request, r_node);
4465
4466 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4467 /*
4468 * Match against @m, not @backoff -- the PG may
4469 * have split on the OSD.
4470 */
4471 if (target_contained_by(&req->r_t, m->begin, m->end)) {
4472 /*
4473 * If no other installed backoff applies,
4474 * resend.
4475 */
4476 send_request(req);
4477 }
4478 }
4479 }
4480}
4481
4482static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4483{
4484 struct ceph_osd_client *osdc = osd->o_osdc;
4485 struct MOSDBackoff m;
4486 int ret;
4487
4488 down_read(&osdc->lock);
4489 if (!osd_registered(osd)) {
4490 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4491 up_read(&osdc->lock);
4492 return;
4493 }
4494 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4495
4496 mutex_lock(&osd->lock);
4497 ret = decode_MOSDBackoff(msg, &m);
4498 if (ret) {
4499 pr_err("failed to decode MOSDBackoff: %d\n", ret);
4500 ceph_msg_dump(msg);
4501 goto out_unlock;
4502 }
4503
4504 switch (m.op) {
4505 case CEPH_OSD_BACKOFF_OP_BLOCK:
4506 handle_backoff_block(osd, &m);
4507 break;
4508 case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4509 handle_backoff_unblock(osd, &m);
4510 break;
4511 default:
4512 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4513 }
4514
4515 free_hoid(m.begin);
4516 free_hoid(m.end);
4517
4518out_unlock:
4519 mutex_unlock(&osd->lock);
4520 up_read(&osdc->lock);
4521}
4522
a40c4f10
YS		 4523/*
4524 * Process osd watch notifications
4525 */
3c663bbd
AE		 4526static void handle_watch_notify(struct ceph_osd_client *osdc,
4527 struct ceph_msg *msg)
a40c4f10 4528{
922dab61
ID		 4529 void *p = msg->front.iov_base;
4530 void *const end = p + msg->front.iov_len;
4531 struct ceph_osd_linger_request *lreq;
4532 struct linger_work *lwork;
4533 u8 proto_ver, opcode;
4534 u64 cookie, notify_id;
4535 u64 notifier_id = 0;
19079203 4536 s32 return_code = 0;
922dab61
ID		 4537 void *payload = NULL;
4538 u32 payload_len = 0;
a40c4f10
YS		 4539
4540 ceph_decode_8_safe(&p, end, proto_ver, bad);
4541 ceph_decode_8_safe(&p, end, opcode, bad);
4542 ceph_decode_64_safe(&p, end, cookie, bad);
922dab61 4543 p += 8; /* skip ver */
a40c4f10
YS		 4544 ceph_decode_64_safe(&p, end, notify_id, bad);
4545
922dab61
ID		 4546 if (proto_ver >= 1) {
4547 ceph_decode_32_safe(&p, end, payload_len, bad);
4548 ceph_decode_need(&p, end, payload_len, bad);
4549 payload = p;
4550 p += payload_len;
4551 }
4552
4553 if (le16_to_cpu(msg->hdr.version) >= 2)
19079203 4554 ceph_decode_32_safe(&p, end, return_code, bad);
922dab61
ID		 4555
4556 if (le16_to_cpu(msg->hdr.version) >= 3)
4557 ceph_decode_64_safe(&p, end, notifier_id, bad);
4558
4559 down_read(&osdc->lock);
4560 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4561 if (!lreq) {
4562 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4563 cookie);
4564 goto out_unlock_osdc;
4565 }
4566
4567 mutex_lock(&lreq->lock);
19079203
ID		 4568 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4569 opcode, cookie, lreq, lreq->is_watch);
922dab61
ID		 4570 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4571 if (!lreq->last_error) {
4572 lreq->last_error = -ENOTCONN;
4573 queue_watch_error(lreq);
4574 }
19079203
ID		 4575 } else if (!lreq->is_watch) {
4576 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4577 if (lreq->notify_id && lreq->notify_id != notify_id) {
4578 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4579 lreq->notify_id, notify_id);
4580 } else if (!completion_done(&lreq->notify_finish_wait)) {
4581 struct ceph_msg_data *data =
0d9c1ab3 4582 msg->num_data_items ? &msg->data[0] : NULL;
19079203
ID		 4583
4584 if (data) {
4585 if (lreq->preply_pages) {
4586 WARN_ON(data->type !=
4587 CEPH_MSG_DATA_PAGES);
4588 *lreq->preply_pages = data->pages;
4589 *lreq->preply_len = data->length;
e8862740 4590 data->own_pages = false;
19079203
ID		 4591 }
4592 }
4593 lreq->notify_finish_error = return_code;
4594 complete_all(&lreq->notify_finish_wait);
4595 }
922dab61
ID		 4596 } else {
4597 /* CEPH_WATCH_EVENT_NOTIFY */
4598 lwork = lwork_alloc(lreq, do_watch_notify);
4599 if (!lwork) {
4600 pr_err("failed to allocate notify-lwork\n");
4601 goto out_unlock_lreq;
a40c4f10 4602 }
a40c4f10 4603
922dab61
ID		 4604 lwork->notify.notify_id = notify_id;
4605 lwork->notify.notifier_id = notifier_id;
4606 lwork->notify.payload = payload;
4607 lwork->notify.payload_len = payload_len;
4608 lwork->notify.msg = ceph_msg_get(msg);
4609 lwork_queue(lwork);
91883cd2 4610 }
a40c4f10 4611
922dab61
ID		 4612out_unlock_lreq:
4613 mutex_unlock(&lreq->lock);
4614out_unlock_osdc:
4615 up_read(&osdc->lock);
a40c4f10
YS		 4616 return;
4617
4618bad:
4619 pr_err("osdc handle_watch_notify corrupt msg\n");
a40c4f10
YS		 4620}
4621
70636773
AE		 4622/*
4623 * Register request, send initial attempt.
4624 */
a8af0d68
JL		 4625void ceph_osdc_start_request(struct ceph_osd_client *osdc,
4626 struct ceph_osd_request *req)
70636773 4627{
5aea3dcd
ID		 4628 down_read(&osdc->lock);
4629 submit_request(req, false);
4630 up_read(&osdc->lock);
f24e9980 4631}
3d14c5d2 4632EXPORT_SYMBOL(ceph_osdc_start_request);
f24e9980 4633
c9f9b93d 4634/*
d0bb883c
ID		 4635 * Unregister request. If @req was registered, it isn't completed:
4636 * r_result isn't set and __complete_request() isn't invoked.
4637 *
4638 * If @req wasn't registered, this call may have raced with
4639 * handle_reply(), in which case r_result would already be set and
4640 * __complete_request() would be getting invoked, possibly even
4641 * concurrently with this call.
c9f9b93d
ID		 4642 */
4643void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4644{
4645 struct ceph_osd_client *osdc = req->r_osdc;
4646
5aea3dcd 4647 down_write(&osdc->lock);
5aea3dcd
ID		 4648 if (req->r_osd)
4649 cancel_request(req);
4650 up_write(&osdc->lock);
c9f9b93d
ID		 4651}
4652EXPORT_SYMBOL(ceph_osdc_cancel_request);
4653
f24e9980 4654/*
42b06965 4655 * @timeout: in jiffies, 0 means "wait forever"
f24e9980 4656 */
42b06965
ID		 4657static int wait_request_timeout(struct ceph_osd_request *req,
4658 unsigned long timeout)
f24e9980 4659{
42b06965 4660 long left;
c9f9b93d 4661
42b06965 4662 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
0e76abf2 4663 left = wait_for_completion_killable_timeout(&req->r_completion,
42b06965
ID		 4664 ceph_timeout_jiffies(timeout));
4665 if (left <= 0) {
4666 left = left ?: -ETIMEDOUT;
c9f9b93d 4667 ceph_osdc_cancel_request(req);
42b06965
ID		 4668 } else {
4669 left = req->r_result; /* completed */
f24e9980
SW		 4670 }
4671
42b06965
ID		 4672 return left;
4673}
4674
4675/*
4676 * wait for a request to complete
4677 */
4678int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4679 struct ceph_osd_request *req)
4680{
4681 return wait_request_timeout(req, 0);
f24e9980 4682}
3d14c5d2 4683EXPORT_SYMBOL(ceph_osdc_wait_request);
f24e9980
SW		 4684
4685/*
4686 * sync - wait for all in-flight requests to flush. avoid starvation.
4687 */
4688void ceph_osdc_sync(struct ceph_osd_client *osdc)
4689{
5aea3dcd
ID		 4690 struct rb_node *n, *p;
4691 u64 last_tid = atomic64_read(&osdc->last_tid);
f24e9980 4692
5aea3dcd
ID		 4693again:
4694 down_read(&osdc->lock);
4695 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4696 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4697
4698 mutex_lock(&osd->lock);
4699 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4700 struct ceph_osd_request *req =
4701 rb_entry(p, struct ceph_osd_request, r_node);
4702
4703 if (req->r_tid > last_tid)
4704 break;
4705
4706 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4707 continue;
f24e9980 4708
5aea3dcd
ID		 4709 ceph_osdc_get_request(req);
4710 mutex_unlock(&osd->lock);
4711 up_read(&osdc->lock);
4712 dout("%s waiting on req %p tid %llu last_tid %llu\n",
4713 __func__, req, req->r_tid, last_tid);
b18b9550 4714 wait_for_completion(&req->r_completion);
5aea3dcd
ID		 4715 ceph_osdc_put_request(req);
4716 goto again;
4717 }
f24e9980 4718
5aea3dcd 4719 mutex_unlock(&osd->lock);
f24e9980 4720 }
5aea3dcd
ID		 4721
4722 up_read(&osdc->lock);
4723 dout("%s done last_tid %llu\n", __func__, last_tid);
f24e9980 4724}
3d14c5d2 4725EXPORT_SYMBOL(ceph_osdc_sync);
f24e9980 4726
922dab61
ID		 4727/*
4728 * Returns a handle, caller owns a ref.
4729 */
4730struct ceph_osd_linger_request *
4731ceph_osdc_watch(struct ceph_osd_client *osdc,
4732 struct ceph_object_id *oid,
4733 struct ceph_object_locator *oloc,
4734 rados_watchcb2_t wcb,
4735 rados_watcherrcb_t errcb,
4736 void *data)
4737{
4738 struct ceph_osd_linger_request *lreq;
4739 int ret;
4740
4741 lreq = linger_alloc(osdc);
4742 if (!lreq)
4743 return ERR_PTR(-ENOMEM);
4744
19079203 4745 lreq->is_watch = true;
922dab61
ID		 4746 lreq->wcb = wcb;
4747 lreq->errcb = errcb;
4748 lreq->data = data;
b07d3c4b 4749 lreq->watch_valid_thru = jiffies;
922dab61
ID		 4750
4751 ceph_oid_copy(&lreq->t.base_oid, oid);
4752 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
54ea0046 4753 lreq->t.flags = CEPH_OSD_FLAG_WRITE;
fac02ddf 4754 ktime_get_real_ts64(&lreq->mtime);
922dab61 4755
81c65213 4756 linger_submit(lreq);
922dab61
ID		 4757 ret = linger_reg_commit_wait(lreq);
4758 if (ret) {
4759 linger_cancel(lreq);
4760 goto err_put_lreq;
4761 }
4762
4763 return lreq;
4764
4765err_put_lreq:
4766 linger_put(lreq);
4767 return ERR_PTR(ret);
4768}
4769EXPORT_SYMBOL(ceph_osdc_watch);
4770
4771/*
4772 * Releases a ref.
4773 *
4774 * Times out after mount_timeout to preserve rbd unmap behaviour
4775 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4776 * with mount_timeout").
4777 */
4778int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4779 struct ceph_osd_linger_request *lreq)
4780{
4781 struct ceph_options *opts = osdc->client->options;
4782 struct ceph_osd_request *req;
4783 int ret;
4784
4785 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4786 if (!req)
4787 return -ENOMEM;
4788
4789 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4790 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
54ea0046 4791 req->r_flags = CEPH_OSD_FLAG_WRITE;
fac02ddf 4792 ktime_get_real_ts64(&req->r_mtime);
75dbb685
ID		 4793 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_UNWATCH,
4794 lreq->linger_id, 0);
922dab61
ID		 4795
4796 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4797 if (ret)
4798 goto out_put_req;
4799
a8af0d68 4800 ceph_osdc_start_request(osdc, req);
922dab61
ID		 4801 linger_cancel(lreq);
4802 linger_put(lreq);
4803 ret = wait_request_timeout(req, opts->mount_timeout);
4804
4805out_put_req:
4806 ceph_osdc_put_request(req);
4807 return ret;
4808}
4809EXPORT_SYMBOL(ceph_osdc_unwatch);
4810
4811static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4812 u64 notify_id, u64 cookie, void *payload,
6d54228f 4813 u32 payload_len)
922dab61
ID		 4814{
4815 struct ceph_osd_req_op *op;
4816 struct ceph_pagelist *pl;
4817 int ret;
4818
042f6498 4819 op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
922dab61 4820
33165d47 4821 pl = ceph_pagelist_alloc(GFP_NOIO);
922dab61
ID		 4822 if (!pl)
4823 return -ENOMEM;
4824
922dab61
ID		 4825 ret = ceph_pagelist_encode_64(pl, notify_id);
4826 ret |= ceph_pagelist_encode_64(pl, cookie);
4827 if (payload) {
4828 ret |= ceph_pagelist_encode_32(pl, payload_len);
4829 ret |= ceph_pagelist_append(pl, payload, payload_len);
4830 } else {
4831 ret |= ceph_pagelist_encode_32(pl, 0);
4832 }
4833 if (ret) {
4834 ceph_pagelist_release(pl);
4835 return -ENOMEM;
4836 }
4837
4838 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4839 op->indata_len = pl->length;
4840 return 0;
4841}
4842
4843int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4844 struct ceph_object_id *oid,
4845 struct ceph_object_locator *oloc,
4846 u64 notify_id,
4847 u64 cookie,
4848 void *payload,
6d54228f 4849 u32 payload_len)
922dab61
ID		 4850{
4851 struct ceph_osd_request *req;
4852 int ret;
4853
4854 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4855 if (!req)
4856 return -ENOMEM;
4857
4858 ceph_oid_copy(&req->r_base_oid, oid);
4859 ceph_oloc_copy(&req->r_base_oloc, oloc);
4860 req->r_flags = CEPH_OSD_FLAG_READ;
4861
26f887e0
ID		 4862 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4863 payload_len);
922dab61
ID		 4864 if (ret)
4865 goto out_put_req;
4866
26f887e0 4867 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
922dab61
ID		 4868 if (ret)
4869 goto out_put_req;
4870
a8af0d68 4871 ceph_osdc_start_request(osdc, req);
922dab61
ID		 4872 ret = ceph_osdc_wait_request(osdc, req);
4873
4874out_put_req:
4875 ceph_osdc_put_request(req);
4876 return ret;
4877}
4878EXPORT_SYMBOL(ceph_osdc_notify_ack);
4879
19079203
ID		 4880/*
4881 * @timeout: in seconds
4882 *
4883 * @preply_{pages,len} are initialized both on success and error.
4884 * The caller is responsible for:
4885 *
4886 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4887 */
4888int ceph_osdc_notify(struct ceph_osd_client *osdc,
4889 struct ceph_object_id *oid,
4890 struct ceph_object_locator *oloc,
4891 void *payload,
6d54228f 4892 u32 payload_len,
19079203
ID		 4893 u32 timeout,
4894 struct page ***preply_pages,
4895 size_t *preply_len)
4896{
4897 struct ceph_osd_linger_request *lreq;
19079203
ID		 4898 int ret;
4899
4900 WARN_ON(!timeout);
4901 if (preply_pages) {
4902 *preply_pages = NULL;
4903 *preply_len = 0;
4904 }
4905
4906 lreq = linger_alloc(osdc);
4907 if (!lreq)
4908 return -ENOMEM;
4909
75dbb685
ID		 4910 lreq->request_pl = ceph_pagelist_alloc(GFP_NOIO);
4911 if (!lreq->request_pl) {
19079203
ID		 4912 ret = -ENOMEM;
4913 goto out_put_lreq;
4914 }
4915
75dbb685
ID		 4916 ret = ceph_pagelist_encode_32(lreq->request_pl, 1); /* prot_ver */
4917 ret |= ceph_pagelist_encode_32(lreq->request_pl, timeout);
4918 ret |= ceph_pagelist_encode_32(lreq->request_pl, payload_len);
4919 ret |= ceph_pagelist_append(lreq->request_pl, payload, payload_len);
4920 if (ret) {
4921 ret = -ENOMEM;
81c65213 4922 goto out_put_lreq;
75dbb685 4923 }
81c65213 4924
19079203 4925 /* for notify_id */
75dbb685
ID		 4926 lreq->notify_id_pages = ceph_alloc_page_vector(1, GFP_NOIO);
4927 if (IS_ERR(lreq->notify_id_pages)) {
4928 ret = PTR_ERR(lreq->notify_id_pages);
4929 lreq->notify_id_pages = NULL;
19079203
ID		 4930 goto out_put_lreq;
4931 }
19079203 4932
75dbb685
ID		 4933 lreq->preply_pages = preply_pages;
4934 lreq->preply_len = preply_len;
4935
4936 ceph_oid_copy(&lreq->t.base_oid, oid);
4937 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4938 lreq->t.flags = CEPH_OSD_FLAG_READ;
26f887e0 4939
81c65213 4940 linger_submit(lreq);
19079203
ID		 4941 ret = linger_reg_commit_wait(lreq);
4942 if (!ret)
e6e28432
ID		 4943 ret = linger_notify_finish_wait(lreq,
4944 msecs_to_jiffies(2 * timeout * MSEC_PER_SEC));
19079203
ID		 4945 else
4946 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4947
4948 linger_cancel(lreq);
4949out_put_lreq:
4950 linger_put(lreq);
4951 return ret;
4952}
4953EXPORT_SYMBOL(ceph_osdc_notify);
4954
b07d3c4b
ID		 4955/*
4956 * Return the number of milliseconds since the watch was last
4957 * confirmed, or an error. If there is an error, the watch is no
4958 * longer valid, and should be destroyed with ceph_osdc_unwatch().
4959 */
4960int ceph_osdc_watch_check(struct ceph_osd_client *osdc,
4961 struct ceph_osd_linger_request *lreq)
4962{
4963 unsigned long stamp, age;
4964 int ret;
4965
4966 down_read(&osdc->lock);
4967 mutex_lock(&lreq->lock);
4968 stamp = lreq->watch_valid_thru;
4969 if (!list_empty(&lreq->pending_lworks)) {
4970 struct linger_work *lwork =
4971 list_first_entry(&lreq->pending_lworks,
4972 struct linger_work,
4973 pending_item);
4974
4975 if (time_before(lwork->queued_stamp, stamp))
4976 stamp = lwork->queued_stamp;
4977 }
4978 age = jiffies - stamp;
4979 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__,
4980 lreq, lreq->linger_id, age, lreq->last_error);
4981 /* we are truncating to msecs, so return a safe upper bound */
4982 ret = lreq->last_error ?: 1 + jiffies_to_msecs(age);
4983
4984 mutex_unlock(&lreq->lock);
4985 up_read(&osdc->lock);
4986 return ret;
4987}
4988
a4ed38d7
DF		 4989static int decode_watcher(void **p, void *end, struct ceph_watch_item *item)
4990{
4991 u8 struct_v;
4992 u32 struct_len;
4993 int ret;
4994
4995 ret = ceph_start_decoding(p, end, 2, "watch_item_t",
4996 &struct_v, &struct_len);
4997 if (ret)
51fc7ab4
JL		 4998 goto bad;
4999
5000 ret = -EINVAL;
5001 ceph_decode_copy_safe(p, end, &item->name, sizeof(item->name), bad);
5002 ceph_decode_64_safe(p, end, item->cookie, bad);
5003 ceph_decode_skip_32(p, end, bad); /* skip timeout seconds */
a4ed38d7 5004
a4ed38d7 5005 if (struct_v >= 2) {
51fc7ab4
JL		 5006 ret = ceph_decode_entity_addr(p, end, &item->addr);
5007 if (ret)
5008 goto bad;
5009 } else {
5010 ret = 0;
a4ed38d7
DF		 5011 }
5012
5013 dout("%s %s%llu cookie %llu addr %s\n", __func__,
5014 ENTITY_NAME(item->name), item->cookie,
b726ec97 5015 ceph_pr_addr(&item->addr));
51fc7ab4
JL		 5016bad:
5017 return ret;
a4ed38d7
DF		 5018}
5019
5020static int decode_watchers(void **p, void *end,
5021 struct ceph_watch_item **watchers,
5022 u32 *num_watchers)
5023{
5024 u8 struct_v;
5025 u32 struct_len;
5026 int i;
5027 int ret;
5028
5029 ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t",
5030 &struct_v, &struct_len);
5031 if (ret)
5032 return ret;
5033
5034 *num_watchers = ceph_decode_32(p);
5035 *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO);
5036 if (!*watchers)
5037 return -ENOMEM;
5038
5039 for (i = 0; i < *num_watchers; i++) {
5040 ret = decode_watcher(p, end, *watchers + i);
5041 if (ret) {
5042 kfree(*watchers);
5043 return ret;
5044 }
5045 }
5046
5047 return 0;
5048}
5049
5050/*
5051 * On success, the caller is responsible for:
5052 *
5053 * kfree(watchers);
5054 */
5055int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5056 struct ceph_object_id *oid,
5057 struct ceph_object_locator *oloc,
5058 struct ceph_watch_item **watchers,
5059 u32 *num_watchers)
5060{
5061 struct ceph_osd_request *req;
5062 struct page **pages;
5063 int ret;
5064
5065 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5066 if (!req)
5067 return -ENOMEM;
5068
5069 ceph_oid_copy(&req->r_base_oid, oid);
5070 ceph_oloc_copy(&req->r_base_oloc, oloc);
5071 req->r_flags = CEPH_OSD_FLAG_READ;
5072
a4ed38d7
DF		 5073 pages = ceph_alloc_page_vector(1, GFP_NOIO);
5074 if (IS_ERR(pages)) {
5075 ret = PTR_ERR(pages);
5076 goto out_put_req;
5077 }
5078
5079 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5080 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5081 response_data),
5082 pages, PAGE_SIZE, 0, false, true);
5083
26f887e0
ID		 5084 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5085 if (ret)
5086 goto out_put_req;
5087
a8af0d68 5088 ceph_osdc_start_request(osdc, req);
a4ed38d7
DF		 5089 ret = ceph_osdc_wait_request(osdc, req);
5090 if (ret >= 0) {
5091 void *p = page_address(pages[0]);
5092 void *const end = p + req->r_ops[0].outdata_len;
5093
5094 ret = decode_watchers(&p, end, watchers, num_watchers);
5095 }
5096
5097out_put_req:
5098 ceph_osdc_put_request(req);
5099 return ret;
5100}
5101EXPORT_SYMBOL(ceph_osdc_list_watchers);
5102
dd935f44
JD		 5103/*
5104 * Call all pending notify callbacks - for use after a watch is
5105 * unregistered, to make sure no more callbacks for it will be invoked
5106 */
f6479449 5107void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
dd935f44 5108{
99d16943 5109 dout("%s osdc %p\n", __func__, osdc);
dd935f44
JD		 5110 flush_workqueue(osdc->notify_wq);
5111}
5112EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5113
7cca78c9
ID		 5114void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5115{
5116 down_read(&osdc->lock);
5117 maybe_request_map(osdc);
5118 up_read(&osdc->lock);
5119}
5120EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
dd935f44 5121
428a7158
DF		 5122/*
5123 * Execute an OSD class method on an object.
5124 *
5125 * @flags: CEPH_OSD_FLAG_*
2544a020 5126 * @resp_len: in/out param for reply length
428a7158
DF		 5127 */
5128int ceph_osdc_call(struct ceph_osd_client *osdc,
5129 struct ceph_object_id *oid,
5130 struct ceph_object_locator *oloc,
5131 const char *class, const char *method,
5132 unsigned int flags,
5133 struct page *req_page, size_t req_len,
68ada915 5134 struct page **resp_pages, size_t *resp_len)
428a7158
DF		 5135{
5136 struct ceph_osd_request *req;
5137 int ret;
5138
68ada915 5139 if (req_len > PAGE_SIZE)
2544a020
ID		 5140 return -E2BIG;
5141
428a7158
DF		 5142 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5143 if (!req)
5144 return -ENOMEM;
5145
5146 ceph_oid_copy(&req->r_base_oid, oid);
5147 ceph_oloc_copy(&req->r_base_oloc, oloc);
5148 req->r_flags = flags;
5149
24639ce5 5150 ret = osd_req_op_cls_init(req, 0, class, method);
fe943d50
CX		 5151 if (ret)
5152 goto out_put_req;
5153
428a7158
DF		 5154 if (req_page)
5155 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5156 0, false, false);
68ada915
ID		 5157 if (resp_pages)
5158 osd_req_op_cls_response_data_pages(req, 0, resp_pages,
2544a020 5159 *resp_len, 0, false, false);
428a7158 5160
26f887e0
ID		 5161 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5162 if (ret)
5163 goto out_put_req;
5164
a8af0d68 5165 ceph_osdc_start_request(osdc, req);
428a7158
DF		 5166 ret = ceph_osdc_wait_request(osdc, req);
5167 if (ret >= 0) {
5168 ret = req->r_ops[0].rval;
68ada915 5169 if (resp_pages)
428a7158
DF		 5170 *resp_len = req->r_ops[0].outdata_len;
5171 }
5172
5173out_put_req:
5174 ceph_osdc_put_request(req);
5175 return ret;
5176}
5177EXPORT_SYMBOL(ceph_osdc_call);
5178
120a75ea
YZ		 5179/*
5180 * reset all osd connections
5181 */
5182void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5183{
5184 struct rb_node *n;
5185
5186 down_write(&osdc->lock);
5187 for (n = rb_first(&osdc->osds); n; ) {
5188 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5189
5190 n = rb_next(n);
5191 if (!reopen_osd(osd))
5192 kick_osd_requests(osd);
5193 }
5194 up_write(&osdc->lock);
5195}
5196
f24e9980
SW		 5197/*
5198 * init, shutdown
5199 */
5200int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5201{
5202 int err;
5203
5204 dout("init\n");
5205 osdc->client = client;
5aea3dcd 5206 init_rwsem(&osdc->lock);
f24e9980 5207 osdc->osds = RB_ROOT;
f5a2041b 5208 INIT_LIST_HEAD(&osdc->osd_lru);
9dd2845c 5209 spin_lock_init(&osdc->osd_lru_lock);
5aea3dcd
ID		 5210 osd_init(&osdc->homeless_osd);
5211 osdc->homeless_osd.o_osdc = osdc;
5212 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
264048af 5213 osdc->last_linger_id = CEPH_LINGER_ID_START;
922dab61 5214 osdc->linger_requests = RB_ROOT;
4609245e
ID		 5215 osdc->map_checks = RB_ROOT;
5216 osdc->linger_map_checks = RB_ROOT;
f24e9980 5217 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
f5a2041b
YS		 5218 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5219
5f44f142 5220 err = -ENOMEM;
e5253a7b
ID		 5221 osdc->osdmap = ceph_osdmap_alloc();
5222 if (!osdc->osdmap)
5223 goto out;
5224
9e767adb
ID		 5225 osdc->req_mempool = mempool_create_slab_pool(10,
5226 ceph_osd_request_cache);
f24e9980 5227 if (!osdc->req_mempool)
e5253a7b 5228 goto out_map;
f24e9980 5229
d50b409f 5230 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
0d9c1ab3 5231 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
f24e9980 5232 if (err < 0)
5f44f142 5233 goto out_mempool;
d50b409f 5234 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
0d9c1ab3
ID		 5235 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5236 "osd_op_reply");
c16e7869
SW		 5237 if (err < 0)
5238 goto out_msgpool;
a40c4f10 5239
dbcae088 5240 err = -ENOMEM;
a40c4f10 5241 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
dbcae088 5242 if (!osdc->notify_wq)
c172ec5c
ID		 5243 goto out_msgpool_reply;
5244
88bc1922
ID		 5245 osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5246 if (!osdc->completion_wq)
5247 goto out_notify_wq;
5248
fbca9635
ID		 5249 schedule_delayed_work(&osdc->timeout_work,
5250 osdc->client->options->osd_keepalive_timeout);
b37ee1b9
ID		 5251 schedule_delayed_work(&osdc->osds_timeout_work,
5252 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5253
f24e9980 5254 return 0;
5f44f142 5255
88bc1922
ID		 5256out_notify_wq:
5257 destroy_workqueue(osdc->notify_wq);
c172ec5c
ID		 5258out_msgpool_reply:
5259 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
c16e7869
SW		 5260out_msgpool:
5261 ceph_msgpool_destroy(&osdc->msgpool_op);
5f44f142
SW		 5262out_mempool:
5263 mempool_destroy(osdc->req_mempool);
e5253a7b
ID		 5264out_map:
5265 ceph_osdmap_destroy(osdc->osdmap);
5f44f142
SW		 5266out:
5267 return err;
f24e9980
SW		 5268}
5269
5270void ceph_osdc_stop(struct ceph_osd_client *osdc)
5271{
88bc1922 5272 destroy_workqueue(osdc->completion_wq);
a40c4f10 5273 destroy_workqueue(osdc->notify_wq);
f24e9980 5274 cancel_delayed_work_sync(&osdc->timeout_work);
f5a2041b 5275 cancel_delayed_work_sync(&osdc->osds_timeout_work);
42a2c09f 5276
5aea3dcd 5277 down_write(&osdc->lock);
42a2c09f
ID		 5278 while (!RB_EMPTY_ROOT(&osdc->osds)) {
5279 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5280 struct ceph_osd, o_node);
5aea3dcd 5281 close_osd(osd);
42a2c09f 5282 }
5aea3dcd 5283 up_write(&osdc->lock);
02113a0f 5284 WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5aea3dcd
ID		 5285 osd_cleanup(&osdc->homeless_osd);
5286
5287 WARN_ON(!list_empty(&osdc->osd_lru));
922dab61 5288 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
4609245e
ID		 5289 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5290 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5aea3dcd
ID		 5291 WARN_ON(atomic_read(&osdc->num_requests));
5292 WARN_ON(atomic_read(&osdc->num_homeless));
42a2c09f 5293
e5253a7b 5294 ceph_osdmap_destroy(osdc->osdmap);
f24e9980
SW		 5295 mempool_destroy(osdc->req_mempool);
5296 ceph_msgpool_destroy(&osdc->msgpool_op);
c16e7869 5297 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
f24e9980
SW		 5298}
5299
aca39d9e
LH		 5300int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5301 u64 src_snapid, u64 src_version,
5302 struct ceph_object_id *src_oid,
5303 struct ceph_object_locator *src_oloc,
5304 u32 src_fadvise_flags,
5305 u32 dst_fadvise_flags,
5306 u32 truncate_seq, u64 truncate_size,
5307 u8 copy_from_flags)
23ddf9be
LH		 5308{
5309 struct ceph_osd_req_op *op;
5310 struct page **pages;
5311 void *p, *end;
5312
5313 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5314 if (IS_ERR(pages))
5315 return PTR_ERR(pages);
5316
042f6498
JL		 5317 op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5318 dst_fadvise_flags);
23ddf9be
LH		 5319 op->copy_from.snapid = src_snapid;
5320 op->copy_from.src_version = src_version;
5321 op->copy_from.flags = copy_from_flags;
5322 op->copy_from.src_fadvise_flags = src_fadvise_flags;
5323
5324 p = page_address(pages[0]);
5325 end = p + PAGE_SIZE;
5326 ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5327 encode_oloc(&p, end, src_oloc);
78beb0ff
LH		 5328 ceph_encode_32(&p, truncate_seq);
5329 ceph_encode_64(&p, truncate_size);
23ddf9be
LH		 5330 op->indata_len = PAGE_SIZE - (end - p);
5331
5332 ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5333 op->indata_len, 0, false, true);
5334 return 0;
5335}
aca39d9e 5336EXPORT_SYMBOL(osd_req_op_copy_from_init);
23ddf9be 5337
57a35dfb 5338int __init ceph_osdc_setup(void)
5522ae0b 5339{
3f1af42a
ID		 5340 size_t size = sizeof(struct ceph_osd_request) +
5341 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5342
5522ae0b 5343 BUG_ON(ceph_osd_request_cache);
3f1af42a
ID		 5344 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5345 0, 0, NULL);
5522ae0b
AE		 5346
5347 return ceph_osd_request_cache ? 0 : -ENOMEM;
5348}
5522ae0b
AE		 5349
5350void ceph_osdc_cleanup(void)
5351{
5352 BUG_ON(!ceph_osd_request_cache);
5353 kmem_cache_destroy(ceph_osd_request_cache);
5354 ceph_osd_request_cache = NULL;
5355}
5522ae0b 5356
f24e9980
SW		 5357/*
5358 * handle incoming message
5359 */
4972cf60 5360static void osd_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
f24e9980
SW		 5361{
5362 struct ceph_osd *osd = con->private;
5aea3dcd 5363 struct ceph_osd_client *osdc = osd->o_osdc;
f24e9980
SW		 5364 int type = le16_to_cpu(msg->hdr.type);
5365
f24e9980
SW		 5366 switch (type) {
5367 case CEPH_MSG_OSD_MAP:
5368 ceph_osdc_handle_map(osdc, msg);
5369 break;
5370 case CEPH_MSG_OSD_OPREPLY:
5aea3dcd 5371 handle_reply(osd, msg);
f24e9980 5372 break;
a02a946d
ID		 5373 case CEPH_MSG_OSD_BACKOFF:
5374 handle_backoff(osd, msg);
5375 break;
a40c4f10
YS		 5376 case CEPH_MSG_WATCH_NOTIFY:
5377 handle_watch_notify(osdc, msg);
5378 break;
f24e9980
SW		 5379
5380 default:
5381 pr_err("received unknown message type %d %s\n", type,
5382 ceph_msg_type_name(type));
5383 }
5aea3dcd 5384
f24e9980
SW		 5385 ceph_msg_put(msg);
5386}
5387
f628d799
JL		 5388/* How much sparse data was requested? */
5389static u64 sparse_data_requested(struct ceph_osd_request *req)
5390{
5391 u64 len = 0;
5392
5393 if (req->r_flags & CEPH_OSD_FLAG_READ) {
5394 int i;
5395
5396 for (i = 0; i < req->r_num_ops; ++i) {
5397 struct ceph_osd_req_op *op = &req->r_ops[i];
5398
5399 if (op->op == CEPH_OSD_OP_SPARSE_READ)
5400 len += op->extent.length;
5401 }
5402 }
5403 return len;
5404}
5405
5b3a4db3 5406/*
d15f9d69
ID		 5407 * Lookup and return message for incoming reply. Don't try to do
5408 * anything about a larger than preallocated data portion of the
5409 * message at the moment - for now, just skip the message.
5b3a4db3
SW		 5410 */
5411static struct ceph_msg *get_reply(struct ceph_connection *con,
2450418c
YS		 5412 struct ceph_msg_header *hdr,
5413 int *skip)
f24e9980
SW		 5414{
5415 struct ceph_osd *osd = con->private;
5416 struct ceph_osd_client *osdc = osd->o_osdc;
5aea3dcd 5417 struct ceph_msg *m = NULL;
0547a9b3 5418 struct ceph_osd_request *req;
3f0a4ac5 5419 int front_len = le32_to_cpu(hdr->front_len);
5b3a4db3 5420 int data_len = le32_to_cpu(hdr->data_len);
5aea3dcd 5421 u64 tid = le64_to_cpu(hdr->tid);
f628d799 5422 u64 srlen;
f24e9980 5423
5aea3dcd
ID		 5424 down_read(&osdc->lock);
5425 if (!osd_registered(osd)) {
5426 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5427 *skip = 1;
5428 goto out_unlock_osdc;
5429 }
5430 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5431
5432 mutex_lock(&osd->lock);
5433 req = lookup_request(&osd->o_requests, tid);
0547a9b3 5434 if (!req) {
cd8140c6
ID		 5435 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5436 osd->o_osd, tid);
d15f9d69 5437 *skip = 1;
5aea3dcd 5438 goto out_unlock_session;
0547a9b3 5439 }
c16e7869 5440
ace6d3a9 5441 ceph_msg_revoke_incoming(req->r_reply);
0547a9b3 5442
f2be82b0 5443 if (front_len > req->r_reply->front_alloc_len) {
d15f9d69
ID		 5444 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5445 __func__, osd->o_osd, req->r_tid, front_len,
5446 req->r_reply->front_alloc_len);
3f0a4ac5
ID		 5447 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5448 false);
a79832f2 5449 if (!m)
5aea3dcd 5450 goto out_unlock_session;
c16e7869
SW		 5451 ceph_msg_put(req->r_reply);
5452 req->r_reply = m;
5453 }
0fff87ec 5454
f628d799
JL		 5455 srlen = sparse_data_requested(req);
5456 if (!srlen && data_len > req->r_reply->data_length) {
d15f9d69
ID		 5457 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5458 __func__, osd->o_osd, req->r_tid, data_len,
5459 req->r_reply->data_length);
5460 m = NULL;
5461 *skip = 1;
5aea3dcd 5462 goto out_unlock_session;
0547a9b3 5463 }
d15f9d69
ID		 5464
5465 m = ceph_msg_get(req->r_reply);
f628d799
JL		 5466 m->sparse_read = (bool)srlen;
5467
c16e7869 5468 dout("get_reply tid %lld %p\n", tid, m);
0547a9b3 5469
5aea3dcd
ID		 5470out_unlock_session:
5471 mutex_unlock(&osd->lock);
5472out_unlock_osdc:
5473 up_read(&osdc->lock);
2450418c 5474 return m;
5b3a4db3
SW		 5475}
5476
19079203
ID		 5477static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5478{
5479 struct ceph_msg *m;
5480 int type = le16_to_cpu(hdr->type);
5481 u32 front_len = le32_to_cpu(hdr->front_len);
5482 u32 data_len = le32_to_cpu(hdr->data_len);
5483
0d9c1ab3 5484 m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
19079203
ID		 5485 if (!m)
5486 return NULL;
5487
5488 if (data_len) {
5489 struct page **pages;
19079203
ID		 5490
5491 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5492 GFP_NOIO);
c22e853a 5493 if (IS_ERR(pages)) {
19079203
ID		 5494 ceph_msg_put(m);
5495 return NULL;
5496 }
5497
e8862740 5498 ceph_msg_data_add_pages(m, pages, data_len, 0, true);
19079203
ID		 5499 }
5500
5501 return m;
5502}
5503
4972cf60
ID		 5504static struct ceph_msg *osd_alloc_msg(struct ceph_connection *con,
5505 struct ceph_msg_header *hdr,
5506 int *skip)
5b3a4db3
SW		 5507{
5508 struct ceph_osd *osd = con->private;
5509 int type = le16_to_cpu(hdr->type);
5b3a4db3 5510
1c20f2d2 5511 *skip = 0;
5b3a4db3
SW		 5512 switch (type) {
5513 case CEPH_MSG_OSD_MAP:
a02a946d 5514 case CEPH_MSG_OSD_BACKOFF:
a40c4f10 5515 case CEPH_MSG_WATCH_NOTIFY:
19079203 5516 return alloc_msg_with_page_vector(hdr);
5b3a4db3
SW		 5517 case CEPH_MSG_OSD_OPREPLY:
5518 return get_reply(con, hdr, skip);
5519 default:
5aea3dcd
ID		 5520 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5521 osd->o_osd, type);
5b3a4db3
SW		 5522 *skip = 1;
5523 return NULL;
5524 }
f24e9980
SW		 5525}
5526
5527/*
5528 * Wrappers to refcount containing ceph_osd struct
5529 */
4972cf60 5530static struct ceph_connection *osd_get_con(struct ceph_connection *con)
f24e9980
SW		 5531{
5532 struct ceph_osd *osd = con->private;
5533 if (get_osd(osd))
5534 return con;
5535 return NULL;
5536}
5537
4972cf60 5538static void osd_put_con(struct ceph_connection *con)
f24e9980
SW		 5539{
5540 struct ceph_osd *osd = con->private;
5541 put_osd(osd);
5542}
5543
4e7a5dcd
SW		 5544/*
5545 * authentication
5546 */
cd1a677c 5547
a3530df3
AE		 5548/*
5549 * Note: returned pointer is the address of a structure that's
5550 * managed separately. Caller must *not* attempt to free it.
5551 */
4972cf60
ID		 5552static struct ceph_auth_handshake *
5553osd_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
4e7a5dcd
SW		 5554{
5555 struct ceph_osd *o = con->private;
5556 struct ceph_osd_client *osdc = o->o_osdc;
5557 struct ceph_auth_client *ac = osdc->client->monc.auth;
74f1869f 5558 struct ceph_auth_handshake *auth = &o->o_auth;
ce287162 5559 int ret;
4e7a5dcd 5560
ce287162
ID		 5561 ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5562 force_new, proto, NULL, NULL);
5563 if (ret)
5564 return ERR_PTR(ret);
74f1869f 5565
a3530df3 5566 return auth;
4e7a5dcd
SW		 5567}
5568
4972cf60 5569static int osd_add_authorizer_challenge(struct ceph_connection *con,
6daca13d
ID		 5570 void *challenge_buf, int challenge_buf_len)
5571{
5572 struct ceph_osd *o = con->private;
5573 struct ceph_osd_client *osdc = o->o_osdc;
5574 struct ceph_auth_client *ac = osdc->client->monc.auth;
5575
5576 return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5577 challenge_buf, challenge_buf_len);
5578}
4e7a5dcd 5579
4972cf60 5580static int osd_verify_authorizer_reply(struct ceph_connection *con)
4e7a5dcd
SW		 5581{
5582 struct ceph_osd *o = con->private;
5583 struct ceph_osd_client *osdc = o->o_osdc;
5584 struct ceph_auth_client *ac = osdc->client->monc.auth;
285ea34f 5585 struct ceph_auth_handshake *auth = &o->o_auth;
4e7a5dcd 5586
285ea34f
ID		 5587 return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5588 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5589 NULL, NULL, NULL, NULL);
4e7a5dcd
SW		 5590}
5591
4972cf60 5592static int osd_invalidate_authorizer(struct ceph_connection *con)
9bd2e6f8
SW		 5593{
5594 struct ceph_osd *o = con->private;
5595 struct ceph_osd_client *osdc = o->o_osdc;
5596 struct ceph_auth_client *ac = osdc->client->monc.auth;
5597
27859f97 5598 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
9bd2e6f8
SW		 5599 return ceph_monc_validate_auth(&osdc->client->monc);
5600}
4e7a5dcd 5601
cd1a677c
ID		 5602static int osd_get_auth_request(struct ceph_connection *con,
5603 void *buf, int *buf_len,
5604 void **authorizer, int *authorizer_len)
5605{
5606 struct ceph_osd *o = con->private;
5607 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5608 struct ceph_auth_handshake *auth = &o->o_auth;
5609 int ret;
5610
5611 ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5612 buf, buf_len);
5613 if (ret)
5614 return ret;
5615
5616 *authorizer = auth->authorizer_buf;
5617 *authorizer_len = auth->authorizer_buf_len;
5618 return 0;
5619}
5620
5621static int osd_handle_auth_reply_more(struct ceph_connection *con,
5622 void *reply, int reply_len,
5623 void *buf, int *buf_len,
5624 void **authorizer, int *authorizer_len)
5625{
5626 struct ceph_osd *o = con->private;
5627 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5628 struct ceph_auth_handshake *auth = &o->o_auth;
5629 int ret;
5630
5631 ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5632 buf, buf_len);
5633 if (ret)
5634 return ret;
5635
5636 *authorizer = auth->authorizer_buf;
5637 *authorizer_len = auth->authorizer_buf_len;
5638 return 0;
5639}
5640
5641static int osd_handle_auth_done(struct ceph_connection *con,
5642 u64 global_id, void *reply, int reply_len,
5643 u8 *session_key, int *session_key_len,
5644 u8 *con_secret, int *con_secret_len)
5645{
5646 struct ceph_osd *o = con->private;
5647 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5648 struct ceph_auth_handshake *auth = &o->o_auth;
5649
5650 return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5651 session_key, session_key_len,
5652 con_secret, con_secret_len);
5653}
5654
5655static int osd_handle_auth_bad_method(struct ceph_connection *con,
5656 int used_proto, int result,
5657 const int *allowed_protos, int proto_cnt,
5658 const int *allowed_modes, int mode_cnt)
5659{
5660 struct ceph_osd *o = con->private;
5661 struct ceph_mon_client *monc = &o->o_osdc->client->monc;
5662 int ret;
5663
5664 if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD,
5665 used_proto, result,
5666 allowed_protos, proto_cnt,
5667 allowed_modes, mode_cnt)) {
5668 ret = ceph_monc_validate_auth(monc);
5669 if (ret)
5670 return ret;
5671 }
5672
5673 return -EACCES;
5674}
5675
8cb441c0
ID		 5676static void osd_reencode_message(struct ceph_msg *msg)
5677{
914902af
ID		 5678 int type = le16_to_cpu(msg->hdr.type);
5679
5680 if (type == CEPH_MSG_OSD_OP)
5681 encode_request_finish(msg);
8cb441c0
ID		 5682}
5683
79dbd1ba 5684static int osd_sign_message(struct ceph_msg *msg)
33d07337 5685{
79dbd1ba 5686 struct ceph_osd *o = msg->con->private;
33d07337 5687 struct ceph_auth_handshake *auth = &o->o_auth;
79dbd1ba 5688
33d07337
YZ		 5689 return ceph_auth_sign_message(auth, msg);
5690}
5691
79dbd1ba 5692static int osd_check_message_signature(struct ceph_msg *msg)
33d07337 5693{
79dbd1ba 5694 struct ceph_osd *o = msg->con->private;
33d07337 5695 struct ceph_auth_handshake *auth = &o->o_auth;
79dbd1ba 5696
33d07337
YZ		 5697 return ceph_auth_check_message_signature(auth, msg);
5698}
5699
f628d799
JL		 5700static void advance_cursor(struct ceph_msg_data_cursor *cursor, size_t len,
5701 bool zero)
5702{
5703 while (len) {
5704 struct page *page;
5705 size_t poff, plen;
5706
5707 page = ceph_msg_data_next(cursor, &poff, &plen);
5708 if (plen > len)
5709 plen = len;
5710 if (zero)
5711 zero_user_segment(page, poff, poff + plen);
5712 len -= plen;
5713 ceph_msg_data_advance(cursor, plen);
5714 }
5715}
5716
5717static int prep_next_sparse_read(struct ceph_connection *con,
5718 struct ceph_msg_data_cursor *cursor)
5719{
5720 struct ceph_osd *o = con->private;
5721 struct ceph_sparse_read *sr = &o->o_sparse_read;
5722 struct ceph_osd_request *req;
5723 struct ceph_osd_req_op *op;
5724
5725 spin_lock(&o->o_requests_lock);
5726 req = lookup_request(&o->o_requests, le64_to_cpu(con->in_msg->hdr.tid));
5727 if (!req) {
5728 spin_unlock(&o->o_requests_lock);
5729 return -EBADR;
5730 }
5731
5732 if (o->o_sparse_op_idx < 0) {
5733 u64 srlen = sparse_data_requested(req);
5734
5735 dout("%s: [%d] starting new sparse read req. srlen=0x%llx\n",
5736 __func__, o->o_osd, srlen);
5737 ceph_msg_data_cursor_init(cursor, con->in_msg, srlen);
5738 } else {
5739 u64 end;
5740
5741 op = &req->r_ops[o->o_sparse_op_idx];
5742
5743 WARN_ON_ONCE(op->extent.sparse_ext);
5744
5745 /* hand back buffer we took earlier */
5746 op->extent.sparse_ext = sr->sr_extent;
5747 sr->sr_extent = NULL;
5748 op->extent.sparse_ext_cnt = sr->sr_count;
5749 sr->sr_ext_len = 0;
5750 dout("%s: [%d] completed extent array len %d cursor->resid %zd\n",
5751 __func__, o->o_osd, op->extent.sparse_ext_cnt, cursor->resid);
5752 /* Advance to end of data for this operation */
5753 end = ceph_sparse_ext_map_end(op);
5754 if (end < sr->sr_req_len)
5755 advance_cursor(cursor, sr->sr_req_len - end, false);
5756 }
5757
5758 ceph_init_sparse_read(sr);
5759
5760 /* find next op in this request (if any) */
5761 while (++o->o_sparse_op_idx < req->r_num_ops) {
5762 op = &req->r_ops[o->o_sparse_op_idx];
5763 if (op->op == CEPH_OSD_OP_SPARSE_READ)
5764 goto found;
5765 }
5766
5767 /* reset for next sparse read request */
5768 spin_unlock(&o->o_requests_lock);
5769 o->o_sparse_op_idx = -1;
5770 return 0;
5771found:
5772 sr->sr_req_off = op->extent.offset;
5773 sr->sr_req_len = op->extent.length;
5774 sr->sr_pos = sr->sr_req_off;
5775 dout("%s: [%d] new sparse read op at idx %d 0x%llx~0x%llx\n", __func__,
5776 o->o_osd, o->o_sparse_op_idx, sr->sr_req_off, sr->sr_req_len);
5777
5778 /* hand off request's sparse extent map buffer */
5779 sr->sr_ext_len = op->extent.sparse_ext_cnt;
5780 op->extent.sparse_ext_cnt = 0;
5781 sr->sr_extent = op->extent.sparse_ext;
5782 op->extent.sparse_ext = NULL;
5783
5784 spin_unlock(&o->o_requests_lock);
5785 return 1;
5786}
5787
5788#ifdef __BIG_ENDIAN
5789static inline void convert_extent_map(struct ceph_sparse_read *sr)
5790{
5791 int i;
5792
5793 for (i = 0; i < sr->sr_count; i++) {
5794 struct ceph_sparse_extent *ext = &sr->sr_extent[i];
5795
5796 ext->off = le64_to_cpu((__force __le64)ext->off);
5797 ext->len = le64_to_cpu((__force __le64)ext->len);
5798 }
5799}
5800#else
5801static inline void convert_extent_map(struct ceph_sparse_read *sr)
5802{
5803}
5804#endif
5805
5806#define MAX_EXTENTS 4096
5807
5808static int osd_sparse_read(struct ceph_connection *con,
5809 struct ceph_msg_data_cursor *cursor,
5810 char **pbuf)
5811{
5812 struct ceph_osd *o = con->private;
5813 struct ceph_sparse_read *sr = &o->o_sparse_read;
5814 u32 count = sr->sr_count;
5815 u64 eoff, elen;
5816 int ret;
5817
5818 switch (sr->sr_state) {
5819 case CEPH_SPARSE_READ_HDR:
5820next_op:
5821 ret = prep_next_sparse_read(con, cursor);
5822 if (ret <= 0)
5823 return ret;
5824
5825 /* number of extents */
5826 ret = sizeof(sr->sr_count);
5827 *pbuf = (char *)&sr->sr_count;
5828 sr->sr_state = CEPH_SPARSE_READ_EXTENTS;
5829 break;
5830 case CEPH_SPARSE_READ_EXTENTS:
5831 /* Convert sr_count to host-endian */
5832 count = le32_to_cpu((__force __le32)sr->sr_count);
5833 sr->sr_count = count;
5834 dout("[%d] got %u extents\n", o->o_osd, count);
5835
5836 if (count > 0) {
5837 if (!sr->sr_extent || count > sr->sr_ext_len) {
5838 /*
5839 * Apply a hard cap to the number of extents.
5840 * If we have more, assume something is wrong.
5841 */
5842 if (count > MAX_EXTENTS) {
5843 dout("%s: OSD returned 0x%x extents in a single reply!\n",
5844 __func__, count);
5845 return -EREMOTEIO;
5846 }
5847
5848 /* no extent array provided, or too short */
5849 kfree(sr->sr_extent);
5850 sr->sr_extent = kmalloc_array(count,
5851 sizeof(*sr->sr_extent),
5852 GFP_NOIO);
5853 if (!sr->sr_extent)
5854 return -ENOMEM;
5855 sr->sr_ext_len = count;
5856 }
5857 ret = count * sizeof(*sr->sr_extent);
5858 *pbuf = (char *)sr->sr_extent;
5859 sr->sr_state = CEPH_SPARSE_READ_DATA_LEN;
5860 break;
5861 }
5862 /* No extents? Read data len */
5863 fallthrough;
5864 case CEPH_SPARSE_READ_DATA_LEN:
5865 convert_extent_map(sr);
5866 ret = sizeof(sr->sr_datalen);
5867 *pbuf = (char *)&sr->sr_datalen;
5868 sr->sr_state = CEPH_SPARSE_READ_DATA;
5869 break;
5870 case CEPH_SPARSE_READ_DATA:
5871 if (sr->sr_index >= count) {
5872 sr->sr_state = CEPH_SPARSE_READ_HDR;
5873 goto next_op;
5874 }
5875
5876 eoff = sr->sr_extent[sr->sr_index].off;
5877 elen = sr->sr_extent[sr->sr_index].len;
5878
5879 dout("[%d] ext %d off 0x%llx len 0x%llx\n",
5880 o->o_osd, sr->sr_index, eoff, elen);
5881
5882 if (elen > INT_MAX) {
5883 dout("Sparse read extent length too long (0x%llx)\n",
5884 elen);
5885 return -EREMOTEIO;
5886 }
5887
5888 /* zero out anything from sr_pos to start of extent */
5889 if (sr->sr_pos < eoff)
5890 advance_cursor(cursor, eoff - sr->sr_pos, true);
5891
5892 /* Set position to end of extent */
5893 sr->sr_pos = eoff + elen;
5894
5895 /* send back the new length and nullify the ptr */
5896 cursor->sr_resid = elen;
5897 ret = elen;
5898 *pbuf = NULL;
5899
5900 /* Bump the array index */
5901 ++sr->sr_index;
5902 break;
5903 }
5904 return ret;
5905}
5906
9e32789f 5907static const struct ceph_connection_operations osd_con_ops = {
4972cf60
ID		 5908 .get = osd_get_con,
5909 .put = osd_put_con,
f628d799 5910 .sparse_read = osd_sparse_read,
4972cf60
ID		 5911 .alloc_msg = osd_alloc_msg,
5912 .dispatch = osd_dispatch,
5913 .fault = osd_fault,
8cb441c0 5914 .reencode_message = osd_reencode_message,
4972cf60
ID		 5915 .get_authorizer = osd_get_authorizer,
5916 .add_authorizer_challenge = osd_add_authorizer_challenge,
5917 .verify_authorizer_reply = osd_verify_authorizer_reply,
5918 .invalidate_authorizer = osd_invalidate_authorizer,
79dbd1ba
ID		 5919 .sign_message = osd_sign_message,
5920 .check_message_signature = osd_check_message_signature,
cd1a677c
ID		 5921 .get_auth_request = osd_get_auth_request,
5922 .handle_auth_reply_more = osd_handle_auth_reply_more,
5923 .handle_auth_done = osd_handle_auth_done,
5924 .handle_auth_bad_method = osd_handle_auth_bad_method,
f24e9980 5925};
Linux 6.x block layer and io_uring tree(s)