projects/linux·blob master · refs

5956 lines · 153.2 KiB

1// SPDX-License-Identifier: GPL-2.0
2
3#include <linux/ceph/ceph_debug.h>
4
5#include <linux/module.h>
6#include <linux/err.h>
7#include <linux/highmem.h>
8#include <linux/mm.h>
9#include <linux/overflow.h>
10#include <linux/pagemap.h>
11#include <linux/slab.h>
12#include <linux/uaccess.h>
13#ifdef CONFIG_BLOCK
14#include <linux/bio.h>
15#endif
16
17#include <linux/ceph/ceph_features.h>
18#include <linux/ceph/libceph.h>
19#include <linux/ceph/osd_client.h>
20#include <linux/ceph/messenger.h>
21#include <linux/ceph/decode.h>
22#include <linux/ceph/auth.h>
23#include <linux/ceph/pagelist.h>
24#include <linux/ceph/striper.h>
25
26#define OSD_OPREPLY_FRONT_LEN 512
27
28static struct kmem_cache *ceph_osd_request_cache;
29
30static const struct ceph_connection_operations osd_con_ops;
31
32/*
33 * Implement client access to distributed object storage cluster.
34 *
35 * All data objects are stored within a cluster/cloud of OSDs, or
36 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
37 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
38 * remote daemons serving up and coordinating consistent and safe
39 * access to storage.
40 *
41 * Cluster membership and the mapping of data objects onto storage devices
42 * are described by the osd map.
43 *
44 * We keep track of pending OSD requests (read, write), resubmit
45 * requests to different OSDs when the cluster topology/data layout
46 * change, or retry the affected requests when the communications
47 * channel with an OSD is reset.
48 */
49
50static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req);
51static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req);
52static void link_linger(struct ceph_osd *osd,
53 struct ceph_osd_linger_request *lreq);
54static void unlink_linger(struct ceph_osd *osd,
55 struct ceph_osd_linger_request *lreq);
56static void clear_backoffs(struct ceph_osd *osd);
57
58#if 1
59static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem)
60{
61 bool wrlocked = true;
62
63 if (unlikely(down_read_trylock(sem))) {
64 wrlocked = false;
65 up_read(sem);
66 }
67
68 return wrlocked;
69}
70static inline void verify_osdc_locked(struct ceph_osd_client *osdc)
71{
72 WARN_ON(!rwsem_is_locked(&osdc->lock));
73}
74static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc)
75{
76 WARN_ON(!rwsem_is_wrlocked(&osdc->lock));
77}
78static inline void verify_osd_locked(struct ceph_osd *osd)
79{
80 struct ceph_osd_client *osdc = osd->o_osdc;
81
82 WARN_ON(!(mutex_is_locked(&osd->lock) &&
83 rwsem_is_locked(&osdc->lock)) &&
84 !rwsem_is_wrlocked(&osdc->lock));
85}
86static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq)
87{
88 WARN_ON(!mutex_is_locked(&lreq->lock));
89}
90#else
91static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { }
92static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { }
93static inline void verify_osd_locked(struct ceph_osd *osd) { }
94static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { }
95#endif
96
97/*
98 * calculate the mapping of a file extent onto an object, and fill out the
99 * request accordingly. shorten extent as necessary if it crosses an
100 * object boundary.
101 *
102 * fill osd op in request message.
103 */
104static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
105 u64 *objnum, u64 *objoff, u64 *objlen)
106{
107 u64 orig_len = *plen;
108 u32 xlen;
109
110 /* object extent? */
111 ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
112 objoff, &xlen);
113 *objlen = xlen;
114 if (*objlen < orig_len) {
115 *plen = *objlen;
116 dout(" skipping last %llu, final file extent %llu~%llu\n",
117 orig_len - *plen, off, *plen);
118 }
119
120 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
121 return 0;
122}
123
124static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
125{
126 memset(osd_data, 0, sizeof (*osd_data));
127 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
128}
129
130/*
131 * Consumes @pages if @own_pages is true.
132 */
133static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
134 struct page **pages, u64 length, u32 alignment,
135 bool pages_from_pool, bool own_pages)
136{
137 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
138 osd_data->pages = pages;
139 osd_data->length = length;
140 osd_data->alignment = alignment;
141 osd_data->pages_from_pool = pages_from_pool;
142 osd_data->own_pages = own_pages;
143}
144
145/*
146 * Consumes a ref on @pagelist.
147 */
148static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
149 struct ceph_pagelist *pagelist)
150{
151 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
152 osd_data->pagelist = pagelist;
153}
154
155#ifdef CONFIG_BLOCK
156static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
157 struct ceph_bio_iter *bio_pos,
158 u32 bio_length)
159{
160 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
161 osd_data->bio_pos = *bio_pos;
162 osd_data->bio_length = bio_length;
163}
164#endif /* CONFIG_BLOCK */
165
166static void ceph_osd_data_bvecs_init(struct ceph_osd_data *osd_data,
167 struct ceph_bvec_iter *bvec_pos,
168 u32 num_bvecs)
169{
170 osd_data->type = CEPH_OSD_DATA_TYPE_BVECS;
171 osd_data->bvec_pos = *bvec_pos;
172 osd_data->num_bvecs = num_bvecs;
173}
174
175static void ceph_osd_iter_init(struct ceph_osd_data *osd_data,
176 struct iov_iter *iter)
177{
178 osd_data->type = CEPH_OSD_DATA_TYPE_ITER;
179 osd_data->iter = *iter;
180}
181
182static struct ceph_osd_data *
183osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
184{
185 BUG_ON(which >= osd_req->r_num_ops);
186
187 return &osd_req->r_ops[which].raw_data_in;
188}
189
190struct ceph_osd_data *
191osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
192 unsigned int which)
193{
194 return osd_req_op_data(osd_req, which, extent, osd_data);
195}
196EXPORT_SYMBOL(osd_req_op_extent_osd_data);
197
198void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
199 unsigned int which, struct page **pages,
200 u64 length, u32 alignment,
201 bool pages_from_pool, bool own_pages)
202{
203 struct ceph_osd_data *osd_data;
204
205 osd_data = osd_req_op_raw_data_in(osd_req, which);
206 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
207 pages_from_pool, own_pages);
208}
209EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
210
211void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
212 unsigned int which, struct page **pages,
213 u64 length, u32 alignment,
214 bool pages_from_pool, bool own_pages)
215{
216 struct ceph_osd_data *osd_data;
217
218 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
219 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
220 pages_from_pool, own_pages);
221}
222EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
223
224#ifdef CONFIG_BLOCK
225void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
226 unsigned int which,
227 struct ceph_bio_iter *bio_pos,
228 u32 bio_length)
229{
230 struct ceph_osd_data *osd_data;
231
232 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
233 ceph_osd_data_bio_init(osd_data, bio_pos, bio_length);
234}
235EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
236#endif /* CONFIG_BLOCK */
237
238void osd_req_op_extent_osd_data_bvecs(struct ceph_osd_request *osd_req,
239 unsigned int which,
240 struct bio_vec *bvecs, u32 num_bvecs,
241 u32 bytes)
242{
243 struct ceph_osd_data *osd_data;
244 struct ceph_bvec_iter it = {
245 .bvecs = bvecs,
246 .iter = { .bi_size = bytes },
247 };
248
249 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
250 ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
251}
252EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvecs);
253
254void osd_req_op_extent_osd_data_bvec_pos(struct ceph_osd_request *osd_req,
255 unsigned int which,
256 struct ceph_bvec_iter *bvec_pos)
257{
258 struct ceph_osd_data *osd_data;
259
260 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
261 ceph_osd_data_bvecs_init(osd_data, bvec_pos, 0);
262}
263EXPORT_SYMBOL(osd_req_op_extent_osd_data_bvec_pos);
264
265/**
266 * osd_req_op_extent_osd_iter - Set up an operation with an iterator buffer
267 * @osd_req: The request to set up
268 * @which: Index of the operation in which to set the iter
269 * @iter: The buffer iterator
270 */
271void osd_req_op_extent_osd_iter(struct ceph_osd_request *osd_req,
272 unsigned int which, struct iov_iter *iter)
273{
274 struct ceph_osd_data *osd_data;
275
276 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
277 ceph_osd_iter_init(osd_data, iter);
278}
279EXPORT_SYMBOL(osd_req_op_extent_osd_iter);
280
281static void osd_req_op_cls_request_info_pagelist(
282 struct ceph_osd_request *osd_req,
283 unsigned int which, struct ceph_pagelist *pagelist)
284{
285 struct ceph_osd_data *osd_data;
286
287 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
288 ceph_osd_data_pagelist_init(osd_data, pagelist);
289}
290
291void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
292 unsigned int which, struct page **pages, u64 length,
293 u32 alignment, bool pages_from_pool, bool own_pages)
294{
295 struct ceph_osd_data *osd_data;
296
297 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
298 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
299 pages_from_pool, own_pages);
300 osd_req->r_ops[which].cls.indata_len += length;
301 osd_req->r_ops[which].indata_len += length;
302}
303EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
304
305void osd_req_op_cls_request_data_bvecs(struct ceph_osd_request *osd_req,
306 unsigned int which,
307 struct bio_vec *bvecs, u32 num_bvecs,
308 u32 bytes)
309{
310 struct ceph_osd_data *osd_data;
311 struct ceph_bvec_iter it = {
312 .bvecs = bvecs,
313 .iter = { .bi_size = bytes },
314 };
315
316 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
317 ceph_osd_data_bvecs_init(osd_data, &it, num_bvecs);
318 osd_req->r_ops[which].cls.indata_len += bytes;
319 osd_req->r_ops[which].indata_len += bytes;
320}
321EXPORT_SYMBOL(osd_req_op_cls_request_data_bvecs);
322
323void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
324 unsigned int which, struct page **pages, u64 length,
325 u32 alignment, bool pages_from_pool, bool own_pages)
326{
327 struct ceph_osd_data *osd_data;
328
329 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
330 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
331 pages_from_pool, own_pages);
332}
333EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
334
335static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
336{
337 switch (osd_data->type) {
338 case CEPH_OSD_DATA_TYPE_NONE:
339 return 0;
340 case CEPH_OSD_DATA_TYPE_PAGES:
341 return osd_data->length;
342 case CEPH_OSD_DATA_TYPE_PAGELIST:
343 return (u64)osd_data->pagelist->length;
344#ifdef CONFIG_BLOCK
345 case CEPH_OSD_DATA_TYPE_BIO:
346 return (u64)osd_data->bio_length;
347#endif /* CONFIG_BLOCK */
348 case CEPH_OSD_DATA_TYPE_BVECS:
349 return osd_data->bvec_pos.iter.bi_size;
350 case CEPH_OSD_DATA_TYPE_ITER:
351 return iov_iter_count(&osd_data->iter);
352 default:
353 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
354 return 0;
355 }
356}
357
358static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
359{
360 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
361 int num_pages;
362
363 num_pages = calc_pages_for((u64)osd_data->alignment,
364 (u64)osd_data->length);
365 ceph_release_page_vector(osd_data->pages, num_pages);
366 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
367 ceph_pagelist_release(osd_data->pagelist);
368 }
369 ceph_osd_data_init(osd_data);
370}
371
372static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
373 unsigned int which)
374{
375 struct ceph_osd_req_op *op;
376
377 BUG_ON(which >= osd_req->r_num_ops);
378 op = &osd_req->r_ops[which];
379
380 switch (op->op) {
381 case CEPH_OSD_OP_READ:
382 case CEPH_OSD_OP_SPARSE_READ:
383 case CEPH_OSD_OP_WRITE:
384 case CEPH_OSD_OP_WRITEFULL:
385 kfree(op->extent.sparse_ext);
386 ceph_osd_data_release(&op->extent.osd_data);
387 break;
388 case CEPH_OSD_OP_CALL:
389 ceph_osd_data_release(&op->cls.request_info);
390 ceph_osd_data_release(&op->cls.request_data);
391 ceph_osd_data_release(&op->cls.response_data);
392 break;
393 case CEPH_OSD_OP_SETXATTR:
394 case CEPH_OSD_OP_CMPXATTR:
395 ceph_osd_data_release(&op->xattr.osd_data);
396 break;
397 case CEPH_OSD_OP_STAT:
398 ceph_osd_data_release(&op->raw_data_in);
399 break;
400 case CEPH_OSD_OP_NOTIFY_ACK:
401 ceph_osd_data_release(&op->notify_ack.request_data);
402 break;
403 case CEPH_OSD_OP_NOTIFY:
404 ceph_osd_data_release(&op->notify.request_data);
405 ceph_osd_data_release(&op->notify.response_data);
406 break;
407 case CEPH_OSD_OP_LIST_WATCHERS:
408 ceph_osd_data_release(&op->list_watchers.response_data);
409 break;
410 case CEPH_OSD_OP_COPY_FROM2:
411 ceph_osd_data_release(&op->copy_from.osd_data);
412 break;
413 default:
414 break;
415 }
416}
417
418/*
419 * Assumes @t is zero-initialized.
420 */
421static void target_init(struct ceph_osd_request_target *t)
422{
423 ceph_oid_init(&t->base_oid);
424 ceph_oloc_init(&t->base_oloc);
425 ceph_oid_init(&t->target_oid);
426 ceph_oloc_init(&t->target_oloc);
427
428 ceph_osds_init(&t->acting);
429 ceph_osds_init(&t->up);
430 t->size = -1;
431 t->min_size = -1;
432
433 t->osd = CEPH_HOMELESS_OSD;
434}
435
436static void target_copy(struct ceph_osd_request_target *dest,
437 const struct ceph_osd_request_target *src)
438{
439 ceph_oid_copy(&dest->base_oid, &src->base_oid);
440 ceph_oloc_copy(&dest->base_oloc, &src->base_oloc);
441 ceph_oid_copy(&dest->target_oid, &src->target_oid);
442 ceph_oloc_copy(&dest->target_oloc, &src->target_oloc);
443
444 dest->pgid = src->pgid; /* struct */
445 dest->spgid = src->spgid; /* struct */
446 dest->pg_num = src->pg_num;
447 dest->pg_num_mask = src->pg_num_mask;
448 ceph_osds_copy(&dest->acting, &src->acting);
449 ceph_osds_copy(&dest->up, &src->up);
450 dest->size = src->size;
451 dest->min_size = src->min_size;
452 dest->sort_bitwise = src->sort_bitwise;
453 dest->recovery_deletes = src->recovery_deletes;
454
455 dest->flags = src->flags;
456 dest->used_replica = src->used_replica;
457 dest->paused = src->paused;
458
459 dest->epoch = src->epoch;
460 dest->last_force_resend = src->last_force_resend;
461
462 dest->osd = src->osd;
463}
464
465static void target_destroy(struct ceph_osd_request_target *t)
466{
467 ceph_oid_destroy(&t->base_oid);
468 ceph_oloc_destroy(&t->base_oloc);
469 ceph_oid_destroy(&t->target_oid);
470 ceph_oloc_destroy(&t->target_oloc);
471}
472
473/*
474 * requests
475 */
476static void request_release_checks(struct ceph_osd_request *req)
477{
478 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
479 WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node));
480 WARN_ON(!list_empty(&req->r_private_item));
481 WARN_ON(req->r_osd);
482}
483
484static void ceph_osdc_release_request(struct kref *kref)
485{
486 struct ceph_osd_request *req = container_of(kref,
487 struct ceph_osd_request, r_kref);
488 unsigned int which;
489
490 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
491 req->r_request, req->r_reply);
492 request_release_checks(req);
493
494 if (req->r_request)
495 ceph_msg_put(req->r_request);
496 if (req->r_reply)
497 ceph_msg_put(req->r_reply);
498
499 for (which = 0; which < req->r_num_ops; which++)
500 osd_req_op_data_release(req, which);
501
502 target_destroy(&req->r_t);
503 ceph_put_snap_context(req->r_snapc);
504
505 if (req->r_mempool)
506 mempool_free(req, req->r_osdc->req_mempool);
507 else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS)
508 kmem_cache_free(ceph_osd_request_cache, req);
509 else
510 kfree(req);
511}
512
513void ceph_osdc_get_request(struct ceph_osd_request *req)
514{
515 dout("%s %p (was %d)\n", __func__, req,
516 kref_read(&req->r_kref));
517 kref_get(&req->r_kref);
518}
519EXPORT_SYMBOL(ceph_osdc_get_request);
520
521void ceph_osdc_put_request(struct ceph_osd_request *req)
522{
523 if (req) {
524 dout("%s %p (was %d)\n", __func__, req,
525 kref_read(&req->r_kref));
526 kref_put(&req->r_kref, ceph_osdc_release_request);
527 }
528}
529EXPORT_SYMBOL(ceph_osdc_put_request);
530
531static void request_init(struct ceph_osd_request *req)
532{
533 /* req only, each op is zeroed in osd_req_op_init() */
534 memset(req, 0, sizeof(*req));
535
536 kref_init(&req->r_kref);
537 init_completion(&req->r_completion);
538 RB_CLEAR_NODE(&req->r_node);
539 RB_CLEAR_NODE(&req->r_mc_node);
540 INIT_LIST_HEAD(&req->r_private_item);
541
542 target_init(&req->r_t);
543}
544
545struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
546 struct ceph_snap_context *snapc,
547 unsigned int num_ops,
548 bool use_mempool,
549 gfp_t gfp_flags)
550{
551 struct ceph_osd_request *req;
552
553 if (use_mempool) {
554 BUG_ON(num_ops > CEPH_OSD_SLAB_OPS);
555 req = mempool_alloc(osdc->req_mempool, gfp_flags);
556 } else if (num_ops <= CEPH_OSD_SLAB_OPS) {
557 req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags);
558 } else {
559 BUG_ON(num_ops > CEPH_OSD_MAX_OPS);
560 req = kmalloc_flex(*req, r_ops, num_ops, gfp_flags);
561 }
562 if (unlikely(!req))
563 return NULL;
564
565 request_init(req);
566 req->r_osdc = osdc;
567 req->r_mempool = use_mempool;
568 req->r_num_ops = num_ops;
569 req->r_snapid = CEPH_NOSNAP;
570 req->r_snapc = ceph_get_snap_context(snapc);
571
572 dout("%s req %p\n", __func__, req);
573 return req;
574}
575EXPORT_SYMBOL(ceph_osdc_alloc_request);
576
577static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc)
578{
579 return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0);
580}
581
582static int __ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp,
583 int num_request_data_items,
584 int num_reply_data_items)
585{
586 struct ceph_osd_client *osdc = req->r_osdc;
587 struct ceph_msg *msg;
588 int msg_size;
589
590 WARN_ON(req->r_request || req->r_reply);
591 WARN_ON(ceph_oid_empty(&req->r_base_oid));
592 WARN_ON(ceph_oloc_empty(&req->r_base_oloc));
593
594 /* create request message */
595 msg_size = CEPH_ENCODING_START_BLK_LEN +
596 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
597 msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */
598 msg_size += CEPH_ENCODING_START_BLK_LEN +
599 sizeof(struct ceph_osd_reqid); /* reqid */
600 msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */
601 msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */
602 msg_size += CEPH_ENCODING_START_BLK_LEN +
603 ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */
604 msg_size += 4 + req->r_base_oid.name_len; /* oid */
605 msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op);
606 msg_size += 8; /* snapid */
607 msg_size += 8; /* snap_seq */
608 msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0);
609 msg_size += 4 + 8; /* retry_attempt, features */
610
611 if (req->r_mempool)
612 msg = ceph_msgpool_get(&osdc->msgpool_op, msg_size,
613 num_request_data_items);
614 else
615 msg = ceph_msg_new2(CEPH_MSG_OSD_OP, msg_size,
616 num_request_data_items, gfp, true);
617 if (!msg)
618 return -ENOMEM;
619
620 memset(msg->front.iov_base, 0, msg->front.iov_len);
621 req->r_request = msg;
622
623 /* create reply message */
624 msg_size = OSD_OPREPLY_FRONT_LEN;
625 msg_size += req->r_base_oid.name_len;
626 msg_size += req->r_num_ops * sizeof(struct ceph_osd_op);
627
628 if (req->r_mempool)
629 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, msg_size,
630 num_reply_data_items);
631 else
632 msg = ceph_msg_new2(CEPH_MSG_OSD_OPREPLY, msg_size,
633 num_reply_data_items, gfp, true);
634 if (!msg)
635 return -ENOMEM;
636
637 req->r_reply = msg;
638
639 return 0;
640}
641
642static bool osd_req_opcode_valid(u16 opcode)
643{
644 switch (opcode) {
645#define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
646__CEPH_FORALL_OSD_OPS(GENERATE_CASE)
647#undef GENERATE_CASE
648 default:
649 return false;
650 }
651}
652
653static void get_num_data_items(struct ceph_osd_request *req,
654 int *num_request_data_items,
655 int *num_reply_data_items)
656{
657 struct ceph_osd_req_op *op;
658
659 *num_request_data_items = 0;
660 *num_reply_data_items = 0;
661
662 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
663 switch (op->op) {
664 /* request */
665 case CEPH_OSD_OP_WRITE:
666 case CEPH_OSD_OP_WRITEFULL:
667 case CEPH_OSD_OP_SETXATTR:
668 case CEPH_OSD_OP_CMPXATTR:
669 case CEPH_OSD_OP_NOTIFY_ACK:
670 case CEPH_OSD_OP_COPY_FROM2:
671 *num_request_data_items += 1;
672 break;
673
674 /* reply */
675 case CEPH_OSD_OP_STAT:
676 case CEPH_OSD_OP_READ:
677 case CEPH_OSD_OP_SPARSE_READ:
678 case CEPH_OSD_OP_LIST_WATCHERS:
679 *num_reply_data_items += 1;
680 break;
681
682 /* both */
683 case CEPH_OSD_OP_NOTIFY:
684 *num_request_data_items += 1;
685 *num_reply_data_items += 1;
686 break;
687 case CEPH_OSD_OP_CALL:
688 *num_request_data_items += 2;
689 *num_reply_data_items += 1;
690 break;
691
692 default:
693 WARN_ON(!osd_req_opcode_valid(op->op));
694 break;
695 }
696 }
697}
698
699/*
700 * oid, oloc and OSD op opcode(s) must be filled in before this function
701 * is called.
702 */
703int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp)
704{
705 int num_request_data_items, num_reply_data_items;
706
707 get_num_data_items(req, &num_request_data_items, &num_reply_data_items);
708 return __ceph_osdc_alloc_messages(req, gfp, num_request_data_items,
709 num_reply_data_items);
710}
711EXPORT_SYMBOL(ceph_osdc_alloc_messages);
712
713/*
714 * This is an osd op init function for opcodes that have no data or
715 * other information associated with them. It also serves as a
716 * common init routine for all the other init functions, below.
717 */
718struct ceph_osd_req_op *
719osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
720 u16 opcode, u32 flags)
721{
722 struct ceph_osd_req_op *op;
723
724 BUG_ON(which >= osd_req->r_num_ops);
725 BUG_ON(!osd_req_opcode_valid(opcode));
726
727 op = &osd_req->r_ops[which];
728 memset(op, 0, sizeof (*op));
729 op->op = opcode;
730 op->flags = flags;
731
732 return op;
733}
734EXPORT_SYMBOL(osd_req_op_init);
735
736void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
737 unsigned int which, u16 opcode,
738 u64 offset, u64 length,
739 u64 truncate_size, u32 truncate_seq)
740{
741 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
742 opcode, 0);
743 size_t payload_len = 0;
744
745 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
746 opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO &&
747 opcode != CEPH_OSD_OP_TRUNCATE && opcode != CEPH_OSD_OP_SPARSE_READ);
748
749 op->extent.offset = offset;
750 op->extent.length = length;
751 op->extent.truncate_size = truncate_size;
752 op->extent.truncate_seq = truncate_seq;
753 if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL)
754 payload_len += length;
755
756 op->indata_len = payload_len;
757}
758EXPORT_SYMBOL(osd_req_op_extent_init);
759
760void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
761 unsigned int which, u64 length)
762{
763 struct ceph_osd_req_op *op;
764 u64 previous;
765
766 BUG_ON(which >= osd_req->r_num_ops);
767 op = &osd_req->r_ops[which];
768 previous = op->extent.length;
769
770 if (length == previous)
771 return; /* Nothing to do */
772 BUG_ON(length > previous);
773
774 op->extent.length = length;
775 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
776 op->indata_len -= previous - length;
777}
778EXPORT_SYMBOL(osd_req_op_extent_update);
779
780void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req,
781 unsigned int which, u64 offset_inc)
782{
783 struct ceph_osd_req_op *op, *prev_op;
784
785 BUG_ON(which + 1 >= osd_req->r_num_ops);
786
787 prev_op = &osd_req->r_ops[which];
788 op = osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags);
789 /* dup previous one */
790 op->indata_len = prev_op->indata_len;
791 op->outdata_len = prev_op->outdata_len;
792 op->extent = prev_op->extent;
793 /* adjust offset */
794 op->extent.offset += offset_inc;
795 op->extent.length -= offset_inc;
796
797 if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL)
798 op->indata_len -= offset_inc;
799}
800EXPORT_SYMBOL(osd_req_op_extent_dup_last);
801
802int osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
803 const char *class, const char *method)
804{
805 struct ceph_osd_req_op *op;
806 struct ceph_pagelist *pagelist;
807 size_t payload_len = 0;
808 size_t size;
809 int ret;
810
811 op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_CALL, 0);
812
813 pagelist = ceph_pagelist_alloc(GFP_NOFS);
814 if (!pagelist)
815 return -ENOMEM;
816
817 op->cls.class_name = class;
818 size = strlen(class);
819 BUG_ON(size > (size_t) U8_MAX);
820 op->cls.class_len = size;
821 ret = ceph_pagelist_append(pagelist, class, size);
822 if (ret)
823 goto err_pagelist_free;
824 payload_len += size;
825
826 op->cls.method_name = method;
827 size = strlen(method);
828 BUG_ON(size > (size_t) U8_MAX);
829 op->cls.method_len = size;
830 ret = ceph_pagelist_append(pagelist, method, size);
831 if (ret)
832 goto err_pagelist_free;
833 payload_len += size;
834
835 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
836 op->indata_len = payload_len;
837 return 0;
838
839err_pagelist_free:
840 ceph_pagelist_release(pagelist);
841 return ret;
842}
843EXPORT_SYMBOL(osd_req_op_cls_init);
844
845int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
846 u16 opcode, const char *name, const void *value,
847 size_t size, u8 cmp_op, u8 cmp_mode)
848{
849 struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which,
850 opcode, 0);
851 struct ceph_pagelist *pagelist;
852 size_t payload_len;
853 int ret;
854
855 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
856
857 pagelist = ceph_pagelist_alloc(GFP_NOFS);
858 if (!pagelist)
859 return -ENOMEM;
860
861 payload_len = strlen(name);
862 op->xattr.name_len = payload_len;
863 ret = ceph_pagelist_append(pagelist, name, payload_len);
864 if (ret)
865 goto err_pagelist_free;
866
867 op->xattr.value_len = size;
868 ret = ceph_pagelist_append(pagelist, value, size);
869 if (ret)
870 goto err_pagelist_free;
871 payload_len += size;
872
873 op->xattr.cmp_op = cmp_op;
874 op->xattr.cmp_mode = cmp_mode;
875
876 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
877 op->indata_len = payload_len;
878 return 0;
879
880err_pagelist_free:
881 ceph_pagelist_release(pagelist);
882 return ret;
883}
884EXPORT_SYMBOL(osd_req_op_xattr_init);
885
886/*
887 * @watch_opcode: CEPH_OSD_WATCH_OP_*
888 */
889static void osd_req_op_watch_init(struct ceph_osd_request *req, int which,
890 u8 watch_opcode, u64 cookie, u32 gen)
891{
892 struct ceph_osd_req_op *op;
893
894 op = osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0);
895 op->watch.cookie = cookie;
896 op->watch.op = watch_opcode;
897 op->watch.gen = gen;
898}
899
900/*
901 * prot_ver, timeout and notify payload (may be empty) should already be
902 * encoded in @request_pl
903 */
904static void osd_req_op_notify_init(struct ceph_osd_request *req, int which,
905 u64 cookie, struct ceph_pagelist *request_pl)
906{
907 struct ceph_osd_req_op *op;
908
909 op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0);
910 op->notify.cookie = cookie;
911
912 ceph_osd_data_pagelist_init(&op->notify.request_data, request_pl);
913 op->indata_len = request_pl->length;
914}
915
916/*
917 * @flags: CEPH_OSD_OP_ALLOC_HINT_FLAG_*
918 */
919void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
920 unsigned int which,
921 u64 expected_object_size,
922 u64 expected_write_size,
923 u32 flags)
924{
925 struct ceph_osd_req_op *op;
926
927 op = osd_req_op_init(osd_req, which, CEPH_OSD_OP_SETALLOCHINT, 0);
928 op->alloc_hint.expected_object_size = expected_object_size;
929 op->alloc_hint.expected_write_size = expected_write_size;
930 op->alloc_hint.flags = flags;
931
932 /*
933 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
934 * not worth a feature bit. Set FAILOK per-op flag to make
935 * sure older osds don't trip over an unsupported opcode.
936 */
937 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
938}
939EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
940
941static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
942 struct ceph_osd_data *osd_data)
943{
944 u64 length = ceph_osd_data_length(osd_data);
945
946 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
947 BUG_ON(length > (u64) SIZE_MAX);
948 if (length)
949 ceph_msg_data_add_pages(msg, osd_data->pages,
950 length, osd_data->alignment, false);
951 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
952 BUG_ON(!length);
953 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
954#ifdef CONFIG_BLOCK
955 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
956 ceph_msg_data_add_bio(msg, &osd_data->bio_pos, length);
957#endif
958 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BVECS) {
959 ceph_msg_data_add_bvecs(msg, &osd_data->bvec_pos);
960 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_ITER) {
961 ceph_msg_data_add_iter(msg, &osd_data->iter);
962 } else {
963 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
964 }
965}
966
967static u32 osd_req_encode_op(struct ceph_osd_op *dst,
968 const struct ceph_osd_req_op *src)
969{
970 switch (src->op) {
971 case CEPH_OSD_OP_STAT:
972 break;
973 case CEPH_OSD_OP_READ:
974 case CEPH_OSD_OP_SPARSE_READ:
975 case CEPH_OSD_OP_WRITE:
976 case CEPH_OSD_OP_WRITEFULL:
977 case CEPH_OSD_OP_ZERO:
978 case CEPH_OSD_OP_TRUNCATE:
979 dst->extent.offset = cpu_to_le64(src->extent.offset);
980 dst->extent.length = cpu_to_le64(src->extent.length);
981 dst->extent.truncate_size =
982 cpu_to_le64(src->extent.truncate_size);
983 dst->extent.truncate_seq =
984 cpu_to_le32(src->extent.truncate_seq);
985 break;
986 case CEPH_OSD_OP_CALL:
987 dst->cls.class_len = src->cls.class_len;
988 dst->cls.method_len = src->cls.method_len;
989 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
990 break;
991 case CEPH_OSD_OP_WATCH:
992 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
993 dst->watch.ver = cpu_to_le64(0);
994 dst->watch.op = src->watch.op;
995 dst->watch.gen = cpu_to_le32(src->watch.gen);
996 break;
997 case CEPH_OSD_OP_NOTIFY_ACK:
998 break;
999 case CEPH_OSD_OP_NOTIFY:
1000 dst->notify.cookie = cpu_to_le64(src->notify.cookie);
1001 break;
1002 case CEPH_OSD_OP_LIST_WATCHERS:
1003 break;
1004 case CEPH_OSD_OP_SETALLOCHINT:
1005 dst->alloc_hint.expected_object_size =
1006 cpu_to_le64(src->alloc_hint.expected_object_size);
1007 dst->alloc_hint.expected_write_size =
1008 cpu_to_le64(src->alloc_hint.expected_write_size);
1009 dst->alloc_hint.flags = cpu_to_le32(src->alloc_hint.flags);
1010 break;
1011 case CEPH_OSD_OP_SETXATTR:
1012 case CEPH_OSD_OP_CMPXATTR:
1013 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
1014 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
1015 dst->xattr.cmp_op = src->xattr.cmp_op;
1016 dst->xattr.cmp_mode = src->xattr.cmp_mode;
1017 break;
1018 case CEPH_OSD_OP_CREATE:
1019 case CEPH_OSD_OP_DELETE:
1020 break;
1021 case CEPH_OSD_OP_COPY_FROM2:
1022 dst->copy_from.snapid = cpu_to_le64(src->copy_from.snapid);
1023 dst->copy_from.src_version =
1024 cpu_to_le64(src->copy_from.src_version);
1025 dst->copy_from.flags = src->copy_from.flags;
1026 dst->copy_from.src_fadvise_flags =
1027 cpu_to_le32(src->copy_from.src_fadvise_flags);
1028 break;
1029 case CEPH_OSD_OP_ASSERT_VER:
1030 dst->assert_ver.unused = cpu_to_le64(0);
1031 dst->assert_ver.ver = cpu_to_le64(src->assert_ver.ver);
1032 break;
1033 default:
1034 pr_err("unsupported osd opcode %s\n",
1035 ceph_osd_op_name(src->op));
1036 WARN_ON(1);
1037
1038 return 0;
1039 }
1040
1041 dst->op = cpu_to_le16(src->op);
1042 dst->flags = cpu_to_le32(src->flags);
1043 dst->payload_len = cpu_to_le32(src->indata_len);
1044
1045 return src->indata_len;
1046}
1047
1048/*
1049 * build new request AND message, calculate layout, and adjust file
1050 * extent as needed.
1051 *
1052 * if the file was recently truncated, we include information about its
1053 * old and new size so that the object can be updated appropriately. (we
1054 * avoid synchronously deleting truncated objects because it's slow.)
1055 */
1056struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
1057 struct ceph_file_layout *layout,
1058 struct ceph_vino vino,
1059 u64 off, u64 *plen,
1060 unsigned int which, int num_ops,
1061 int opcode, int flags,
1062 struct ceph_snap_context *snapc,
1063 u32 truncate_seq,
1064 u64 truncate_size,
1065 bool use_mempool)
1066{
1067 struct ceph_osd_request *req;
1068 u64 objnum = 0;
1069 u64 objoff = 0;
1070 u64 objlen = 0;
1071 int r;
1072
1073 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
1074 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
1075 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE &&
1076 opcode != CEPH_OSD_OP_SPARSE_READ);
1077
1078 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
1079 GFP_NOFS);
1080 if (!req) {
1081 r = -ENOMEM;
1082 goto fail;
1083 }
1084
1085 /* calculate max write size */
1086 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
1087 if (r)
1088 goto fail;
1089
1090 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
1091 osd_req_op_init(req, which, opcode, 0);
1092 } else {
1093 u32 object_size = layout->object_size;
1094 u32 object_base = off - objoff;
1095 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
1096 if (truncate_size <= object_base) {
1097 truncate_size = 0;
1098 } else {
1099 truncate_size -= object_base;
1100 if (truncate_size > object_size)
1101 truncate_size = object_size;
1102 }
1103 }
1104 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
1105 truncate_size, truncate_seq);
1106 }
1107
1108 req->r_base_oloc.pool = layout->pool_id;
1109 req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns);
1110 ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum);
1111 req->r_flags = flags | osdc->client->options->read_from_replica;
1112
1113 req->r_snapid = vino.snap;
1114 if (flags & CEPH_OSD_FLAG_WRITE)
1115 req->r_data_offset = off;
1116
1117 if (num_ops > 1) {
1118 int num_req_ops, num_rep_ops;
1119
1120 /*
1121 * If this is a multi-op write request, assume that we'll need
1122 * request ops. If it's a multi-op read then assume we'll need
1123 * reply ops. Anything else and call it -EINVAL.
1124 */
1125 if (flags & CEPH_OSD_FLAG_WRITE) {
1126 num_req_ops = num_ops;
1127 num_rep_ops = 0;
1128 } else if (flags & CEPH_OSD_FLAG_READ) {
1129 num_req_ops = 0;
1130 num_rep_ops = num_ops;
1131 } else {
1132 r = -EINVAL;
1133 goto fail;
1134 }
1135
1136 r = __ceph_osdc_alloc_messages(req, GFP_NOFS, num_req_ops,
1137 num_rep_ops);
1138 } else {
1139 r = ceph_osdc_alloc_messages(req, GFP_NOFS);
1140 }
1141 if (r)
1142 goto fail;
1143
1144 return req;
1145
1146fail:
1147 ceph_osdc_put_request(req);
1148 return ERR_PTR(r);
1149}
1150EXPORT_SYMBOL(ceph_osdc_new_request);
1151
1152int __ceph_alloc_sparse_ext_map(struct ceph_osd_req_op *op, int cnt)
1153{
1154 WARN_ON(op->op != CEPH_OSD_OP_SPARSE_READ);
1155
1156 op->extent.sparse_ext_cnt = cnt;
1157 op->extent.sparse_ext = kmalloc_objs(*op->extent.sparse_ext, cnt,
1158 GFP_NOFS);
1159 if (!op->extent.sparse_ext)
1160 return -ENOMEM;
1161 return 0;
1162}
1163EXPORT_SYMBOL(__ceph_alloc_sparse_ext_map);
1164
1165/*
1166 * We keep osd requests in an rbtree, sorted by ->r_tid.
1167 */
1168DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node)
1169DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node)
1170
1171/*
1172 * Call @fn on each OSD request as long as @fn returns 0.
1173 */
1174static void for_each_request(struct ceph_osd_client *osdc,
1175 int (*fn)(struct ceph_osd_request *req, void *arg),
1176 void *arg)
1177{
1178 struct rb_node *n, *p;
1179
1180 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
1181 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
1182
1183 for (p = rb_first(&osd->o_requests); p; ) {
1184 struct ceph_osd_request *req =
1185 rb_entry(p, struct ceph_osd_request, r_node);
1186
1187 p = rb_next(p);
1188 if (fn(req, arg))
1189 return;
1190 }
1191 }
1192
1193 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
1194 struct ceph_osd_request *req =
1195 rb_entry(p, struct ceph_osd_request, r_node);
1196
1197 p = rb_next(p);
1198 if (fn(req, arg))
1199 return;
1200 }
1201}
1202
1203static bool osd_homeless(struct ceph_osd *osd)
1204{
1205 return osd->o_osd == CEPH_HOMELESS_OSD;
1206}
1207
1208static bool osd_registered(struct ceph_osd *osd)
1209{
1210 verify_osdc_locked(osd->o_osdc);
1211
1212 return !RB_EMPTY_NODE(&osd->o_node);
1213}
1214
1215/*
1216 * Assumes @osd is zero-initialized.
1217 */
1218static void osd_init(struct ceph_osd *osd)
1219{
1220 refcount_set(&osd->o_ref, 1);
1221 RB_CLEAR_NODE(&osd->o_node);
1222 spin_lock_init(&osd->o_requests_lock);
1223 osd->o_requests = RB_ROOT;
1224 osd->o_linger_requests = RB_ROOT;
1225 osd->o_backoff_mappings = RB_ROOT;
1226 osd->o_backoffs_by_id = RB_ROOT;
1227 INIT_LIST_HEAD(&osd->o_osd_lru);
1228 INIT_LIST_HEAD(&osd->o_keepalive_item);
1229 osd->o_incarnation = 1;
1230 mutex_init(&osd->lock);
1231}
1232
1233static void ceph_init_sparse_read(struct ceph_sparse_read *sr)
1234{
1235 kfree(sr->sr_extent);
1236 memset(sr, '\0', sizeof(*sr));
1237 sr->sr_state = CEPH_SPARSE_READ_HDR;
1238}
1239
1240static void osd_cleanup(struct ceph_osd *osd)
1241{
1242 WARN_ON(!RB_EMPTY_NODE(&osd->o_node));
1243 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
1244 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
1245 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoff_mappings));
1246 WARN_ON(!RB_EMPTY_ROOT(&osd->o_backoffs_by_id));
1247 WARN_ON(!list_empty(&osd->o_osd_lru));
1248 WARN_ON(!list_empty(&osd->o_keepalive_item));
1249
1250 ceph_init_sparse_read(&osd->o_sparse_read);
1251
1252 if (osd->o_auth.authorizer) {
1253 WARN_ON(osd_homeless(osd));
1254 ceph_auth_destroy_authorizer(osd->o_auth.authorizer);
1255 }
1256}
1257
1258/*
1259 * Track open sessions with osds.
1260 */
1261static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
1262{
1263 struct ceph_osd *osd;
1264
1265 WARN_ON(onum == CEPH_HOMELESS_OSD);
1266
1267 osd = kzalloc_obj(*osd, GFP_NOIO | __GFP_NOFAIL);
1268 osd_init(osd);
1269 osd->o_osdc = osdc;
1270 osd->o_osd = onum;
1271 osd->o_sparse_op_idx = -1;
1272
1273 ceph_init_sparse_read(&osd->o_sparse_read);
1274
1275 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
1276
1277 return osd;
1278}
1279
1280static struct ceph_osd *get_osd(struct ceph_osd *osd)
1281{
1282 if (refcount_inc_not_zero(&osd->o_ref)) {
1283 dout("get_osd %p -> %d\n", osd, refcount_read(&osd->o_ref));
1284 return osd;
1285 } else {
1286 dout("get_osd %p FAIL\n", osd);
1287 return NULL;
1288 }
1289}
1290
1291static void put_osd(struct ceph_osd *osd)
1292{
1293 dout("put_osd %p -> %d\n", osd, refcount_read(&osd->o_ref) - 1);
1294 if (refcount_dec_and_test(&osd->o_ref)) {
1295 osd_cleanup(osd);
1296 kfree(osd);
1297 }
1298}
1299
1300DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node)
1301
1302static void __move_osd_to_lru(struct ceph_osd *osd)
1303{
1304 struct ceph_osd_client *osdc = osd->o_osdc;
1305
1306 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1307 BUG_ON(!list_empty(&osd->o_osd_lru));
1308
1309 spin_lock(&osdc->osd_lru_lock);
1310 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1311 spin_unlock(&osdc->osd_lru_lock);
1312
1313 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
1314}
1315
1316static void maybe_move_osd_to_lru(struct ceph_osd *osd)
1317{
1318 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1319 RB_EMPTY_ROOT(&osd->o_linger_requests))
1320 __move_osd_to_lru(osd);
1321}
1322
1323static void __remove_osd_from_lru(struct ceph_osd *osd)
1324{
1325 struct ceph_osd_client *osdc = osd->o_osdc;
1326
1327 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1328
1329 spin_lock(&osdc->osd_lru_lock);
1330 if (!list_empty(&osd->o_osd_lru))
1331 list_del_init(&osd->o_osd_lru);
1332 spin_unlock(&osdc->osd_lru_lock);
1333}
1334
1335/*
1336 * Close the connection and assign any leftover requests to the
1337 * homeless session.
1338 */
1339static void close_osd(struct ceph_osd *osd)
1340{
1341 struct ceph_osd_client *osdc = osd->o_osdc;
1342 struct rb_node *n;
1343
1344 verify_osdc_wrlocked(osdc);
1345 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1346
1347 ceph_con_close(&osd->o_con);
1348
1349 for (n = rb_first(&osd->o_requests); n; ) {
1350 struct ceph_osd_request *req =
1351 rb_entry(n, struct ceph_osd_request, r_node);
1352
1353 n = rb_next(n); /* unlink_request() */
1354
1355 dout(" reassigning req %p tid %llu\n", req, req->r_tid);
1356 unlink_request(osd, req);
1357 link_request(&osdc->homeless_osd, req);
1358 }
1359 for (n = rb_first(&osd->o_linger_requests); n; ) {
1360 struct ceph_osd_linger_request *lreq =
1361 rb_entry(n, struct ceph_osd_linger_request, node);
1362
1363 n = rb_next(n); /* unlink_linger() */
1364
1365 dout(" reassigning lreq %p linger_id %llu\n", lreq,
1366 lreq->linger_id);
1367 unlink_linger(osd, lreq);
1368 link_linger(&osdc->homeless_osd, lreq);
1369 }
1370 clear_backoffs(osd);
1371
1372 __remove_osd_from_lru(osd);
1373 erase_osd(&osdc->osds, osd);
1374 put_osd(osd);
1375}
1376
1377/*
1378 * reset osd connect
1379 */
1380static int reopen_osd(struct ceph_osd *osd)
1381{
1382 struct ceph_entity_addr *peer_addr;
1383
1384 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
1385
1386 if (RB_EMPTY_ROOT(&osd->o_requests) &&
1387 RB_EMPTY_ROOT(&osd->o_linger_requests)) {
1388 close_osd(osd);
1389 return -ENODEV;
1390 }
1391
1392 peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd];
1393 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1394 !ceph_con_opened(&osd->o_con)) {
1395 struct rb_node *n;
1396
1397 dout("osd addr hasn't changed and connection never opened, "
1398 "letting msgr retry\n");
1399 /* touch each r_stamp for handle_timeout()'s benfit */
1400 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
1401 struct ceph_osd_request *req =
1402 rb_entry(n, struct ceph_osd_request, r_node);
1403 req->r_stamp = jiffies;
1404 }
1405
1406 return -EAGAIN;
1407 }
1408
1409 ceph_con_close(&osd->o_con);
1410 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1411 osd->o_incarnation++;
1412
1413 return 0;
1414}
1415
1416static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o,
1417 bool wrlocked)
1418{
1419 struct ceph_osd *osd;
1420
1421 if (wrlocked)
1422 verify_osdc_wrlocked(osdc);
1423 else
1424 verify_osdc_locked(osdc);
1425
1426 if (o != CEPH_HOMELESS_OSD)
1427 osd = lookup_osd(&osdc->osds, o);
1428 else
1429 osd = &osdc->homeless_osd;
1430 if (!osd) {
1431 if (!wrlocked)
1432 return ERR_PTR(-EAGAIN);
1433
1434 osd = create_osd(osdc, o);
1435 insert_osd(&osdc->osds, osd);
1436 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd,
1437 &osdc->osdmap->osd_addr[osd->o_osd]);
1438 }
1439
1440 dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd);
1441 return osd;
1442}
1443
1444/*
1445 * Create request <-> OSD session relation.
1446 *
1447 * @req has to be assigned a tid, @osd may be homeless.
1448 */
1449static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1450{
1451 verify_osd_locked(osd);
1452 WARN_ON(!req->r_tid || req->r_osd);
1453 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1454 req, req->r_tid);
1455
1456 if (!osd_homeless(osd))
1457 __remove_osd_from_lru(osd);
1458 else
1459 atomic_inc(&osd->o_osdc->num_homeless);
1460
1461 get_osd(osd);
1462 spin_lock(&osd->o_requests_lock);
1463 insert_request(&osd->o_requests, req);
1464 spin_unlock(&osd->o_requests_lock);
1465 req->r_osd = osd;
1466}
1467
1468static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req)
1469{
1470 verify_osd_locked(osd);
1471 WARN_ON(req->r_osd != osd);
1472 dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd,
1473 req, req->r_tid);
1474
1475 req->r_osd = NULL;
1476 spin_lock(&osd->o_requests_lock);
1477 erase_request(&osd->o_requests, req);
1478 spin_unlock(&osd->o_requests_lock);
1479 put_osd(osd);
1480
1481 if (!osd_homeless(osd))
1482 maybe_move_osd_to_lru(osd);
1483 else
1484 atomic_dec(&osd->o_osdc->num_homeless);
1485}
1486
1487static bool __pool_full(struct ceph_pg_pool_info *pi)
1488{
1489 return pi->flags & CEPH_POOL_FLAG_FULL;
1490}
1491
1492static bool have_pool_full(struct ceph_osd_client *osdc)
1493{
1494 struct rb_node *n;
1495
1496 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
1497 struct ceph_pg_pool_info *pi =
1498 rb_entry(n, struct ceph_pg_pool_info, node);
1499
1500 if (__pool_full(pi))
1501 return true;
1502 }
1503
1504 return false;
1505}
1506
1507static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id)
1508{
1509 struct ceph_pg_pool_info *pi;
1510
1511 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
1512 if (!pi)
1513 return false;
1514
1515 return __pool_full(pi);
1516}
1517
1518/*
1519 * Returns whether a request should be blocked from being sent
1520 * based on the current osdmap and osd_client settings.
1521 */
1522static bool target_should_be_paused(struct ceph_osd_client *osdc,
1523 const struct ceph_osd_request_target *t,
1524 struct ceph_pg_pool_info *pi)
1525{
1526 bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
1527 bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
1528 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
1529 __pool_full(pi);
1530
1531 WARN_ON(pi->id != t->target_oloc.pool);
1532 return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) ||
1533 ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) ||
1534 (osdc->osdmap->epoch < osdc->epoch_barrier);
1535}
1536
1537static int pick_random_replica(const struct ceph_osds *acting)
1538{
1539 int i = get_random_u32_below(acting->size);
1540
1541 dout("%s picked osd%d, primary osd%d\n", __func__,
1542 acting->osds[i], acting->primary);
1543 return i;
1544}
1545
1546/*
1547 * Picks the closest replica based on client's location given by
1548 * crush_location option. Prefers the primary if the locality is
1549 * the same.
1550 */
1551static int pick_closest_replica(struct ceph_osd_client *osdc,
1552 const struct ceph_osds *acting)
1553{
1554 struct ceph_options *opt = osdc->client->options;
1555 int best_i, best_locality;
1556 int i = 0, locality;
1557
1558 do {
1559 locality = ceph_get_crush_locality(osdc->osdmap,
1560 acting->osds[i],
1561 &opt->crush_locs);
1562 if (i == 0 ||
1563 (locality >= 0 && best_locality < 0) ||
1564 (locality >= 0 && best_locality >= 0 &&
1565 locality < best_locality)) {
1566 best_i = i;
1567 best_locality = locality;
1568 }
1569 } while (++i < acting->size);
1570
1571 dout("%s picked osd%d with locality %d, primary osd%d\n", __func__,
1572 acting->osds[best_i], best_locality, acting->primary);
1573 return best_i;
1574}
1575
1576enum calc_target_result {
1577 CALC_TARGET_NO_ACTION = 0,
1578 CALC_TARGET_NEED_RESEND,
1579 CALC_TARGET_POOL_DNE,
1580};
1581
1582static enum calc_target_result calc_target(struct ceph_osd_client *osdc,
1583 struct ceph_osd_request_target *t,
1584 bool any_change)
1585{
1586 struct ceph_pg_pool_info *pi;
1587 struct ceph_pg pgid, last_pgid;
1588 struct ceph_osds up, acting;
1589 bool should_be_paused;
1590 bool is_read = t->flags & CEPH_OSD_FLAG_READ;
1591 bool is_write = t->flags & CEPH_OSD_FLAG_WRITE;
1592 bool force_resend = false;
1593 bool unpaused = false;
1594 bool legacy_change = false;
1595 bool split = false;
1596 bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE);
1597 bool recovery_deletes = ceph_osdmap_flag(osdc,
1598 CEPH_OSDMAP_RECOVERY_DELETES);
1599 enum calc_target_result ct_res;
1600
1601 t->epoch = osdc->osdmap->epoch;
1602 pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool);
1603 if (!pi) {
1604 t->osd = CEPH_HOMELESS_OSD;
1605 ct_res = CALC_TARGET_POOL_DNE;
1606 goto out;
1607 }
1608
1609 if (osdc->osdmap->epoch == pi->last_force_request_resend) {
1610 if (t->last_force_resend < pi->last_force_request_resend) {
1611 t->last_force_resend = pi->last_force_request_resend;
1612 force_resend = true;
1613 } else if (t->last_force_resend == 0) {
1614 force_resend = true;
1615 }
1616 }
1617
1618 /* apply tiering */
1619 ceph_oid_copy(&t->target_oid, &t->base_oid);
1620 ceph_oloc_copy(&t->target_oloc, &t->base_oloc);
1621 if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
1622 if (is_read && pi->read_tier >= 0)
1623 t->target_oloc.pool = pi->read_tier;
1624 if (is_write && pi->write_tier >= 0)
1625 t->target_oloc.pool = pi->write_tier;
1626
1627 pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool);
1628 if (!pi) {
1629 t->osd = CEPH_HOMELESS_OSD;
1630 ct_res = CALC_TARGET_POOL_DNE;
1631 goto out;
1632 }
1633 }
1634
1635 __ceph_object_locator_to_pg(pi, &t->target_oid, &t->target_oloc, &pgid);
1636 last_pgid.pool = pgid.pool;
1637 last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask);
1638
1639 ceph_pg_to_up_acting_osds(osdc->osdmap, pi, &pgid, &up, &acting);
1640 if (any_change &&
1641 ceph_is_new_interval(&t->acting,
1642 &acting,
1643 &t->up,
1644 &up,
1645 t->size,
1646 pi->size,
1647 t->min_size,
1648 pi->min_size,
1649 t->pg_num,
1650 pi->pg_num,
1651 t->sort_bitwise,
1652 sort_bitwise,
1653 t->recovery_deletes,
1654 recovery_deletes,
1655 &last_pgid))
1656 force_resend = true;
1657
1658 should_be_paused = target_should_be_paused(osdc, t, pi);
1659 if (t->paused && !should_be_paused) {
1660 unpaused = true;
1661 }
1662 if (t->paused != should_be_paused) {
1663 dout("%s t %p paused %d -> %d\n", __func__, t, t->paused,
1664 should_be_paused);
1665 t->paused = should_be_paused;
1666 }
1667
1668 legacy_change = ceph_pg_compare(&t->pgid, &pgid) ||
1669 ceph_osds_changed(&t->acting, &acting,
1670 t->used_replica || any_change);
1671 if (t->pg_num)
1672 split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num);
1673
1674 if (legacy_change || force_resend || split) {
1675 t->pgid = pgid; /* struct */
1676 ceph_pg_to_primary_shard(osdc->osdmap, pi, &pgid, &t->spgid);
1677 ceph_osds_copy(&t->acting, &acting);
1678 ceph_osds_copy(&t->up, &up);
1679 t->size = pi->size;
1680 t->min_size = pi->min_size;
1681 t->pg_num = pi->pg_num;
1682 t->pg_num_mask = pi->pg_num_mask;
1683 t->sort_bitwise = sort_bitwise;
1684 t->recovery_deletes = recovery_deletes;
1685
1686 if ((t->flags & (CEPH_OSD_FLAG_BALANCE_READS |
1687 CEPH_OSD_FLAG_LOCALIZE_READS)) &&
1688 !is_write && pi->type == CEPH_POOL_TYPE_REP &&
1689 acting.size > 1) {
1690 int pos;
1691
1692 WARN_ON(!is_read || acting.osds[0] != acting.primary);
1693 if (t->flags & CEPH_OSD_FLAG_BALANCE_READS) {
1694 pos = pick_random_replica(&acting);
1695 } else {
1696 pos = pick_closest_replica(osdc, &acting);
1697 }
1698 t->osd = acting.osds[pos];
1699 t->used_replica = pos > 0;
1700 } else {
1701 t->osd = acting.primary;
1702 t->used_replica = false;
1703 }
1704 }
1705
1706 if (unpaused || legacy_change || force_resend || split)
1707 ct_res = CALC_TARGET_NEED_RESEND;
1708 else
1709 ct_res = CALC_TARGET_NO_ACTION;
1710
1711out:
1712 dout("%s t %p -> %d%d%d%d ct_res %d osd%d\n", __func__, t, unpaused,
1713 legacy_change, force_resend, split, ct_res, t->osd);
1714 return ct_res;
1715}
1716
1717static struct ceph_spg_mapping *alloc_spg_mapping(void)
1718{
1719 struct ceph_spg_mapping *spg;
1720
1721 spg = kmalloc_obj(*spg, GFP_NOIO);
1722 if (!spg)
1723 return NULL;
1724
1725 RB_CLEAR_NODE(&spg->node);
1726 spg->backoffs = RB_ROOT;
1727 return spg;
1728}
1729
1730static void free_spg_mapping(struct ceph_spg_mapping *spg)
1731{
1732 WARN_ON(!RB_EMPTY_NODE(&spg->node));
1733 WARN_ON(!RB_EMPTY_ROOT(&spg->backoffs));
1734
1735 kfree(spg);
1736}
1737
1738/*
1739 * rbtree of ceph_spg_mapping for handling map<spg_t, ...>, similar to
1740 * ceph_pg_mapping. Used to track OSD backoffs -- a backoff [range] is
1741 * defined only within a specific spgid; it does not pass anything to
1742 * children on split, or to another primary.
1743 */
1744DEFINE_RB_FUNCS2(spg_mapping, struct ceph_spg_mapping, spgid, ceph_spg_compare,
1745 RB_BYPTR, const struct ceph_spg *, node)
1746
1747static u64 hoid_get_bitwise_key(const struct ceph_hobject_id *hoid)
1748{
1749 return hoid->is_max ? 0x100000000ull : hoid->hash_reverse_bits;
1750}
1751
1752static void hoid_get_effective_key(const struct ceph_hobject_id *hoid,
1753 void **pkey, size_t *pkey_len)
1754{
1755 if (hoid->key_len) {
1756 *pkey = hoid->key;
1757 *pkey_len = hoid->key_len;
1758 } else {
1759 *pkey = hoid->oid;
1760 *pkey_len = hoid->oid_len;
1761 }
1762}
1763
1764static int compare_names(const void *name1, size_t name1_len,
1765 const void *name2, size_t name2_len)
1766{
1767 int ret;
1768
1769 ret = memcmp(name1, name2, min(name1_len, name2_len));
1770 if (!ret) {
1771 if (name1_len < name2_len)
1772 ret = -1;
1773 else if (name1_len > name2_len)
1774 ret = 1;
1775 }
1776 return ret;
1777}
1778
1779static int hoid_compare(const struct ceph_hobject_id *lhs,
1780 const struct ceph_hobject_id *rhs)
1781{
1782 void *effective_key1, *effective_key2;
1783 size_t effective_key1_len, effective_key2_len;
1784 int ret;
1785
1786 if (lhs->is_max < rhs->is_max)
1787 return -1;
1788 if (lhs->is_max > rhs->is_max)
1789 return 1;
1790
1791 if (lhs->pool < rhs->pool)
1792 return -1;
1793 if (lhs->pool > rhs->pool)
1794 return 1;
1795
1796 if (hoid_get_bitwise_key(lhs) < hoid_get_bitwise_key(rhs))
1797 return -1;
1798 if (hoid_get_bitwise_key(lhs) > hoid_get_bitwise_key(rhs))
1799 return 1;
1800
1801 ret = compare_names(lhs->nspace, lhs->nspace_len,
1802 rhs->nspace, rhs->nspace_len);
1803 if (ret)
1804 return ret;
1805
1806 hoid_get_effective_key(lhs, &effective_key1, &effective_key1_len);
1807 hoid_get_effective_key(rhs, &effective_key2, &effective_key2_len);
1808 ret = compare_names(effective_key1, effective_key1_len,
1809 effective_key2, effective_key2_len);
1810 if (ret)
1811 return ret;
1812
1813 ret = compare_names(lhs->oid, lhs->oid_len, rhs->oid, rhs->oid_len);
1814 if (ret)
1815 return ret;
1816
1817 if (lhs->snapid < rhs->snapid)
1818 return -1;
1819 if (lhs->snapid > rhs->snapid)
1820 return 1;
1821
1822 return 0;
1823}
1824
1825/*
1826 * For decoding ->begin and ->end of MOSDBackoff only -- no MIN/MAX
1827 * compat stuff here.
1828 *
1829 * Assumes @hoid is zero-initialized.
1830 */
1831static int decode_hoid(void **p, void *end, struct ceph_hobject_id *hoid)
1832{
1833 u8 struct_v;
1834 u32 struct_len;
1835 int ret;
1836
1837 ret = ceph_start_decoding(p, end, 4, "hobject_t", &struct_v,
1838 &struct_len);
1839 if (ret)
1840 return ret;
1841
1842 if (struct_v < 4) {
1843 pr_err("got struct_v %d < 4 of hobject_t\n", struct_v);
1844 goto e_inval;
1845 }
1846
1847 hoid->key = ceph_extract_encoded_string(p, end, &hoid->key_len,
1848 GFP_NOIO);
1849 if (IS_ERR(hoid->key)) {
1850 ret = PTR_ERR(hoid->key);
1851 hoid->key = NULL;
1852 return ret;
1853 }
1854
1855 hoid->oid = ceph_extract_encoded_string(p, end, &hoid->oid_len,
1856 GFP_NOIO);
1857 if (IS_ERR(hoid->oid)) {
1858 ret = PTR_ERR(hoid->oid);
1859 hoid->oid = NULL;
1860 return ret;
1861 }
1862
1863 ceph_decode_64_safe(p, end, hoid->snapid, e_inval);
1864 ceph_decode_32_safe(p, end, hoid->hash, e_inval);
1865 ceph_decode_8_safe(p, end, hoid->is_max, e_inval);
1866
1867 hoid->nspace = ceph_extract_encoded_string(p, end, &hoid->nspace_len,
1868 GFP_NOIO);
1869 if (IS_ERR(hoid->nspace)) {
1870 ret = PTR_ERR(hoid->nspace);
1871 hoid->nspace = NULL;
1872 return ret;
1873 }
1874
1875 ceph_decode_64_safe(p, end, hoid->pool, e_inval);
1876
1877 ceph_hoid_build_hash_cache(hoid);
1878 return 0;
1879
1880e_inval:
1881 return -EINVAL;
1882}
1883
1884static int hoid_encoding_size(const struct ceph_hobject_id *hoid)
1885{
1886 return 8 + 4 + 1 + 8 + /* snapid, hash, is_max, pool */
1887 4 + hoid->key_len + 4 + hoid->oid_len + 4 + hoid->nspace_len;
1888}
1889
1890static void encode_hoid(void **p, void *end, const struct ceph_hobject_id *hoid)
1891{
1892 ceph_start_encoding(p, 4, 3, hoid_encoding_size(hoid));
1893 ceph_encode_string(p, end, hoid->key, hoid->key_len);
1894 ceph_encode_string(p, end, hoid->oid, hoid->oid_len);
1895 ceph_encode_64(p, hoid->snapid);
1896 ceph_encode_32(p, hoid->hash);
1897 ceph_encode_8(p, hoid->is_max);
1898 ceph_encode_string(p, end, hoid->nspace, hoid->nspace_len);
1899 ceph_encode_64(p, hoid->pool);
1900}
1901
1902static void free_hoid(struct ceph_hobject_id *hoid)
1903{
1904 if (hoid) {
1905 kfree(hoid->key);
1906 kfree(hoid->oid);
1907 kfree(hoid->nspace);
1908 kfree(hoid);
1909 }
1910}
1911
1912static struct ceph_osd_backoff *alloc_backoff(void)
1913{
1914 struct ceph_osd_backoff *backoff;
1915
1916 backoff = kzalloc_obj(*backoff, GFP_NOIO);
1917 if (!backoff)
1918 return NULL;
1919
1920 RB_CLEAR_NODE(&backoff->spg_node);
1921 RB_CLEAR_NODE(&backoff->id_node);
1922 return backoff;
1923}
1924
1925static void free_backoff(struct ceph_osd_backoff *backoff)
1926{
1927 WARN_ON(!RB_EMPTY_NODE(&backoff->spg_node));
1928 WARN_ON(!RB_EMPTY_NODE(&backoff->id_node));
1929
1930 free_hoid(backoff->begin);
1931 free_hoid(backoff->end);
1932 kfree(backoff);
1933}
1934
1935/*
1936 * Within a specific spgid, backoffs are managed by ->begin hoid.
1937 */
1938DEFINE_RB_INSDEL_FUNCS2(backoff, struct ceph_osd_backoff, begin, hoid_compare,
1939 RB_BYVAL, spg_node);
1940
1941static struct ceph_osd_backoff *lookup_containing_backoff(struct rb_root *root,
1942 const struct ceph_hobject_id *hoid)
1943{
1944 struct rb_node *n = root->rb_node;
1945
1946 while (n) {
1947 struct ceph_osd_backoff *cur =
1948 rb_entry(n, struct ceph_osd_backoff, spg_node);
1949 int cmp;
1950
1951 cmp = hoid_compare(hoid, cur->begin);
1952 if (cmp < 0) {
1953 n = n->rb_left;
1954 } else if (cmp > 0) {
1955 if (hoid_compare(hoid, cur->end) < 0)
1956 return cur;
1957
1958 n = n->rb_right;
1959 } else {
1960 return cur;
1961 }
1962 }
1963
1964 return NULL;
1965}
1966
1967/*
1968 * Each backoff has a unique id within its OSD session.
1969 */
1970DEFINE_RB_FUNCS(backoff_by_id, struct ceph_osd_backoff, id, id_node)
1971
1972static void clear_backoffs(struct ceph_osd *osd)
1973{
1974 while (!RB_EMPTY_ROOT(&osd->o_backoff_mappings)) {
1975 struct ceph_spg_mapping *spg =
1976 rb_entry(rb_first(&osd->o_backoff_mappings),
1977 struct ceph_spg_mapping, node);
1978
1979 while (!RB_EMPTY_ROOT(&spg->backoffs)) {
1980 struct ceph_osd_backoff *backoff =
1981 rb_entry(rb_first(&spg->backoffs),
1982 struct ceph_osd_backoff, spg_node);
1983
1984 erase_backoff(&spg->backoffs, backoff);
1985 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
1986 free_backoff(backoff);
1987 }
1988 erase_spg_mapping(&osd->o_backoff_mappings, spg);
1989 free_spg_mapping(spg);
1990 }
1991}
1992
1993/*
1994 * Set up a temporary, non-owning view into @t.
1995 */
1996static void hoid_fill_from_target(struct ceph_hobject_id *hoid,
1997 const struct ceph_osd_request_target *t)
1998{
1999 hoid->key = NULL;
2000 hoid->key_len = 0;
2001 hoid->oid = t->target_oid.name;
2002 hoid->oid_len = t->target_oid.name_len;
2003 hoid->snapid = CEPH_NOSNAP;
2004 hoid->hash = t->pgid.seed;
2005 hoid->is_max = false;
2006 if (t->target_oloc.pool_ns) {
2007 hoid->nspace = t->target_oloc.pool_ns->str;
2008 hoid->nspace_len = t->target_oloc.pool_ns->len;
2009 } else {
2010 hoid->nspace = NULL;
2011 hoid->nspace_len = 0;
2012 }
2013 hoid->pool = t->target_oloc.pool;
2014 ceph_hoid_build_hash_cache(hoid);
2015}
2016
2017static bool should_plug_request(struct ceph_osd_request *req)
2018{
2019 struct ceph_osd *osd = req->r_osd;
2020 struct ceph_spg_mapping *spg;
2021 struct ceph_osd_backoff *backoff;
2022 struct ceph_hobject_id hoid;
2023
2024 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &req->r_t.spgid);
2025 if (!spg)
2026 return false;
2027
2028 hoid_fill_from_target(&hoid, &req->r_t);
2029 backoff = lookup_containing_backoff(&spg->backoffs, &hoid);
2030 if (!backoff)
2031 return false;
2032
2033 dout("%s req %p tid %llu backoff osd%d spgid %llu.%xs%d id %llu\n",
2034 __func__, req, req->r_tid, osd->o_osd, backoff->spgid.pgid.pool,
2035 backoff->spgid.pgid.seed, backoff->spgid.shard, backoff->id);
2036 return true;
2037}
2038
2039/*
2040 * Keep get_num_data_items() in sync with this function.
2041 */
2042static void setup_request_data(struct ceph_osd_request *req)
2043{
2044 struct ceph_msg *request_msg = req->r_request;
2045 struct ceph_msg *reply_msg = req->r_reply;
2046 struct ceph_osd_req_op *op;
2047
2048 if (req->r_request->num_data_items || req->r_reply->num_data_items)
2049 return;
2050
2051 WARN_ON(request_msg->data_length || reply_msg->data_length);
2052 for (op = req->r_ops; op != &req->r_ops[req->r_num_ops]; op++) {
2053 switch (op->op) {
2054 /* request */
2055 case CEPH_OSD_OP_WRITE:
2056 case CEPH_OSD_OP_WRITEFULL:
2057 WARN_ON(op->indata_len != op->extent.length);
2058 ceph_osdc_msg_data_add(request_msg,
2059 &op->extent.osd_data);
2060 break;
2061 case CEPH_OSD_OP_SETXATTR:
2062 case CEPH_OSD_OP_CMPXATTR:
2063 WARN_ON(op->indata_len != op->xattr.name_len +
2064 op->xattr.value_len);
2065 ceph_osdc_msg_data_add(request_msg,
2066 &op->xattr.osd_data);
2067 break;
2068 case CEPH_OSD_OP_NOTIFY_ACK:
2069 ceph_osdc_msg_data_add(request_msg,
2070 &op->notify_ack.request_data);
2071 break;
2072 case CEPH_OSD_OP_COPY_FROM2:
2073 ceph_osdc_msg_data_add(request_msg,
2074 &op->copy_from.osd_data);
2075 break;
2076
2077 /* reply */
2078 case CEPH_OSD_OP_STAT:
2079 ceph_osdc_msg_data_add(reply_msg,
2080 &op->raw_data_in);
2081 break;
2082 case CEPH_OSD_OP_READ:
2083 case CEPH_OSD_OP_SPARSE_READ:
2084 ceph_osdc_msg_data_add(reply_msg,
2085 &op->extent.osd_data);
2086 break;
2087 case CEPH_OSD_OP_LIST_WATCHERS:
2088 ceph_osdc_msg_data_add(reply_msg,
2089 &op->list_watchers.response_data);
2090 break;
2091
2092 /* both */
2093 case CEPH_OSD_OP_CALL:
2094 WARN_ON(op->indata_len != op->cls.class_len +
2095 op->cls.method_len +
2096 op->cls.indata_len);
2097 ceph_osdc_msg_data_add(request_msg,
2098 &op->cls.request_info);
2099 /* optional, can be NONE */
2100 ceph_osdc_msg_data_add(request_msg,
2101 &op->cls.request_data);
2102 /* optional, can be NONE */
2103 ceph_osdc_msg_data_add(reply_msg,
2104 &op->cls.response_data);
2105 break;
2106 case CEPH_OSD_OP_NOTIFY:
2107 ceph_osdc_msg_data_add(request_msg,
2108 &op->notify.request_data);
2109 ceph_osdc_msg_data_add(reply_msg,
2110 &op->notify.response_data);
2111 break;
2112 }
2113 }
2114}
2115
2116static void encode_pgid(void **p, const struct ceph_pg *pgid)
2117{
2118 ceph_encode_8(p, 1);
2119 ceph_encode_64(p, pgid->pool);
2120 ceph_encode_32(p, pgid->seed);
2121 ceph_encode_32(p, -1); /* preferred */
2122}
2123
2124static void encode_spgid(void **p, const struct ceph_spg *spgid)
2125{
2126 ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1);
2127 encode_pgid(p, &spgid->pgid);
2128 ceph_encode_8(p, spgid->shard);
2129}
2130
2131static void encode_oloc(void **p, void *end,
2132 const struct ceph_object_locator *oloc)
2133{
2134 ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc));
2135 ceph_encode_64(p, oloc->pool);
2136 ceph_encode_32(p, -1); /* preferred */
2137 ceph_encode_32(p, 0); /* key len */
2138 if (oloc->pool_ns)
2139 ceph_encode_string(p, end, oloc->pool_ns->str,
2140 oloc->pool_ns->len);
2141 else
2142 ceph_encode_32(p, 0);
2143}
2144
2145static void encode_request_partial(struct ceph_osd_request *req,
2146 struct ceph_msg *msg)
2147{
2148 void *p = msg->front.iov_base;
2149 void *const end = p + msg->front_alloc_len;
2150 u32 data_len = 0;
2151 int i;
2152
2153 if (req->r_flags & CEPH_OSD_FLAG_WRITE) {
2154 /* snapshots aren't writeable */
2155 WARN_ON(req->r_snapid != CEPH_NOSNAP);
2156 } else {
2157 WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec ||
2158 req->r_data_offset || req->r_snapc);
2159 }
2160
2161 setup_request_data(req);
2162
2163 encode_spgid(&p, &req->r_t.spgid); /* actual spg */
2164 ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */
2165 ceph_encode_32(&p, req->r_osdc->osdmap->epoch);
2166 ceph_encode_32(&p, req->r_flags);
2167
2168 /* reqid */
2169 ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid));
2170 memset(p, 0, sizeof(struct ceph_osd_reqid));
2171 p += sizeof(struct ceph_osd_reqid);
2172
2173 /* trace */
2174 memset(p, 0, sizeof(struct ceph_blkin_trace_info));
2175 p += sizeof(struct ceph_blkin_trace_info);
2176
2177 ceph_encode_32(&p, 0); /* client_inc, always 0 */
2178 ceph_encode_timespec64(p, &req->r_mtime);
2179 p += sizeof(struct ceph_timespec);
2180
2181 encode_oloc(&p, end, &req->r_t.target_oloc);
2182 ceph_encode_string(&p, end, req->r_t.target_oid.name,
2183 req->r_t.target_oid.name_len);
2184
2185 /* ops, can imply data */
2186 ceph_encode_16(&p, req->r_num_ops);
2187 for (i = 0; i < req->r_num_ops; i++) {
2188 data_len += osd_req_encode_op(p, &req->r_ops[i]);
2189 p += sizeof(struct ceph_osd_op);
2190 }
2191
2192 ceph_encode_64(&p, req->r_snapid); /* snapid */
2193 if (req->r_snapc) {
2194 ceph_encode_64(&p, req->r_snapc->seq);
2195 ceph_encode_32(&p, req->r_snapc->num_snaps);
2196 for (i = 0; i < req->r_snapc->num_snaps; i++)
2197 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2198 } else {
2199 ceph_encode_64(&p, 0); /* snap_seq */
2200 ceph_encode_32(&p, 0); /* snaps len */
2201 }
2202
2203 ceph_encode_32(&p, req->r_attempts); /* retry_attempt */
2204 BUG_ON(p > end - 8); /* space for features */
2205
2206 msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */
2207 /* front_len is finalized in encode_request_finish() */
2208 msg->front.iov_len = p - msg->front.iov_base;
2209 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2210 msg->hdr.data_len = cpu_to_le32(data_len);
2211 /*
2212 * The header "data_off" is a hint to the receiver allowing it
2213 * to align received data into its buffers such that there's no
2214 * need to re-copy it before writing it to disk (direct I/O).
2215 */
2216 msg->hdr.data_off = cpu_to_le16(req->r_data_offset);
2217
2218 dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg,
2219 req->r_t.target_oid.name, req->r_t.target_oid.name_len);
2220}
2221
2222static void encode_request_finish(struct ceph_msg *msg)
2223{
2224 void *p = msg->front.iov_base;
2225 void *const partial_end = p + msg->front.iov_len;
2226 void *const end = p + msg->front_alloc_len;
2227
2228 if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) {
2229 /* luminous OSD -- encode features and be done */
2230 p = partial_end;
2231 ceph_encode_64(&p, msg->con->peer_features);
2232 } else {
2233 struct {
2234 char spgid[CEPH_ENCODING_START_BLK_LEN +
2235 CEPH_PGID_ENCODING_LEN + 1];
2236 __le32 hash;
2237 __le32 epoch;
2238 __le32 flags;
2239 char reqid[CEPH_ENCODING_START_BLK_LEN +
2240 sizeof(struct ceph_osd_reqid)];
2241 char trace[sizeof(struct ceph_blkin_trace_info)];
2242 __le32 client_inc;
2243 struct ceph_timespec mtime;
2244 } __packed head;
2245 struct ceph_pg pgid;
2246 void *oloc, *oid, *tail;
2247 int oloc_len, oid_len, tail_len;
2248 int len;
2249
2250 /*
2251 * Pre-luminous OSD -- reencode v8 into v4 using @head
2252 * as a temporary buffer. Encode the raw PG; the rest
2253 * is just a matter of moving oloc, oid and tail blobs
2254 * around.
2255 */
2256 memcpy(&head, p, sizeof(head));
2257 p += sizeof(head);
2258
2259 oloc = p;
2260 p += CEPH_ENCODING_START_BLK_LEN;
2261 pgid.pool = ceph_decode_64(&p);
2262 p += 4 + 4; /* preferred, key len */
2263 len = ceph_decode_32(&p);
2264 p += len; /* nspace */
2265 oloc_len = p - oloc;
2266
2267 oid = p;
2268 len = ceph_decode_32(&p);
2269 p += len;
2270 oid_len = p - oid;
2271
2272 tail = p;
2273 tail_len = partial_end - p;
2274
2275 p = msg->front.iov_base;
2276 ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc));
2277 ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch));
2278 ceph_encode_copy(&p, &head.flags, sizeof(head.flags));
2279 ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime));
2280
2281 /* reassert_version */
2282 memset(p, 0, sizeof(struct ceph_eversion));
2283 p += sizeof(struct ceph_eversion);
2284
2285 BUG_ON(p >= oloc);
2286 memmove(p, oloc, oloc_len);
2287 p += oloc_len;
2288
2289 pgid.seed = le32_to_cpu(head.hash);
2290 encode_pgid(&p, &pgid); /* raw pg */
2291
2292 BUG_ON(p >= oid);
2293 memmove(p, oid, oid_len);
2294 p += oid_len;
2295
2296 /* tail -- ops, snapid, snapc, retry_attempt */
2297 BUG_ON(p >= tail);
2298 memmove(p, tail, tail_len);
2299 p += tail_len;
2300
2301 msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */
2302 }
2303
2304 BUG_ON(p > end);
2305 msg->front.iov_len = p - msg->front.iov_base;
2306 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2307
2308 dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg,
2309 le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len),
2310 le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len),
2311 le16_to_cpu(msg->hdr.version));
2312}
2313
2314/*
2315 * @req has to be assigned a tid and registered.
2316 */
2317static void send_request(struct ceph_osd_request *req)
2318{
2319 struct ceph_osd *osd = req->r_osd;
2320
2321 verify_osd_locked(osd);
2322 WARN_ON(osd->o_osd != req->r_t.osd);
2323
2324 /* backoff? */
2325 if (should_plug_request(req))
2326 return;
2327
2328 /*
2329 * We may have a previously queued request message hanging
2330 * around. Cancel it to avoid corrupting the msgr.
2331 */
2332 if (req->r_sent)
2333 ceph_msg_revoke(req->r_request);
2334
2335 req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR;
2336 if (req->r_attempts)
2337 req->r_flags |= CEPH_OSD_FLAG_RETRY;
2338 else
2339 WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY);
2340
2341 encode_request_partial(req, req->r_request);
2342
2343 dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n",
2344 __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed,
2345 req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed,
2346 req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags,
2347 req->r_attempts);
2348
2349 req->r_t.paused = false;
2350 req->r_stamp = jiffies;
2351 req->r_attempts++;
2352
2353 req->r_sent = osd->o_incarnation;
2354 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
2355 ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request));
2356}
2357
2358static void maybe_request_map(struct ceph_osd_client *osdc)
2359{
2360 bool continuous = false;
2361
2362 verify_osdc_locked(osdc);
2363 WARN_ON(!osdc->osdmap->epoch);
2364
2365 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2366 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) ||
2367 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2368 dout("%s osdc %p continuous\n", __func__, osdc);
2369 continuous = true;
2370 } else {
2371 dout("%s osdc %p onetime\n", __func__, osdc);
2372 }
2373
2374 if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
2375 osdc->osdmap->epoch + 1, continuous))
2376 ceph_monc_renew_subs(&osdc->client->monc);
2377}
2378
2379static void complete_request(struct ceph_osd_request *req, int err);
2380static void send_map_check(struct ceph_osd_request *req);
2381
2382static void __submit_request(struct ceph_osd_request *req, bool wrlocked)
2383{
2384 struct ceph_osd_client *osdc = req->r_osdc;
2385 struct ceph_osd *osd;
2386 enum calc_target_result ct_res;
2387 int err = 0;
2388 bool need_send = false;
2389 bool promoted = false;
2390
2391 WARN_ON(req->r_tid);
2392 dout("%s req %p wrlocked %d\n", __func__, req, wrlocked);
2393
2394again:
2395 ct_res = calc_target(osdc, &req->r_t, false);
2396 if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked)
2397 goto promote;
2398
2399 osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked);
2400 if (IS_ERR(osd)) {
2401 WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked);
2402 goto promote;
2403 }
2404
2405 if (osdc->abort_err) {
2406 dout("req %p abort_err %d\n", req, osdc->abort_err);
2407 err = osdc->abort_err;
2408 } else if (osdc->osdmap->epoch < osdc->epoch_barrier) {
2409 dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch,
2410 osdc->epoch_barrier);
2411 req->r_t.paused = true;
2412 maybe_request_map(osdc);
2413 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2414 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) {
2415 dout("req %p pausewr\n", req);
2416 req->r_t.paused = true;
2417 maybe_request_map(osdc);
2418 } else if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
2419 ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
2420 dout("req %p pauserd\n", req);
2421 req->r_t.paused = true;
2422 maybe_request_map(osdc);
2423 } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2424 !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY |
2425 CEPH_OSD_FLAG_FULL_FORCE)) &&
2426 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2427 pool_full(osdc, req->r_t.base_oloc.pool))) {
2428 dout("req %p full/pool_full\n", req);
2429 if (ceph_test_opt(osdc->client, ABORT_ON_FULL)) {
2430 err = -ENOSPC;
2431 } else {
2432 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL))
2433 pr_warn_ratelimited("cluster is full (osdmap FULL)\n");
2434 else
2435 pr_warn_ratelimited("pool %lld is full or reached quota\n",
2436 req->r_t.base_oloc.pool);
2437 req->r_t.paused = true;
2438 maybe_request_map(osdc);
2439 }
2440 } else if (!osd_homeless(osd)) {
2441 need_send = true;
2442 } else {
2443 maybe_request_map(osdc);
2444 }
2445
2446 mutex_lock(&osd->lock);
2447 /*
2448 * Assign the tid atomically with send_request() to protect
2449 * multiple writes to the same object from racing with each
2450 * other, resulting in out of order ops on the OSDs.
2451 */
2452 req->r_tid = atomic64_inc_return(&osdc->last_tid);
2453 link_request(osd, req);
2454 if (need_send)
2455 send_request(req);
2456 else if (err)
2457 complete_request(req, err);
2458 mutex_unlock(&osd->lock);
2459
2460 if (!err && ct_res == CALC_TARGET_POOL_DNE)
2461 send_map_check(req);
2462
2463 if (promoted)
2464 downgrade_write(&osdc->lock);
2465 return;
2466
2467promote:
2468 up_read(&osdc->lock);
2469 down_write(&osdc->lock);
2470 wrlocked = true;
2471 promoted = true;
2472 goto again;
2473}
2474
2475static void account_request(struct ceph_osd_request *req)
2476{
2477 WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK));
2478 WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE)));
2479
2480 req->r_flags |= CEPH_OSD_FLAG_ONDISK;
2481 atomic_inc(&req->r_osdc->num_requests);
2482
2483 req->r_start_stamp = jiffies;
2484 req->r_start_latency = ktime_get();
2485}
2486
2487static void submit_request(struct ceph_osd_request *req, bool wrlocked)
2488{
2489 ceph_osdc_get_request(req);
2490 account_request(req);
2491 __submit_request(req, wrlocked);
2492}
2493
2494static void finish_request(struct ceph_osd_request *req)
2495{
2496 struct ceph_osd_client *osdc = req->r_osdc;
2497
2498 WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid));
2499 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2500
2501 req->r_end_latency = ktime_get();
2502
2503 if (req->r_osd) {
2504 ceph_init_sparse_read(&req->r_osd->o_sparse_read);
2505 unlink_request(req->r_osd, req);
2506 }
2507 atomic_dec(&osdc->num_requests);
2508
2509 /*
2510 * If an OSD has failed or returned and a request has been sent
2511 * twice, it's possible to get a reply and end up here while the
2512 * request message is queued for delivery. We will ignore the
2513 * reply, so not a big deal, but better to try and catch it.
2514 */
2515 ceph_msg_revoke(req->r_request);
2516 ceph_msg_revoke_incoming(req->r_reply);
2517}
2518
2519static void __complete_request(struct ceph_osd_request *req)
2520{
2521 dout("%s req %p tid %llu cb %ps result %d\n", __func__, req,
2522 req->r_tid, req->r_callback, req->r_result);
2523
2524 if (req->r_callback)
2525 req->r_callback(req);
2526 complete_all(&req->r_completion);
2527 ceph_osdc_put_request(req);
2528}
2529
2530static void complete_request_workfn(struct work_struct *work)
2531{
2532 struct ceph_osd_request *req =
2533 container_of(work, struct ceph_osd_request, r_complete_work);
2534
2535 __complete_request(req);
2536}
2537
2538/*
2539 * This is open-coded in handle_reply().
2540 */
2541static void complete_request(struct ceph_osd_request *req, int err)
2542{
2543 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2544
2545 req->r_result = err;
2546 finish_request(req);
2547
2548 INIT_WORK(&req->r_complete_work, complete_request_workfn);
2549 queue_work(req->r_osdc->completion_wq, &req->r_complete_work);
2550}
2551
2552static void cancel_map_check(struct ceph_osd_request *req)
2553{
2554 struct ceph_osd_client *osdc = req->r_osdc;
2555 struct ceph_osd_request *lookup_req;
2556
2557 verify_osdc_wrlocked(osdc);
2558
2559 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2560 if (!lookup_req)
2561 return;
2562
2563 WARN_ON(lookup_req != req);
2564 erase_request_mc(&osdc->map_checks, req);
2565 ceph_osdc_put_request(req);
2566}
2567
2568static void cancel_request(struct ceph_osd_request *req)
2569{
2570 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
2571
2572 cancel_map_check(req);
2573 finish_request(req);
2574 complete_all(&req->r_completion);
2575 ceph_osdc_put_request(req);
2576}
2577
2578static void abort_request(struct ceph_osd_request *req, int err)
2579{
2580 dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
2581
2582 cancel_map_check(req);
2583 complete_request(req, err);
2584}
2585
2586static int abort_fn(struct ceph_osd_request *req, void *arg)
2587{
2588 int err = *(int *)arg;
2589
2590 abort_request(req, err);
2591 return 0; /* continue iteration */
2592}
2593
2594/*
2595 * Abort all in-flight requests with @err and arrange for all future
2596 * requests to be failed immediately.
2597 */
2598void ceph_osdc_abort_requests(struct ceph_osd_client *osdc, int err)
2599{
2600 dout("%s osdc %p err %d\n", __func__, osdc, err);
2601 down_write(&osdc->lock);
2602 for_each_request(osdc, abort_fn, &err);
2603 osdc->abort_err = err;
2604 up_write(&osdc->lock);
2605}
2606EXPORT_SYMBOL(ceph_osdc_abort_requests);
2607
2608void ceph_osdc_clear_abort_err(struct ceph_osd_client *osdc)
2609{
2610 down_write(&osdc->lock);
2611 osdc->abort_err = 0;
2612 up_write(&osdc->lock);
2613}
2614EXPORT_SYMBOL(ceph_osdc_clear_abort_err);
2615
2616static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2617{
2618 if (likely(eb > osdc->epoch_barrier)) {
2619 dout("updating epoch_barrier from %u to %u\n",
2620 osdc->epoch_barrier, eb);
2621 osdc->epoch_barrier = eb;
2622 /* Request map if we're not to the barrier yet */
2623 if (eb > osdc->osdmap->epoch)
2624 maybe_request_map(osdc);
2625 }
2626}
2627
2628void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb)
2629{
2630 down_read(&osdc->lock);
2631 if (unlikely(eb > osdc->epoch_barrier)) {
2632 up_read(&osdc->lock);
2633 down_write(&osdc->lock);
2634 update_epoch_barrier(osdc, eb);
2635 up_write(&osdc->lock);
2636 } else {
2637 up_read(&osdc->lock);
2638 }
2639}
2640EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier);
2641
2642/*
2643 * We can end up releasing caps as a result of abort_request().
2644 * In that case, we probably want to ensure that the cap release message
2645 * has an updated epoch barrier in it, so set the epoch barrier prior to
2646 * aborting the first request.
2647 */
2648static int abort_on_full_fn(struct ceph_osd_request *req, void *arg)
2649{
2650 struct ceph_osd_client *osdc = req->r_osdc;
2651 bool *victims = arg;
2652
2653 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
2654 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
2655 pool_full(osdc, req->r_t.base_oloc.pool))) {
2656 if (!*victims) {
2657 update_epoch_barrier(osdc, osdc->osdmap->epoch);
2658 *victims = true;
2659 }
2660 abort_request(req, -ENOSPC);
2661 }
2662
2663 return 0; /* continue iteration */
2664}
2665
2666/*
2667 * Drop all pending requests that are stalled waiting on a full condition to
2668 * clear, and complete them with ENOSPC as the return code. Set the
2669 * osdc->epoch_barrier to the latest map epoch that we've seen if any were
2670 * cancelled.
2671 */
2672static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc)
2673{
2674 bool victims = false;
2675
2676 if (ceph_test_opt(osdc->client, ABORT_ON_FULL) &&
2677 (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || have_pool_full(osdc)))
2678 for_each_request(osdc, abort_on_full_fn, &victims);
2679}
2680
2681static void check_pool_dne(struct ceph_osd_request *req)
2682{
2683 struct ceph_osd_client *osdc = req->r_osdc;
2684 struct ceph_osdmap *map = osdc->osdmap;
2685
2686 verify_osdc_wrlocked(osdc);
2687 WARN_ON(!map->epoch);
2688
2689 if (req->r_attempts) {
2690 /*
2691 * We sent a request earlier, which means that
2692 * previously the pool existed, and now it does not
2693 * (i.e., it was deleted).
2694 */
2695 req->r_map_dne_bound = map->epoch;
2696 dout("%s req %p tid %llu pool disappeared\n", __func__, req,
2697 req->r_tid);
2698 } else {
2699 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__,
2700 req, req->r_tid, req->r_map_dne_bound, map->epoch);
2701 }
2702
2703 if (req->r_map_dne_bound) {
2704 if (map->epoch >= req->r_map_dne_bound) {
2705 /* we had a new enough map */
2706 pr_info_ratelimited("tid %llu pool does not exist\n",
2707 req->r_tid);
2708 complete_request(req, -ENOENT);
2709 }
2710 } else {
2711 send_map_check(req);
2712 }
2713}
2714
2715static void map_check_cb(struct ceph_mon_generic_request *greq)
2716{
2717 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
2718 struct ceph_osd_request *req;
2719 u64 tid = greq->private_data;
2720
2721 WARN_ON(greq->result || !greq->u.newest);
2722
2723 down_write(&osdc->lock);
2724 req = lookup_request_mc(&osdc->map_checks, tid);
2725 if (!req) {
2726 dout("%s tid %llu dne\n", __func__, tid);
2727 goto out_unlock;
2728 }
2729
2730 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__,
2731 req, req->r_tid, req->r_map_dne_bound, greq->u.newest);
2732 if (!req->r_map_dne_bound)
2733 req->r_map_dne_bound = greq->u.newest;
2734 erase_request_mc(&osdc->map_checks, req);
2735 check_pool_dne(req);
2736
2737 ceph_osdc_put_request(req);
2738out_unlock:
2739 up_write(&osdc->lock);
2740}
2741
2742static void send_map_check(struct ceph_osd_request *req)
2743{
2744 struct ceph_osd_client *osdc = req->r_osdc;
2745 struct ceph_osd_request *lookup_req;
2746 int ret;
2747
2748 verify_osdc_wrlocked(osdc);
2749
2750 lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid);
2751 if (lookup_req) {
2752 WARN_ON(lookup_req != req);
2753 return;
2754 }
2755
2756 ceph_osdc_get_request(req);
2757 insert_request_mc(&osdc->map_checks, req);
2758 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
2759 map_check_cb, req->r_tid);
2760 WARN_ON(ret);
2761}
2762
2763/*
2764 * lingering requests, watch/notify v2 infrastructure
2765 */
2766static void linger_release(struct kref *kref)
2767{
2768 struct ceph_osd_linger_request *lreq =
2769 container_of(kref, struct ceph_osd_linger_request, kref);
2770
2771 dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq,
2772 lreq->reg_req, lreq->ping_req);
2773 WARN_ON(!RB_EMPTY_NODE(&lreq->node));
2774 WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node));
2775 WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node));
2776 WARN_ON(!list_empty(&lreq->scan_item));
2777 WARN_ON(!list_empty(&lreq->pending_lworks));
2778 WARN_ON(lreq->osd);
2779
2780 if (lreq->request_pl)
2781 ceph_pagelist_release(lreq->request_pl);
2782 if (lreq->notify_id_pages)
2783 ceph_release_page_vector(lreq->notify_id_pages, 1);
2784
2785 ceph_osdc_put_request(lreq->reg_req);
2786 ceph_osdc_put_request(lreq->ping_req);
2787 target_destroy(&lreq->t);
2788 kfree(lreq);
2789}
2790
2791static void linger_put(struct ceph_osd_linger_request *lreq)
2792{
2793 if (lreq)
2794 kref_put(&lreq->kref, linger_release);
2795}
2796
2797static struct ceph_osd_linger_request *
2798linger_get(struct ceph_osd_linger_request *lreq)
2799{
2800 kref_get(&lreq->kref);
2801 return lreq;
2802}
2803
2804static struct ceph_osd_linger_request *
2805linger_alloc(struct ceph_osd_client *osdc)
2806{
2807 struct ceph_osd_linger_request *lreq;
2808
2809 lreq = kzalloc_obj(*lreq, GFP_NOIO);
2810 if (!lreq)
2811 return NULL;
2812
2813 kref_init(&lreq->kref);
2814 mutex_init(&lreq->lock);
2815 RB_CLEAR_NODE(&lreq->node);
2816 RB_CLEAR_NODE(&lreq->osdc_node);
2817 RB_CLEAR_NODE(&lreq->mc_node);
2818 INIT_LIST_HEAD(&lreq->scan_item);
2819 INIT_LIST_HEAD(&lreq->pending_lworks);
2820 init_completion(&lreq->reg_commit_wait);
2821 init_completion(&lreq->notify_finish_wait);
2822
2823 lreq->osdc = osdc;
2824 target_init(&lreq->t);
2825
2826 dout("%s lreq %p\n", __func__, lreq);
2827 return lreq;
2828}
2829
2830DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node)
2831DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node)
2832DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node)
2833
2834/*
2835 * Create linger request <-> OSD session relation.
2836 *
2837 * @lreq has to be registered, @osd may be homeless.
2838 */
2839static void link_linger(struct ceph_osd *osd,
2840 struct ceph_osd_linger_request *lreq)
2841{
2842 verify_osd_locked(osd);
2843 WARN_ON(!lreq->linger_id || lreq->osd);
2844 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2845 osd->o_osd, lreq, lreq->linger_id);
2846
2847 if (!osd_homeless(osd))
2848 __remove_osd_from_lru(osd);
2849 else
2850 atomic_inc(&osd->o_osdc->num_homeless);
2851
2852 get_osd(osd);
2853 insert_linger(&osd->o_linger_requests, lreq);
2854 lreq->osd = osd;
2855}
2856
2857static void unlink_linger(struct ceph_osd *osd,
2858 struct ceph_osd_linger_request *lreq)
2859{
2860 verify_osd_locked(osd);
2861 WARN_ON(lreq->osd != osd);
2862 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd,
2863 osd->o_osd, lreq, lreq->linger_id);
2864
2865 lreq->osd = NULL;
2866 erase_linger(&osd->o_linger_requests, lreq);
2867 put_osd(osd);
2868
2869 if (!osd_homeless(osd))
2870 maybe_move_osd_to_lru(osd);
2871 else
2872 atomic_dec(&osd->o_osdc->num_homeless);
2873}
2874
2875static bool __linger_registered(struct ceph_osd_linger_request *lreq)
2876{
2877 verify_osdc_locked(lreq->osdc);
2878
2879 return !RB_EMPTY_NODE(&lreq->osdc_node);
2880}
2881
2882static bool linger_registered(struct ceph_osd_linger_request *lreq)
2883{
2884 struct ceph_osd_client *osdc = lreq->osdc;
2885 bool registered;
2886
2887 down_read(&osdc->lock);
2888 registered = __linger_registered(lreq);
2889 up_read(&osdc->lock);
2890
2891 return registered;
2892}
2893
2894static void linger_register(struct ceph_osd_linger_request *lreq)
2895{
2896 struct ceph_osd_client *osdc = lreq->osdc;
2897
2898 verify_osdc_wrlocked(osdc);
2899 WARN_ON(lreq->linger_id);
2900
2901 linger_get(lreq);
2902 lreq->linger_id = ++osdc->last_linger_id;
2903 insert_linger_osdc(&osdc->linger_requests, lreq);
2904}
2905
2906static void linger_unregister(struct ceph_osd_linger_request *lreq)
2907{
2908 struct ceph_osd_client *osdc = lreq->osdc;
2909
2910 verify_osdc_wrlocked(osdc);
2911
2912 erase_linger_osdc(&osdc->linger_requests, lreq);
2913 linger_put(lreq);
2914}
2915
2916static void cancel_linger_request(struct ceph_osd_request *req)
2917{
2918 struct ceph_osd_linger_request *lreq = req->r_priv;
2919
2920 WARN_ON(!req->r_linger);
2921 cancel_request(req);
2922 linger_put(lreq);
2923}
2924
2925struct linger_work {
2926 struct work_struct work;
2927 struct ceph_osd_linger_request *lreq;
2928 struct list_head pending_item;
2929 unsigned long queued_stamp;
2930
2931 union {
2932 struct {
2933 u64 notify_id;
2934 u64 notifier_id;
2935 void *payload; /* points into @msg front */
2936 size_t payload_len;
2937
2938 struct ceph_msg *msg; /* for ceph_msg_put() */
2939 } notify;
2940 struct {
2941 int err;
2942 } error;
2943 };
2944};
2945
2946static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq,
2947 work_func_t workfn)
2948{
2949 struct linger_work *lwork;
2950
2951 lwork = kzalloc_obj(*lwork, GFP_NOIO);
2952 if (!lwork)
2953 return NULL;
2954
2955 INIT_WORK(&lwork->work, workfn);
2956 INIT_LIST_HEAD(&lwork->pending_item);
2957 lwork->lreq = linger_get(lreq);
2958
2959 return lwork;
2960}
2961
2962static void lwork_free(struct linger_work *lwork)
2963{
2964 struct ceph_osd_linger_request *lreq = lwork->lreq;
2965
2966 mutex_lock(&lreq->lock);
2967 list_del(&lwork->pending_item);
2968 mutex_unlock(&lreq->lock);
2969
2970 linger_put(lreq);
2971 kfree(lwork);
2972}
2973
2974static void lwork_queue(struct linger_work *lwork)
2975{
2976 struct ceph_osd_linger_request *lreq = lwork->lreq;
2977 struct ceph_osd_client *osdc = lreq->osdc;
2978
2979 verify_lreq_locked(lreq);
2980 WARN_ON(!list_empty(&lwork->pending_item));
2981
2982 lwork->queued_stamp = jiffies;
2983 list_add_tail(&lwork->pending_item, &lreq->pending_lworks);
2984 queue_work(osdc->notify_wq, &lwork->work);
2985}
2986
2987static void do_watch_notify(struct work_struct *w)
2988{
2989 struct linger_work *lwork = container_of(w, struct linger_work, work);
2990 struct ceph_osd_linger_request *lreq = lwork->lreq;
2991
2992 if (!linger_registered(lreq)) {
2993 dout("%s lreq %p not registered\n", __func__, lreq);
2994 goto out;
2995 }
2996
2997 WARN_ON(!lreq->is_watch);
2998 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2999 __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id,
3000 lwork->notify.payload_len);
3001 lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id,
3002 lwork->notify.notifier_id, lwork->notify.payload,
3003 lwork->notify.payload_len);
3004
3005out:
3006 ceph_msg_put(lwork->notify.msg);
3007 lwork_free(lwork);
3008}
3009
3010static void do_watch_error(struct work_struct *w)
3011{
3012 struct linger_work *lwork = container_of(w, struct linger_work, work);
3013 struct ceph_osd_linger_request *lreq = lwork->lreq;
3014
3015 if (!linger_registered(lreq)) {
3016 dout("%s lreq %p not registered\n", __func__, lreq);
3017 goto out;
3018 }
3019
3020 dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err);
3021 lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err);
3022
3023out:
3024 lwork_free(lwork);
3025}
3026
3027static void queue_watch_error(struct ceph_osd_linger_request *lreq)
3028{
3029 struct linger_work *lwork;
3030
3031 lwork = lwork_alloc(lreq, do_watch_error);
3032 if (!lwork) {
3033 pr_err("failed to allocate error-lwork\n");
3034 return;
3035 }
3036
3037 lwork->error.err = lreq->last_error;
3038 lwork_queue(lwork);
3039}
3040
3041static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq,
3042 int result)
3043{
3044 if (!completion_done(&lreq->reg_commit_wait)) {
3045 lreq->reg_commit_error = (result <= 0 ? result : 0);
3046 complete_all(&lreq->reg_commit_wait);
3047 }
3048}
3049
3050static void linger_commit_cb(struct ceph_osd_request *req)
3051{
3052 struct ceph_osd_linger_request *lreq = req->r_priv;
3053
3054 mutex_lock(&lreq->lock);
3055 if (req != lreq->reg_req) {
3056 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3057 __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3058 goto out;
3059 }
3060
3061 dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq,
3062 lreq->linger_id, req->r_result);
3063 linger_reg_commit_complete(lreq, req->r_result);
3064 lreq->committed = true;
3065
3066 if (!lreq->is_watch) {
3067 struct ceph_osd_data *osd_data =
3068 osd_req_op_data(req, 0, notify, response_data);
3069 void *p = page_address(osd_data->pages[0]);
3070
3071 WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY ||
3072 osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
3073
3074 /* make note of the notify_id */
3075 if (req->r_ops[0].outdata_len >= sizeof(u64)) {
3076 lreq->notify_id = ceph_decode_64(&p);
3077 dout("lreq %p notify_id %llu\n", lreq,
3078 lreq->notify_id);
3079 } else {
3080 dout("lreq %p no notify_id\n", lreq);
3081 }
3082 }
3083
3084out:
3085 mutex_unlock(&lreq->lock);
3086 linger_put(lreq);
3087}
3088
3089static int normalize_watch_error(int err)
3090{
3091 /*
3092 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
3093 * notification and a failure to reconnect because we raced with
3094 * the delete appear the same to the user.
3095 */
3096 if (err == -ENOENT)
3097 err = -ENOTCONN;
3098
3099 return err;
3100}
3101
3102static void linger_reconnect_cb(struct ceph_osd_request *req)
3103{
3104 struct ceph_osd_linger_request *lreq = req->r_priv;
3105
3106 mutex_lock(&lreq->lock);
3107 if (req != lreq->reg_req) {
3108 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3109 __func__, lreq, lreq->linger_id, req, lreq->reg_req);
3110 goto out;
3111 }
3112
3113 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__,
3114 lreq, lreq->linger_id, req->r_result, lreq->last_error);
3115 if (req->r_result < 0) {
3116 if (!lreq->last_error) {
3117 lreq->last_error = normalize_watch_error(req->r_result);
3118 queue_watch_error(lreq);
3119 }
3120 }
3121
3122out:
3123 mutex_unlock(&lreq->lock);
3124 linger_put(lreq);
3125}
3126
3127static void send_linger(struct ceph_osd_linger_request *lreq)
3128{
3129 struct ceph_osd_client *osdc = lreq->osdc;
3130 struct ceph_osd_request *req;
3131 int ret;
3132
3133 verify_osdc_wrlocked(osdc);
3134 mutex_lock(&lreq->lock);
3135 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3136
3137 if (lreq->reg_req) {
3138 if (lreq->reg_req->r_osd)
3139 cancel_linger_request(lreq->reg_req);
3140 ceph_osdc_put_request(lreq->reg_req);
3141 }
3142
3143 req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3144 BUG_ON(!req);
3145
3146 target_copy(&req->r_t, &lreq->t);
3147 req->r_mtime = lreq->mtime;
3148
3149 if (lreq->is_watch && lreq->committed) {
3150 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_RECONNECT,
3151 lreq->linger_id, ++lreq->register_gen);
3152 dout("lreq %p reconnect register_gen %u\n", lreq,
3153 req->r_ops[0].watch.gen);
3154 req->r_callback = linger_reconnect_cb;
3155 } else {
3156 if (lreq->is_watch) {
3157 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_WATCH,
3158 lreq->linger_id, 0);
3159 } else {
3160 lreq->notify_id = 0;
3161
3162 refcount_inc(&lreq->request_pl->refcnt);
3163 osd_req_op_notify_init(req, 0, lreq->linger_id,
3164 lreq->request_pl);
3165 ceph_osd_data_pages_init(
3166 osd_req_op_data(req, 0, notify, response_data),
3167 lreq->notify_id_pages, PAGE_SIZE, 0, false, false);
3168 }
3169 dout("lreq %p register\n", lreq);
3170 req->r_callback = linger_commit_cb;
3171 }
3172
3173 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3174 BUG_ON(ret);
3175
3176 req->r_priv = linger_get(lreq);
3177 req->r_linger = true;
3178 lreq->reg_req = req;
3179 mutex_unlock(&lreq->lock);
3180
3181 submit_request(req, true);
3182}
3183
3184static void linger_ping_cb(struct ceph_osd_request *req)
3185{
3186 struct ceph_osd_linger_request *lreq = req->r_priv;
3187
3188 mutex_lock(&lreq->lock);
3189 if (req != lreq->ping_req) {
3190 dout("%s lreq %p linger_id %llu unknown req (%p != %p)\n",
3191 __func__, lreq, lreq->linger_id, req, lreq->ping_req);
3192 goto out;
3193 }
3194
3195 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
3196 __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent,
3197 lreq->last_error);
3198 if (lreq->register_gen == req->r_ops[0].watch.gen) {
3199 if (!req->r_result) {
3200 lreq->watch_valid_thru = lreq->ping_sent;
3201 } else if (!lreq->last_error) {
3202 lreq->last_error = normalize_watch_error(req->r_result);
3203 queue_watch_error(lreq);
3204 }
3205 } else {
3206 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq,
3207 lreq->register_gen, req->r_ops[0].watch.gen);
3208 }
3209
3210out:
3211 mutex_unlock(&lreq->lock);
3212 linger_put(lreq);
3213}
3214
3215static void send_linger_ping(struct ceph_osd_linger_request *lreq)
3216{
3217 struct ceph_osd_client *osdc = lreq->osdc;
3218 struct ceph_osd_request *req;
3219 int ret;
3220
3221 if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) {
3222 dout("%s PAUSERD\n", __func__);
3223 return;
3224 }
3225
3226 lreq->ping_sent = jiffies;
3227 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
3228 __func__, lreq, lreq->linger_id, lreq->ping_sent,
3229 lreq->register_gen);
3230
3231 if (lreq->ping_req) {
3232 if (lreq->ping_req->r_osd)
3233 cancel_linger_request(lreq->ping_req);
3234 ceph_osdc_put_request(lreq->ping_req);
3235 }
3236
3237 req = ceph_osdc_alloc_request(osdc, NULL, 1, true, GFP_NOIO);
3238 BUG_ON(!req);
3239
3240 target_copy(&req->r_t, &lreq->t);
3241 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_PING, lreq->linger_id,
3242 lreq->register_gen);
3243 req->r_callback = linger_ping_cb;
3244
3245 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
3246 BUG_ON(ret);
3247
3248 req->r_priv = linger_get(lreq);
3249 req->r_linger = true;
3250 lreq->ping_req = req;
3251
3252 ceph_osdc_get_request(req);
3253 account_request(req);
3254 req->r_tid = atomic64_inc_return(&osdc->last_tid);
3255 link_request(lreq->osd, req);
3256 send_request(req);
3257}
3258
3259static void linger_submit(struct ceph_osd_linger_request *lreq)
3260{
3261 struct ceph_osd_client *osdc = lreq->osdc;
3262 struct ceph_osd *osd;
3263
3264 down_write(&osdc->lock);
3265 linger_register(lreq);
3266
3267 calc_target(osdc, &lreq->t, false);
3268 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3269 link_linger(osd, lreq);
3270
3271 send_linger(lreq);
3272 up_write(&osdc->lock);
3273}
3274
3275static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq)
3276{
3277 struct ceph_osd_client *osdc = lreq->osdc;
3278 struct ceph_osd_linger_request *lookup_lreq;
3279
3280 verify_osdc_wrlocked(osdc);
3281
3282 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3283 lreq->linger_id);
3284 if (!lookup_lreq)
3285 return;
3286
3287 WARN_ON(lookup_lreq != lreq);
3288 erase_linger_mc(&osdc->linger_map_checks, lreq);
3289 linger_put(lreq);
3290}
3291
3292/*
3293 * @lreq has to be both registered and linked.
3294 */
3295static void __linger_cancel(struct ceph_osd_linger_request *lreq)
3296{
3297 if (lreq->ping_req && lreq->ping_req->r_osd)
3298 cancel_linger_request(lreq->ping_req);
3299 if (lreq->reg_req && lreq->reg_req->r_osd)
3300 cancel_linger_request(lreq->reg_req);
3301 cancel_linger_map_check(lreq);
3302 unlink_linger(lreq->osd, lreq);
3303 linger_unregister(lreq);
3304}
3305
3306static void linger_cancel(struct ceph_osd_linger_request *lreq)
3307{
3308 struct ceph_osd_client *osdc = lreq->osdc;
3309
3310 down_write(&osdc->lock);
3311 if (__linger_registered(lreq))
3312 __linger_cancel(lreq);
3313 up_write(&osdc->lock);
3314}
3315
3316static void send_linger_map_check(struct ceph_osd_linger_request *lreq);
3317
3318static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq)
3319{
3320 struct ceph_osd_client *osdc = lreq->osdc;
3321 struct ceph_osdmap *map = osdc->osdmap;
3322
3323 verify_osdc_wrlocked(osdc);
3324 WARN_ON(!map->epoch);
3325
3326 if (lreq->register_gen) {
3327 lreq->map_dne_bound = map->epoch;
3328 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__,
3329 lreq, lreq->linger_id);
3330 } else {
3331 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
3332 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3333 map->epoch);
3334 }
3335
3336 if (lreq->map_dne_bound) {
3337 if (map->epoch >= lreq->map_dne_bound) {
3338 /* we had a new enough map */
3339 pr_info("linger_id %llu pool does not exist\n",
3340 lreq->linger_id);
3341 linger_reg_commit_complete(lreq, -ENOENT);
3342 __linger_cancel(lreq);
3343 }
3344 } else {
3345 send_linger_map_check(lreq);
3346 }
3347}
3348
3349static void linger_map_check_cb(struct ceph_mon_generic_request *greq)
3350{
3351 struct ceph_osd_client *osdc = &greq->monc->client->osdc;
3352 struct ceph_osd_linger_request *lreq;
3353 u64 linger_id = greq->private_data;
3354
3355 WARN_ON(greq->result || !greq->u.newest);
3356
3357 down_write(&osdc->lock);
3358 lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id);
3359 if (!lreq) {
3360 dout("%s linger_id %llu dne\n", __func__, linger_id);
3361 goto out_unlock;
3362 }
3363
3364 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
3365 __func__, lreq, lreq->linger_id, lreq->map_dne_bound,
3366 greq->u.newest);
3367 if (!lreq->map_dne_bound)
3368 lreq->map_dne_bound = greq->u.newest;
3369 erase_linger_mc(&osdc->linger_map_checks, lreq);
3370 check_linger_pool_dne(lreq);
3371
3372 linger_put(lreq);
3373out_unlock:
3374 up_write(&osdc->lock);
3375}
3376
3377static void send_linger_map_check(struct ceph_osd_linger_request *lreq)
3378{
3379 struct ceph_osd_client *osdc = lreq->osdc;
3380 struct ceph_osd_linger_request *lookup_lreq;
3381 int ret;
3382
3383 verify_osdc_wrlocked(osdc);
3384
3385 lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks,
3386 lreq->linger_id);
3387 if (lookup_lreq) {
3388 WARN_ON(lookup_lreq != lreq);
3389 return;
3390 }
3391
3392 linger_get(lreq);
3393 insert_linger_mc(&osdc->linger_map_checks, lreq);
3394 ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap",
3395 linger_map_check_cb, lreq->linger_id);
3396 WARN_ON(ret);
3397}
3398
3399static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq)
3400{
3401 int ret;
3402
3403 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3404 ret = wait_for_completion_killable(&lreq->reg_commit_wait);
3405 return ret ?: lreq->reg_commit_error;
3406}
3407
3408static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq,
3409 unsigned long timeout)
3410{
3411 long left;
3412
3413 dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id);
3414 left = wait_for_completion_killable_timeout(&lreq->notify_finish_wait,
3415 ceph_timeout_jiffies(timeout));
3416 if (left <= 0)
3417 left = left ?: -ETIMEDOUT;
3418 else
3419 left = lreq->notify_finish_error; /* completed */
3420
3421 return left;
3422}
3423
3424/*
3425 * Timeout callback, called every N seconds. When 1 or more OSD
3426 * requests has been active for more than N seconds, we send a keepalive
3427 * (tag + timestamp) to its OSD to ensure any communications channel
3428 * reset is detected.
3429 */
3430static void handle_timeout(struct work_struct *work)
3431{
3432 struct ceph_osd_client *osdc =
3433 container_of(work, struct ceph_osd_client, timeout_work.work);
3434 struct ceph_options *opts = osdc->client->options;
3435 unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
3436 unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
3437 LIST_HEAD(slow_osds);
3438 struct rb_node *n, *p;
3439
3440 dout("%s osdc %p\n", __func__, osdc);
3441 down_write(&osdc->lock);
3442
3443 /*
3444 * ping osds that are a bit slow. this ensures that if there
3445 * is a break in the TCP connection we will notice, and reopen
3446 * a connection with that osd (from the fault callback).
3447 */
3448 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
3449 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
3450 bool found = false;
3451
3452 for (p = rb_first(&osd->o_requests); p; ) {
3453 struct ceph_osd_request *req =
3454 rb_entry(p, struct ceph_osd_request, r_node);
3455
3456 p = rb_next(p); /* abort_request() */
3457
3458 if (time_before(req->r_stamp, cutoff)) {
3459 dout(" req %p tid %llu on osd%d is laggy\n",
3460 req, req->r_tid, osd->o_osd);
3461 found = true;
3462 }
3463 if (opts->osd_request_timeout &&
3464 time_before(req->r_start_stamp, expiry_cutoff)) {
3465 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3466 req->r_tid, osd->o_osd);
3467 abort_request(req, -ETIMEDOUT);
3468 }
3469 }
3470 for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
3471 struct ceph_osd_linger_request *lreq =
3472 rb_entry(p, struct ceph_osd_linger_request, node);
3473
3474 dout(" lreq %p linger_id %llu is served by osd%d\n",
3475 lreq, lreq->linger_id, osd->o_osd);
3476 found = true;
3477
3478 mutex_lock(&lreq->lock);
3479 if (lreq->is_watch && lreq->committed && !lreq->last_error)
3480 send_linger_ping(lreq);
3481 mutex_unlock(&lreq->lock);
3482 }
3483
3484 if (found)
3485 list_move_tail(&osd->o_keepalive_item, &slow_osds);
3486 }
3487
3488 if (opts->osd_request_timeout) {
3489 for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
3490 struct ceph_osd_request *req =
3491 rb_entry(p, struct ceph_osd_request, r_node);
3492
3493 p = rb_next(p); /* abort_request() */
3494
3495 if (time_before(req->r_start_stamp, expiry_cutoff)) {
3496 pr_err_ratelimited("tid %llu on osd%d timeout\n",
3497 req->r_tid, osdc->homeless_osd.o_osd);
3498 abort_request(req, -ETIMEDOUT);
3499 }
3500 }
3501 }
3502
3503 if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
3504 maybe_request_map(osdc);
3505
3506 while (!list_empty(&slow_osds)) {
3507 struct ceph_osd *osd = list_first_entry(&slow_osds,
3508 struct ceph_osd,
3509 o_keepalive_item);
3510 list_del_init(&osd->o_keepalive_item);
3511 ceph_con_keepalive(&osd->o_con);
3512 }
3513
3514 up_write(&osdc->lock);
3515 schedule_delayed_work(&osdc->timeout_work,
3516 osdc->client->options->osd_keepalive_timeout);
3517}
3518
3519static void handle_osds_timeout(struct work_struct *work)
3520{
3521 struct ceph_osd_client *osdc =
3522 container_of(work, struct ceph_osd_client,
3523 osds_timeout_work.work);
3524 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
3525 struct ceph_osd *osd, *nosd;
3526
3527 dout("%s osdc %p\n", __func__, osdc);
3528 down_write(&osdc->lock);
3529 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
3530 if (time_before(jiffies, osd->lru_ttl))
3531 break;
3532
3533 WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests));
3534 WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests));
3535 close_osd(osd);
3536 }
3537
3538 up_write(&osdc->lock);
3539 schedule_delayed_work(&osdc->osds_timeout_work,
3540 round_jiffies_relative(delay));
3541}
3542
3543static int ceph_oloc_decode(void **p, void *end,
3544 struct ceph_object_locator *oloc)
3545{
3546 u8 struct_v, struct_cv;
3547 u32 len;
3548 void *struct_end;
3549 int ret = 0;
3550
3551 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3552 struct_v = ceph_decode_8(p);
3553 struct_cv = ceph_decode_8(p);
3554 if (struct_v < 3) {
3555 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
3556 struct_v, struct_cv);
3557 goto e_inval;
3558 }
3559 if (struct_cv > 6) {
3560 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
3561 struct_v, struct_cv);
3562 goto e_inval;
3563 }
3564 len = ceph_decode_32(p);
3565 ceph_decode_need(p, end, len, e_inval);
3566 struct_end = *p + len;
3567
3568 oloc->pool = ceph_decode_64(p);
3569 *p += 4; /* skip preferred */
3570
3571 len = ceph_decode_32(p);
3572 if (len > 0) {
3573 pr_warn("ceph_object_locator::key is set\n");
3574 goto e_inval;
3575 }
3576
3577 if (struct_v >= 5) {
3578 bool changed = false;
3579
3580 len = ceph_decode_32(p);
3581 if (len > 0) {
3582 ceph_decode_need(p, end, len, e_inval);
3583 if (!oloc->pool_ns ||
3584 ceph_compare_string(oloc->pool_ns, *p, len))
3585 changed = true;
3586 *p += len;
3587 } else {
3588 if (oloc->pool_ns)
3589 changed = true;
3590 }
3591 if (changed) {
3592 /* redirect changes namespace */
3593 pr_warn("ceph_object_locator::nspace is changed\n");
3594 goto e_inval;
3595 }
3596 }
3597
3598 if (struct_v >= 6) {
3599 s64 hash = ceph_decode_64(p);
3600 if (hash != -1) {
3601 pr_warn("ceph_object_locator::hash is set\n");
3602 goto e_inval;
3603 }
3604 }
3605
3606 /* skip the rest */
3607 *p = struct_end;
3608out:
3609 return ret;
3610
3611e_inval:
3612 ret = -EINVAL;
3613 goto out;
3614}
3615
3616static int ceph_redirect_decode(void **p, void *end,
3617 struct ceph_request_redirect *redir)
3618{
3619 u8 struct_v, struct_cv;
3620 u32 len;
3621 void *struct_end;
3622 int ret;
3623
3624 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
3625 struct_v = ceph_decode_8(p);
3626 struct_cv = ceph_decode_8(p);
3627 if (struct_cv > 1) {
3628 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
3629 struct_v, struct_cv);
3630 goto e_inval;
3631 }
3632 len = ceph_decode_32(p);
3633 ceph_decode_need(p, end, len, e_inval);
3634 struct_end = *p + len;
3635
3636 ret = ceph_oloc_decode(p, end, &redir->oloc);
3637 if (ret)
3638 goto out;
3639
3640 len = ceph_decode_32(p);
3641 if (len > 0) {
3642 pr_warn("ceph_request_redirect::object_name is set\n");
3643 goto e_inval;
3644 }
3645
3646 /* skip the rest */
3647 *p = struct_end;
3648out:
3649 return ret;
3650
3651e_inval:
3652 ret = -EINVAL;
3653 goto out;
3654}
3655
3656struct MOSDOpReply {
3657 struct ceph_pg pgid;
3658 u64 flags;
3659 int result;
3660 u32 epoch;
3661 int num_ops;
3662 u32 outdata_len[CEPH_OSD_MAX_OPS];
3663 s32 rval[CEPH_OSD_MAX_OPS];
3664 int retry_attempt;
3665 struct ceph_eversion replay_version;
3666 u64 user_version;
3667 struct ceph_request_redirect redirect;
3668};
3669
3670static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m)
3671{
3672 void *p = msg->front.iov_base;
3673 void *const end = p + msg->front.iov_len;
3674 u16 version = le16_to_cpu(msg->hdr.version);
3675 struct ceph_eversion bad_replay_version;
3676 u8 decode_redir;
3677 u32 len;
3678 int ret;
3679 int i;
3680
3681 ceph_decode_32_safe(&p, end, len, e_inval);
3682 ceph_decode_need(&p, end, len, e_inval);
3683 p += len; /* skip oid */
3684
3685 ret = ceph_decode_pgid(&p, end, &m->pgid);
3686 if (ret)
3687 return ret;
3688
3689 ceph_decode_64_safe(&p, end, m->flags, e_inval);
3690 ceph_decode_32_safe(&p, end, m->result, e_inval);
3691 ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval);
3692 memcpy(&bad_replay_version, p, sizeof(bad_replay_version));
3693 p += sizeof(bad_replay_version);
3694 ceph_decode_32_safe(&p, end, m->epoch, e_inval);
3695
3696 ceph_decode_32_safe(&p, end, m->num_ops, e_inval);
3697 if (m->num_ops > ARRAY_SIZE(m->outdata_len))
3698 goto e_inval;
3699
3700 ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op),
3701 e_inval);
3702 for (i = 0; i < m->num_ops; i++) {
3703 struct ceph_osd_op *op = p;
3704
3705 m->outdata_len[i] = le32_to_cpu(op->payload_len);
3706 p += sizeof(*op);
3707 }
3708
3709 ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval);
3710 for (i = 0; i < m->num_ops; i++)
3711 ceph_decode_32_safe(&p, end, m->rval[i], e_inval);
3712
3713 if (version >= 5) {
3714 ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval);
3715 memcpy(&m->replay_version, p, sizeof(m->replay_version));
3716 p += sizeof(m->replay_version);
3717 ceph_decode_64_safe(&p, end, m->user_version, e_inval);
3718 } else {
3719 m->replay_version = bad_replay_version; /* struct */
3720 m->user_version = le64_to_cpu(m->replay_version.version);
3721 }
3722
3723 if (version >= 6) {
3724 if (version >= 7)
3725 ceph_decode_8_safe(&p, end, decode_redir, e_inval);
3726 else
3727 decode_redir = 1;
3728 } else {
3729 decode_redir = 0;
3730 }
3731
3732 if (decode_redir) {
3733 ret = ceph_redirect_decode(&p, end, &m->redirect);
3734 if (ret)
3735 return ret;
3736 } else {
3737 ceph_oloc_init(&m->redirect.oloc);
3738 }
3739
3740 return 0;
3741
3742e_inval:
3743 return -EINVAL;
3744}
3745
3746/*
3747 * Handle MOSDOpReply. Set ->r_result and call the callback if it is
3748 * specified.
3749 */
3750static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
3751{
3752 struct ceph_osd_client *osdc = osd->o_osdc;
3753 struct ceph_osd_request *req;
3754 struct MOSDOpReply m;
3755 u64 tid = le64_to_cpu(msg->hdr.tid);
3756 u32 data_len = 0;
3757 int ret;
3758 int i;
3759
3760 dout("%s msg %p tid %llu\n", __func__, msg, tid);
3761
3762 down_read(&osdc->lock);
3763 if (!osd_registered(osd)) {
3764 dout("%s osd%d unknown\n", __func__, osd->o_osd);
3765 goto out_unlock_osdc;
3766 }
3767 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
3768
3769 mutex_lock(&osd->lock);
3770 req = lookup_request(&osd->o_requests, tid);
3771 if (!req) {
3772 dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid);
3773 goto out_unlock_session;
3774 }
3775
3776 m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns;
3777 ret = decode_MOSDOpReply(msg, &m);
3778 m.redirect.oloc.pool_ns = NULL;
3779 if (ret) {
3780 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
3781 req->r_tid, ret);
3782 ceph_msg_dump(msg);
3783 goto fail_request;
3784 }
3785 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
3786 __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed,
3787 m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch),
3788 le64_to_cpu(m.replay_version.version), m.user_version);
3789
3790 if (m.retry_attempt >= 0) {
3791 if (m.retry_attempt != req->r_attempts - 1) {
3792 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
3793 req, req->r_tid, m.retry_attempt,
3794 req->r_attempts - 1);
3795 goto out_unlock_session;
3796 }
3797 } else {
3798 WARN_ON(1); /* MOSDOpReply v4 is assumed */
3799 }
3800
3801 if (!ceph_oloc_empty(&m.redirect.oloc)) {
3802 dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid,
3803 m.redirect.oloc.pool);
3804 unlink_request(osd, req);
3805 mutex_unlock(&osd->lock);
3806
3807 /*
3808 * Not ceph_oloc_copy() - changing pool_ns is not
3809 * supported.
3810 */
3811 req->r_t.target_oloc.pool = m.redirect.oloc.pool;
3812 req->r_flags |= CEPH_OSD_FLAG_REDIRECTED |
3813 CEPH_OSD_FLAG_IGNORE_OVERLAY |
3814 CEPH_OSD_FLAG_IGNORE_CACHE;
3815 req->r_tid = 0;
3816 __submit_request(req, false);
3817 goto out_unlock_osdc;
3818 }
3819
3820 if (m.result == -EAGAIN) {
3821 dout("req %p tid %llu EAGAIN\n", req, req->r_tid);
3822 unlink_request(osd, req);
3823 mutex_unlock(&osd->lock);
3824
3825 /*
3826 * The object is missing on the replica or not (yet)
3827 * readable. Clear pgid to force a resend to the primary
3828 * via legacy_change.
3829 */
3830 req->r_t.pgid.pool = 0;
3831 req->r_t.pgid.seed = 0;
3832 WARN_ON(!req->r_t.used_replica);
3833 req->r_flags &= ~(CEPH_OSD_FLAG_BALANCE_READS |
3834 CEPH_OSD_FLAG_LOCALIZE_READS);
3835 req->r_tid = 0;
3836 __submit_request(req, false);
3837 goto out_unlock_osdc;
3838 }
3839
3840 if (m.num_ops != req->r_num_ops) {
3841 pr_err("num_ops %d != %d for tid %llu\n", m.num_ops,
3842 req->r_num_ops, req->r_tid);
3843 goto fail_request;
3844 }
3845 for (i = 0; i < req->r_num_ops; i++) {
3846 dout(" req %p tid %llu op %d rval %d len %u\n", req,
3847 req->r_tid, i, m.rval[i], m.outdata_len[i]);
3848 req->r_ops[i].rval = m.rval[i];
3849 req->r_ops[i].outdata_len = m.outdata_len[i];
3850 data_len += m.outdata_len[i];
3851 }
3852 if (data_len != le32_to_cpu(msg->hdr.data_len)) {
3853 pr_err("sum of lens %u != %u for tid %llu\n", data_len,
3854 le32_to_cpu(msg->hdr.data_len), req->r_tid);
3855 goto fail_request;
3856 }
3857 dout("%s req %p tid %llu result %d data_len %u\n", __func__,
3858 req, req->r_tid, m.result, data_len);
3859
3860 /*
3861 * Since we only ever request ONDISK, we should only ever get
3862 * one (type of) reply back.
3863 */
3864 WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
3865 req->r_version = m.user_version;
3866 req->r_result = m.result ?: data_len;
3867 finish_request(req);
3868 mutex_unlock(&osd->lock);
3869 up_read(&osdc->lock);
3870
3871 __complete_request(req);
3872 return;
3873
3874fail_request:
3875 complete_request(req, -EIO);
3876out_unlock_session:
3877 mutex_unlock(&osd->lock);
3878out_unlock_osdc:
3879 up_read(&osdc->lock);
3880}
3881
3882static void set_pool_was_full(struct ceph_osd_client *osdc)
3883{
3884 struct rb_node *n;
3885
3886 for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) {
3887 struct ceph_pg_pool_info *pi =
3888 rb_entry(n, struct ceph_pg_pool_info, node);
3889
3890 pi->was_full = __pool_full(pi);
3891 }
3892}
3893
3894static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id)
3895{
3896 struct ceph_pg_pool_info *pi;
3897
3898 pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id);
3899 if (!pi)
3900 return false;
3901
3902 return pi->was_full && !__pool_full(pi);
3903}
3904
3905static enum calc_target_result
3906recalc_linger_target(struct ceph_osd_linger_request *lreq)
3907{
3908 struct ceph_osd_client *osdc = lreq->osdc;
3909 enum calc_target_result ct_res;
3910
3911 ct_res = calc_target(osdc, &lreq->t, true);
3912 if (ct_res == CALC_TARGET_NEED_RESEND) {
3913 struct ceph_osd *osd;
3914
3915 osd = lookup_create_osd(osdc, lreq->t.osd, true);
3916 if (osd != lreq->osd) {
3917 unlink_linger(lreq->osd, lreq);
3918 link_linger(osd, lreq);
3919 }
3920 }
3921
3922 return ct_res;
3923}
3924
3925/*
3926 * Requeue requests whose mapping to an OSD has changed.
3927 */
3928static void scan_requests(struct ceph_osd *osd,
3929 bool force_resend,
3930 bool cleared_full,
3931 bool check_pool_cleared_full,
3932 struct rb_root *need_resend,
3933 struct list_head *need_resend_linger)
3934{
3935 struct ceph_osd_client *osdc = osd->o_osdc;
3936 struct rb_node *n;
3937 bool force_resend_writes;
3938
3939 for (n = rb_first(&osd->o_linger_requests); n; ) {
3940 struct ceph_osd_linger_request *lreq =
3941 rb_entry(n, struct ceph_osd_linger_request, node);
3942 enum calc_target_result ct_res;
3943
3944 n = rb_next(n); /* recalc_linger_target() */
3945
3946 dout("%s lreq %p linger_id %llu\n", __func__, lreq,
3947 lreq->linger_id);
3948 ct_res = recalc_linger_target(lreq);
3949 switch (ct_res) {
3950 case CALC_TARGET_NO_ACTION:
3951 force_resend_writes = cleared_full ||
3952 (check_pool_cleared_full &&
3953 pool_cleared_full(osdc, lreq->t.base_oloc.pool));
3954 if (!force_resend && !force_resend_writes)
3955 break;
3956
3957 fallthrough;
3958 case CALC_TARGET_NEED_RESEND:
3959 cancel_linger_map_check(lreq);
3960 /*
3961 * scan_requests() for the previous epoch(s)
3962 * may have already added it to the list, since
3963 * it's not unlinked here.
3964 */
3965 if (list_empty(&lreq->scan_item))
3966 list_add_tail(&lreq->scan_item, need_resend_linger);
3967 break;
3968 case CALC_TARGET_POOL_DNE:
3969 list_del_init(&lreq->scan_item);
3970 check_linger_pool_dne(lreq);
3971 break;
3972 }
3973 }
3974
3975 for (n = rb_first(&osd->o_requests); n; ) {
3976 struct ceph_osd_request *req =
3977 rb_entry(n, struct ceph_osd_request, r_node);
3978 enum calc_target_result ct_res;
3979
3980 n = rb_next(n); /* unlink_request(), check_pool_dne() */
3981
3982 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
3983 ct_res = calc_target(osdc, &req->r_t, false);
3984 switch (ct_res) {
3985 case CALC_TARGET_NO_ACTION:
3986 force_resend_writes = cleared_full ||
3987 (check_pool_cleared_full &&
3988 pool_cleared_full(osdc, req->r_t.base_oloc.pool));
3989 if (!force_resend &&
3990 (!(req->r_flags & CEPH_OSD_FLAG_WRITE) ||
3991 !force_resend_writes))
3992 break;
3993
3994 fallthrough;
3995 case CALC_TARGET_NEED_RESEND:
3996 cancel_map_check(req);
3997 unlink_request(osd, req);
3998 insert_request(need_resend, req);
3999 break;
4000 case CALC_TARGET_POOL_DNE:
4001 check_pool_dne(req);
4002 break;
4003 }
4004 }
4005}
4006
4007static int handle_one_map(struct ceph_osd_client *osdc,
4008 void *p, void *end, bool incremental,
4009 struct rb_root *need_resend,
4010 struct list_head *need_resend_linger)
4011{
4012 struct ceph_osdmap *newmap;
4013 struct rb_node *n;
4014 bool skipped_map = false;
4015 bool was_full;
4016
4017 was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
4018 set_pool_was_full(osdc);
4019
4020 if (incremental)
4021 newmap = osdmap_apply_incremental(&p, end,
4022 ceph_msgr2(osdc->client),
4023 osdc->osdmap);
4024 else
4025 newmap = ceph_osdmap_decode(&p, end, ceph_msgr2(osdc->client));
4026 if (IS_ERR(newmap))
4027 return PTR_ERR(newmap);
4028
4029 if (newmap != osdc->osdmap) {
4030 /*
4031 * Preserve ->was_full before destroying the old map.
4032 * For pools that weren't in the old map, ->was_full
4033 * should be false.
4034 */
4035 for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) {
4036 struct ceph_pg_pool_info *pi =
4037 rb_entry(n, struct ceph_pg_pool_info, node);
4038 struct ceph_pg_pool_info *old_pi;
4039
4040 old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id);
4041 if (old_pi)
4042 pi->was_full = old_pi->was_full;
4043 else
4044 WARN_ON(pi->was_full);
4045 }
4046
4047 if (osdc->osdmap->epoch &&
4048 osdc->osdmap->epoch + 1 < newmap->epoch) {
4049 WARN_ON(incremental);
4050 skipped_map = true;
4051 }
4052
4053 ceph_osdmap_destroy(osdc->osdmap);
4054 osdc->osdmap = newmap;
4055 }
4056
4057 was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL);
4058 scan_requests(&osdc->homeless_osd, skipped_map, was_full, true,
4059 need_resend, need_resend_linger);
4060
4061 for (n = rb_first(&osdc->osds); n; ) {
4062 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4063
4064 n = rb_next(n); /* close_osd() */
4065
4066 scan_requests(osd, skipped_map, was_full, true, need_resend,
4067 need_resend_linger);
4068 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
4069 memcmp(&osd->o_con.peer_addr,
4070 ceph_osd_addr(osdc->osdmap, osd->o_osd),
4071 sizeof(struct ceph_entity_addr)))
4072 close_osd(osd);
4073 }
4074
4075 return 0;
4076}
4077
4078static void kick_requests(struct ceph_osd_client *osdc,
4079 struct rb_root *need_resend,
4080 struct list_head *need_resend_linger)
4081{
4082 struct ceph_osd_linger_request *lreq, *nlreq;
4083 enum calc_target_result ct_res;
4084 struct rb_node *n;
4085
4086 /* make sure need_resend targets reflect latest map */
4087 for (n = rb_first(need_resend); n; ) {
4088 struct ceph_osd_request *req =
4089 rb_entry(n, struct ceph_osd_request, r_node);
4090
4091 n = rb_next(n);
4092
4093 if (req->r_t.epoch < osdc->osdmap->epoch) {
4094 ct_res = calc_target(osdc, &req->r_t, false);
4095 if (ct_res == CALC_TARGET_POOL_DNE) {
4096 erase_request(need_resend, req);
4097 check_pool_dne(req);
4098 }
4099 }
4100 }
4101
4102 for (n = rb_first(need_resend); n; ) {
4103 struct ceph_osd_request *req =
4104 rb_entry(n, struct ceph_osd_request, r_node);
4105 struct ceph_osd *osd;
4106
4107 n = rb_next(n);
4108 erase_request(need_resend, req); /* before link_request() */
4109
4110 osd = lookup_create_osd(osdc, req->r_t.osd, true);
4111 link_request(osd, req);
4112 if (!req->r_linger) {
4113 if (!osd_homeless(osd) && !req->r_t.paused)
4114 send_request(req);
4115 } else {
4116 cancel_linger_request(req);
4117 }
4118 }
4119
4120 list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) {
4121 if (!osd_homeless(lreq->osd))
4122 send_linger(lreq);
4123
4124 list_del_init(&lreq->scan_item);
4125 }
4126}
4127
4128/*
4129 * Process updated osd map.
4130 *
4131 * The message contains any number of incremental and full maps, normally
4132 * indicating some sort of topology change in the cluster. Kick requests
4133 * off to different OSDs as needed.
4134 */
4135void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
4136{
4137 void *p = msg->front.iov_base;
4138 void *const end = p + msg->front.iov_len;
4139 u32 nr_maps, maplen;
4140 u32 epoch;
4141 struct ceph_fsid fsid;
4142 struct rb_root need_resend = RB_ROOT;
4143 LIST_HEAD(need_resend_linger);
4144 bool handled_incremental = false;
4145 bool was_pauserd, was_pausewr;
4146 bool pauserd, pausewr;
4147 int err;
4148
4149 dout("%s have %u\n", __func__, osdc->osdmap->epoch);
4150 down_write(&osdc->lock);
4151
4152 /* verify fsid */
4153 ceph_decode_need(&p, end, sizeof(fsid), bad);
4154 ceph_decode_copy(&p, &fsid, sizeof(fsid));
4155 if (ceph_check_fsid(osdc->client, &fsid) < 0)
4156 goto bad;
4157
4158 was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4159 was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4160 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4161 have_pool_full(osdc);
4162
4163 /* incremental maps */
4164 ceph_decode_32_safe(&p, end, nr_maps, bad);
4165 dout(" %d inc maps\n", nr_maps);
4166 while (nr_maps > 0) {
4167 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4168 epoch = ceph_decode_32(&p);
4169 maplen = ceph_decode_32(&p);
4170 ceph_decode_need(&p, end, maplen, bad);
4171 if (osdc->osdmap->epoch &&
4172 osdc->osdmap->epoch + 1 == epoch) {
4173 dout("applying incremental map %u len %d\n",
4174 epoch, maplen);
4175 err = handle_one_map(osdc, p, p + maplen, true,
4176 &need_resend, &need_resend_linger);
4177 if (err)
4178 goto bad;
4179 handled_incremental = true;
4180 } else {
4181 dout("ignoring incremental map %u len %d\n",
4182 epoch, maplen);
4183 }
4184 p += maplen;
4185 nr_maps--;
4186 }
4187 if (handled_incremental)
4188 goto done;
4189
4190 /* full maps */
4191 ceph_decode_32_safe(&p, end, nr_maps, bad);
4192 dout(" %d full maps\n", nr_maps);
4193 while (nr_maps) {
4194 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4195 epoch = ceph_decode_32(&p);
4196 maplen = ceph_decode_32(&p);
4197 ceph_decode_need(&p, end, maplen, bad);
4198 if (nr_maps > 1) {
4199 dout("skipping non-latest full map %u len %d\n",
4200 epoch, maplen);
4201 } else if (osdc->osdmap->epoch >= epoch) {
4202 dout("skipping full map %u len %d, "
4203 "older than our %u\n", epoch, maplen,
4204 osdc->osdmap->epoch);
4205 } else {
4206 dout("taking full map %u len %d\n", epoch, maplen);
4207 err = handle_one_map(osdc, p, p + maplen, false,
4208 &need_resend, &need_resend_linger);
4209 if (err)
4210 goto bad;
4211 }
4212 p += maplen;
4213 nr_maps--;
4214 }
4215
4216done:
4217 /*
4218 * subscribe to subsequent osdmap updates if full to ensure
4219 * we find out when we are no longer full and stop returning
4220 * ENOSPC.
4221 */
4222 pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD);
4223 pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) ||
4224 ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) ||
4225 have_pool_full(osdc);
4226 if (was_pauserd || was_pausewr || pauserd || pausewr ||
4227 osdc->osdmap->epoch < osdc->epoch_barrier)
4228 maybe_request_map(osdc);
4229
4230 kick_requests(osdc, &need_resend, &need_resend_linger);
4231
4232 ceph_osdc_abort_on_full(osdc);
4233 ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP,
4234 osdc->osdmap->epoch);
4235 up_write(&osdc->lock);
4236 wake_up_all(&osdc->client->auth_wq);
4237 return;
4238
4239bad:
4240 pr_err("osdc handle_map corrupt msg\n");
4241 ceph_msg_dump(msg);
4242 up_write(&osdc->lock);
4243}
4244
4245/*
4246 * Resubmit requests pending on the given osd.
4247 */
4248static void kick_osd_requests(struct ceph_osd *osd)
4249{
4250 struct rb_node *n;
4251
4252 clear_backoffs(osd);
4253
4254 for (n = rb_first(&osd->o_requests); n; ) {
4255 struct ceph_osd_request *req =
4256 rb_entry(n, struct ceph_osd_request, r_node);
4257
4258 n = rb_next(n); /* cancel_linger_request() */
4259
4260 if (!req->r_linger) {
4261 if (!req->r_t.paused)
4262 send_request(req);
4263 } else {
4264 cancel_linger_request(req);
4265 }
4266 }
4267 for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) {
4268 struct ceph_osd_linger_request *lreq =
4269 rb_entry(n, struct ceph_osd_linger_request, node);
4270
4271 send_linger(lreq);
4272 }
4273}
4274
4275/*
4276 * If the osd connection drops, we need to resubmit all requests.
4277 */
4278static void osd_fault(struct ceph_connection *con)
4279{
4280 struct ceph_osd *osd = con->private;
4281 struct ceph_osd_client *osdc = osd->o_osdc;
4282
4283 dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd);
4284
4285 down_write(&osdc->lock);
4286 if (!osd_registered(osd)) {
4287 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4288 goto out_unlock;
4289 }
4290
4291 osd->o_sparse_op_idx = -1;
4292 ceph_init_sparse_read(&osd->o_sparse_read);
4293
4294 if (!reopen_osd(osd))
4295 kick_osd_requests(osd);
4296 maybe_request_map(osdc);
4297
4298out_unlock:
4299 up_write(&osdc->lock);
4300}
4301
4302struct MOSDBackoff {
4303 struct ceph_spg spgid;
4304 u32 map_epoch;
4305 u8 op;
4306 u64 id;
4307 struct ceph_hobject_id *begin;
4308 struct ceph_hobject_id *end;
4309};
4310
4311static int decode_MOSDBackoff(const struct ceph_msg *msg, struct MOSDBackoff *m)
4312{
4313 void *p = msg->front.iov_base;
4314 void *const end = p + msg->front.iov_len;
4315 u8 struct_v;
4316 u32 struct_len;
4317 int ret;
4318
4319 ret = ceph_start_decoding(&p, end, 1, "spg_t", &struct_v, &struct_len);
4320 if (ret)
4321 return ret;
4322
4323 ret = ceph_decode_pgid(&p, end, &m->spgid.pgid);
4324 if (ret)
4325 return ret;
4326
4327 ceph_decode_8_safe(&p, end, m->spgid.shard, e_inval);
4328 ceph_decode_32_safe(&p, end, m->map_epoch, e_inval);
4329 ceph_decode_8_safe(&p, end, m->op, e_inval);
4330 ceph_decode_64_safe(&p, end, m->id, e_inval);
4331
4332 m->begin = kzalloc_obj(*m->begin, GFP_NOIO);
4333 if (!m->begin)
4334 return -ENOMEM;
4335
4336 ret = decode_hoid(&p, end, m->begin);
4337 if (ret) {
4338 free_hoid(m->begin);
4339 return ret;
4340 }
4341
4342 m->end = kzalloc_obj(*m->end, GFP_NOIO);
4343 if (!m->end) {
4344 free_hoid(m->begin);
4345 return -ENOMEM;
4346 }
4347
4348 ret = decode_hoid(&p, end, m->end);
4349 if (ret) {
4350 free_hoid(m->begin);
4351 free_hoid(m->end);
4352 return ret;
4353 }
4354
4355 return 0;
4356
4357e_inval:
4358 return -EINVAL;
4359}
4360
4361static struct ceph_msg *create_backoff_message(
4362 const struct ceph_osd_backoff *backoff,
4363 u32 map_epoch)
4364{
4365 struct ceph_msg *msg;
4366 void *p, *end;
4367 int msg_size;
4368
4369 msg_size = CEPH_ENCODING_START_BLK_LEN +
4370 CEPH_PGID_ENCODING_LEN + 1; /* spgid */
4371 msg_size += 4 + 1 + 8; /* map_epoch, op, id */
4372 msg_size += CEPH_ENCODING_START_BLK_LEN +
4373 hoid_encoding_size(backoff->begin);
4374 msg_size += CEPH_ENCODING_START_BLK_LEN +
4375 hoid_encoding_size(backoff->end);
4376
4377 msg = ceph_msg_new(CEPH_MSG_OSD_BACKOFF, msg_size, GFP_NOIO, true);
4378 if (!msg)
4379 return NULL;
4380
4381 p = msg->front.iov_base;
4382 end = p + msg->front_alloc_len;
4383
4384 encode_spgid(&p, &backoff->spgid);
4385 ceph_encode_32(&p, map_epoch);
4386 ceph_encode_8(&p, CEPH_OSD_BACKOFF_OP_ACK_BLOCK);
4387 ceph_encode_64(&p, backoff->id);
4388 encode_hoid(&p, end, backoff->begin);
4389 encode_hoid(&p, end, backoff->end);
4390 BUG_ON(p != end);
4391
4392 msg->front.iov_len = p - msg->front.iov_base;
4393 msg->hdr.version = cpu_to_le16(1); /* MOSDBackoff v1 */
4394 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
4395
4396 return msg;
4397}
4398
4399static void handle_backoff_block(struct ceph_osd *osd, struct MOSDBackoff *m)
4400{
4401 struct ceph_spg_mapping *spg;
4402 struct ceph_osd_backoff *backoff;
4403 struct ceph_msg *msg;
4404
4405 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4406 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4407
4408 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &m->spgid);
4409 if (!spg) {
4410 spg = alloc_spg_mapping();
4411 if (!spg) {
4412 pr_err("%s failed to allocate spg\n", __func__);
4413 return;
4414 }
4415 spg->spgid = m->spgid; /* struct */
4416 insert_spg_mapping(&osd->o_backoff_mappings, spg);
4417 }
4418
4419 backoff = alloc_backoff();
4420 if (!backoff) {
4421 pr_err("%s failed to allocate backoff\n", __func__);
4422 return;
4423 }
4424 backoff->spgid = m->spgid; /* struct */
4425 backoff->id = m->id;
4426 backoff->begin = m->begin;
4427 m->begin = NULL; /* backoff now owns this */
4428 backoff->end = m->end;
4429 m->end = NULL; /* ditto */
4430
4431 insert_backoff(&spg->backoffs, backoff);
4432 insert_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4433
4434 /*
4435 * Ack with original backoff's epoch so that the OSD can
4436 * discard this if there was a PG split.
4437 */
4438 msg = create_backoff_message(backoff, m->map_epoch);
4439 if (!msg) {
4440 pr_err("%s failed to allocate msg\n", __func__);
4441 return;
4442 }
4443 ceph_con_send(&osd->o_con, msg);
4444}
4445
4446static bool target_contained_by(const struct ceph_osd_request_target *t,
4447 const struct ceph_hobject_id *begin,
4448 const struct ceph_hobject_id *end)
4449{
4450 struct ceph_hobject_id hoid;
4451 int cmp;
4452
4453 hoid_fill_from_target(&hoid, t);
4454 cmp = hoid_compare(&hoid, begin);
4455 return !cmp || (cmp > 0 && hoid_compare(&hoid, end) < 0);
4456}
4457
4458static void handle_backoff_unblock(struct ceph_osd *osd,
4459 const struct MOSDBackoff *m)
4460{
4461 struct ceph_spg_mapping *spg;
4462 struct ceph_osd_backoff *backoff;
4463 struct rb_node *n;
4464
4465 dout("%s osd%d spgid %llu.%xs%d id %llu\n", __func__, osd->o_osd,
4466 m->spgid.pgid.pool, m->spgid.pgid.seed, m->spgid.shard, m->id);
4467
4468 backoff = lookup_backoff_by_id(&osd->o_backoffs_by_id, m->id);
4469 if (!backoff) {
4470 pr_err("%s osd%d spgid %llu.%xs%d id %llu backoff dne\n",
4471 __func__, osd->o_osd, m->spgid.pgid.pool,
4472 m->spgid.pgid.seed, m->spgid.shard, m->id);
4473 return;
4474 }
4475
4476 if (hoid_compare(backoff->begin, m->begin) &&
4477 hoid_compare(backoff->end, m->end)) {
4478 pr_err("%s osd%d spgid %llu.%xs%d id %llu bad range?\n",
4479 __func__, osd->o_osd, m->spgid.pgid.pool,
4480 m->spgid.pgid.seed, m->spgid.shard, m->id);
4481 /* unblock it anyway... */
4482 }
4483
4484 spg = lookup_spg_mapping(&osd->o_backoff_mappings, &backoff->spgid);
4485 BUG_ON(!spg);
4486
4487 erase_backoff(&spg->backoffs, backoff);
4488 erase_backoff_by_id(&osd->o_backoffs_by_id, backoff);
4489 free_backoff(backoff);
4490
4491 if (RB_EMPTY_ROOT(&spg->backoffs)) {
4492 erase_spg_mapping(&osd->o_backoff_mappings, spg);
4493 free_spg_mapping(spg);
4494 }
4495
4496 for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) {
4497 struct ceph_osd_request *req =
4498 rb_entry(n, struct ceph_osd_request, r_node);
4499
4500 if (!ceph_spg_compare(&req->r_t.spgid, &m->spgid)) {
4501 /*
4502 * Match against @m, not @backoff -- the PG may
4503 * have split on the OSD.
4504 */
4505 if (target_contained_by(&req->r_t, m->begin, m->end)) {
4506 /*
4507 * If no other installed backoff applies,
4508 * resend.
4509 */
4510 send_request(req);
4511 }
4512 }
4513 }
4514}
4515
4516static void handle_backoff(struct ceph_osd *osd, struct ceph_msg *msg)
4517{
4518 struct ceph_osd_client *osdc = osd->o_osdc;
4519 struct MOSDBackoff m;
4520 int ret;
4521
4522 down_read(&osdc->lock);
4523 if (!osd_registered(osd)) {
4524 dout("%s osd%d unknown\n", __func__, osd->o_osd);
4525 up_read(&osdc->lock);
4526 return;
4527 }
4528 WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num));
4529
4530 mutex_lock(&osd->lock);
4531 ret = decode_MOSDBackoff(msg, &m);
4532 if (ret) {
4533 pr_err("failed to decode MOSDBackoff: %d\n", ret);
4534 ceph_msg_dump(msg);
4535 goto out_unlock;
4536 }
4537
4538 switch (m.op) {
4539 case CEPH_OSD_BACKOFF_OP_BLOCK:
4540 handle_backoff_block(osd, &m);
4541 break;
4542 case CEPH_OSD_BACKOFF_OP_UNBLOCK:
4543 handle_backoff_unblock(osd, &m);
4544 break;
4545 default:
4546 pr_err("%s osd%d unknown op %d\n", __func__, osd->o_osd, m.op);
4547 }
4548
4549 free_hoid(m.begin);
4550 free_hoid(m.end);
4551
4552out_unlock:
4553 mutex_unlock(&osd->lock);
4554 up_read(&osdc->lock);
4555}
4556
4557/*
4558 * Process osd watch notifications
4559 */
4560static void handle_watch_notify(struct ceph_osd_client *osdc,
4561 struct ceph_msg *msg)
4562{
4563 void *p = msg->front.iov_base;
4564 void *const end = p + msg->front.iov_len;
4565 struct ceph_osd_linger_request *lreq;
4566 struct linger_work *lwork;
4567 u8 proto_ver, opcode;
4568 u64 cookie, notify_id;
4569 u64 notifier_id = 0;
4570 s32 return_code = 0;
4571 void *payload = NULL;
4572 u32 payload_len = 0;
4573
4574 ceph_decode_8_safe(&p, end, proto_ver, bad);
4575 ceph_decode_8_safe(&p, end, opcode, bad);
4576 ceph_decode_64_safe(&p, end, cookie, bad);
4577 p += 8; /* skip ver */
4578 ceph_decode_64_safe(&p, end, notify_id, bad);
4579
4580 if (proto_ver >= 1) {
4581 ceph_decode_32_safe(&p, end, payload_len, bad);
4582 ceph_decode_need(&p, end, payload_len, bad);
4583 payload = p;
4584 p += payload_len;
4585 }
4586
4587 if (le16_to_cpu(msg->hdr.version) >= 2)
4588 ceph_decode_32_safe(&p, end, return_code, bad);
4589
4590 if (le16_to_cpu(msg->hdr.version) >= 3)
4591 ceph_decode_64_safe(&p, end, notifier_id, bad);
4592
4593 down_read(&osdc->lock);
4594 lreq = lookup_linger_osdc(&osdc->linger_requests, cookie);
4595 if (!lreq) {
4596 dout("%s opcode %d cookie %llu dne\n", __func__, opcode,
4597 cookie);
4598 goto out_unlock_osdc;
4599 }
4600
4601 mutex_lock(&lreq->lock);
4602 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__,
4603 opcode, cookie, lreq, lreq->is_watch);
4604 if (opcode == CEPH_WATCH_EVENT_DISCONNECT) {
4605 if (!lreq->last_error) {
4606 lreq->last_error = -ENOTCONN;
4607 queue_watch_error(lreq);
4608 }
4609 } else if (!lreq->is_watch) {
4610 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
4611 if (lreq->notify_id && lreq->notify_id != notify_id) {
4612 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq,
4613 lreq->notify_id, notify_id);
4614 } else if (!completion_done(&lreq->notify_finish_wait)) {
4615 struct ceph_msg_data *data =
4616 msg->num_data_items ? &msg->data[0] : NULL;
4617
4618 if (data) {
4619 if (lreq->preply_pages) {
4620 WARN_ON(data->type !=
4621 CEPH_MSG_DATA_PAGES);
4622 *lreq->preply_pages = data->pages;
4623 *lreq->preply_len = data->length;
4624 data->own_pages = false;
4625 }
4626 }
4627 lreq->notify_finish_error = return_code;
4628 complete_all(&lreq->notify_finish_wait);
4629 }
4630 } else {
4631 /* CEPH_WATCH_EVENT_NOTIFY */
4632 lwork = lwork_alloc(lreq, do_watch_notify);
4633 if (!lwork) {
4634 pr_err("failed to allocate notify-lwork\n");
4635 goto out_unlock_lreq;
4636 }
4637
4638 lwork->notify.notify_id = notify_id;
4639 lwork->notify.notifier_id = notifier_id;
4640 lwork->notify.payload = payload;
4641 lwork->notify.payload_len = payload_len;
4642 lwork->notify.msg = ceph_msg_get(msg);
4643 lwork_queue(lwork);
4644 }
4645
4646out_unlock_lreq:
4647 mutex_unlock(&lreq->lock);
4648out_unlock_osdc:
4649 up_read(&osdc->lock);
4650 return;
4651
4652bad:
4653 pr_err("osdc handle_watch_notify corrupt msg\n");
4654}
4655
4656/*
4657 * Register request, send initial attempt.
4658 */
4659void ceph_osdc_start_request(struct ceph_osd_client *osdc,
4660 struct ceph_osd_request *req)
4661{
4662 down_read(&osdc->lock);
4663 submit_request(req, false);
4664 up_read(&osdc->lock);
4665}
4666EXPORT_SYMBOL(ceph_osdc_start_request);
4667
4668/*
4669 * Unregister request. If @req was registered, it isn't completed:
4670 * r_result isn't set and __complete_request() isn't invoked.
4671 *
4672 * If @req wasn't registered, this call may have raced with
4673 * handle_reply(), in which case r_result would already be set and
4674 * __complete_request() would be getting invoked, possibly even
4675 * concurrently with this call.
4676 */
4677void ceph_osdc_cancel_request(struct ceph_osd_request *req)
4678{
4679 struct ceph_osd_client *osdc = req->r_osdc;
4680
4681 down_write(&osdc->lock);
4682 if (req->r_osd)
4683 cancel_request(req);
4684 up_write(&osdc->lock);
4685}
4686EXPORT_SYMBOL(ceph_osdc_cancel_request);
4687
4688/*
4689 * @timeout: in jiffies, 0 means "wait forever"
4690 */
4691static int wait_request_timeout(struct ceph_osd_request *req,
4692 unsigned long timeout)
4693{
4694 long left;
4695
4696 dout("%s req %p tid %llu\n", __func__, req, req->r_tid);
4697 left = wait_for_completion_killable_timeout(&req->r_completion,
4698 ceph_timeout_jiffies(timeout));
4699 if (left <= 0) {
4700 left = left ?: -ETIMEDOUT;
4701 ceph_osdc_cancel_request(req);
4702 } else {
4703 left = req->r_result; /* completed */
4704 }
4705
4706 return left;
4707}
4708
4709/*
4710 * wait for a request to complete
4711 */
4712int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
4713 struct ceph_osd_request *req)
4714{
4715 return wait_request_timeout(req, 0);
4716}
4717EXPORT_SYMBOL(ceph_osdc_wait_request);
4718
4719/*
4720 * sync - wait for all in-flight requests to flush. avoid starvation.
4721 */
4722void ceph_osdc_sync(struct ceph_osd_client *osdc)
4723{
4724 struct rb_node *n, *p;
4725 u64 last_tid = atomic64_read(&osdc->last_tid);
4726
4727again:
4728 down_read(&osdc->lock);
4729 for (n = rb_first(&osdc->osds); n; n = rb_next(n)) {
4730 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
4731
4732 mutex_lock(&osd->lock);
4733 for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
4734 struct ceph_osd_request *req =
4735 rb_entry(p, struct ceph_osd_request, r_node);
4736
4737 if (req->r_tid > last_tid)
4738 break;
4739
4740 if (!(req->r_flags & CEPH_OSD_FLAG_WRITE))
4741 continue;
4742
4743 ceph_osdc_get_request(req);
4744 mutex_unlock(&osd->lock);
4745 up_read(&osdc->lock);
4746 dout("%s waiting on req %p tid %llu last_tid %llu\n",
4747 __func__, req, req->r_tid, last_tid);
4748 wait_for_completion(&req->r_completion);
4749 ceph_osdc_put_request(req);
4750 goto again;
4751 }
4752
4753 mutex_unlock(&osd->lock);
4754 }
4755
4756 up_read(&osdc->lock);
4757 dout("%s done last_tid %llu\n", __func__, last_tid);
4758}
4759EXPORT_SYMBOL(ceph_osdc_sync);
4760
4761/*
4762 * Returns a handle, caller owns a ref.
4763 */
4764struct ceph_osd_linger_request *
4765ceph_osdc_watch(struct ceph_osd_client *osdc,
4766 struct ceph_object_id *oid,
4767 struct ceph_object_locator *oloc,
4768 rados_watchcb2_t wcb,
4769 rados_watcherrcb_t errcb,
4770 void *data)
4771{
4772 struct ceph_osd_linger_request *lreq;
4773 int ret;
4774
4775 lreq = linger_alloc(osdc);
4776 if (!lreq)
4777 return ERR_PTR(-ENOMEM);
4778
4779 lreq->is_watch = true;
4780 lreq->wcb = wcb;
4781 lreq->errcb = errcb;
4782 lreq->data = data;
4783 lreq->watch_valid_thru = jiffies;
4784
4785 ceph_oid_copy(&lreq->t.base_oid, oid);
4786 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4787 lreq->t.flags = CEPH_OSD_FLAG_WRITE;
4788 ktime_get_real_ts64(&lreq->mtime);
4789
4790 linger_submit(lreq);
4791 ret = linger_reg_commit_wait(lreq);
4792 if (ret) {
4793 linger_cancel(lreq);
4794 goto err_put_lreq;
4795 }
4796
4797 return lreq;
4798
4799err_put_lreq:
4800 linger_put(lreq);
4801 return ERR_PTR(ret);
4802}
4803EXPORT_SYMBOL(ceph_osdc_watch);
4804
4805/*
4806 * Releases a ref.
4807 *
4808 * Times out after mount_timeout to preserve rbd unmap behaviour
4809 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
4810 * with mount_timeout").
4811 */
4812int ceph_osdc_unwatch(struct ceph_osd_client *osdc,
4813 struct ceph_osd_linger_request *lreq)
4814{
4815 struct ceph_options *opts = osdc->client->options;
4816 struct ceph_osd_request *req;
4817 int ret;
4818
4819 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4820 if (!req)
4821 return -ENOMEM;
4822
4823 ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid);
4824 ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc);
4825 req->r_flags = CEPH_OSD_FLAG_WRITE;
4826 ktime_get_real_ts64(&req->r_mtime);
4827 osd_req_op_watch_init(req, 0, CEPH_OSD_WATCH_OP_UNWATCH,
4828 lreq->linger_id, 0);
4829
4830 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4831 if (ret)
4832 goto out_put_req;
4833
4834 ceph_osdc_start_request(osdc, req);
4835 linger_cancel(lreq);
4836 linger_put(lreq);
4837 ret = wait_request_timeout(req, opts->mount_timeout);
4838
4839out_put_req:
4840 ceph_osdc_put_request(req);
4841 return ret;
4842}
4843EXPORT_SYMBOL(ceph_osdc_unwatch);
4844
4845static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which,
4846 u64 notify_id, u64 cookie, void *payload,
4847 u32 payload_len)
4848{
4849 struct ceph_osd_req_op *op;
4850 struct ceph_pagelist *pl;
4851 int ret;
4852
4853 op = osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0);
4854
4855 pl = ceph_pagelist_alloc(GFP_NOIO);
4856 if (!pl)
4857 return -ENOMEM;
4858
4859 ret = ceph_pagelist_encode_64(pl, notify_id);
4860 ret |= ceph_pagelist_encode_64(pl, cookie);
4861 if (payload) {
4862 ret |= ceph_pagelist_encode_32(pl, payload_len);
4863 ret |= ceph_pagelist_append(pl, payload, payload_len);
4864 } else {
4865 ret |= ceph_pagelist_encode_32(pl, 0);
4866 }
4867 if (ret) {
4868 ceph_pagelist_release(pl);
4869 return -ENOMEM;
4870 }
4871
4872 ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl);
4873 op->indata_len = pl->length;
4874 return 0;
4875}
4876
4877int ceph_osdc_notify_ack(struct ceph_osd_client *osdc,
4878 struct ceph_object_id *oid,
4879 struct ceph_object_locator *oloc,
4880 u64 notify_id,
4881 u64 cookie,
4882 void *payload,
4883 u32 payload_len)
4884{
4885 struct ceph_osd_request *req;
4886 int ret;
4887
4888 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
4889 if (!req)
4890 return -ENOMEM;
4891
4892 ceph_oid_copy(&req->r_base_oid, oid);
4893 ceph_oloc_copy(&req->r_base_oloc, oloc);
4894 req->r_flags = CEPH_OSD_FLAG_READ;
4895
4896 ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload,
4897 payload_len);
4898 if (ret)
4899 goto out_put_req;
4900
4901 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
4902 if (ret)
4903 goto out_put_req;
4904
4905 ceph_osdc_start_request(osdc, req);
4906 ret = ceph_osdc_wait_request(osdc, req);
4907
4908out_put_req:
4909 ceph_osdc_put_request(req);
4910 return ret;
4911}
4912EXPORT_SYMBOL(ceph_osdc_notify_ack);
4913
4914/*
4915 * @timeout: in seconds
4916 *
4917 * @preply_{pages,len} are initialized both on success and error.
4918 * The caller is responsible for:
4919 *
4920 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
4921 */
4922int ceph_osdc_notify(struct ceph_osd_client *osdc,
4923 struct ceph_object_id *oid,
4924 struct ceph_object_locator *oloc,
4925 void *payload,
4926 u32 payload_len,
4927 u32 timeout,
4928 struct page ***preply_pages,
4929 size_t *preply_len)
4930{
4931 struct ceph_osd_linger_request *lreq;
4932 int ret;
4933
4934 WARN_ON(!timeout);
4935 if (preply_pages) {
4936 *preply_pages = NULL;
4937 *preply_len = 0;
4938 }
4939
4940 lreq = linger_alloc(osdc);
4941 if (!lreq)
4942 return -ENOMEM;
4943
4944 lreq->request_pl = ceph_pagelist_alloc(GFP_NOIO);
4945 if (!lreq->request_pl) {
4946 ret = -ENOMEM;
4947 goto out_put_lreq;
4948 }
4949
4950 ret = ceph_pagelist_encode_32(lreq->request_pl, 1); /* prot_ver */
4951 ret |= ceph_pagelist_encode_32(lreq->request_pl, timeout);
4952 ret |= ceph_pagelist_encode_32(lreq->request_pl, payload_len);
4953 ret |= ceph_pagelist_append(lreq->request_pl, payload, payload_len);
4954 if (ret) {
4955 ret = -ENOMEM;
4956 goto out_put_lreq;
4957 }
4958
4959 /* for notify_id */
4960 lreq->notify_id_pages = ceph_alloc_page_vector(1, GFP_NOIO);
4961 if (IS_ERR(lreq->notify_id_pages)) {
4962 ret = PTR_ERR(lreq->notify_id_pages);
4963 lreq->notify_id_pages = NULL;
4964 goto out_put_lreq;
4965 }
4966
4967 lreq->preply_pages = preply_pages;
4968 lreq->preply_len = preply_len;
4969
4970 ceph_oid_copy(&lreq->t.base_oid, oid);
4971 ceph_oloc_copy(&lreq->t.base_oloc, oloc);
4972 lreq->t.flags = CEPH_OSD_FLAG_READ;
4973
4974 linger_submit(lreq);
4975 ret = linger_reg_commit_wait(lreq);
4976 if (!ret)
4977 ret = linger_notify_finish_wait(lreq,
4978 msecs_to_jiffies(2 * timeout * MSEC_PER_SEC));
4979 else
4980 dout("lreq %p failed to initiate notify %d\n", lreq, ret);
4981
4982 linger_cancel(lreq);
4983out_put_lreq:
4984 linger_put(lreq);
4985 return ret;
4986}
4987EXPORT_SYMBOL(ceph_osdc_notify);
4988
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)
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 */
5004
5005 if (struct_v >= 2) {
5006 ret = ceph_decode_entity_addr(p, end, &item->addr);
5007 if (ret)
5008 goto bad;
5009 } else {
5010 ret = 0;
5011 }
5012
5013 dout("%s %s%llu cookie %llu addr %s\n", __func__,
5014 ENTITY_NAME(item->name), item->cookie,
5015 ceph_pr_addr(&item->addr));
5016bad:
5017 return ret;
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 ceph_decode_32_safe(p, end, *num_watchers, bad);
5035 *watchers = kzalloc_objs(**watchers, *num_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
5049bad:
5050 return -EINVAL;
5051}
5052
5053/*
5054 * On success, the caller is responsible for:
5055 *
5056 * kfree(watchers);
5057 */
5058int ceph_osdc_list_watchers(struct ceph_osd_client *osdc,
5059 struct ceph_object_id *oid,
5060 struct ceph_object_locator *oloc,
5061 struct ceph_watch_item **watchers,
5062 u32 *num_watchers)
5063{
5064 struct ceph_osd_request *req;
5065 struct page **pages;
5066 int ret;
5067
5068 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5069 if (!req)
5070 return -ENOMEM;
5071
5072 ceph_oid_copy(&req->r_base_oid, oid);
5073 ceph_oloc_copy(&req->r_base_oloc, oloc);
5074 req->r_flags = CEPH_OSD_FLAG_READ;
5075
5076 pages = ceph_alloc_page_vector(1, GFP_NOIO);
5077 if (IS_ERR(pages)) {
5078 ret = PTR_ERR(pages);
5079 goto out_put_req;
5080 }
5081
5082 osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0);
5083 ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers,
5084 response_data),
5085 pages, PAGE_SIZE, 0, false, true);
5086
5087 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5088 if (ret)
5089 goto out_put_req;
5090
5091 ceph_osdc_start_request(osdc, req);
5092 ret = ceph_osdc_wait_request(osdc, req);
5093 if (ret >= 0) {
5094 void *p = page_address(pages[0]);
5095 void *const end = p + req->r_ops[0].outdata_len;
5096
5097 ret = decode_watchers(&p, end, watchers, num_watchers);
5098 }
5099
5100out_put_req:
5101 ceph_osdc_put_request(req);
5102 return ret;
5103}
5104EXPORT_SYMBOL(ceph_osdc_list_watchers);
5105
5106/*
5107 * Call all pending notify callbacks - for use after a watch is
5108 * unregistered, to make sure no more callbacks for it will be invoked
5109 */
5110void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
5111{
5112 dout("%s osdc %p\n", __func__, osdc);
5113 flush_workqueue(osdc->notify_wq);
5114}
5115EXPORT_SYMBOL(ceph_osdc_flush_notifies);
5116
5117void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc)
5118{
5119 down_read(&osdc->lock);
5120 maybe_request_map(osdc);
5121 up_read(&osdc->lock);
5122}
5123EXPORT_SYMBOL(ceph_osdc_maybe_request_map);
5124
5125/*
5126 * Execute an OSD class method on an object.
5127 *
5128 * @flags: CEPH_OSD_FLAG_*
5129 * @resp_len: in/out param for reply length
5130 */
5131int ceph_osdc_call(struct ceph_osd_client *osdc,
5132 struct ceph_object_id *oid,
5133 struct ceph_object_locator *oloc,
5134 const char *class, const char *method,
5135 unsigned int flags,
5136 struct page *req_page, size_t req_len,
5137 struct page **resp_pages, size_t *resp_len)
5138{
5139 struct ceph_osd_request *req;
5140 int ret;
5141
5142 if (req_len > PAGE_SIZE)
5143 return -E2BIG;
5144
5145 req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO);
5146 if (!req)
5147 return -ENOMEM;
5148
5149 ceph_oid_copy(&req->r_base_oid, oid);
5150 ceph_oloc_copy(&req->r_base_oloc, oloc);
5151 req->r_flags = flags;
5152
5153 ret = osd_req_op_cls_init(req, 0, class, method);
5154 if (ret)
5155 goto out_put_req;
5156
5157 if (req_page)
5158 osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len,
5159 0, false, false);
5160 if (resp_pages)
5161 osd_req_op_cls_response_data_pages(req, 0, resp_pages,
5162 *resp_len, 0, false, false);
5163
5164 ret = ceph_osdc_alloc_messages(req, GFP_NOIO);
5165 if (ret)
5166 goto out_put_req;
5167
5168 ceph_osdc_start_request(osdc, req);
5169 ret = ceph_osdc_wait_request(osdc, req);
5170 if (ret >= 0) {
5171 ret = req->r_ops[0].rval;
5172 if (resp_pages)
5173 *resp_len = req->r_ops[0].outdata_len;
5174 }
5175
5176out_put_req:
5177 ceph_osdc_put_request(req);
5178 return ret;
5179}
5180EXPORT_SYMBOL(ceph_osdc_call);
5181
5182/*
5183 * reset all osd connections
5184 */
5185void ceph_osdc_reopen_osds(struct ceph_osd_client *osdc)
5186{
5187 struct rb_node *n;
5188
5189 down_write(&osdc->lock);
5190 for (n = rb_first(&osdc->osds); n; ) {
5191 struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
5192
5193 n = rb_next(n);
5194 if (!reopen_osd(osd))
5195 kick_osd_requests(osd);
5196 }
5197 up_write(&osdc->lock);
5198}
5199
5200/*
5201 * init, shutdown
5202 */
5203int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
5204{
5205 int err;
5206
5207 dout("init\n");
5208 osdc->client = client;
5209 init_rwsem(&osdc->lock);
5210 osdc->osds = RB_ROOT;
5211 INIT_LIST_HEAD(&osdc->osd_lru);
5212 spin_lock_init(&osdc->osd_lru_lock);
5213 osd_init(&osdc->homeless_osd);
5214 osdc->homeless_osd.o_osdc = osdc;
5215 osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD;
5216 osdc->last_linger_id = CEPH_LINGER_ID_START;
5217 osdc->linger_requests = RB_ROOT;
5218 osdc->map_checks = RB_ROOT;
5219 osdc->linger_map_checks = RB_ROOT;
5220 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
5221 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
5222
5223 err = -ENOMEM;
5224 osdc->osdmap = ceph_osdmap_alloc();
5225 if (!osdc->osdmap)
5226 goto out;
5227
5228 osdc->req_mempool = mempool_create_slab_pool(10,
5229 ceph_osd_request_cache);
5230 if (!osdc->req_mempool)
5231 goto out_map;
5232
5233 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
5234 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10, "osd_op");
5235 if (err < 0)
5236 goto out_mempool;
5237 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
5238 PAGE_SIZE, CEPH_OSD_SLAB_OPS, 10,
5239 "osd_op_reply");
5240 if (err < 0)
5241 goto out_msgpool;
5242
5243 err = -ENOMEM;
5244 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
5245 if (!osdc->notify_wq)
5246 goto out_msgpool_reply;
5247
5248 osdc->completion_wq = create_singlethread_workqueue("ceph-completion");
5249 if (!osdc->completion_wq)
5250 goto out_notify_wq;
5251
5252 schedule_delayed_work(&osdc->timeout_work,
5253 osdc->client->options->osd_keepalive_timeout);
5254 schedule_delayed_work(&osdc->osds_timeout_work,
5255 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
5256
5257 return 0;
5258
5259out_notify_wq:
5260 destroy_workqueue(osdc->notify_wq);
5261out_msgpool_reply:
5262 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5263out_msgpool:
5264 ceph_msgpool_destroy(&osdc->msgpool_op);
5265out_mempool:
5266 mempool_destroy(osdc->req_mempool);
5267out_map:
5268 ceph_osdmap_destroy(osdc->osdmap);
5269out:
5270 return err;
5271}
5272
5273void ceph_osdc_stop(struct ceph_osd_client *osdc)
5274{
5275 destroy_workqueue(osdc->completion_wq);
5276 destroy_workqueue(osdc->notify_wq);
5277 cancel_delayed_work_sync(&osdc->timeout_work);
5278 cancel_delayed_work_sync(&osdc->osds_timeout_work);
5279
5280 down_write(&osdc->lock);
5281 while (!RB_EMPTY_ROOT(&osdc->osds)) {
5282 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
5283 struct ceph_osd, o_node);
5284 close_osd(osd);
5285 }
5286 up_write(&osdc->lock);
5287 WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1);
5288 osd_cleanup(&osdc->homeless_osd);
5289
5290 WARN_ON(!list_empty(&osdc->osd_lru));
5291 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests));
5292 WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks));
5293 WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks));
5294 WARN_ON(atomic_read(&osdc->num_requests));
5295 WARN_ON(atomic_read(&osdc->num_homeless));
5296
5297 ceph_osdmap_destroy(osdc->osdmap);
5298 mempool_destroy(osdc->req_mempool);
5299 ceph_msgpool_destroy(&osdc->msgpool_op);
5300 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
5301}
5302
5303int osd_req_op_copy_from_init(struct ceph_osd_request *req,
5304 u64 src_snapid, u64 src_version,
5305 struct ceph_object_id *src_oid,
5306 struct ceph_object_locator *src_oloc,
5307 u32 src_fadvise_flags,
5308 u32 dst_fadvise_flags,
5309 u32 truncate_seq, u64 truncate_size,
5310 u8 copy_from_flags)
5311{
5312 struct ceph_osd_req_op *op;
5313 struct page **pages;
5314 void *p, *end;
5315
5316 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
5317 if (IS_ERR(pages))
5318 return PTR_ERR(pages);
5319
5320 op = osd_req_op_init(req, 0, CEPH_OSD_OP_COPY_FROM2,
5321 dst_fadvise_flags);
5322 op->copy_from.snapid = src_snapid;
5323 op->copy_from.src_version = src_version;
5324 op->copy_from.flags = copy_from_flags;
5325 op->copy_from.src_fadvise_flags = src_fadvise_flags;
5326
5327 p = page_address(pages[0]);
5328 end = p + PAGE_SIZE;
5329 ceph_encode_string(&p, end, src_oid->name, src_oid->name_len);
5330 encode_oloc(&p, end, src_oloc);
5331 ceph_encode_32(&p, truncate_seq);
5332 ceph_encode_64(&p, truncate_size);
5333 op->indata_len = PAGE_SIZE - (end - p);
5334
5335 ceph_osd_data_pages_init(&op->copy_from.osd_data, pages,
5336 op->indata_len, 0, false, true);
5337 return 0;
5338}
5339EXPORT_SYMBOL(osd_req_op_copy_from_init);
5340
5341int __init ceph_osdc_setup(void)
5342{
5343 size_t size = sizeof(struct ceph_osd_request) +
5344 CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op);
5345
5346 BUG_ON(ceph_osd_request_cache);
5347 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size,
5348 0, 0, NULL);
5349
5350 return ceph_osd_request_cache ? 0 : -ENOMEM;
5351}
5352
5353void ceph_osdc_cleanup(void)
5354{
5355 BUG_ON(!ceph_osd_request_cache);
5356 kmem_cache_destroy(ceph_osd_request_cache);
5357 ceph_osd_request_cache = NULL;
5358}
5359
5360/*
5361 * handle incoming message
5362 */
5363static void osd_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5364{
5365 struct ceph_osd *osd = con->private;
5366 struct ceph_osd_client *osdc = osd->o_osdc;
5367 int type = le16_to_cpu(msg->hdr.type);
5368
5369 switch (type) {
5370 case CEPH_MSG_OSD_MAP:
5371 ceph_osdc_handle_map(osdc, msg);
5372 break;
5373 case CEPH_MSG_OSD_OPREPLY:
5374 handle_reply(osd, msg);
5375 break;
5376 case CEPH_MSG_OSD_BACKOFF:
5377 handle_backoff(osd, msg);
5378 break;
5379 case CEPH_MSG_WATCH_NOTIFY:
5380 handle_watch_notify(osdc, msg);
5381 break;
5382
5383 default:
5384 pr_err("received unknown message type %d %s\n", type,
5385 ceph_msg_type_name(type));
5386 }
5387
5388 ceph_msg_put(msg);
5389}
5390
5391/* How much sparse data was requested? */
5392static u64 sparse_data_requested(struct ceph_osd_request *req)
5393{
5394 u64 len = 0;
5395
5396 if (req->r_flags & CEPH_OSD_FLAG_READ) {
5397 int i;
5398
5399 for (i = 0; i < req->r_num_ops; ++i) {
5400 struct ceph_osd_req_op *op = &req->r_ops[i];
5401
5402 if (op->op == CEPH_OSD_OP_SPARSE_READ)
5403 len += op->extent.length;
5404 }
5405 }
5406 return len;
5407}
5408
5409/*
5410 * Lookup and return message for incoming reply. Don't try to do
5411 * anything about a larger than preallocated data portion of the
5412 * message at the moment - for now, just skip the message.
5413 */
5414static struct ceph_msg *get_reply(struct ceph_connection *con,
5415 struct ceph_msg_header *hdr,
5416 int *skip)
5417{
5418 struct ceph_osd *osd = con->private;
5419 struct ceph_osd_client *osdc = osd->o_osdc;
5420 struct ceph_msg *m = NULL;
5421 struct ceph_osd_request *req;
5422 int front_len = le32_to_cpu(hdr->front_len);
5423 int data_len = le32_to_cpu(hdr->data_len);
5424 u64 tid = le64_to_cpu(hdr->tid);
5425 u64 srlen;
5426
5427 down_read(&osdc->lock);
5428 if (!osd_registered(osd)) {
5429 dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd);
5430 *skip = 1;
5431 goto out_unlock_osdc;
5432 }
5433 WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num));
5434
5435 mutex_lock(&osd->lock);
5436 req = lookup_request(&osd->o_requests, tid);
5437 if (!req) {
5438 dout("%s osd%d tid %llu unknown, skipping\n", __func__,
5439 osd->o_osd, tid);
5440 *skip = 1;
5441 goto out_unlock_session;
5442 }
5443
5444 ceph_msg_revoke_incoming(req->r_reply);
5445
5446 if (front_len > req->r_reply->front_alloc_len) {
5447 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
5448 __func__, osd->o_osd, req->r_tid, front_len,
5449 req->r_reply->front_alloc_len);
5450 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
5451 false);
5452 if (!m)
5453 goto out_unlock_session;
5454 ceph_msg_put(req->r_reply);
5455 req->r_reply = m;
5456 }
5457
5458 srlen = sparse_data_requested(req);
5459 if (!srlen && data_len > req->r_reply->data_length) {
5460 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
5461 __func__, osd->o_osd, req->r_tid, data_len,
5462 req->r_reply->data_length);
5463 m = NULL;
5464 *skip = 1;
5465 goto out_unlock_session;
5466 }
5467
5468 m = ceph_msg_get(req->r_reply);
5469 m->sparse_read_total = srlen;
5470
5471 dout("get_reply tid %lld %p\n", tid, m);
5472
5473out_unlock_session:
5474 mutex_unlock(&osd->lock);
5475out_unlock_osdc:
5476 up_read(&osdc->lock);
5477 return m;
5478}
5479
5480static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr)
5481{
5482 struct ceph_msg *m;
5483 int type = le16_to_cpu(hdr->type);
5484 u32 front_len = le32_to_cpu(hdr->front_len);
5485 u32 data_len = le32_to_cpu(hdr->data_len);
5486
5487 m = ceph_msg_new2(type, front_len, 1, GFP_NOIO, false);
5488 if (!m)
5489 return NULL;
5490
5491 if (data_len) {
5492 struct page **pages;
5493
5494 pages = ceph_alloc_page_vector(calc_pages_for(0, data_len),
5495 GFP_NOIO);
5496 if (IS_ERR(pages)) {
5497 ceph_msg_put(m);
5498 return NULL;
5499 }
5500
5501 ceph_msg_data_add_pages(m, pages, data_len, 0, true);
5502 }
5503
5504 return m;
5505}
5506
5507static struct ceph_msg *osd_alloc_msg(struct ceph_connection *con,
5508 struct ceph_msg_header *hdr,
5509 int *skip)
5510{
5511 struct ceph_osd *osd = con->private;
5512 int type = le16_to_cpu(hdr->type);
5513
5514 *skip = 0;
5515 switch (type) {
5516 case CEPH_MSG_OSD_MAP:
5517 case CEPH_MSG_OSD_BACKOFF:
5518 case CEPH_MSG_WATCH_NOTIFY:
5519 return alloc_msg_with_page_vector(hdr);
5520 case CEPH_MSG_OSD_OPREPLY:
5521 return get_reply(con, hdr, skip);
5522 default:
5523 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__,
5524 osd->o_osd, type);
5525 *skip = 1;
5526 return NULL;
5527 }
5528}
5529
5530/*
5531 * Wrappers to refcount containing ceph_osd struct
5532 */
5533static struct ceph_connection *osd_get_con(struct ceph_connection *con)
5534{
5535 struct ceph_osd *osd = con->private;
5536 if (get_osd(osd))
5537 return con;
5538 return NULL;
5539}
5540
5541static void osd_put_con(struct ceph_connection *con)
5542{
5543 struct ceph_osd *osd = con->private;
5544 put_osd(osd);
5545}
5546
5547/*
5548 * authentication
5549 */
5550
5551/*
5552 * Note: returned pointer is the address of a structure that's
5553 * managed separately. Caller must *not* attempt to free it.
5554 */
5555static struct ceph_auth_handshake *
5556osd_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5557{
5558 struct ceph_osd *o = con->private;
5559 struct ceph_osd_client *osdc = o->o_osdc;
5560 struct ceph_auth_client *ac = osdc->client->monc.auth;
5561 struct ceph_auth_handshake *auth = &o->o_auth;
5562 int ret;
5563
5564 ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5565 force_new, proto, NULL, NULL);
5566 if (ret)
5567 return ERR_PTR(ret);
5568
5569 return auth;
5570}
5571
5572static int osd_add_authorizer_challenge(struct ceph_connection *con,
5573 void *challenge_buf, int challenge_buf_len)
5574{
5575 struct ceph_osd *o = con->private;
5576 struct ceph_osd_client *osdc = o->o_osdc;
5577 struct ceph_auth_client *ac = osdc->client->monc.auth;
5578
5579 return ceph_auth_add_authorizer_challenge(ac, o->o_auth.authorizer,
5580 challenge_buf, challenge_buf_len);
5581}
5582
5583static int osd_verify_authorizer_reply(struct ceph_connection *con)
5584{
5585 struct ceph_osd *o = con->private;
5586 struct ceph_osd_client *osdc = o->o_osdc;
5587 struct ceph_auth_client *ac = osdc->client->monc.auth;
5588 struct ceph_auth_handshake *auth = &o->o_auth;
5589
5590 return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5591 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5592 NULL, NULL, NULL, NULL);
5593}
5594
5595static int osd_invalidate_authorizer(struct ceph_connection *con)
5596{
5597 struct ceph_osd *o = con->private;
5598 struct ceph_osd_client *osdc = o->o_osdc;
5599 struct ceph_auth_client *ac = osdc->client->monc.auth;
5600
5601 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
5602 return ceph_monc_validate_auth(&osdc->client->monc);
5603}
5604
5605static int osd_get_auth_request(struct ceph_connection *con,
5606 void *buf, int *buf_len,
5607 void **authorizer, int *authorizer_len)
5608{
5609 struct ceph_osd *o = con->private;
5610 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5611 struct ceph_auth_handshake *auth = &o->o_auth;
5612 int ret;
5613
5614 ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_OSD,
5615 buf, buf_len);
5616 if (ret)
5617 return ret;
5618
5619 *authorizer = auth->authorizer_buf;
5620 *authorizer_len = auth->authorizer_buf_len;
5621 return 0;
5622}
5623
5624static int osd_handle_auth_reply_more(struct ceph_connection *con,
5625 void *reply, int reply_len,
5626 void *buf, int *buf_len,
5627 void **authorizer, int *authorizer_len)
5628{
5629 struct ceph_osd *o = con->private;
5630 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5631 struct ceph_auth_handshake *auth = &o->o_auth;
5632 int ret;
5633
5634 ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5635 buf, buf_len);
5636 if (ret)
5637 return ret;
5638
5639 *authorizer = auth->authorizer_buf;
5640 *authorizer_len = auth->authorizer_buf_len;
5641 return 0;
5642}
5643
5644static int osd_handle_auth_done(struct ceph_connection *con,
5645 u64 global_id, void *reply, int reply_len,
5646 u8 *session_key, int *session_key_len,
5647 u8 *con_secret, int *con_secret_len)
5648{
5649 struct ceph_osd *o = con->private;
5650 struct ceph_auth_client *ac = o->o_osdc->client->monc.auth;
5651 struct ceph_auth_handshake *auth = &o->o_auth;
5652
5653 return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5654 session_key, session_key_len,
5655 con_secret, con_secret_len);
5656}
5657
5658static int osd_handle_auth_bad_method(struct ceph_connection *con,
5659 int used_proto, int result,
5660 const int *allowed_protos, int proto_cnt,
5661 const int *allowed_modes, int mode_cnt)
5662{
5663 struct ceph_osd *o = con->private;
5664 struct ceph_mon_client *monc = &o->o_osdc->client->monc;
5665 int ret;
5666
5667 if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_OSD,
5668 used_proto, result,
5669 allowed_protos, proto_cnt,
5670 allowed_modes, mode_cnt)) {
5671 ret = ceph_monc_validate_auth(monc);
5672 if (ret)
5673 return ret;
5674 }
5675
5676 return -EACCES;
5677}
5678
5679static void osd_reencode_message(struct ceph_msg *msg)
5680{
5681 int type = le16_to_cpu(msg->hdr.type);
5682
5683 if (type == CEPH_MSG_OSD_OP)
5684 encode_request_finish(msg);
5685}
5686
5687static int osd_sign_message(struct ceph_msg *msg)
5688{
5689 struct ceph_osd *o = msg->con->private;
5690 struct ceph_auth_handshake *auth = &o->o_auth;
5691
5692 return ceph_auth_sign_message(auth, msg);
5693}
5694
5695static int osd_check_message_signature(struct ceph_msg *msg)
5696{
5697 struct ceph_osd *o = msg->con->private;
5698 struct ceph_auth_handshake *auth = &o->o_auth;
5699
5700 return ceph_auth_check_message_signature(auth, msg);
5701}
5702
5703static void advance_cursor(struct ceph_msg_data_cursor *cursor, size_t len,
5704 bool zero)
5705{
5706 while (len) {
5707 struct page *page;
5708 size_t poff, plen;
5709
5710 page = ceph_msg_data_next(cursor, &poff, &plen);
5711 if (plen > len)
5712 plen = len;
5713 if (zero)
5714 zero_user_segment(page, poff, poff + plen);
5715 len -= plen;
5716 ceph_msg_data_advance(cursor, plen);
5717 }
5718}
5719
5720static int prep_next_sparse_read(struct ceph_connection *con,
5721 struct ceph_msg_data_cursor *cursor)
5722{
5723 struct ceph_osd *o = con->private;
5724 struct ceph_sparse_read *sr = &o->o_sparse_read;
5725 struct ceph_osd_request *req;
5726 struct ceph_osd_req_op *op;
5727
5728 spin_lock(&o->o_requests_lock);
5729 req = lookup_request(&o->o_requests, le64_to_cpu(con->in_msg->hdr.tid));
5730 if (!req) {
5731 spin_unlock(&o->o_requests_lock);
5732 return -EBADR;
5733 }
5734
5735 if (o->o_sparse_op_idx < 0) {
5736 dout("%s: [%d] starting new sparse read req\n",
5737 __func__, o->o_osd);
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
5806static bool sparse_extent_map_valid(struct ceph_sparse_read *sr)
5807{
5808 u64 req_end, pos;
5809 int i;
5810
5811 if (check_add_overflow(sr->sr_req_off, sr->sr_req_len, &req_end))
5812 return false;
5813
5814 pos = sr->sr_req_off;
5815 for (i = 0; i < sr->sr_count; i++) {
5816 struct ceph_sparse_extent *ext = &sr->sr_extent[i];
5817 u64 end;
5818
5819 if (ext->off < pos)
5820 return false;
5821 if (check_add_overflow(ext->off, ext->len, &end))
5822 return false;
5823 if (end > req_end)
5824 return false;
5825 pos = end;
5826 }
5827
5828 return true;
5829}
5830
5831static int osd_sparse_read(struct ceph_connection *con,
5832 struct ceph_msg_data_cursor *cursor,
5833 char **pbuf)
5834{
5835 struct ceph_osd *o = con->private;
5836 struct ceph_sparse_read *sr = &o->o_sparse_read;
5837 u32 count = sr->sr_count;
5838 u64 eoff, elen, len = 0;
5839 int i, ret;
5840
5841 switch (sr->sr_state) {
5842 case CEPH_SPARSE_READ_HDR:
5843next_op:
5844 ret = prep_next_sparse_read(con, cursor);
5845 if (ret <= 0)
5846 return ret;
5847
5848 /* number of extents */
5849 ret = sizeof(sr->sr_count);
5850 *pbuf = (char *)&sr->sr_count;
5851 sr->sr_state = CEPH_SPARSE_READ_EXTENTS;
5852 break;
5853 case CEPH_SPARSE_READ_EXTENTS:
5854 /* Convert sr_count to host-endian */
5855 count = le32_to_cpu((__force __le32)sr->sr_count);
5856 sr->sr_count = count;
5857 dout("[%d] got %u extents\n", o->o_osd, count);
5858
5859 if (count > 0) {
5860 if (!sr->sr_extent || count > sr->sr_ext_len) {
5861 /* no extent array provided, or too short */
5862 kfree(sr->sr_extent);
5863 sr->sr_extent = kmalloc_objs(*sr->sr_extent,
5864 count, GFP_NOIO);
5865 if (!sr->sr_extent) {
5866 pr_err("%s: failed to allocate %u extents\n",
5867 __func__, count);
5868 return -ENOMEM;
5869 }
5870 sr->sr_ext_len = count;
5871 }
5872 ret = count * sizeof(*sr->sr_extent);
5873 *pbuf = (char *)sr->sr_extent;
5874 sr->sr_state = CEPH_SPARSE_READ_DATA_LEN;
5875 break;
5876 }
5877 /* No extents? Read data len */
5878 fallthrough;
5879 case CEPH_SPARSE_READ_DATA_LEN:
5880 convert_extent_map(sr);
5881 if (!sparse_extent_map_valid(sr)) {
5882 pr_warn_ratelimited("invalid sparse extent map\n");
5883 return -EREMOTEIO;
5884 }
5885 ret = sizeof(sr->sr_datalen);
5886 *pbuf = (char *)&sr->sr_datalen;
5887 sr->sr_state = CEPH_SPARSE_READ_DATA_PRE;
5888 break;
5889 case CEPH_SPARSE_READ_DATA_PRE:
5890 /* Convert sr_datalen to host-endian */
5891 sr->sr_datalen = le32_to_cpu((__force __le32)sr->sr_datalen);
5892 for (i = 0; i < count; i++)
5893 len += sr->sr_extent[i].len;
5894 if (sr->sr_datalen != len) {
5895 pr_warn_ratelimited("data len %u != extent len %llu\n",
5896 sr->sr_datalen, len);
5897 return -EREMOTEIO;
5898 }
5899 sr->sr_state = CEPH_SPARSE_READ_DATA;
5900 fallthrough;
5901 case CEPH_SPARSE_READ_DATA:
5902 if (sr->sr_index >= count) {
5903 sr->sr_state = CEPH_SPARSE_READ_HDR;
5904 goto next_op;
5905 }
5906
5907 eoff = sr->sr_extent[sr->sr_index].off;
5908 elen = sr->sr_extent[sr->sr_index].len;
5909
5910 dout("[%d] ext %d off 0x%llx len 0x%llx\n",
5911 o->o_osd, sr->sr_index, eoff, elen);
5912
5913 if (elen > INT_MAX) {
5914 dout("Sparse read extent length too long (0x%llx)\n",
5915 elen);
5916 return -EREMOTEIO;
5917 }
5918
5919 /* zero out anything from sr_pos to start of extent */
5920 if (sr->sr_pos < eoff)
5921 advance_cursor(cursor, eoff - sr->sr_pos, true);
5922
5923 /* Set position to end of extent */
5924 sr->sr_pos = eoff + elen;
5925
5926 /* send back the new length and nullify the ptr */
5927 cursor->sr_resid = elen;
5928 ret = elen;
5929 *pbuf = NULL;
5930
5931 /* Bump the array index */
5932 ++sr->sr_index;
5933 break;
5934 }
5935 return ret;
5936}
5937
5938static const struct ceph_connection_operations osd_con_ops = {
5939 .get = osd_get_con,
5940 .put = osd_put_con,
5941 .sparse_read = osd_sparse_read,
5942 .alloc_msg = osd_alloc_msg,
5943 .dispatch = osd_dispatch,
5944 .fault = osd_fault,
5945 .reencode_message = osd_reencode_message,
5946 .get_authorizer = osd_get_authorizer,
5947 .add_authorizer_challenge = osd_add_authorizer_challenge,
5948 .verify_authorizer_reply = osd_verify_authorizer_reply,
5949 .invalidate_authorizer = osd_invalidate_authorizer,
5950 .sign_message = osd_sign_message,
5951 .check_message_signature = osd_check_message_signature,
5952 .get_auth_request = osd_get_auth_request,
5953 .handle_auth_reply_more = osd_handle_auth_reply_more,
5954 .handle_auth_done = osd_handle_auth_done,
5955 .handle_auth_bad_method = osd_handle_auth_bad_method,
5956};