2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
6 #include <linux/highmem.h>
8 #include <linux/pagemap.h>
9 #include <linux/slab.h>
10 #include <linux/uaccess.h>
12 #include <linux/bio.h>
15 #include <linux/ceph/libceph.h>
16 #include <linux/ceph/osd_client.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/auth.h>
20 #include <linux/ceph/pagelist.h>
22 #define OSD_OP_FRONT_LEN 4096
23 #define OSD_OPREPLY_FRONT_LEN 512
25 static struct kmem_cache
*ceph_osd_request_cache
;
27 static const struct ceph_connection_operations osd_con_ops
;
29 static void __send_queued(struct ceph_osd_client
*osdc
);
30 static int __reset_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
);
31 static void __register_request(struct ceph_osd_client
*osdc
,
32 struct ceph_osd_request
*req
);
33 static void __unregister_linger_request(struct ceph_osd_client
*osdc
,
34 struct ceph_osd_request
*req
);
35 static void __send_request(struct ceph_osd_client
*osdc
,
36 struct ceph_osd_request
*req
);
39 * Implement client access to distributed object storage cluster.
41 * All data objects are stored within a cluster/cloud of OSDs, or
42 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
43 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
44 * remote daemons serving up and coordinating consistent and safe
47 * Cluster membership and the mapping of data objects onto storage devices
48 * are described by the osd map.
50 * We keep track of pending OSD requests (read, write), resubmit
51 * requests to different OSDs when the cluster topology/data layout
52 * change, or retry the affected requests when the communications
53 * channel with an OSD is reset.
57 * calculate the mapping of a file extent onto an object, and fill out the
58 * request accordingly. shorten extent as necessary if it crosses an
61 * fill osd op in request message.
63 static int calc_layout(struct ceph_file_layout
*layout
, u64 off
, u64
*plen
,
64 u64
*objnum
, u64
*objoff
, u64
*objlen
)
70 r
= ceph_calc_file_object_mapping(layout
, off
, orig_len
, objnum
,
74 if (*objlen
< orig_len
) {
76 dout(" skipping last %llu, final file extent %llu~%llu\n",
77 orig_len
- *plen
, off
, *plen
);
80 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum
, *objoff
, *objlen
);
85 static void ceph_osd_data_init(struct ceph_osd_data
*osd_data
)
87 memset(osd_data
, 0, sizeof (*osd_data
));
88 osd_data
->type
= CEPH_OSD_DATA_TYPE_NONE
;
91 static void ceph_osd_data_pages_init(struct ceph_osd_data
*osd_data
,
92 struct page
**pages
, u64 length
, u32 alignment
,
93 bool pages_from_pool
, bool own_pages
)
95 osd_data
->type
= CEPH_OSD_DATA_TYPE_PAGES
;
96 osd_data
->pages
= pages
;
97 osd_data
->length
= length
;
98 osd_data
->alignment
= alignment
;
99 osd_data
->pages_from_pool
= pages_from_pool
;
100 osd_data
->own_pages
= own_pages
;
103 static void ceph_osd_data_pagelist_init(struct ceph_osd_data
*osd_data
,
104 struct ceph_pagelist
*pagelist
)
106 osd_data
->type
= CEPH_OSD_DATA_TYPE_PAGELIST
;
107 osd_data
->pagelist
= pagelist
;
111 static void ceph_osd_data_bio_init(struct ceph_osd_data
*osd_data
,
112 struct bio
*bio
, size_t bio_length
)
114 osd_data
->type
= CEPH_OSD_DATA_TYPE_BIO
;
116 osd_data
->bio_length
= bio_length
;
118 #endif /* CONFIG_BLOCK */
120 #define osd_req_op_data(oreq, whch, typ, fld) \
122 BUG_ON(whch >= (oreq)->r_num_ops); \
123 &(oreq)->r_ops[whch].typ.fld; \
126 static struct ceph_osd_data
*
127 osd_req_op_raw_data_in(struct ceph_osd_request
*osd_req
, unsigned int which
)
129 BUG_ON(which
>= osd_req
->r_num_ops
);
131 return &osd_req
->r_ops
[which
].raw_data_in
;
134 struct ceph_osd_data
*
135 osd_req_op_extent_osd_data(struct ceph_osd_request
*osd_req
,
138 return osd_req_op_data(osd_req
, which
, extent
, osd_data
);
140 EXPORT_SYMBOL(osd_req_op_extent_osd_data
);
142 struct ceph_osd_data
*
143 osd_req_op_cls_response_data(struct ceph_osd_request
*osd_req
,
146 return osd_req_op_data(osd_req
, which
, cls
, response_data
);
148 EXPORT_SYMBOL(osd_req_op_cls_response_data
); /* ??? */
150 void osd_req_op_raw_data_in_pages(struct ceph_osd_request
*osd_req
,
151 unsigned int which
, struct page
**pages
,
152 u64 length
, u32 alignment
,
153 bool pages_from_pool
, bool own_pages
)
155 struct ceph_osd_data
*osd_data
;
157 osd_data
= osd_req_op_raw_data_in(osd_req
, which
);
158 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
159 pages_from_pool
, own_pages
);
161 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages
);
163 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request
*osd_req
,
164 unsigned int which
, struct page
**pages
,
165 u64 length
, u32 alignment
,
166 bool pages_from_pool
, bool own_pages
)
168 struct ceph_osd_data
*osd_data
;
170 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
171 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
172 pages_from_pool
, own_pages
);
174 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages
);
176 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request
*osd_req
,
177 unsigned int which
, struct ceph_pagelist
*pagelist
)
179 struct ceph_osd_data
*osd_data
;
181 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
182 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
184 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist
);
187 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request
*osd_req
,
188 unsigned int which
, struct bio
*bio
, size_t bio_length
)
190 struct ceph_osd_data
*osd_data
;
192 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
193 ceph_osd_data_bio_init(osd_data
, bio
, bio_length
);
195 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio
);
196 #endif /* CONFIG_BLOCK */
198 static void osd_req_op_cls_request_info_pagelist(
199 struct ceph_osd_request
*osd_req
,
200 unsigned int which
, struct ceph_pagelist
*pagelist
)
202 struct ceph_osd_data
*osd_data
;
204 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_info
);
205 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
208 void osd_req_op_cls_request_data_pagelist(
209 struct ceph_osd_request
*osd_req
,
210 unsigned int which
, struct ceph_pagelist
*pagelist
)
212 struct ceph_osd_data
*osd_data
;
214 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_data
);
215 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
217 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist
);
219 void osd_req_op_cls_request_data_pages(struct ceph_osd_request
*osd_req
,
220 unsigned int which
, struct page
**pages
, u64 length
,
221 u32 alignment
, bool pages_from_pool
, bool own_pages
)
223 struct ceph_osd_data
*osd_data
;
225 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_data
);
226 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
227 pages_from_pool
, own_pages
);
229 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages
);
231 void osd_req_op_cls_response_data_pages(struct ceph_osd_request
*osd_req
,
232 unsigned int which
, struct page
**pages
, u64 length
,
233 u32 alignment
, bool pages_from_pool
, bool own_pages
)
235 struct ceph_osd_data
*osd_data
;
237 osd_data
= osd_req_op_data(osd_req
, which
, cls
, response_data
);
238 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
239 pages_from_pool
, own_pages
);
241 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages
);
243 static u64
ceph_osd_data_length(struct ceph_osd_data
*osd_data
)
245 switch (osd_data
->type
) {
246 case CEPH_OSD_DATA_TYPE_NONE
:
248 case CEPH_OSD_DATA_TYPE_PAGES
:
249 return osd_data
->length
;
250 case CEPH_OSD_DATA_TYPE_PAGELIST
:
251 return (u64
)osd_data
->pagelist
->length
;
253 case CEPH_OSD_DATA_TYPE_BIO
:
254 return (u64
)osd_data
->bio_length
;
255 #endif /* CONFIG_BLOCK */
257 WARN(true, "unrecognized data type %d\n", (int)osd_data
->type
);
262 static void ceph_osd_data_release(struct ceph_osd_data
*osd_data
)
264 if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGES
&& osd_data
->own_pages
) {
267 num_pages
= calc_pages_for((u64
)osd_data
->alignment
,
268 (u64
)osd_data
->length
);
269 ceph_release_page_vector(osd_data
->pages
, num_pages
);
271 ceph_osd_data_init(osd_data
);
274 static void osd_req_op_data_release(struct ceph_osd_request
*osd_req
,
277 struct ceph_osd_req_op
*op
;
279 BUG_ON(which
>= osd_req
->r_num_ops
);
280 op
= &osd_req
->r_ops
[which
];
283 case CEPH_OSD_OP_READ
:
284 case CEPH_OSD_OP_WRITE
:
285 ceph_osd_data_release(&op
->extent
.osd_data
);
287 case CEPH_OSD_OP_CALL
:
288 ceph_osd_data_release(&op
->cls
.request_info
);
289 ceph_osd_data_release(&op
->cls
.request_data
);
290 ceph_osd_data_release(&op
->cls
.response_data
);
300 void ceph_osdc_release_request(struct kref
*kref
)
302 struct ceph_osd_request
*req
;
305 req
= container_of(kref
, struct ceph_osd_request
, r_kref
);
307 ceph_msg_put(req
->r_request
);
309 ceph_msg_revoke_incoming(req
->r_reply
);
310 ceph_msg_put(req
->r_reply
);
313 for (which
= 0; which
< req
->r_num_ops
; which
++)
314 osd_req_op_data_release(req
, which
);
316 ceph_put_snap_context(req
->r_snapc
);
318 mempool_free(req
, req
->r_osdc
->req_mempool
);
320 kmem_cache_free(ceph_osd_request_cache
, req
);
323 EXPORT_SYMBOL(ceph_osdc_release_request
);
325 struct ceph_osd_request
*ceph_osdc_alloc_request(struct ceph_osd_client
*osdc
,
326 struct ceph_snap_context
*snapc
,
327 unsigned int num_ops
,
331 struct ceph_osd_request
*req
;
332 struct ceph_msg
*msg
;
335 BUILD_BUG_ON(CEPH_OSD_MAX_OP
> U16_MAX
);
336 BUG_ON(num_ops
> CEPH_OSD_MAX_OP
);
338 msg_size
= 4 + 4 + 8 + 8 + 4+8;
339 msg_size
+= 2 + 4 + 8 + 4 + 4; /* oloc */
340 msg_size
+= 1 + 8 + 4 + 4; /* pg_t */
341 msg_size
+= 4 + MAX_OBJ_NAME_SIZE
;
342 msg_size
+= 2 + num_ops
*sizeof(struct ceph_osd_op
);
343 msg_size
+= 8; /* snapid */
344 msg_size
+= 8; /* snap_seq */
345 msg_size
+= 8 * (snapc
? snapc
->num_snaps
: 0); /* snaps */
349 req
= mempool_alloc(osdc
->req_mempool
, gfp_flags
);
350 memset(req
, 0, sizeof(*req
));
352 req
= kmem_cache_zalloc(ceph_osd_request_cache
, gfp_flags
);
358 req
->r_mempool
= use_mempool
;
359 req
->r_num_ops
= num_ops
;
361 kref_init(&req
->r_kref
);
362 init_completion(&req
->r_completion
);
363 init_completion(&req
->r_safe_completion
);
364 RB_CLEAR_NODE(&req
->r_node
);
365 INIT_LIST_HEAD(&req
->r_unsafe_item
);
366 INIT_LIST_HEAD(&req
->r_linger_item
);
367 INIT_LIST_HEAD(&req
->r_linger_osd
);
368 INIT_LIST_HEAD(&req
->r_req_lru_item
);
369 INIT_LIST_HEAD(&req
->r_osd_item
);
371 /* create reply message */
373 msg
= ceph_msgpool_get(&osdc
->msgpool_op_reply
, 0);
375 msg
= ceph_msg_new(CEPH_MSG_OSD_OPREPLY
,
376 OSD_OPREPLY_FRONT_LEN
, gfp_flags
, true);
378 ceph_osdc_put_request(req
);
383 /* create request message; allow space for oid */
385 msg
= ceph_msgpool_get(&osdc
->msgpool_op
, 0);
387 msg
= ceph_msg_new(CEPH_MSG_OSD_OP
, msg_size
, gfp_flags
, true);
389 ceph_osdc_put_request(req
);
393 memset(msg
->front
.iov_base
, 0, msg
->front
.iov_len
);
395 req
->r_request
= msg
;
399 EXPORT_SYMBOL(ceph_osdc_alloc_request
);
401 static bool osd_req_opcode_valid(u16 opcode
)
404 case CEPH_OSD_OP_READ
:
405 case CEPH_OSD_OP_STAT
:
406 case CEPH_OSD_OP_MAPEXT
:
407 case CEPH_OSD_OP_MASKTRUNC
:
408 case CEPH_OSD_OP_SPARSE_READ
:
409 case CEPH_OSD_OP_NOTIFY
:
410 case CEPH_OSD_OP_NOTIFY_ACK
:
411 case CEPH_OSD_OP_ASSERT_VER
:
412 case CEPH_OSD_OP_WRITE
:
413 case CEPH_OSD_OP_WRITEFULL
:
414 case CEPH_OSD_OP_TRUNCATE
:
415 case CEPH_OSD_OP_ZERO
:
416 case CEPH_OSD_OP_DELETE
:
417 case CEPH_OSD_OP_APPEND
:
418 case CEPH_OSD_OP_STARTSYNC
:
419 case CEPH_OSD_OP_SETTRUNC
:
420 case CEPH_OSD_OP_TRIMTRUNC
:
421 case CEPH_OSD_OP_TMAPUP
:
422 case CEPH_OSD_OP_TMAPPUT
:
423 case CEPH_OSD_OP_TMAPGET
:
424 case CEPH_OSD_OP_CREATE
:
425 case CEPH_OSD_OP_ROLLBACK
:
426 case CEPH_OSD_OP_WATCH
:
427 case CEPH_OSD_OP_OMAPGETKEYS
:
428 case CEPH_OSD_OP_OMAPGETVALS
:
429 case CEPH_OSD_OP_OMAPGETHEADER
:
430 case CEPH_OSD_OP_OMAPGETVALSBYKEYS
:
431 case CEPH_OSD_OP_OMAPSETVALS
:
432 case CEPH_OSD_OP_OMAPSETHEADER
:
433 case CEPH_OSD_OP_OMAPCLEAR
:
434 case CEPH_OSD_OP_OMAPRMKEYS
:
435 case CEPH_OSD_OP_OMAP_CMP
:
436 case CEPH_OSD_OP_CLONERANGE
:
437 case CEPH_OSD_OP_ASSERT_SRC_VERSION
:
438 case CEPH_OSD_OP_SRC_CMPXATTR
:
439 case CEPH_OSD_OP_GETXATTR
:
440 case CEPH_OSD_OP_GETXATTRS
:
441 case CEPH_OSD_OP_CMPXATTR
:
442 case CEPH_OSD_OP_SETXATTR
:
443 case CEPH_OSD_OP_SETXATTRS
:
444 case CEPH_OSD_OP_RESETXATTRS
:
445 case CEPH_OSD_OP_RMXATTR
:
446 case CEPH_OSD_OP_PULL
:
447 case CEPH_OSD_OP_PUSH
:
448 case CEPH_OSD_OP_BALANCEREADS
:
449 case CEPH_OSD_OP_UNBALANCEREADS
:
450 case CEPH_OSD_OP_SCRUB
:
451 case CEPH_OSD_OP_SCRUB_RESERVE
:
452 case CEPH_OSD_OP_SCRUB_UNRESERVE
:
453 case CEPH_OSD_OP_SCRUB_STOP
:
454 case CEPH_OSD_OP_SCRUB_MAP
:
455 case CEPH_OSD_OP_WRLOCK
:
456 case CEPH_OSD_OP_WRUNLOCK
:
457 case CEPH_OSD_OP_RDLOCK
:
458 case CEPH_OSD_OP_RDUNLOCK
:
459 case CEPH_OSD_OP_UPLOCK
:
460 case CEPH_OSD_OP_DNLOCK
:
461 case CEPH_OSD_OP_CALL
:
462 case CEPH_OSD_OP_PGLS
:
463 case CEPH_OSD_OP_PGLS_FILTER
:
471 * This is an osd op init function for opcodes that have no data or
472 * other information associated with them. It also serves as a
473 * common init routine for all the other init functions, below.
475 static struct ceph_osd_req_op
*
476 _osd_req_op_init(struct ceph_osd_request
*osd_req
, unsigned int which
,
479 struct ceph_osd_req_op
*op
;
481 BUG_ON(which
>= osd_req
->r_num_ops
);
482 BUG_ON(!osd_req_opcode_valid(opcode
));
484 op
= &osd_req
->r_ops
[which
];
485 memset(op
, 0, sizeof (*op
));
491 void osd_req_op_init(struct ceph_osd_request
*osd_req
,
492 unsigned int which
, u16 opcode
)
494 (void)_osd_req_op_init(osd_req
, which
, opcode
);
496 EXPORT_SYMBOL(osd_req_op_init
);
498 void osd_req_op_extent_init(struct ceph_osd_request
*osd_req
,
499 unsigned int which
, u16 opcode
,
500 u64 offset
, u64 length
,
501 u64 truncate_size
, u32 truncate_seq
)
503 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
, opcode
);
504 size_t payload_len
= 0;
506 BUG_ON(opcode
!= CEPH_OSD_OP_READ
&& opcode
!= CEPH_OSD_OP_WRITE
&&
507 opcode
!= CEPH_OSD_OP_DELETE
&& opcode
!= CEPH_OSD_OP_ZERO
&&
508 opcode
!= CEPH_OSD_OP_TRUNCATE
);
510 op
->extent
.offset
= offset
;
511 op
->extent
.length
= length
;
512 op
->extent
.truncate_size
= truncate_size
;
513 op
->extent
.truncate_seq
= truncate_seq
;
514 if (opcode
== CEPH_OSD_OP_WRITE
)
515 payload_len
+= length
;
517 op
->payload_len
= payload_len
;
519 EXPORT_SYMBOL(osd_req_op_extent_init
);
521 void osd_req_op_extent_update(struct ceph_osd_request
*osd_req
,
522 unsigned int which
, u64 length
)
524 struct ceph_osd_req_op
*op
;
527 BUG_ON(which
>= osd_req
->r_num_ops
);
528 op
= &osd_req
->r_ops
[which
];
529 previous
= op
->extent
.length
;
531 if (length
== previous
)
532 return; /* Nothing to do */
533 BUG_ON(length
> previous
);
535 op
->extent
.length
= length
;
536 op
->payload_len
-= previous
- length
;
538 EXPORT_SYMBOL(osd_req_op_extent_update
);
540 void osd_req_op_cls_init(struct ceph_osd_request
*osd_req
, unsigned int which
,
541 u16 opcode
, const char *class, const char *method
)
543 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
, opcode
);
544 struct ceph_pagelist
*pagelist
;
545 size_t payload_len
= 0;
548 BUG_ON(opcode
!= CEPH_OSD_OP_CALL
);
550 pagelist
= kmalloc(sizeof (*pagelist
), GFP_NOFS
);
552 ceph_pagelist_init(pagelist
);
554 op
->cls
.class_name
= class;
555 size
= strlen(class);
556 BUG_ON(size
> (size_t) U8_MAX
);
557 op
->cls
.class_len
= size
;
558 ceph_pagelist_append(pagelist
, class, size
);
561 op
->cls
.method_name
= method
;
562 size
= strlen(method
);
563 BUG_ON(size
> (size_t) U8_MAX
);
564 op
->cls
.method_len
= size
;
565 ceph_pagelist_append(pagelist
, method
, size
);
568 osd_req_op_cls_request_info_pagelist(osd_req
, which
, pagelist
);
570 op
->cls
.argc
= 0; /* currently unused */
572 op
->payload_len
= payload_len
;
574 EXPORT_SYMBOL(osd_req_op_cls_init
);
576 void osd_req_op_watch_init(struct ceph_osd_request
*osd_req
,
577 unsigned int which
, u16 opcode
,
578 u64 cookie
, u64 version
, int flag
)
580 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
, opcode
);
582 BUG_ON(opcode
!= CEPH_OSD_OP_NOTIFY_ACK
&& opcode
!= CEPH_OSD_OP_WATCH
);
584 op
->watch
.cookie
= cookie
;
585 op
->watch
.ver
= version
;
586 if (opcode
== CEPH_OSD_OP_WATCH
&& flag
)
587 op
->watch
.flag
= (u8
)1;
589 EXPORT_SYMBOL(osd_req_op_watch_init
);
591 static void ceph_osdc_msg_data_add(struct ceph_msg
*msg
,
592 struct ceph_osd_data
*osd_data
)
594 u64 length
= ceph_osd_data_length(osd_data
);
596 if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGES
) {
597 BUG_ON(length
> (u64
) SIZE_MAX
);
599 ceph_msg_data_add_pages(msg
, osd_data
->pages
,
600 length
, osd_data
->alignment
);
601 } else if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGELIST
) {
603 ceph_msg_data_add_pagelist(msg
, osd_data
->pagelist
);
605 } else if (osd_data
->type
== CEPH_OSD_DATA_TYPE_BIO
) {
606 ceph_msg_data_add_bio(msg
, osd_data
->bio
, length
);
609 BUG_ON(osd_data
->type
!= CEPH_OSD_DATA_TYPE_NONE
);
613 static u64
osd_req_encode_op(struct ceph_osd_request
*req
,
614 struct ceph_osd_op
*dst
, unsigned int which
)
616 struct ceph_osd_req_op
*src
;
617 struct ceph_osd_data
*osd_data
;
618 u64 request_data_len
= 0;
621 BUG_ON(which
>= req
->r_num_ops
);
622 src
= &req
->r_ops
[which
];
623 if (WARN_ON(!osd_req_opcode_valid(src
->op
))) {
624 pr_err("unrecognized osd opcode %d\n", src
->op
);
630 case CEPH_OSD_OP_STAT
:
631 osd_data
= &src
->raw_data_in
;
632 ceph_osdc_msg_data_add(req
->r_reply
, osd_data
);
634 case CEPH_OSD_OP_READ
:
635 case CEPH_OSD_OP_WRITE
:
636 case CEPH_OSD_OP_ZERO
:
637 case CEPH_OSD_OP_DELETE
:
638 case CEPH_OSD_OP_TRUNCATE
:
639 if (src
->op
== CEPH_OSD_OP_WRITE
)
640 request_data_len
= src
->extent
.length
;
641 dst
->extent
.offset
= cpu_to_le64(src
->extent
.offset
);
642 dst
->extent
.length
= cpu_to_le64(src
->extent
.length
);
643 dst
->extent
.truncate_size
=
644 cpu_to_le64(src
->extent
.truncate_size
);
645 dst
->extent
.truncate_seq
=
646 cpu_to_le32(src
->extent
.truncate_seq
);
647 osd_data
= &src
->extent
.osd_data
;
648 if (src
->op
== CEPH_OSD_OP_WRITE
)
649 ceph_osdc_msg_data_add(req
->r_request
, osd_data
);
651 ceph_osdc_msg_data_add(req
->r_reply
, osd_data
);
653 case CEPH_OSD_OP_CALL
:
654 dst
->cls
.class_len
= src
->cls
.class_len
;
655 dst
->cls
.method_len
= src
->cls
.method_len
;
656 osd_data
= &src
->cls
.request_info
;
657 ceph_osdc_msg_data_add(req
->r_request
, osd_data
);
658 BUG_ON(osd_data
->type
!= CEPH_OSD_DATA_TYPE_PAGELIST
);
659 request_data_len
= osd_data
->pagelist
->length
;
661 osd_data
= &src
->cls
.request_data
;
662 data_length
= ceph_osd_data_length(osd_data
);
664 BUG_ON(osd_data
->type
== CEPH_OSD_DATA_TYPE_NONE
);
665 dst
->cls
.indata_len
= cpu_to_le32(data_length
);
666 ceph_osdc_msg_data_add(req
->r_request
, osd_data
);
667 src
->payload_len
+= data_length
;
668 request_data_len
+= data_length
;
670 osd_data
= &src
->cls
.response_data
;
671 ceph_osdc_msg_data_add(req
->r_reply
, osd_data
);
673 case CEPH_OSD_OP_STARTSYNC
:
675 case CEPH_OSD_OP_NOTIFY_ACK
:
676 case CEPH_OSD_OP_WATCH
:
677 dst
->watch
.cookie
= cpu_to_le64(src
->watch
.cookie
);
678 dst
->watch
.ver
= cpu_to_le64(src
->watch
.ver
);
679 dst
->watch
.flag
= src
->watch
.flag
;
682 pr_err("unsupported osd opcode %s\n",
683 ceph_osd_op_name(src
->op
));
688 dst
->op
= cpu_to_le16(src
->op
);
689 dst
->payload_len
= cpu_to_le32(src
->payload_len
);
691 return request_data_len
;
695 * build new request AND message, calculate layout, and adjust file
698 * if the file was recently truncated, we include information about its
699 * old and new size so that the object can be updated appropriately. (we
700 * avoid synchronously deleting truncated objects because it's slow.)
702 * if @do_sync, include a 'startsync' command so that the osd will flush
705 struct ceph_osd_request
*ceph_osdc_new_request(struct ceph_osd_client
*osdc
,
706 struct ceph_file_layout
*layout
,
707 struct ceph_vino vino
,
708 u64 off
, u64
*plen
, int num_ops
,
709 int opcode
, int flags
,
710 struct ceph_snap_context
*snapc
,
715 struct ceph_osd_request
*req
;
723 BUG_ON(opcode
!= CEPH_OSD_OP_READ
&& opcode
!= CEPH_OSD_OP_WRITE
&&
724 opcode
!= CEPH_OSD_OP_DELETE
&& opcode
!= CEPH_OSD_OP_ZERO
&&
725 opcode
!= CEPH_OSD_OP_TRUNCATE
);
727 req
= ceph_osdc_alloc_request(osdc
, snapc
, num_ops
, use_mempool
,
730 return ERR_PTR(-ENOMEM
);
732 req
->r_flags
= flags
;
734 /* calculate max write size */
735 r
= calc_layout(layout
, off
, plen
, &objnum
, &objoff
, &objlen
);
737 ceph_osdc_put_request(req
);
741 object_size
= le32_to_cpu(layout
->fl_object_size
);
742 object_base
= off
- objoff
;
743 if (!(truncate_seq
== 1 && truncate_size
== -1ULL)) {
744 if (truncate_size
<= object_base
) {
747 truncate_size
-= object_base
;
748 if (truncate_size
> object_size
)
749 truncate_size
= object_size
;
753 osd_req_op_extent_init(req
, 0, opcode
, objoff
, objlen
,
754 truncate_size
, truncate_seq
);
757 * A second op in the ops array means the caller wants to
758 * also issue a include a 'startsync' command so that the
759 * osd will flush data quickly.
762 osd_req_op_init(req
, 1, CEPH_OSD_OP_STARTSYNC
);
764 req
->r_file_layout
= *layout
; /* keep a copy */
766 snprintf(req
->r_oid
, sizeof(req
->r_oid
), "%llx.%08llx",
768 req
->r_oid_len
= strlen(req
->r_oid
);
772 EXPORT_SYMBOL(ceph_osdc_new_request
);
775 * We keep osd requests in an rbtree, sorted by ->r_tid.
777 static void __insert_request(struct ceph_osd_client
*osdc
,
778 struct ceph_osd_request
*new)
780 struct rb_node
**p
= &osdc
->requests
.rb_node
;
781 struct rb_node
*parent
= NULL
;
782 struct ceph_osd_request
*req
= NULL
;
786 req
= rb_entry(parent
, struct ceph_osd_request
, r_node
);
787 if (new->r_tid
< req
->r_tid
)
789 else if (new->r_tid
> req
->r_tid
)
795 rb_link_node(&new->r_node
, parent
, p
);
796 rb_insert_color(&new->r_node
, &osdc
->requests
);
799 static struct ceph_osd_request
*__lookup_request(struct ceph_osd_client
*osdc
,
802 struct ceph_osd_request
*req
;
803 struct rb_node
*n
= osdc
->requests
.rb_node
;
806 req
= rb_entry(n
, struct ceph_osd_request
, r_node
);
807 if (tid
< req
->r_tid
)
809 else if (tid
> req
->r_tid
)
817 static struct ceph_osd_request
*
818 __lookup_request_ge(struct ceph_osd_client
*osdc
,
821 struct ceph_osd_request
*req
;
822 struct rb_node
*n
= osdc
->requests
.rb_node
;
825 req
= rb_entry(n
, struct ceph_osd_request
, r_node
);
826 if (tid
< req
->r_tid
) {
830 } else if (tid
> req
->r_tid
) {
840 * Resubmit requests pending on the given osd.
842 static void __kick_osd_requests(struct ceph_osd_client
*osdc
,
843 struct ceph_osd
*osd
)
845 struct ceph_osd_request
*req
, *nreq
;
849 dout("__kick_osd_requests osd%d\n", osd
->o_osd
);
850 err
= __reset_osd(osdc
, osd
);
854 * Build up a list of requests to resend by traversing the
855 * osd's list of requests. Requests for a given object are
856 * sent in tid order, and that is also the order they're
857 * kept on this list. Therefore all requests that are in
858 * flight will be found first, followed by all requests that
859 * have not yet been sent. And to resend requests while
860 * preserving this order we will want to put any sent
861 * requests back on the front of the osd client's unsent
864 * So we build a separate ordered list of already-sent
865 * requests for the affected osd and splice it onto the
866 * front of the osd client's unsent list. Once we've seen a
867 * request that has not yet been sent we're done. Those
868 * requests are already sitting right where they belong.
870 list_for_each_entry(req
, &osd
->o_requests
, r_osd_item
) {
873 list_move_tail(&req
->r_req_lru_item
, &resend
);
874 dout("requeueing %p tid %llu osd%d\n", req
, req
->r_tid
,
877 req
->r_flags
|= CEPH_OSD_FLAG_RETRY
;
879 list_splice(&resend
, &osdc
->req_unsent
);
882 * Linger requests are re-registered before sending, which
883 * sets up a new tid for each. We add them to the unsent
884 * list at the end to keep things in tid order.
886 list_for_each_entry_safe(req
, nreq
, &osd
->o_linger_requests
,
889 * reregister request prior to unregistering linger so
890 * that r_osd is preserved.
892 BUG_ON(!list_empty(&req
->r_req_lru_item
));
893 __register_request(osdc
, req
);
894 list_add_tail(&req
->r_req_lru_item
, &osdc
->req_unsent
);
895 list_add_tail(&req
->r_osd_item
, &req
->r_osd
->o_requests
);
896 __unregister_linger_request(osdc
, req
);
897 dout("requeued lingering %p tid %llu osd%d\n", req
, req
->r_tid
,
903 * If the osd connection drops, we need to resubmit all requests.
905 static void osd_reset(struct ceph_connection
*con
)
907 struct ceph_osd
*osd
= con
->private;
908 struct ceph_osd_client
*osdc
;
912 dout("osd_reset osd%d\n", osd
->o_osd
);
914 down_read(&osdc
->map_sem
);
915 mutex_lock(&osdc
->request_mutex
);
916 __kick_osd_requests(osdc
, osd
);
918 mutex_unlock(&osdc
->request_mutex
);
919 up_read(&osdc
->map_sem
);
923 * Track open sessions with osds.
925 static struct ceph_osd
*create_osd(struct ceph_osd_client
*osdc
, int onum
)
927 struct ceph_osd
*osd
;
929 osd
= kzalloc(sizeof(*osd
), GFP_NOFS
);
933 atomic_set(&osd
->o_ref
, 1);
936 RB_CLEAR_NODE(&osd
->o_node
);
937 INIT_LIST_HEAD(&osd
->o_requests
);
938 INIT_LIST_HEAD(&osd
->o_linger_requests
);
939 INIT_LIST_HEAD(&osd
->o_osd_lru
);
940 osd
->o_incarnation
= 1;
942 ceph_con_init(&osd
->o_con
, osd
, &osd_con_ops
, &osdc
->client
->msgr
);
944 INIT_LIST_HEAD(&osd
->o_keepalive_item
);
948 static struct ceph_osd
*get_osd(struct ceph_osd
*osd
)
950 if (atomic_inc_not_zero(&osd
->o_ref
)) {
951 dout("get_osd %p %d -> %d\n", osd
, atomic_read(&osd
->o_ref
)-1,
952 atomic_read(&osd
->o_ref
));
955 dout("get_osd %p FAIL\n", osd
);
960 static void put_osd(struct ceph_osd
*osd
)
962 dout("put_osd %p %d -> %d\n", osd
, atomic_read(&osd
->o_ref
),
963 atomic_read(&osd
->o_ref
) - 1);
964 if (atomic_dec_and_test(&osd
->o_ref
) && osd
->o_auth
.authorizer
) {
965 struct ceph_auth_client
*ac
= osd
->o_osdc
->client
->monc
.auth
;
967 ceph_auth_destroy_authorizer(ac
, osd
->o_auth
.authorizer
);
973 * remove an osd from our map
975 static void __remove_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
)
977 dout("__remove_osd %p\n", osd
);
978 BUG_ON(!list_empty(&osd
->o_requests
));
979 rb_erase(&osd
->o_node
, &osdc
->osds
);
980 list_del_init(&osd
->o_osd_lru
);
981 ceph_con_close(&osd
->o_con
);
985 static void remove_all_osds(struct ceph_osd_client
*osdc
)
987 dout("%s %p\n", __func__
, osdc
);
988 mutex_lock(&osdc
->request_mutex
);
989 while (!RB_EMPTY_ROOT(&osdc
->osds
)) {
990 struct ceph_osd
*osd
= rb_entry(rb_first(&osdc
->osds
),
991 struct ceph_osd
, o_node
);
992 __remove_osd(osdc
, osd
);
994 mutex_unlock(&osdc
->request_mutex
);
997 static void __move_osd_to_lru(struct ceph_osd_client
*osdc
,
998 struct ceph_osd
*osd
)
1000 dout("__move_osd_to_lru %p\n", osd
);
1001 BUG_ON(!list_empty(&osd
->o_osd_lru
));
1002 list_add_tail(&osd
->o_osd_lru
, &osdc
->osd_lru
);
1003 osd
->lru_ttl
= jiffies
+ osdc
->client
->options
->osd_idle_ttl
* HZ
;
1006 static void __remove_osd_from_lru(struct ceph_osd
*osd
)
1008 dout("__remove_osd_from_lru %p\n", osd
);
1009 if (!list_empty(&osd
->o_osd_lru
))
1010 list_del_init(&osd
->o_osd_lru
);
1013 static void remove_old_osds(struct ceph_osd_client
*osdc
)
1015 struct ceph_osd
*osd
, *nosd
;
1017 dout("__remove_old_osds %p\n", osdc
);
1018 mutex_lock(&osdc
->request_mutex
);
1019 list_for_each_entry_safe(osd
, nosd
, &osdc
->osd_lru
, o_osd_lru
) {
1020 if (time_before(jiffies
, osd
->lru_ttl
))
1022 __remove_osd(osdc
, osd
);
1024 mutex_unlock(&osdc
->request_mutex
);
1030 static int __reset_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
)
1032 struct ceph_entity_addr
*peer_addr
;
1034 dout("__reset_osd %p osd%d\n", osd
, osd
->o_osd
);
1035 if (list_empty(&osd
->o_requests
) &&
1036 list_empty(&osd
->o_linger_requests
)) {
1037 __remove_osd(osdc
, osd
);
1042 peer_addr
= &osdc
->osdmap
->osd_addr
[osd
->o_osd
];
1043 if (!memcmp(peer_addr
, &osd
->o_con
.peer_addr
, sizeof (*peer_addr
)) &&
1044 !ceph_con_opened(&osd
->o_con
)) {
1045 struct ceph_osd_request
*req
;
1047 dout(" osd addr hasn't changed and connection never opened,"
1048 " letting msgr retry");
1049 /* touch each r_stamp for handle_timeout()'s benfit */
1050 list_for_each_entry(req
, &osd
->o_requests
, r_osd_item
)
1051 req
->r_stamp
= jiffies
;
1056 ceph_con_close(&osd
->o_con
);
1057 ceph_con_open(&osd
->o_con
, CEPH_ENTITY_TYPE_OSD
, osd
->o_osd
, peer_addr
);
1058 osd
->o_incarnation
++;
1063 static void __insert_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*new)
1065 struct rb_node
**p
= &osdc
->osds
.rb_node
;
1066 struct rb_node
*parent
= NULL
;
1067 struct ceph_osd
*osd
= NULL
;
1069 dout("__insert_osd %p osd%d\n", new, new->o_osd
);
1072 osd
= rb_entry(parent
, struct ceph_osd
, o_node
);
1073 if (new->o_osd
< osd
->o_osd
)
1075 else if (new->o_osd
> osd
->o_osd
)
1076 p
= &(*p
)->rb_right
;
1081 rb_link_node(&new->o_node
, parent
, p
);
1082 rb_insert_color(&new->o_node
, &osdc
->osds
);
1085 static struct ceph_osd
*__lookup_osd(struct ceph_osd_client
*osdc
, int o
)
1087 struct ceph_osd
*osd
;
1088 struct rb_node
*n
= osdc
->osds
.rb_node
;
1091 osd
= rb_entry(n
, struct ceph_osd
, o_node
);
1094 else if (o
> osd
->o_osd
)
1102 static void __schedule_osd_timeout(struct ceph_osd_client
*osdc
)
1104 schedule_delayed_work(&osdc
->timeout_work
,
1105 osdc
->client
->options
->osd_keepalive_timeout
* HZ
);
1108 static void __cancel_osd_timeout(struct ceph_osd_client
*osdc
)
1110 cancel_delayed_work(&osdc
->timeout_work
);
1114 * Register request, assign tid. If this is the first request, set up
1115 * the timeout event.
1117 static void __register_request(struct ceph_osd_client
*osdc
,
1118 struct ceph_osd_request
*req
)
1120 req
->r_tid
= ++osdc
->last_tid
;
1121 req
->r_request
->hdr
.tid
= cpu_to_le64(req
->r_tid
);
1122 dout("__register_request %p tid %lld\n", req
, req
->r_tid
);
1123 __insert_request(osdc
, req
);
1124 ceph_osdc_get_request(req
);
1125 osdc
->num_requests
++;
1126 if (osdc
->num_requests
== 1) {
1127 dout(" first request, scheduling timeout\n");
1128 __schedule_osd_timeout(osdc
);
1133 * called under osdc->request_mutex
1135 static void __unregister_request(struct ceph_osd_client
*osdc
,
1136 struct ceph_osd_request
*req
)
1138 if (RB_EMPTY_NODE(&req
->r_node
)) {
1139 dout("__unregister_request %p tid %lld not registered\n",
1144 dout("__unregister_request %p tid %lld\n", req
, req
->r_tid
);
1145 rb_erase(&req
->r_node
, &osdc
->requests
);
1146 osdc
->num_requests
--;
1149 /* make sure the original request isn't in flight. */
1150 ceph_msg_revoke(req
->r_request
);
1152 list_del_init(&req
->r_osd_item
);
1153 if (list_empty(&req
->r_osd
->o_requests
) &&
1154 list_empty(&req
->r_osd
->o_linger_requests
)) {
1155 dout("moving osd to %p lru\n", req
->r_osd
);
1156 __move_osd_to_lru(osdc
, req
->r_osd
);
1158 if (list_empty(&req
->r_linger_item
))
1162 list_del_init(&req
->r_req_lru_item
);
1163 ceph_osdc_put_request(req
);
1165 if (osdc
->num_requests
== 0) {
1166 dout(" no requests, canceling timeout\n");
1167 __cancel_osd_timeout(osdc
);
1172 * Cancel a previously queued request message
1174 static void __cancel_request(struct ceph_osd_request
*req
)
1176 if (req
->r_sent
&& req
->r_osd
) {
1177 ceph_msg_revoke(req
->r_request
);
1182 static void __register_linger_request(struct ceph_osd_client
*osdc
,
1183 struct ceph_osd_request
*req
)
1185 dout("__register_linger_request %p\n", req
);
1186 ceph_osdc_get_request(req
);
1187 list_add_tail(&req
->r_linger_item
, &osdc
->req_linger
);
1189 list_add_tail(&req
->r_linger_osd
,
1190 &req
->r_osd
->o_linger_requests
);
1193 static void __unregister_linger_request(struct ceph_osd_client
*osdc
,
1194 struct ceph_osd_request
*req
)
1196 dout("__unregister_linger_request %p\n", req
);
1197 list_del_init(&req
->r_linger_item
);
1199 list_del_init(&req
->r_linger_osd
);
1201 if (list_empty(&req
->r_osd
->o_requests
) &&
1202 list_empty(&req
->r_osd
->o_linger_requests
)) {
1203 dout("moving osd to %p lru\n", req
->r_osd
);
1204 __move_osd_to_lru(osdc
, req
->r_osd
);
1206 if (list_empty(&req
->r_osd_item
))
1209 ceph_osdc_put_request(req
);
1212 void ceph_osdc_unregister_linger_request(struct ceph_osd_client
*osdc
,
1213 struct ceph_osd_request
*req
)
1215 mutex_lock(&osdc
->request_mutex
);
1216 if (req
->r_linger
) {
1218 __unregister_linger_request(osdc
, req
);
1220 mutex_unlock(&osdc
->request_mutex
);
1222 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request
);
1224 void ceph_osdc_set_request_linger(struct ceph_osd_client
*osdc
,
1225 struct ceph_osd_request
*req
)
1227 if (!req
->r_linger
) {
1228 dout("set_request_linger %p\n", req
);
1232 EXPORT_SYMBOL(ceph_osdc_set_request_linger
);
1235 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1236 * (as needed), and set the request r_osd appropriately. If there is
1237 * no up osd, set r_osd to NULL. Move the request to the appropriate list
1238 * (unsent, homeless) or leave on in-flight lru.
1240 * Return 0 if unchanged, 1 if changed, or negative on error.
1242 * Caller should hold map_sem for read and request_mutex.
1244 static int __map_request(struct ceph_osd_client
*osdc
,
1245 struct ceph_osd_request
*req
, int force_resend
)
1247 struct ceph_pg pgid
;
1248 int acting
[CEPH_PG_MAX_SIZE
];
1249 int o
= -1, num
= 0;
1252 dout("map_request %p tid %lld\n", req
, req
->r_tid
);
1253 err
= ceph_calc_ceph_pg(&pgid
, req
->r_oid
, osdc
->osdmap
,
1254 ceph_file_layout_pg_pool(req
->r_file_layout
));
1256 list_move(&req
->r_req_lru_item
, &osdc
->req_notarget
);
1261 err
= ceph_calc_pg_acting(osdc
->osdmap
, pgid
, acting
);
1267 if ((!force_resend
&&
1268 req
->r_osd
&& req
->r_osd
->o_osd
== o
&&
1269 req
->r_sent
>= req
->r_osd
->o_incarnation
&&
1270 req
->r_num_pg_osds
== num
&&
1271 memcmp(req
->r_pg_osds
, acting
, sizeof(acting
[0])*num
) == 0) ||
1272 (req
->r_osd
== NULL
&& o
== -1))
1273 return 0; /* no change */
1275 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1276 req
->r_tid
, pgid
.pool
, pgid
.seed
, o
,
1277 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1279 /* record full pg acting set */
1280 memcpy(req
->r_pg_osds
, acting
, sizeof(acting
[0]) * num
);
1281 req
->r_num_pg_osds
= num
;
1284 __cancel_request(req
);
1285 list_del_init(&req
->r_osd_item
);
1289 req
->r_osd
= __lookup_osd(osdc
, o
);
1290 if (!req
->r_osd
&& o
>= 0) {
1292 req
->r_osd
= create_osd(osdc
, o
);
1294 list_move(&req
->r_req_lru_item
, &osdc
->req_notarget
);
1298 dout("map_request osd %p is osd%d\n", req
->r_osd
, o
);
1299 __insert_osd(osdc
, req
->r_osd
);
1301 ceph_con_open(&req
->r_osd
->o_con
,
1302 CEPH_ENTITY_TYPE_OSD
, o
,
1303 &osdc
->osdmap
->osd_addr
[o
]);
1307 __remove_osd_from_lru(req
->r_osd
);
1308 list_add_tail(&req
->r_osd_item
, &req
->r_osd
->o_requests
);
1309 list_move_tail(&req
->r_req_lru_item
, &osdc
->req_unsent
);
1311 list_move_tail(&req
->r_req_lru_item
, &osdc
->req_notarget
);
1313 err
= 1; /* osd or pg changed */
1320 * caller should hold map_sem (for read) and request_mutex
1322 static void __send_request(struct ceph_osd_client
*osdc
,
1323 struct ceph_osd_request
*req
)
1327 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1328 req
, req
->r_tid
, req
->r_osd
->o_osd
, req
->r_flags
,
1329 (unsigned long long)req
->r_pgid
.pool
, req
->r_pgid
.seed
);
1331 /* fill in message content that changes each time we send it */
1332 put_unaligned_le32(osdc
->osdmap
->epoch
, req
->r_request_osdmap_epoch
);
1333 put_unaligned_le32(req
->r_flags
, req
->r_request_flags
);
1334 put_unaligned_le64(req
->r_pgid
.pool
, req
->r_request_pool
);
1335 p
= req
->r_request_pgid
;
1336 ceph_encode_64(&p
, req
->r_pgid
.pool
);
1337 ceph_encode_32(&p
, req
->r_pgid
.seed
);
1338 put_unaligned_le64(1, req
->r_request_attempts
); /* FIXME */
1339 memcpy(req
->r_request_reassert_version
, &req
->r_reassert_version
,
1340 sizeof(req
->r_reassert_version
));
1342 req
->r_stamp
= jiffies
;
1343 list_move_tail(&req
->r_req_lru_item
, &osdc
->req_lru
);
1345 ceph_msg_get(req
->r_request
); /* send consumes a ref */
1347 req
->r_sent
= req
->r_osd
->o_incarnation
;
1349 ceph_con_send(&req
->r_osd
->o_con
, req
->r_request
);
1353 * Send any requests in the queue (req_unsent).
1355 static void __send_queued(struct ceph_osd_client
*osdc
)
1357 struct ceph_osd_request
*req
, *tmp
;
1359 dout("__send_queued\n");
1360 list_for_each_entry_safe(req
, tmp
, &osdc
->req_unsent
, r_req_lru_item
)
1361 __send_request(osdc
, req
);
1365 * Timeout callback, called every N seconds when 1 or more osd
1366 * requests has been active for more than N seconds. When this
1367 * happens, we ping all OSDs with requests who have timed out to
1368 * ensure any communications channel reset is detected. Reset the
1369 * request timeouts another N seconds in the future as we go.
1370 * Reschedule the timeout event another N seconds in future (unless
1371 * there are no open requests).
1373 static void handle_timeout(struct work_struct
*work
)
1375 struct ceph_osd_client
*osdc
=
1376 container_of(work
, struct ceph_osd_client
, timeout_work
.work
);
1377 struct ceph_osd_request
*req
;
1378 struct ceph_osd
*osd
;
1379 unsigned long keepalive
=
1380 osdc
->client
->options
->osd_keepalive_timeout
* HZ
;
1381 struct list_head slow_osds
;
1383 down_read(&osdc
->map_sem
);
1385 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1387 mutex_lock(&osdc
->request_mutex
);
1390 * ping osds that are a bit slow. this ensures that if there
1391 * is a break in the TCP connection we will notice, and reopen
1392 * a connection with that osd (from the fault callback).
1394 INIT_LIST_HEAD(&slow_osds
);
1395 list_for_each_entry(req
, &osdc
->req_lru
, r_req_lru_item
) {
1396 if (time_before(jiffies
, req
->r_stamp
+ keepalive
))
1401 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1402 req
->r_tid
, osd
->o_osd
);
1403 list_move_tail(&osd
->o_keepalive_item
, &slow_osds
);
1405 while (!list_empty(&slow_osds
)) {
1406 osd
= list_entry(slow_osds
.next
, struct ceph_osd
,
1408 list_del_init(&osd
->o_keepalive_item
);
1409 ceph_con_keepalive(&osd
->o_con
);
1412 __schedule_osd_timeout(osdc
);
1413 __send_queued(osdc
);
1414 mutex_unlock(&osdc
->request_mutex
);
1415 up_read(&osdc
->map_sem
);
1418 static void handle_osds_timeout(struct work_struct
*work
)
1420 struct ceph_osd_client
*osdc
=
1421 container_of(work
, struct ceph_osd_client
,
1422 osds_timeout_work
.work
);
1423 unsigned long delay
=
1424 osdc
->client
->options
->osd_idle_ttl
* HZ
>> 2;
1426 dout("osds timeout\n");
1427 down_read(&osdc
->map_sem
);
1428 remove_old_osds(osdc
);
1429 up_read(&osdc
->map_sem
);
1431 schedule_delayed_work(&osdc
->osds_timeout_work
,
1432 round_jiffies_relative(delay
));
1435 static void complete_request(struct ceph_osd_request
*req
)
1437 complete_all(&req
->r_safe_completion
); /* fsync waiter */
1441 * handle osd op reply. either call the callback if it is specified,
1442 * or do the completion to wake up the waiting thread.
1444 static void handle_reply(struct ceph_osd_client
*osdc
, struct ceph_msg
*msg
,
1445 struct ceph_connection
*con
)
1448 struct ceph_osd_request
*req
;
1451 unsigned int numops
;
1452 int payload_len
, flags
;
1458 u64 reassert_version
;
1460 int already_completed
;
1464 tid
= le64_to_cpu(msg
->hdr
.tid
);
1465 dout("handle_reply %p tid %llu\n", msg
, tid
);
1467 p
= msg
->front
.iov_base
;
1468 end
= p
+ msg
->front
.iov_len
;
1470 ceph_decode_need(&p
, end
, 4, bad
);
1471 object_len
= ceph_decode_32(&p
);
1472 ceph_decode_need(&p
, end
, object_len
, bad
);
1475 err
= ceph_decode_pgid(&p
, end
, &pg
);
1479 ceph_decode_need(&p
, end
, 8 + 4 + 4 + 8 + 4, bad
);
1480 flags
= ceph_decode_64(&p
);
1481 result
= ceph_decode_32(&p
);
1482 reassert_epoch
= ceph_decode_32(&p
);
1483 reassert_version
= ceph_decode_64(&p
);
1484 osdmap_epoch
= ceph_decode_32(&p
);
1487 mutex_lock(&osdc
->request_mutex
);
1488 req
= __lookup_request(osdc
, tid
);
1490 dout("handle_reply tid %llu dne\n", tid
);
1493 ceph_osdc_get_request(req
);
1495 dout("handle_reply %p tid %llu req %p result %d\n", msg
, tid
,
1498 ceph_decode_need(&p
, end
, 4, bad_put
);
1499 numops
= ceph_decode_32(&p
);
1500 if (numops
> CEPH_OSD_MAX_OP
)
1502 if (numops
!= req
->r_num_ops
)
1505 ceph_decode_need(&p
, end
, numops
* sizeof(struct ceph_osd_op
), bad_put
);
1506 for (i
= 0; i
< numops
; i
++) {
1507 struct ceph_osd_op
*op
= p
;
1510 len
= le32_to_cpu(op
->payload_len
);
1511 req
->r_reply_op_len
[i
] = len
;
1512 dout(" op %d has %d bytes\n", i
, len
);
1516 bytes
= le32_to_cpu(msg
->hdr
.data_len
);
1517 if (payload_len
!= bytes
) {
1518 pr_warning("sum of op payload lens %d != data_len %d",
1519 payload_len
, bytes
);
1523 ceph_decode_need(&p
, end
, 4 + numops
* 4, bad_put
);
1524 retry_attempt
= ceph_decode_32(&p
);
1525 for (i
= 0; i
< numops
; i
++)
1526 req
->r_reply_op_result
[i
] = ceph_decode_32(&p
);
1528 already_completed
= req
->r_got_reply
;
1530 if (!req
->r_got_reply
) {
1532 req
->r_result
= result
;
1533 dout("handle_reply result %d bytes %d\n", req
->r_result
,
1535 if (req
->r_result
== 0)
1536 req
->r_result
= bytes
;
1538 /* in case this is a write and we need to replay, */
1539 req
->r_reassert_version
.epoch
= cpu_to_le32(reassert_epoch
);
1540 req
->r_reassert_version
.version
= cpu_to_le64(reassert_version
);
1542 req
->r_got_reply
= 1;
1543 } else if ((flags
& CEPH_OSD_FLAG_ONDISK
) == 0) {
1544 dout("handle_reply tid %llu dup ack\n", tid
);
1545 mutex_unlock(&osdc
->request_mutex
);
1549 dout("handle_reply tid %llu flags %d\n", tid
, flags
);
1551 if (req
->r_linger
&& (flags
& CEPH_OSD_FLAG_ONDISK
))
1552 __register_linger_request(osdc
, req
);
1554 /* either this is a read, or we got the safe response */
1556 (flags
& CEPH_OSD_FLAG_ONDISK
) ||
1557 ((flags
& CEPH_OSD_FLAG_WRITE
) == 0))
1558 __unregister_request(osdc
, req
);
1560 mutex_unlock(&osdc
->request_mutex
);
1562 if (!already_completed
) {
1563 if (req
->r_unsafe_callback
&&
1564 result
>= 0 && !(flags
& CEPH_OSD_FLAG_ONDISK
))
1565 req
->r_unsafe_callback(req
, true);
1566 if (req
->r_callback
)
1567 req
->r_callback(req
, msg
);
1569 complete_all(&req
->r_completion
);
1572 if (flags
& CEPH_OSD_FLAG_ONDISK
) {
1573 if (req
->r_unsafe_callback
&& already_completed
)
1574 req
->r_unsafe_callback(req
, false);
1575 complete_request(req
);
1579 dout("req=%p req->r_linger=%d\n", req
, req
->r_linger
);
1580 ceph_osdc_put_request(req
);
1584 ceph_osdc_put_request(req
);
1586 mutex_unlock(&osdc
->request_mutex
);
1588 pr_err("corrupt osd_op_reply got %d %d\n",
1589 (int)msg
->front
.iov_len
, le32_to_cpu(msg
->hdr
.front_len
));
1593 static void reset_changed_osds(struct ceph_osd_client
*osdc
)
1595 struct rb_node
*p
, *n
;
1597 for (p
= rb_first(&osdc
->osds
); p
; p
= n
) {
1598 struct ceph_osd
*osd
= rb_entry(p
, struct ceph_osd
, o_node
);
1601 if (!ceph_osd_is_up(osdc
->osdmap
, osd
->o_osd
) ||
1602 memcmp(&osd
->o_con
.peer_addr
,
1603 ceph_osd_addr(osdc
->osdmap
,
1605 sizeof(struct ceph_entity_addr
)) != 0)
1606 __reset_osd(osdc
, osd
);
1611 * Requeue requests whose mapping to an OSD has changed. If requests map to
1612 * no osd, request a new map.
1614 * Caller should hold map_sem for read.
1616 static void kick_requests(struct ceph_osd_client
*osdc
, int force_resend
)
1618 struct ceph_osd_request
*req
, *nreq
;
1623 dout("kick_requests %s\n", force_resend
? " (force resend)" : "");
1624 mutex_lock(&osdc
->request_mutex
);
1625 for (p
= rb_first(&osdc
->requests
); p
; ) {
1626 req
= rb_entry(p
, struct ceph_osd_request
, r_node
);
1630 * For linger requests that have not yet been
1631 * registered, move them to the linger list; they'll
1632 * be sent to the osd in the loop below. Unregister
1633 * the request before re-registering it as a linger
1634 * request to ensure the __map_request() below
1635 * will decide it needs to be sent.
1637 if (req
->r_linger
&& list_empty(&req
->r_linger_item
)) {
1638 dout("%p tid %llu restart on osd%d\n",
1640 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1641 ceph_osdc_get_request(req
);
1642 __unregister_request(osdc
, req
);
1643 __register_linger_request(osdc
, req
);
1644 ceph_osdc_put_request(req
);
1648 err
= __map_request(osdc
, req
, force_resend
);
1650 continue; /* error */
1651 if (req
->r_osd
== NULL
) {
1652 dout("%p tid %llu maps to no osd\n", req
, req
->r_tid
);
1653 needmap
++; /* request a newer map */
1654 } else if (err
> 0) {
1655 if (!req
->r_linger
) {
1656 dout("%p tid %llu requeued on osd%d\n", req
,
1658 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1659 req
->r_flags
|= CEPH_OSD_FLAG_RETRY
;
1664 list_for_each_entry_safe(req
, nreq
, &osdc
->req_linger
,
1666 dout("linger req=%p req->r_osd=%p\n", req
, req
->r_osd
);
1668 err
= __map_request(osdc
, req
, force_resend
);
1669 dout("__map_request returned %d\n", err
);
1671 continue; /* no change and no osd was specified */
1673 continue; /* hrm! */
1674 if (req
->r_osd
== NULL
) {
1675 dout("tid %llu maps to no valid osd\n", req
->r_tid
);
1676 needmap
++; /* request a newer map */
1680 dout("kicking lingering %p tid %llu osd%d\n", req
, req
->r_tid
,
1681 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1682 __register_request(osdc
, req
);
1683 __unregister_linger_request(osdc
, req
);
1685 reset_changed_osds(osdc
);
1686 mutex_unlock(&osdc
->request_mutex
);
1689 dout("%d requests for down osds, need new map\n", needmap
);
1690 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1696 * Process updated osd map.
1698 * The message contains any number of incremental and full maps, normally
1699 * indicating some sort of topology change in the cluster. Kick requests
1700 * off to different OSDs as needed.
1702 void ceph_osdc_handle_map(struct ceph_osd_client
*osdc
, struct ceph_msg
*msg
)
1704 void *p
, *end
, *next
;
1705 u32 nr_maps
, maplen
;
1707 struct ceph_osdmap
*newmap
= NULL
, *oldmap
;
1709 struct ceph_fsid fsid
;
1711 dout("handle_map have %u\n", osdc
->osdmap
? osdc
->osdmap
->epoch
: 0);
1712 p
= msg
->front
.iov_base
;
1713 end
= p
+ msg
->front
.iov_len
;
1716 ceph_decode_need(&p
, end
, sizeof(fsid
), bad
);
1717 ceph_decode_copy(&p
, &fsid
, sizeof(fsid
));
1718 if (ceph_check_fsid(osdc
->client
, &fsid
) < 0)
1721 down_write(&osdc
->map_sem
);
1723 /* incremental maps */
1724 ceph_decode_32_safe(&p
, end
, nr_maps
, bad
);
1725 dout(" %d inc maps\n", nr_maps
);
1726 while (nr_maps
> 0) {
1727 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
1728 epoch
= ceph_decode_32(&p
);
1729 maplen
= ceph_decode_32(&p
);
1730 ceph_decode_need(&p
, end
, maplen
, bad
);
1732 if (osdc
->osdmap
&& osdc
->osdmap
->epoch
+1 == epoch
) {
1733 dout("applying incremental map %u len %d\n",
1735 newmap
= osdmap_apply_incremental(&p
, next
,
1737 &osdc
->client
->msgr
);
1738 if (IS_ERR(newmap
)) {
1739 err
= PTR_ERR(newmap
);
1743 if (newmap
!= osdc
->osdmap
) {
1744 ceph_osdmap_destroy(osdc
->osdmap
);
1745 osdc
->osdmap
= newmap
;
1747 kick_requests(osdc
, 0);
1749 dout("ignoring incremental map %u len %d\n",
1759 ceph_decode_32_safe(&p
, end
, nr_maps
, bad
);
1760 dout(" %d full maps\n", nr_maps
);
1762 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
1763 epoch
= ceph_decode_32(&p
);
1764 maplen
= ceph_decode_32(&p
);
1765 ceph_decode_need(&p
, end
, maplen
, bad
);
1767 dout("skipping non-latest full map %u len %d\n",
1769 } else if (osdc
->osdmap
&& osdc
->osdmap
->epoch
>= epoch
) {
1770 dout("skipping full map %u len %d, "
1771 "older than our %u\n", epoch
, maplen
,
1772 osdc
->osdmap
->epoch
);
1774 int skipped_map
= 0;
1776 dout("taking full map %u len %d\n", epoch
, maplen
);
1777 newmap
= osdmap_decode(&p
, p
+maplen
);
1778 if (IS_ERR(newmap
)) {
1779 err
= PTR_ERR(newmap
);
1783 oldmap
= osdc
->osdmap
;
1784 osdc
->osdmap
= newmap
;
1786 if (oldmap
->epoch
+ 1 < newmap
->epoch
)
1788 ceph_osdmap_destroy(oldmap
);
1790 kick_requests(osdc
, skipped_map
);
1799 downgrade_write(&osdc
->map_sem
);
1800 ceph_monc_got_osdmap(&osdc
->client
->monc
, osdc
->osdmap
->epoch
);
1803 * subscribe to subsequent osdmap updates if full to ensure
1804 * we find out when we are no longer full and stop returning
1807 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
))
1808 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1810 mutex_lock(&osdc
->request_mutex
);
1811 __send_queued(osdc
);
1812 mutex_unlock(&osdc
->request_mutex
);
1813 up_read(&osdc
->map_sem
);
1814 wake_up_all(&osdc
->client
->auth_wq
);
1818 pr_err("osdc handle_map corrupt msg\n");
1820 up_write(&osdc
->map_sem
);
1825 * watch/notify callback event infrastructure
1827 * These callbacks are used both for watch and notify operations.
1829 static void __release_event(struct kref
*kref
)
1831 struct ceph_osd_event
*event
=
1832 container_of(kref
, struct ceph_osd_event
, kref
);
1834 dout("__release_event %p\n", event
);
1838 static void get_event(struct ceph_osd_event
*event
)
1840 kref_get(&event
->kref
);
1843 void ceph_osdc_put_event(struct ceph_osd_event
*event
)
1845 kref_put(&event
->kref
, __release_event
);
1847 EXPORT_SYMBOL(ceph_osdc_put_event
);
1849 static void __insert_event(struct ceph_osd_client
*osdc
,
1850 struct ceph_osd_event
*new)
1852 struct rb_node
**p
= &osdc
->event_tree
.rb_node
;
1853 struct rb_node
*parent
= NULL
;
1854 struct ceph_osd_event
*event
= NULL
;
1858 event
= rb_entry(parent
, struct ceph_osd_event
, node
);
1859 if (new->cookie
< event
->cookie
)
1861 else if (new->cookie
> event
->cookie
)
1862 p
= &(*p
)->rb_right
;
1867 rb_link_node(&new->node
, parent
, p
);
1868 rb_insert_color(&new->node
, &osdc
->event_tree
);
1871 static struct ceph_osd_event
*__find_event(struct ceph_osd_client
*osdc
,
1874 struct rb_node
**p
= &osdc
->event_tree
.rb_node
;
1875 struct rb_node
*parent
= NULL
;
1876 struct ceph_osd_event
*event
= NULL
;
1880 event
= rb_entry(parent
, struct ceph_osd_event
, node
);
1881 if (cookie
< event
->cookie
)
1883 else if (cookie
> event
->cookie
)
1884 p
= &(*p
)->rb_right
;
1891 static void __remove_event(struct ceph_osd_event
*event
)
1893 struct ceph_osd_client
*osdc
= event
->osdc
;
1895 if (!RB_EMPTY_NODE(&event
->node
)) {
1896 dout("__remove_event removed %p\n", event
);
1897 rb_erase(&event
->node
, &osdc
->event_tree
);
1898 ceph_osdc_put_event(event
);
1900 dout("__remove_event didn't remove %p\n", event
);
1904 int ceph_osdc_create_event(struct ceph_osd_client
*osdc
,
1905 void (*event_cb
)(u64
, u64
, u8
, void *),
1906 void *data
, struct ceph_osd_event
**pevent
)
1908 struct ceph_osd_event
*event
;
1910 event
= kmalloc(sizeof(*event
), GFP_NOIO
);
1914 dout("create_event %p\n", event
);
1915 event
->cb
= event_cb
;
1916 event
->one_shot
= 0;
1919 INIT_LIST_HEAD(&event
->osd_node
);
1920 RB_CLEAR_NODE(&event
->node
);
1921 kref_init(&event
->kref
); /* one ref for us */
1922 kref_get(&event
->kref
); /* one ref for the caller */
1924 spin_lock(&osdc
->event_lock
);
1925 event
->cookie
= ++osdc
->event_count
;
1926 __insert_event(osdc
, event
);
1927 spin_unlock(&osdc
->event_lock
);
1932 EXPORT_SYMBOL(ceph_osdc_create_event
);
1934 void ceph_osdc_cancel_event(struct ceph_osd_event
*event
)
1936 struct ceph_osd_client
*osdc
= event
->osdc
;
1938 dout("cancel_event %p\n", event
);
1939 spin_lock(&osdc
->event_lock
);
1940 __remove_event(event
);
1941 spin_unlock(&osdc
->event_lock
);
1942 ceph_osdc_put_event(event
); /* caller's */
1944 EXPORT_SYMBOL(ceph_osdc_cancel_event
);
1947 static void do_event_work(struct work_struct
*work
)
1949 struct ceph_osd_event_work
*event_work
=
1950 container_of(work
, struct ceph_osd_event_work
, work
);
1951 struct ceph_osd_event
*event
= event_work
->event
;
1952 u64 ver
= event_work
->ver
;
1953 u64 notify_id
= event_work
->notify_id
;
1954 u8 opcode
= event_work
->opcode
;
1956 dout("do_event_work completing %p\n", event
);
1957 event
->cb(ver
, notify_id
, opcode
, event
->data
);
1958 dout("do_event_work completed %p\n", event
);
1959 ceph_osdc_put_event(event
);
1965 * Process osd watch notifications
1967 static void handle_watch_notify(struct ceph_osd_client
*osdc
,
1968 struct ceph_msg
*msg
)
1972 u64 cookie
, ver
, notify_id
;
1974 struct ceph_osd_event
*event
;
1975 struct ceph_osd_event_work
*event_work
;
1977 p
= msg
->front
.iov_base
;
1978 end
= p
+ msg
->front
.iov_len
;
1980 ceph_decode_8_safe(&p
, end
, proto_ver
, bad
);
1981 ceph_decode_8_safe(&p
, end
, opcode
, bad
);
1982 ceph_decode_64_safe(&p
, end
, cookie
, bad
);
1983 ceph_decode_64_safe(&p
, end
, ver
, bad
);
1984 ceph_decode_64_safe(&p
, end
, notify_id
, bad
);
1986 spin_lock(&osdc
->event_lock
);
1987 event
= __find_event(osdc
, cookie
);
1989 BUG_ON(event
->one_shot
);
1992 spin_unlock(&osdc
->event_lock
);
1993 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1994 cookie
, ver
, event
);
1996 event_work
= kmalloc(sizeof(*event_work
), GFP_NOIO
);
1998 dout("ERROR: could not allocate event_work\n");
2001 INIT_WORK(&event_work
->work
, do_event_work
);
2002 event_work
->event
= event
;
2003 event_work
->ver
= ver
;
2004 event_work
->notify_id
= notify_id
;
2005 event_work
->opcode
= opcode
;
2006 if (!queue_work(osdc
->notify_wq
, &event_work
->work
)) {
2007 dout("WARNING: failed to queue notify event work\n");
2015 ceph_osdc_put_event(event
);
2019 pr_err("osdc handle_watch_notify corrupt msg\n");
2024 * build new request AND message
2027 void ceph_osdc_build_request(struct ceph_osd_request
*req
, u64 off
,
2028 struct ceph_snap_context
*snapc
, u64 snap_id
,
2029 struct timespec
*mtime
)
2031 struct ceph_msg
*msg
= req
->r_request
;
2034 int flags
= req
->r_flags
;
2038 req
->r_snapid
= snap_id
;
2039 req
->r_snapc
= ceph_get_snap_context(snapc
);
2041 /* encode request */
2042 msg
->hdr
.version
= cpu_to_le16(4);
2044 p
= msg
->front
.iov_base
;
2045 ceph_encode_32(&p
, 1); /* client_inc is always 1 */
2046 req
->r_request_osdmap_epoch
= p
;
2048 req
->r_request_flags
= p
;
2050 if (req
->r_flags
& CEPH_OSD_FLAG_WRITE
)
2051 ceph_encode_timespec(p
, mtime
);
2052 p
+= sizeof(struct ceph_timespec
);
2053 req
->r_request_reassert_version
= p
;
2054 p
+= sizeof(struct ceph_eversion
); /* will get filled in */
2057 ceph_encode_8(&p
, 4);
2058 ceph_encode_8(&p
, 4);
2059 ceph_encode_32(&p
, 8 + 4 + 4);
2060 req
->r_request_pool
= p
;
2062 ceph_encode_32(&p
, -1); /* preferred */
2063 ceph_encode_32(&p
, 0); /* key len */
2065 ceph_encode_8(&p
, 1);
2066 req
->r_request_pgid
= p
;
2068 ceph_encode_32(&p
, -1); /* preferred */
2071 ceph_encode_32(&p
, req
->r_oid_len
);
2072 memcpy(p
, req
->r_oid
, req
->r_oid_len
);
2073 dout("oid '%.*s' len %d\n", req
->r_oid_len
, req
->r_oid
, req
->r_oid_len
);
2074 p
+= req
->r_oid_len
;
2076 /* ops--can imply data */
2077 ceph_encode_16(&p
, (u16
)req
->r_num_ops
);
2079 for (i
= 0; i
< req
->r_num_ops
; i
++) {
2080 data_len
+= osd_req_encode_op(req
, p
, i
);
2081 p
+= sizeof(struct ceph_osd_op
);
2085 ceph_encode_64(&p
, req
->r_snapid
);
2086 ceph_encode_64(&p
, req
->r_snapc
? req
->r_snapc
->seq
: 0);
2087 ceph_encode_32(&p
, req
->r_snapc
? req
->r_snapc
->num_snaps
: 0);
2089 for (i
= 0; i
< snapc
->num_snaps
; i
++) {
2090 ceph_encode_64(&p
, req
->r_snapc
->snaps
[i
]);
2094 req
->r_request_attempts
= p
;
2098 if (flags
& CEPH_OSD_FLAG_WRITE
) {
2102 * The header "data_off" is a hint to the receiver
2103 * allowing it to align received data into its
2104 * buffers such that there's no need to re-copy
2105 * it before writing it to disk (direct I/O).
2107 data_off
= (u16
) (off
& 0xffff);
2108 req
->r_request
->hdr
.data_off
= cpu_to_le16(data_off
);
2110 req
->r_request
->hdr
.data_len
= cpu_to_le32(data_len
);
2112 BUG_ON(p
> msg
->front
.iov_base
+ msg
->front
.iov_len
);
2113 msg_size
= p
- msg
->front
.iov_base
;
2114 msg
->front
.iov_len
= msg_size
;
2115 msg
->hdr
.front_len
= cpu_to_le32(msg_size
);
2117 dout("build_request msg_size was %d\n", (int)msg_size
);
2119 EXPORT_SYMBOL(ceph_osdc_build_request
);
2122 * Register request, send initial attempt.
2124 int ceph_osdc_start_request(struct ceph_osd_client
*osdc
,
2125 struct ceph_osd_request
*req
,
2130 down_read(&osdc
->map_sem
);
2131 mutex_lock(&osdc
->request_mutex
);
2132 __register_request(osdc
, req
);
2134 req
->r_got_reply
= 0;
2135 rc
= __map_request(osdc
, req
, 0);
2138 dout("osdc_start_request failed map, "
2139 " will retry %lld\n", req
->r_tid
);
2142 __unregister_request(osdc
, req
);
2146 if (req
->r_osd
== NULL
) {
2147 dout("send_request %p no up osds in pg\n", req
);
2148 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
2150 __send_queued(osdc
);
2154 mutex_unlock(&osdc
->request_mutex
);
2155 up_read(&osdc
->map_sem
);
2158 EXPORT_SYMBOL(ceph_osdc_start_request
);
2161 * wait for a request to complete
2163 int ceph_osdc_wait_request(struct ceph_osd_client
*osdc
,
2164 struct ceph_osd_request
*req
)
2168 rc
= wait_for_completion_interruptible(&req
->r_completion
);
2170 mutex_lock(&osdc
->request_mutex
);
2171 __cancel_request(req
);
2172 __unregister_request(osdc
, req
);
2173 mutex_unlock(&osdc
->request_mutex
);
2174 complete_request(req
);
2175 dout("wait_request tid %llu canceled/timed out\n", req
->r_tid
);
2179 dout("wait_request tid %llu result %d\n", req
->r_tid
, req
->r_result
);
2180 return req
->r_result
;
2182 EXPORT_SYMBOL(ceph_osdc_wait_request
);
2185 * sync - wait for all in-flight requests to flush. avoid starvation.
2187 void ceph_osdc_sync(struct ceph_osd_client
*osdc
)
2189 struct ceph_osd_request
*req
;
2190 u64 last_tid
, next_tid
= 0;
2192 mutex_lock(&osdc
->request_mutex
);
2193 last_tid
= osdc
->last_tid
;
2195 req
= __lookup_request_ge(osdc
, next_tid
);
2198 if (req
->r_tid
> last_tid
)
2201 next_tid
= req
->r_tid
+ 1;
2202 if ((req
->r_flags
& CEPH_OSD_FLAG_WRITE
) == 0)
2205 ceph_osdc_get_request(req
);
2206 mutex_unlock(&osdc
->request_mutex
);
2207 dout("sync waiting on tid %llu (last is %llu)\n",
2208 req
->r_tid
, last_tid
);
2209 wait_for_completion(&req
->r_safe_completion
);
2210 mutex_lock(&osdc
->request_mutex
);
2211 ceph_osdc_put_request(req
);
2213 mutex_unlock(&osdc
->request_mutex
);
2214 dout("sync done (thru tid %llu)\n", last_tid
);
2216 EXPORT_SYMBOL(ceph_osdc_sync
);
2219 * Call all pending notify callbacks - for use after a watch is
2220 * unregistered, to make sure no more callbacks for it will be invoked
2222 extern void ceph_osdc_flush_notifies(struct ceph_osd_client
*osdc
)
2224 flush_workqueue(osdc
->notify_wq
);
2226 EXPORT_SYMBOL(ceph_osdc_flush_notifies
);
2232 int ceph_osdc_init(struct ceph_osd_client
*osdc
, struct ceph_client
*client
)
2237 osdc
->client
= client
;
2238 osdc
->osdmap
= NULL
;
2239 init_rwsem(&osdc
->map_sem
);
2240 init_completion(&osdc
->map_waiters
);
2241 osdc
->last_requested_map
= 0;
2242 mutex_init(&osdc
->request_mutex
);
2244 osdc
->osds
= RB_ROOT
;
2245 INIT_LIST_HEAD(&osdc
->osd_lru
);
2246 osdc
->requests
= RB_ROOT
;
2247 INIT_LIST_HEAD(&osdc
->req_lru
);
2248 INIT_LIST_HEAD(&osdc
->req_unsent
);
2249 INIT_LIST_HEAD(&osdc
->req_notarget
);
2250 INIT_LIST_HEAD(&osdc
->req_linger
);
2251 osdc
->num_requests
= 0;
2252 INIT_DELAYED_WORK(&osdc
->timeout_work
, handle_timeout
);
2253 INIT_DELAYED_WORK(&osdc
->osds_timeout_work
, handle_osds_timeout
);
2254 spin_lock_init(&osdc
->event_lock
);
2255 osdc
->event_tree
= RB_ROOT
;
2256 osdc
->event_count
= 0;
2258 schedule_delayed_work(&osdc
->osds_timeout_work
,
2259 round_jiffies_relative(osdc
->client
->options
->osd_idle_ttl
* HZ
));
2262 osdc
->req_mempool
= mempool_create_kmalloc_pool(10,
2263 sizeof(struct ceph_osd_request
));
2264 if (!osdc
->req_mempool
)
2267 err
= ceph_msgpool_init(&osdc
->msgpool_op
, CEPH_MSG_OSD_OP
,
2268 OSD_OP_FRONT_LEN
, 10, true,
2272 err
= ceph_msgpool_init(&osdc
->msgpool_op_reply
, CEPH_MSG_OSD_OPREPLY
,
2273 OSD_OPREPLY_FRONT_LEN
, 10, true,
2279 osdc
->notify_wq
= create_singlethread_workqueue("ceph-watch-notify");
2280 if (!osdc
->notify_wq
)
2285 ceph_msgpool_destroy(&osdc
->msgpool_op
);
2287 mempool_destroy(osdc
->req_mempool
);
2292 void ceph_osdc_stop(struct ceph_osd_client
*osdc
)
2294 flush_workqueue(osdc
->notify_wq
);
2295 destroy_workqueue(osdc
->notify_wq
);
2296 cancel_delayed_work_sync(&osdc
->timeout_work
);
2297 cancel_delayed_work_sync(&osdc
->osds_timeout_work
);
2299 ceph_osdmap_destroy(osdc
->osdmap
);
2300 osdc
->osdmap
= NULL
;
2302 remove_all_osds(osdc
);
2303 mempool_destroy(osdc
->req_mempool
);
2304 ceph_msgpool_destroy(&osdc
->msgpool_op
);
2305 ceph_msgpool_destroy(&osdc
->msgpool_op_reply
);
2309 * Read some contiguous pages. If we cross a stripe boundary, shorten
2310 * *plen. Return number of bytes read, or error.
2312 int ceph_osdc_readpages(struct ceph_osd_client
*osdc
,
2313 struct ceph_vino vino
, struct ceph_file_layout
*layout
,
2315 u32 truncate_seq
, u64 truncate_size
,
2316 struct page
**pages
, int num_pages
, int page_align
)
2318 struct ceph_osd_request
*req
;
2321 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino
.ino
,
2322 vino
.snap
, off
, *plen
);
2323 req
= ceph_osdc_new_request(osdc
, layout
, vino
, off
, plen
, 1,
2324 CEPH_OSD_OP_READ
, CEPH_OSD_FLAG_READ
,
2325 NULL
, truncate_seq
, truncate_size
,
2328 return PTR_ERR(req
);
2330 /* it may be a short read due to an object boundary */
2332 osd_req_op_extent_osd_data_pages(req
, 0,
2333 pages
, *plen
, page_align
, false, false);
2335 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
2336 off
, *plen
, *plen
, page_align
);
2338 ceph_osdc_build_request(req
, off
, NULL
, vino
.snap
, NULL
);
2340 rc
= ceph_osdc_start_request(osdc
, req
, false);
2342 rc
= ceph_osdc_wait_request(osdc
, req
);
2344 ceph_osdc_put_request(req
);
2345 dout("readpages result %d\n", rc
);
2348 EXPORT_SYMBOL(ceph_osdc_readpages
);
2351 * do a synchronous write on N pages
2353 int ceph_osdc_writepages(struct ceph_osd_client
*osdc
, struct ceph_vino vino
,
2354 struct ceph_file_layout
*layout
,
2355 struct ceph_snap_context
*snapc
,
2357 u32 truncate_seq
, u64 truncate_size
,
2358 struct timespec
*mtime
,
2359 struct page
**pages
, int num_pages
)
2361 struct ceph_osd_request
*req
;
2363 int page_align
= off
& ~PAGE_MASK
;
2365 BUG_ON(vino
.snap
!= CEPH_NOSNAP
); /* snapshots aren't writeable */
2366 req
= ceph_osdc_new_request(osdc
, layout
, vino
, off
, &len
, 1,
2368 CEPH_OSD_FLAG_ONDISK
| CEPH_OSD_FLAG_WRITE
,
2369 snapc
, truncate_seq
, truncate_size
,
2372 return PTR_ERR(req
);
2374 /* it may be a short write due to an object boundary */
2375 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, page_align
,
2377 dout("writepages %llu~%llu (%llu bytes)\n", off
, len
, len
);
2379 ceph_osdc_build_request(req
, off
, snapc
, CEPH_NOSNAP
, mtime
);
2381 rc
= ceph_osdc_start_request(osdc
, req
, true);
2383 rc
= ceph_osdc_wait_request(osdc
, req
);
2385 ceph_osdc_put_request(req
);
2388 dout("writepages result %d\n", rc
);
2391 EXPORT_SYMBOL(ceph_osdc_writepages
);
2393 int ceph_osdc_setup(void)
2395 BUG_ON(ceph_osd_request_cache
);
2396 ceph_osd_request_cache
= kmem_cache_create("ceph_osd_request",
2397 sizeof (struct ceph_osd_request
),
2398 __alignof__(struct ceph_osd_request
),
2401 return ceph_osd_request_cache
? 0 : -ENOMEM
;
2403 EXPORT_SYMBOL(ceph_osdc_setup
);
2405 void ceph_osdc_cleanup(void)
2407 BUG_ON(!ceph_osd_request_cache
);
2408 kmem_cache_destroy(ceph_osd_request_cache
);
2409 ceph_osd_request_cache
= NULL
;
2411 EXPORT_SYMBOL(ceph_osdc_cleanup
);
2414 * handle incoming message
2416 static void dispatch(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2418 struct ceph_osd
*osd
= con
->private;
2419 struct ceph_osd_client
*osdc
;
2420 int type
= le16_to_cpu(msg
->hdr
.type
);
2427 case CEPH_MSG_OSD_MAP
:
2428 ceph_osdc_handle_map(osdc
, msg
);
2430 case CEPH_MSG_OSD_OPREPLY
:
2431 handle_reply(osdc
, msg
, con
);
2433 case CEPH_MSG_WATCH_NOTIFY
:
2434 handle_watch_notify(osdc
, msg
);
2438 pr_err("received unknown message type %d %s\n", type
,
2439 ceph_msg_type_name(type
));
2446 * lookup and return message for incoming reply. set up reply message
2449 static struct ceph_msg
*get_reply(struct ceph_connection
*con
,
2450 struct ceph_msg_header
*hdr
,
2453 struct ceph_osd
*osd
= con
->private;
2454 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
2456 struct ceph_osd_request
*req
;
2457 int front
= le32_to_cpu(hdr
->front_len
);
2458 int data_len
= le32_to_cpu(hdr
->data_len
);
2461 tid
= le64_to_cpu(hdr
->tid
);
2462 mutex_lock(&osdc
->request_mutex
);
2463 req
= __lookup_request(osdc
, tid
);
2467 dout("get_reply unknown tid %llu from osd%d\n", tid
,
2472 if (req
->r_reply
->con
)
2473 dout("%s revoking msg %p from old con %p\n", __func__
,
2474 req
->r_reply
, req
->r_reply
->con
);
2475 ceph_msg_revoke_incoming(req
->r_reply
);
2477 if (front
> req
->r_reply
->front
.iov_len
) {
2478 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2479 front
, (int)req
->r_reply
->front
.iov_len
,
2480 (unsigned int)con
->peer_name
.type
,
2481 le64_to_cpu(con
->peer_name
.num
));
2482 m
= ceph_msg_new(CEPH_MSG_OSD_OPREPLY
, front
, GFP_NOFS
, false);
2485 ceph_msg_put(req
->r_reply
);
2488 m
= ceph_msg_get(req
->r_reply
);
2491 struct ceph_osd_data
*osd_data
;
2494 * XXX This is assuming there is only one op containing
2495 * XXX page data. Probably OK for reads, but this
2496 * XXX ought to be done more generally.
2498 osd_data
= osd_req_op_extent_osd_data(req
, 0);
2499 if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGES
) {
2500 if (osd_data
->pages
&&
2501 unlikely(osd_data
->length
< data_len
)) {
2503 pr_warning("tid %lld reply has %d bytes "
2504 "we had only %llu bytes ready\n",
2505 tid
, data_len
, osd_data
->length
);
2514 dout("get_reply tid %lld %p\n", tid
, m
);
2517 mutex_unlock(&osdc
->request_mutex
);
2522 static struct ceph_msg
*alloc_msg(struct ceph_connection
*con
,
2523 struct ceph_msg_header
*hdr
,
2526 struct ceph_osd
*osd
= con
->private;
2527 int type
= le16_to_cpu(hdr
->type
);
2528 int front
= le32_to_cpu(hdr
->front_len
);
2532 case CEPH_MSG_OSD_MAP
:
2533 case CEPH_MSG_WATCH_NOTIFY
:
2534 return ceph_msg_new(type
, front
, GFP_NOFS
, false);
2535 case CEPH_MSG_OSD_OPREPLY
:
2536 return get_reply(con
, hdr
, skip
);
2538 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type
,
2546 * Wrappers to refcount containing ceph_osd struct
2548 static struct ceph_connection
*get_osd_con(struct ceph_connection
*con
)
2550 struct ceph_osd
*osd
= con
->private;
2556 static void put_osd_con(struct ceph_connection
*con
)
2558 struct ceph_osd
*osd
= con
->private;
2566 * Note: returned pointer is the address of a structure that's
2567 * managed separately. Caller must *not* attempt to free it.
2569 static struct ceph_auth_handshake
*get_authorizer(struct ceph_connection
*con
,
2570 int *proto
, int force_new
)
2572 struct ceph_osd
*o
= con
->private;
2573 struct ceph_osd_client
*osdc
= o
->o_osdc
;
2574 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
2575 struct ceph_auth_handshake
*auth
= &o
->o_auth
;
2577 if (force_new
&& auth
->authorizer
) {
2578 ceph_auth_destroy_authorizer(ac
, auth
->authorizer
);
2579 auth
->authorizer
= NULL
;
2581 if (!auth
->authorizer
) {
2582 int ret
= ceph_auth_create_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
,
2585 return ERR_PTR(ret
);
2587 int ret
= ceph_auth_update_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
,
2590 return ERR_PTR(ret
);
2592 *proto
= ac
->protocol
;
2598 static int verify_authorizer_reply(struct ceph_connection
*con
, int len
)
2600 struct ceph_osd
*o
= con
->private;
2601 struct ceph_osd_client
*osdc
= o
->o_osdc
;
2602 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
2604 return ceph_auth_verify_authorizer_reply(ac
, o
->o_auth
.authorizer
, len
);
2607 static int invalidate_authorizer(struct ceph_connection
*con
)
2609 struct ceph_osd
*o
= con
->private;
2610 struct ceph_osd_client
*osdc
= o
->o_osdc
;
2611 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
2613 ceph_auth_invalidate_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
);
2614 return ceph_monc_validate_auth(&osdc
->client
->monc
);
2617 static const struct ceph_connection_operations osd_con_ops
= {
2620 .dispatch
= dispatch
,
2621 .get_authorizer
= get_authorizer
,
2622 .verify_authorizer_reply
= verify_authorizer_reply
,
2623 .invalidate_authorizer
= invalidate_authorizer
,
2624 .alloc_msg
= alloc_msg
,