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_OPREPLY_FRONT_LEN 512
24 static struct kmem_cache
*ceph_osd_request_cache
;
26 static const struct ceph_connection_operations osd_con_ops
;
28 static void __send_queued(struct ceph_osd_client
*osdc
);
29 static int __reset_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
);
30 static void __register_request(struct ceph_osd_client
*osdc
,
31 struct ceph_osd_request
*req
);
32 static void __unregister_request(struct ceph_osd_client
*osdc
,
33 struct ceph_osd_request
*req
);
34 static void __unregister_linger_request(struct ceph_osd_client
*osdc
,
35 struct ceph_osd_request
*req
);
36 static void __enqueue_request(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 struct ceph_osd_request *__oreq = (oreq); \
123 unsigned int __whch = (whch); \
124 BUG_ON(__whch >= __oreq->r_num_ops); \
125 &__oreq->r_ops[__whch].typ.fld; \
128 static struct ceph_osd_data
*
129 osd_req_op_raw_data_in(struct ceph_osd_request
*osd_req
, unsigned int which
)
131 BUG_ON(which
>= osd_req
->r_num_ops
);
133 return &osd_req
->r_ops
[which
].raw_data_in
;
136 struct ceph_osd_data
*
137 osd_req_op_extent_osd_data(struct ceph_osd_request
*osd_req
,
140 return osd_req_op_data(osd_req
, which
, extent
, osd_data
);
142 EXPORT_SYMBOL(osd_req_op_extent_osd_data
);
144 void osd_req_op_raw_data_in_pages(struct ceph_osd_request
*osd_req
,
145 unsigned int which
, struct page
**pages
,
146 u64 length
, u32 alignment
,
147 bool pages_from_pool
, bool own_pages
)
149 struct ceph_osd_data
*osd_data
;
151 osd_data
= osd_req_op_raw_data_in(osd_req
, which
);
152 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
153 pages_from_pool
, own_pages
);
155 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages
);
157 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request
*osd_req
,
158 unsigned int which
, struct page
**pages
,
159 u64 length
, u32 alignment
,
160 bool pages_from_pool
, bool own_pages
)
162 struct ceph_osd_data
*osd_data
;
164 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
165 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
166 pages_from_pool
, own_pages
);
168 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages
);
170 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request
*osd_req
,
171 unsigned int which
, struct ceph_pagelist
*pagelist
)
173 struct ceph_osd_data
*osd_data
;
175 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
176 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
178 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist
);
181 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request
*osd_req
,
182 unsigned int which
, struct bio
*bio
, size_t bio_length
)
184 struct ceph_osd_data
*osd_data
;
186 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
187 ceph_osd_data_bio_init(osd_data
, bio
, bio_length
);
189 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio
);
190 #endif /* CONFIG_BLOCK */
192 static void osd_req_op_cls_request_info_pagelist(
193 struct ceph_osd_request
*osd_req
,
194 unsigned int which
, struct ceph_pagelist
*pagelist
)
196 struct ceph_osd_data
*osd_data
;
198 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_info
);
199 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
202 void osd_req_op_cls_request_data_pagelist(
203 struct ceph_osd_request
*osd_req
,
204 unsigned int which
, struct ceph_pagelist
*pagelist
)
206 struct ceph_osd_data
*osd_data
;
208 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_data
);
209 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
210 osd_req
->r_ops
[which
].cls
.indata_len
+= pagelist
->length
;
211 osd_req
->r_ops
[which
].indata_len
+= pagelist
->length
;
213 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist
);
215 void osd_req_op_cls_request_data_pages(struct ceph_osd_request
*osd_req
,
216 unsigned int which
, struct page
**pages
, u64 length
,
217 u32 alignment
, bool pages_from_pool
, bool own_pages
)
219 struct ceph_osd_data
*osd_data
;
221 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_data
);
222 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
223 pages_from_pool
, own_pages
);
224 osd_req
->r_ops
[which
].cls
.indata_len
+= length
;
225 osd_req
->r_ops
[which
].indata_len
+= length
;
227 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages
);
229 void osd_req_op_cls_response_data_pages(struct ceph_osd_request
*osd_req
,
230 unsigned int which
, struct page
**pages
, u64 length
,
231 u32 alignment
, bool pages_from_pool
, bool own_pages
)
233 struct ceph_osd_data
*osd_data
;
235 osd_data
= osd_req_op_data(osd_req
, which
, cls
, response_data
);
236 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
237 pages_from_pool
, own_pages
);
239 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages
);
241 static u64
ceph_osd_data_length(struct ceph_osd_data
*osd_data
)
243 switch (osd_data
->type
) {
244 case CEPH_OSD_DATA_TYPE_NONE
:
246 case CEPH_OSD_DATA_TYPE_PAGES
:
247 return osd_data
->length
;
248 case CEPH_OSD_DATA_TYPE_PAGELIST
:
249 return (u64
)osd_data
->pagelist
->length
;
251 case CEPH_OSD_DATA_TYPE_BIO
:
252 return (u64
)osd_data
->bio_length
;
253 #endif /* CONFIG_BLOCK */
255 WARN(true, "unrecognized data type %d\n", (int)osd_data
->type
);
260 static void ceph_osd_data_release(struct ceph_osd_data
*osd_data
)
262 if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGES
&& osd_data
->own_pages
) {
265 num_pages
= calc_pages_for((u64
)osd_data
->alignment
,
266 (u64
)osd_data
->length
);
267 ceph_release_page_vector(osd_data
->pages
, num_pages
);
269 ceph_osd_data_init(osd_data
);
272 static void osd_req_op_data_release(struct ceph_osd_request
*osd_req
,
275 struct ceph_osd_req_op
*op
;
277 BUG_ON(which
>= osd_req
->r_num_ops
);
278 op
= &osd_req
->r_ops
[which
];
281 case CEPH_OSD_OP_READ
:
282 case CEPH_OSD_OP_WRITE
:
283 case CEPH_OSD_OP_WRITEFULL
:
284 ceph_osd_data_release(&op
->extent
.osd_data
);
286 case CEPH_OSD_OP_CALL
:
287 ceph_osd_data_release(&op
->cls
.request_info
);
288 ceph_osd_data_release(&op
->cls
.request_data
);
289 ceph_osd_data_release(&op
->cls
.response_data
);
291 case CEPH_OSD_OP_SETXATTR
:
292 case CEPH_OSD_OP_CMPXATTR
:
293 ceph_osd_data_release(&op
->xattr
.osd_data
);
295 case CEPH_OSD_OP_STAT
:
296 ceph_osd_data_release(&op
->raw_data_in
);
304 * Assumes @t is zero-initialized.
306 static void target_init(struct ceph_osd_request_target
*t
)
308 ceph_oid_init(&t
->base_oid
);
309 ceph_oloc_init(&t
->base_oloc
);
310 ceph_oid_init(&t
->target_oid
);
311 ceph_oloc_init(&t
->target_oloc
);
313 ceph_osds_init(&t
->acting
);
314 ceph_osds_init(&t
->up
);
318 t
->osd
= CEPH_HOMELESS_OSD
;
321 static void target_destroy(struct ceph_osd_request_target
*t
)
323 ceph_oid_destroy(&t
->base_oid
);
324 ceph_oid_destroy(&t
->target_oid
);
330 static void ceph_osdc_release_request(struct kref
*kref
)
332 struct ceph_osd_request
*req
= container_of(kref
,
333 struct ceph_osd_request
, r_kref
);
336 dout("%s %p (r_request %p r_reply %p)\n", __func__
, req
,
337 req
->r_request
, req
->r_reply
);
338 WARN_ON(!RB_EMPTY_NODE(&req
->r_node
));
339 WARN_ON(!list_empty(&req
->r_req_lru_item
));
340 WARN_ON(!list_empty(&req
->r_osd_item
));
341 WARN_ON(!list_empty(&req
->r_linger_item
));
342 WARN_ON(!list_empty(&req
->r_linger_osd_item
));
346 ceph_msg_put(req
->r_request
);
348 ceph_msg_revoke_incoming(req
->r_reply
);
349 ceph_msg_put(req
->r_reply
);
352 for (which
= 0; which
< req
->r_num_ops
; which
++)
353 osd_req_op_data_release(req
, which
);
355 target_destroy(&req
->r_t
);
356 ceph_put_snap_context(req
->r_snapc
);
359 mempool_free(req
, req
->r_osdc
->req_mempool
);
360 else if (req
->r_num_ops
<= CEPH_OSD_SLAB_OPS
)
361 kmem_cache_free(ceph_osd_request_cache
, req
);
366 void ceph_osdc_get_request(struct ceph_osd_request
*req
)
368 dout("%s %p (was %d)\n", __func__
, req
,
369 atomic_read(&req
->r_kref
.refcount
));
370 kref_get(&req
->r_kref
);
372 EXPORT_SYMBOL(ceph_osdc_get_request
);
374 void ceph_osdc_put_request(struct ceph_osd_request
*req
)
377 dout("%s %p (was %d)\n", __func__
, req
,
378 atomic_read(&req
->r_kref
.refcount
));
379 kref_put(&req
->r_kref
, ceph_osdc_release_request
);
382 EXPORT_SYMBOL(ceph_osdc_put_request
);
384 struct ceph_osd_request
*ceph_osdc_alloc_request(struct ceph_osd_client
*osdc
,
385 struct ceph_snap_context
*snapc
,
386 unsigned int num_ops
,
390 struct ceph_osd_request
*req
;
393 BUG_ON(num_ops
> CEPH_OSD_SLAB_OPS
);
394 req
= mempool_alloc(osdc
->req_mempool
, gfp_flags
);
395 } else if (num_ops
<= CEPH_OSD_SLAB_OPS
) {
396 req
= kmem_cache_alloc(ceph_osd_request_cache
, gfp_flags
);
398 BUG_ON(num_ops
> CEPH_OSD_MAX_OPS
);
399 req
= kmalloc(sizeof(*req
) + num_ops
* sizeof(req
->r_ops
[0]),
405 /* req only, each op is zeroed in _osd_req_op_init() */
406 memset(req
, 0, sizeof(*req
));
409 req
->r_mempool
= use_mempool
;
410 req
->r_num_ops
= num_ops
;
411 req
->r_snapid
= CEPH_NOSNAP
;
412 req
->r_snapc
= ceph_get_snap_context(snapc
);
414 kref_init(&req
->r_kref
);
415 init_completion(&req
->r_completion
);
416 init_completion(&req
->r_safe_completion
);
417 RB_CLEAR_NODE(&req
->r_node
);
418 INIT_LIST_HEAD(&req
->r_unsafe_item
);
419 INIT_LIST_HEAD(&req
->r_linger_item
);
420 INIT_LIST_HEAD(&req
->r_linger_osd_item
);
421 INIT_LIST_HEAD(&req
->r_req_lru_item
);
422 INIT_LIST_HEAD(&req
->r_osd_item
);
424 target_init(&req
->r_t
);
426 dout("%s req %p\n", __func__
, req
);
429 EXPORT_SYMBOL(ceph_osdc_alloc_request
);
431 int ceph_osdc_alloc_messages(struct ceph_osd_request
*req
, gfp_t gfp
)
433 struct ceph_osd_client
*osdc
= req
->r_osdc
;
434 struct ceph_msg
*msg
;
437 WARN_ON(ceph_oid_empty(&req
->r_base_oid
));
439 /* create request message */
440 msg_size
= 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */
441 msg_size
+= 4 + 4 + 4 + 8; /* mtime, reassert_version */
442 msg_size
+= 2 + 4 + 8 + 4 + 4; /* oloc */
443 msg_size
+= 1 + 8 + 4 + 4; /* pgid */
444 msg_size
+= 4 + req
->r_base_oid
.name_len
; /* oid */
445 msg_size
+= 2 + req
->r_num_ops
* sizeof(struct ceph_osd_op
);
446 msg_size
+= 8; /* snapid */
447 msg_size
+= 8; /* snap_seq */
448 msg_size
+= 4 + 8 * (req
->r_snapc
? req
->r_snapc
->num_snaps
: 0);
449 msg_size
+= 4; /* retry_attempt */
452 msg
= ceph_msgpool_get(&osdc
->msgpool_op
, 0);
454 msg
= ceph_msg_new(CEPH_MSG_OSD_OP
, msg_size
, gfp
, true);
458 memset(msg
->front
.iov_base
, 0, msg
->front
.iov_len
);
459 req
->r_request
= msg
;
461 /* create reply message */
462 msg_size
= OSD_OPREPLY_FRONT_LEN
;
463 msg_size
+= req
->r_base_oid
.name_len
;
464 msg_size
+= req
->r_num_ops
* sizeof(struct ceph_osd_op
);
467 msg
= ceph_msgpool_get(&osdc
->msgpool_op_reply
, 0);
469 msg
= ceph_msg_new(CEPH_MSG_OSD_OPREPLY
, msg_size
, gfp
, true);
477 EXPORT_SYMBOL(ceph_osdc_alloc_messages
);
479 static bool osd_req_opcode_valid(u16 opcode
)
482 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
483 __CEPH_FORALL_OSD_OPS(GENERATE_CASE
)
491 * This is an osd op init function for opcodes that have no data or
492 * other information associated with them. It also serves as a
493 * common init routine for all the other init functions, below.
495 static struct ceph_osd_req_op
*
496 _osd_req_op_init(struct ceph_osd_request
*osd_req
, unsigned int which
,
497 u16 opcode
, u32 flags
)
499 struct ceph_osd_req_op
*op
;
501 BUG_ON(which
>= osd_req
->r_num_ops
);
502 BUG_ON(!osd_req_opcode_valid(opcode
));
504 op
= &osd_req
->r_ops
[which
];
505 memset(op
, 0, sizeof (*op
));
512 void osd_req_op_init(struct ceph_osd_request
*osd_req
,
513 unsigned int which
, u16 opcode
, u32 flags
)
515 (void)_osd_req_op_init(osd_req
, which
, opcode
, flags
);
517 EXPORT_SYMBOL(osd_req_op_init
);
519 void osd_req_op_extent_init(struct ceph_osd_request
*osd_req
,
520 unsigned int which
, u16 opcode
,
521 u64 offset
, u64 length
,
522 u64 truncate_size
, u32 truncate_seq
)
524 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
526 size_t payload_len
= 0;
528 BUG_ON(opcode
!= CEPH_OSD_OP_READ
&& opcode
!= CEPH_OSD_OP_WRITE
&&
529 opcode
!= CEPH_OSD_OP_WRITEFULL
&& opcode
!= CEPH_OSD_OP_ZERO
&&
530 opcode
!= CEPH_OSD_OP_TRUNCATE
);
532 op
->extent
.offset
= offset
;
533 op
->extent
.length
= length
;
534 op
->extent
.truncate_size
= truncate_size
;
535 op
->extent
.truncate_seq
= truncate_seq
;
536 if (opcode
== CEPH_OSD_OP_WRITE
|| opcode
== CEPH_OSD_OP_WRITEFULL
)
537 payload_len
+= length
;
539 op
->indata_len
= payload_len
;
541 EXPORT_SYMBOL(osd_req_op_extent_init
);
543 void osd_req_op_extent_update(struct ceph_osd_request
*osd_req
,
544 unsigned int which
, u64 length
)
546 struct ceph_osd_req_op
*op
;
549 BUG_ON(which
>= osd_req
->r_num_ops
);
550 op
= &osd_req
->r_ops
[which
];
551 previous
= op
->extent
.length
;
553 if (length
== previous
)
554 return; /* Nothing to do */
555 BUG_ON(length
> previous
);
557 op
->extent
.length
= length
;
558 op
->indata_len
-= previous
- length
;
560 EXPORT_SYMBOL(osd_req_op_extent_update
);
562 void osd_req_op_extent_dup_last(struct ceph_osd_request
*osd_req
,
563 unsigned int which
, u64 offset_inc
)
565 struct ceph_osd_req_op
*op
, *prev_op
;
567 BUG_ON(which
+ 1 >= osd_req
->r_num_ops
);
569 prev_op
= &osd_req
->r_ops
[which
];
570 op
= _osd_req_op_init(osd_req
, which
+ 1, prev_op
->op
, prev_op
->flags
);
571 /* dup previous one */
572 op
->indata_len
= prev_op
->indata_len
;
573 op
->outdata_len
= prev_op
->outdata_len
;
574 op
->extent
= prev_op
->extent
;
576 op
->extent
.offset
+= offset_inc
;
577 op
->extent
.length
-= offset_inc
;
579 if (op
->op
== CEPH_OSD_OP_WRITE
|| op
->op
== CEPH_OSD_OP_WRITEFULL
)
580 op
->indata_len
-= offset_inc
;
582 EXPORT_SYMBOL(osd_req_op_extent_dup_last
);
584 void osd_req_op_cls_init(struct ceph_osd_request
*osd_req
, unsigned int which
,
585 u16 opcode
, const char *class, const char *method
)
587 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
589 struct ceph_pagelist
*pagelist
;
590 size_t payload_len
= 0;
593 BUG_ON(opcode
!= CEPH_OSD_OP_CALL
);
595 pagelist
= kmalloc(sizeof (*pagelist
), GFP_NOFS
);
597 ceph_pagelist_init(pagelist
);
599 op
->cls
.class_name
= class;
600 size
= strlen(class);
601 BUG_ON(size
> (size_t) U8_MAX
);
602 op
->cls
.class_len
= size
;
603 ceph_pagelist_append(pagelist
, class, size
);
606 op
->cls
.method_name
= method
;
607 size
= strlen(method
);
608 BUG_ON(size
> (size_t) U8_MAX
);
609 op
->cls
.method_len
= size
;
610 ceph_pagelist_append(pagelist
, method
, size
);
613 osd_req_op_cls_request_info_pagelist(osd_req
, which
, pagelist
);
615 op
->indata_len
= payload_len
;
617 EXPORT_SYMBOL(osd_req_op_cls_init
);
619 int osd_req_op_xattr_init(struct ceph_osd_request
*osd_req
, unsigned int which
,
620 u16 opcode
, const char *name
, const void *value
,
621 size_t size
, u8 cmp_op
, u8 cmp_mode
)
623 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
625 struct ceph_pagelist
*pagelist
;
628 BUG_ON(opcode
!= CEPH_OSD_OP_SETXATTR
&& opcode
!= CEPH_OSD_OP_CMPXATTR
);
630 pagelist
= kmalloc(sizeof(*pagelist
), GFP_NOFS
);
634 ceph_pagelist_init(pagelist
);
636 payload_len
= strlen(name
);
637 op
->xattr
.name_len
= payload_len
;
638 ceph_pagelist_append(pagelist
, name
, payload_len
);
640 op
->xattr
.value_len
= size
;
641 ceph_pagelist_append(pagelist
, value
, size
);
644 op
->xattr
.cmp_op
= cmp_op
;
645 op
->xattr
.cmp_mode
= cmp_mode
;
647 ceph_osd_data_pagelist_init(&op
->xattr
.osd_data
, pagelist
);
648 op
->indata_len
= payload_len
;
651 EXPORT_SYMBOL(osd_req_op_xattr_init
);
653 void osd_req_op_watch_init(struct ceph_osd_request
*osd_req
,
654 unsigned int which
, u16 opcode
,
655 u64 cookie
, u64 version
, int flag
)
657 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
660 BUG_ON(opcode
!= CEPH_OSD_OP_NOTIFY_ACK
&& opcode
!= CEPH_OSD_OP_WATCH
);
662 op
->watch
.cookie
= cookie
;
663 op
->watch
.ver
= version
;
664 if (opcode
== CEPH_OSD_OP_WATCH
&& flag
)
665 op
->watch
.flag
= (u8
)1;
667 EXPORT_SYMBOL(osd_req_op_watch_init
);
669 void osd_req_op_alloc_hint_init(struct ceph_osd_request
*osd_req
,
671 u64 expected_object_size
,
672 u64 expected_write_size
)
674 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
675 CEPH_OSD_OP_SETALLOCHINT
,
678 op
->alloc_hint
.expected_object_size
= expected_object_size
;
679 op
->alloc_hint
.expected_write_size
= expected_write_size
;
682 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
683 * not worth a feature bit. Set FAILOK per-op flag to make
684 * sure older osds don't trip over an unsupported opcode.
686 op
->flags
|= CEPH_OSD_OP_FLAG_FAILOK
;
688 EXPORT_SYMBOL(osd_req_op_alloc_hint_init
);
690 static void ceph_osdc_msg_data_add(struct ceph_msg
*msg
,
691 struct ceph_osd_data
*osd_data
)
693 u64 length
= ceph_osd_data_length(osd_data
);
695 if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGES
) {
696 BUG_ON(length
> (u64
) SIZE_MAX
);
698 ceph_msg_data_add_pages(msg
, osd_data
->pages
,
699 length
, osd_data
->alignment
);
700 } else if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGELIST
) {
702 ceph_msg_data_add_pagelist(msg
, osd_data
->pagelist
);
704 } else if (osd_data
->type
== CEPH_OSD_DATA_TYPE_BIO
) {
705 ceph_msg_data_add_bio(msg
, osd_data
->bio
, length
);
708 BUG_ON(osd_data
->type
!= CEPH_OSD_DATA_TYPE_NONE
);
712 static u32
osd_req_encode_op(struct ceph_osd_op
*dst
,
713 const struct ceph_osd_req_op
*src
)
715 if (WARN_ON(!osd_req_opcode_valid(src
->op
))) {
716 pr_err("unrecognized osd opcode %d\n", src
->op
);
722 case CEPH_OSD_OP_STAT
:
724 case CEPH_OSD_OP_READ
:
725 case CEPH_OSD_OP_WRITE
:
726 case CEPH_OSD_OP_WRITEFULL
:
727 case CEPH_OSD_OP_ZERO
:
728 case CEPH_OSD_OP_TRUNCATE
:
729 dst
->extent
.offset
= cpu_to_le64(src
->extent
.offset
);
730 dst
->extent
.length
= cpu_to_le64(src
->extent
.length
);
731 dst
->extent
.truncate_size
=
732 cpu_to_le64(src
->extent
.truncate_size
);
733 dst
->extent
.truncate_seq
=
734 cpu_to_le32(src
->extent
.truncate_seq
);
736 case CEPH_OSD_OP_CALL
:
737 dst
->cls
.class_len
= src
->cls
.class_len
;
738 dst
->cls
.method_len
= src
->cls
.method_len
;
739 dst
->cls
.indata_len
= cpu_to_le32(src
->cls
.indata_len
);
741 case CEPH_OSD_OP_STARTSYNC
:
743 case CEPH_OSD_OP_NOTIFY_ACK
:
744 case CEPH_OSD_OP_WATCH
:
745 dst
->watch
.cookie
= cpu_to_le64(src
->watch
.cookie
);
746 dst
->watch
.ver
= cpu_to_le64(src
->watch
.ver
);
747 dst
->watch
.flag
= src
->watch
.flag
;
749 case CEPH_OSD_OP_SETALLOCHINT
:
750 dst
->alloc_hint
.expected_object_size
=
751 cpu_to_le64(src
->alloc_hint
.expected_object_size
);
752 dst
->alloc_hint
.expected_write_size
=
753 cpu_to_le64(src
->alloc_hint
.expected_write_size
);
755 case CEPH_OSD_OP_SETXATTR
:
756 case CEPH_OSD_OP_CMPXATTR
:
757 dst
->xattr
.name_len
= cpu_to_le32(src
->xattr
.name_len
);
758 dst
->xattr
.value_len
= cpu_to_le32(src
->xattr
.value_len
);
759 dst
->xattr
.cmp_op
= src
->xattr
.cmp_op
;
760 dst
->xattr
.cmp_mode
= src
->xattr
.cmp_mode
;
762 case CEPH_OSD_OP_CREATE
:
763 case CEPH_OSD_OP_DELETE
:
766 pr_err("unsupported osd opcode %s\n",
767 ceph_osd_op_name(src
->op
));
773 dst
->op
= cpu_to_le16(src
->op
);
774 dst
->flags
= cpu_to_le32(src
->flags
);
775 dst
->payload_len
= cpu_to_le32(src
->indata_len
);
777 return src
->indata_len
;
781 * build new request AND message, calculate layout, and adjust file
784 * if the file was recently truncated, we include information about its
785 * old and new size so that the object can be updated appropriately. (we
786 * avoid synchronously deleting truncated objects because it's slow.)
788 * if @do_sync, include a 'startsync' command so that the osd will flush
791 struct ceph_osd_request
*ceph_osdc_new_request(struct ceph_osd_client
*osdc
,
792 struct ceph_file_layout
*layout
,
793 struct ceph_vino vino
,
795 unsigned int which
, int num_ops
,
796 int opcode
, int flags
,
797 struct ceph_snap_context
*snapc
,
802 struct ceph_osd_request
*req
;
808 BUG_ON(opcode
!= CEPH_OSD_OP_READ
&& opcode
!= CEPH_OSD_OP_WRITE
&&
809 opcode
!= CEPH_OSD_OP_ZERO
&& opcode
!= CEPH_OSD_OP_TRUNCATE
&&
810 opcode
!= CEPH_OSD_OP_CREATE
&& opcode
!= CEPH_OSD_OP_DELETE
);
812 req
= ceph_osdc_alloc_request(osdc
, snapc
, num_ops
, use_mempool
,
819 /* calculate max write size */
820 r
= calc_layout(layout
, off
, plen
, &objnum
, &objoff
, &objlen
);
824 if (opcode
== CEPH_OSD_OP_CREATE
|| opcode
== CEPH_OSD_OP_DELETE
) {
825 osd_req_op_init(req
, which
, opcode
, 0);
827 u32 object_size
= le32_to_cpu(layout
->fl_object_size
);
828 u32 object_base
= off
- objoff
;
829 if (!(truncate_seq
== 1 && truncate_size
== -1ULL)) {
830 if (truncate_size
<= object_base
) {
833 truncate_size
-= object_base
;
834 if (truncate_size
> object_size
)
835 truncate_size
= object_size
;
838 osd_req_op_extent_init(req
, which
, opcode
, objoff
, objlen
,
839 truncate_size
, truncate_seq
);
842 req
->r_flags
= flags
;
843 req
->r_base_oloc
.pool
= ceph_file_layout_pg_pool(*layout
);
844 ceph_oid_printf(&req
->r_base_oid
, "%llx.%08llx", vino
.ino
, objnum
);
846 req
->r_snapid
= vino
.snap
;
847 if (flags
& CEPH_OSD_FLAG_WRITE
)
848 req
->r_data_offset
= off
;
850 r
= ceph_osdc_alloc_messages(req
, GFP_NOFS
);
857 ceph_osdc_put_request(req
);
860 EXPORT_SYMBOL(ceph_osdc_new_request
);
863 * We keep osd requests in an rbtree, sorted by ->r_tid.
865 DEFINE_RB_FUNCS(request
, struct ceph_osd_request
, r_tid
, r_node
)
867 static bool osd_homeless(struct ceph_osd
*osd
)
869 return osd
->o_osd
== CEPH_HOMELESS_OSD
;
872 static struct ceph_osd_request
*
873 __lookup_request_ge(struct ceph_osd_client
*osdc
,
876 struct ceph_osd_request
*req
;
877 struct rb_node
*n
= osdc
->requests
.rb_node
;
880 req
= rb_entry(n
, struct ceph_osd_request
, r_node
);
881 if (tid
< req
->r_tid
) {
885 } else if (tid
> req
->r_tid
) {
894 static void __kick_linger_request(struct ceph_osd_request
*req
)
896 struct ceph_osd_client
*osdc
= req
->r_osdc
;
897 struct ceph_osd
*osd
= req
->r_osd
;
900 * Linger requests need to be resent with a new tid to avoid
901 * the dup op detection logic on the OSDs. Achieve this with
902 * a re-register dance instead of open-coding.
904 ceph_osdc_get_request(req
);
905 if (!list_empty(&req
->r_linger_item
))
906 __unregister_linger_request(osdc
, req
);
908 __unregister_request(osdc
, req
);
909 __register_request(osdc
, req
);
910 ceph_osdc_put_request(req
);
913 * Unless request has been registered as both normal and
914 * lingering, __unregister{,_linger}_request clears r_osd.
915 * However, here we need to preserve r_osd to make sure we
916 * requeue on the same OSD.
918 WARN_ON(req
->r_osd
|| !osd
);
921 dout("%s requeueing %p tid %llu\n", __func__
, req
, req
->r_tid
);
922 __enqueue_request(req
);
926 * Resubmit requests pending on the given osd.
928 static void __kick_osd_requests(struct ceph_osd_client
*osdc
,
929 struct ceph_osd
*osd
)
931 struct ceph_osd_request
*req
, *nreq
;
933 LIST_HEAD(resend_linger
);
936 dout("%s osd%d\n", __func__
, osd
->o_osd
);
937 err
= __reset_osd(osdc
, osd
);
942 * Build up a list of requests to resend by traversing the
943 * osd's list of requests. Requests for a given object are
944 * sent in tid order, and that is also the order they're
945 * kept on this list. Therefore all requests that are in
946 * flight will be found first, followed by all requests that
947 * have not yet been sent. And to resend requests while
948 * preserving this order we will want to put any sent
949 * requests back on the front of the osd client's unsent
952 * So we build a separate ordered list of already-sent
953 * requests for the affected osd and splice it onto the
954 * front of the osd client's unsent list. Once we've seen a
955 * request that has not yet been sent we're done. Those
956 * requests are already sitting right where they belong.
958 list_for_each_entry(req
, &osd
->o_requests
, r_osd_item
) {
962 if (!req
->r_linger
) {
963 dout("%s requeueing %p tid %llu\n", __func__
, req
,
965 list_move_tail(&req
->r_req_lru_item
, &resend
);
966 req
->r_flags
|= CEPH_OSD_FLAG_RETRY
;
968 list_move_tail(&req
->r_req_lru_item
, &resend_linger
);
971 list_splice(&resend
, &osdc
->req_unsent
);
974 * Both registered and not yet registered linger requests are
975 * enqueued with a new tid on the same OSD. We add/move them
976 * to req_unsent/o_requests at the end to keep things in tid
979 list_for_each_entry_safe(req
, nreq
, &osd
->o_linger_requests
,
981 WARN_ON(!list_empty(&req
->r_req_lru_item
));
982 __kick_linger_request(req
);
985 list_for_each_entry_safe(req
, nreq
, &resend_linger
, r_req_lru_item
)
986 __kick_linger_request(req
);
990 * If the osd connection drops, we need to resubmit all requests.
992 static void osd_reset(struct ceph_connection
*con
)
994 struct ceph_osd
*osd
= con
->private;
995 struct ceph_osd_client
*osdc
;
999 dout("osd_reset osd%d\n", osd
->o_osd
);
1001 down_read(&osdc
->map_sem
);
1002 mutex_lock(&osdc
->request_mutex
);
1003 __kick_osd_requests(osdc
, osd
);
1004 __send_queued(osdc
);
1005 mutex_unlock(&osdc
->request_mutex
);
1006 up_read(&osdc
->map_sem
);
1010 * Assumes @osd is zero-initialized.
1012 static void osd_init(struct ceph_osd
*osd
)
1014 atomic_set(&osd
->o_ref
, 1);
1015 RB_CLEAR_NODE(&osd
->o_node
);
1016 INIT_LIST_HEAD(&osd
->o_requests
);
1017 INIT_LIST_HEAD(&osd
->o_linger_requests
);
1018 INIT_LIST_HEAD(&osd
->o_osd_lru
);
1019 INIT_LIST_HEAD(&osd
->o_keepalive_item
);
1020 osd
->o_incarnation
= 1;
1023 static void osd_cleanup(struct ceph_osd
*osd
)
1025 WARN_ON(!RB_EMPTY_NODE(&osd
->o_node
));
1026 WARN_ON(!list_empty(&osd
->o_requests
));
1027 WARN_ON(!list_empty(&osd
->o_linger_requests
));
1028 WARN_ON(!list_empty(&osd
->o_osd_lru
));
1029 WARN_ON(!list_empty(&osd
->o_keepalive_item
));
1031 if (osd
->o_auth
.authorizer
) {
1032 WARN_ON(osd_homeless(osd
));
1033 ceph_auth_destroy_authorizer(osd
->o_auth
.authorizer
);
1038 * Track open sessions with osds.
1040 static struct ceph_osd
*create_osd(struct ceph_osd_client
*osdc
, int onum
)
1042 struct ceph_osd
*osd
;
1044 WARN_ON(onum
== CEPH_HOMELESS_OSD
);
1046 osd
= kzalloc(sizeof(*osd
), GFP_NOIO
| __GFP_NOFAIL
);
1051 ceph_con_init(&osd
->o_con
, osd
, &osd_con_ops
, &osdc
->client
->msgr
);
1056 static struct ceph_osd
*get_osd(struct ceph_osd
*osd
)
1058 if (atomic_inc_not_zero(&osd
->o_ref
)) {
1059 dout("get_osd %p %d -> %d\n", osd
, atomic_read(&osd
->o_ref
)-1,
1060 atomic_read(&osd
->o_ref
));
1063 dout("get_osd %p FAIL\n", osd
);
1068 static void put_osd(struct ceph_osd
*osd
)
1070 dout("put_osd %p %d -> %d\n", osd
, atomic_read(&osd
->o_ref
),
1071 atomic_read(&osd
->o_ref
) - 1);
1072 if (atomic_dec_and_test(&osd
->o_ref
)) {
1078 DEFINE_RB_FUNCS(osd
, struct ceph_osd
, o_osd
, o_node
)
1081 * remove an osd from our map
1083 static void __remove_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
)
1085 dout("%s %p osd%d\n", __func__
, osd
, osd
->o_osd
);
1086 WARN_ON(!list_empty(&osd
->o_requests
));
1087 WARN_ON(!list_empty(&osd
->o_linger_requests
));
1089 list_del_init(&osd
->o_osd_lru
);
1090 erase_osd(&osdc
->osds
, osd
);
1093 static void remove_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
)
1095 dout("%s %p osd%d\n", __func__
, osd
, osd
->o_osd
);
1097 if (!RB_EMPTY_NODE(&osd
->o_node
)) {
1098 ceph_con_close(&osd
->o_con
);
1099 __remove_osd(osdc
, osd
);
1104 static void __move_osd_to_lru(struct ceph_osd
*osd
)
1106 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
1108 dout("%s osd %p osd%d\n", __func__
, osd
, osd
->o_osd
);
1109 BUG_ON(!list_empty(&osd
->o_osd_lru
));
1111 spin_lock(&osdc
->osd_lru_lock
);
1112 list_add_tail(&osd
->o_osd_lru
, &osdc
->osd_lru
);
1113 spin_unlock(&osdc
->osd_lru_lock
);
1115 osd
->lru_ttl
= jiffies
+ osdc
->client
->options
->osd_idle_ttl
;
1118 static void maybe_move_osd_to_lru(struct ceph_osd
*osd
)
1120 if (list_empty(&osd
->o_requests
) &&
1121 list_empty(&osd
->o_linger_requests
))
1122 __move_osd_to_lru(osd
);
1125 static void __remove_osd_from_lru(struct ceph_osd
*osd
)
1127 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
1129 dout("%s osd %p osd%d\n", __func__
, osd
, osd
->o_osd
);
1131 spin_lock(&osdc
->osd_lru_lock
);
1132 if (!list_empty(&osd
->o_osd_lru
))
1133 list_del_init(&osd
->o_osd_lru
);
1134 spin_unlock(&osdc
->osd_lru_lock
);
1140 static int __reset_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
)
1142 struct ceph_entity_addr
*peer_addr
;
1144 dout("__reset_osd %p osd%d\n", osd
, osd
->o_osd
);
1145 if (list_empty(&osd
->o_requests
) &&
1146 list_empty(&osd
->o_linger_requests
)) {
1147 remove_osd(osdc
, osd
);
1151 peer_addr
= &osdc
->osdmap
->osd_addr
[osd
->o_osd
];
1152 if (!memcmp(peer_addr
, &osd
->o_con
.peer_addr
, sizeof (*peer_addr
)) &&
1153 !ceph_con_opened(&osd
->o_con
)) {
1154 struct ceph_osd_request
*req
;
1156 dout("osd addr hasn't changed and connection never opened, "
1157 "letting msgr retry\n");
1158 /* touch each r_stamp for handle_timeout()'s benfit */
1159 list_for_each_entry(req
, &osd
->o_requests
, r_osd_item
)
1160 req
->r_stamp
= jiffies
;
1165 ceph_con_close(&osd
->o_con
);
1166 ceph_con_open(&osd
->o_con
, CEPH_ENTITY_TYPE_OSD
, osd
->o_osd
, peer_addr
);
1167 osd
->o_incarnation
++;
1173 * Register request, assign tid. If this is the first request, set up
1174 * the timeout event.
1176 static void __register_request(struct ceph_osd_client
*osdc
,
1177 struct ceph_osd_request
*req
)
1179 req
->r_tid
= ++osdc
->last_tid
;
1180 req
->r_request
->hdr
.tid
= cpu_to_le64(req
->r_tid
);
1181 dout("__register_request %p tid %lld\n", req
, req
->r_tid
);
1182 insert_request(&osdc
->requests
, req
);
1183 ceph_osdc_get_request(req
);
1184 osdc
->num_requests
++;
1188 * called under osdc->request_mutex
1190 static void __unregister_request(struct ceph_osd_client
*osdc
,
1191 struct ceph_osd_request
*req
)
1193 if (RB_EMPTY_NODE(&req
->r_node
)) {
1194 dout("__unregister_request %p tid %lld not registered\n",
1199 dout("__unregister_request %p tid %lld\n", req
, req
->r_tid
);
1200 erase_request(&osdc
->requests
, req
);
1201 osdc
->num_requests
--;
1204 /* make sure the original request isn't in flight. */
1205 ceph_msg_revoke(req
->r_request
);
1207 list_del_init(&req
->r_osd_item
);
1208 maybe_move_osd_to_lru(req
->r_osd
);
1209 if (list_empty(&req
->r_linger_osd_item
))
1213 list_del_init(&req
->r_req_lru_item
);
1214 ceph_osdc_put_request(req
);
1218 * Cancel a previously queued request message
1220 static void __cancel_request(struct ceph_osd_request
*req
)
1222 if (req
->r_sent
&& req
->r_osd
) {
1223 ceph_msg_revoke(req
->r_request
);
1228 static void __register_linger_request(struct ceph_osd_client
*osdc
,
1229 struct ceph_osd_request
*req
)
1231 dout("%s %p tid %llu\n", __func__
, req
, req
->r_tid
);
1232 WARN_ON(!req
->r_linger
);
1234 ceph_osdc_get_request(req
);
1235 list_add_tail(&req
->r_linger_item
, &osdc
->req_linger
);
1237 list_add_tail(&req
->r_linger_osd_item
,
1238 &req
->r_osd
->o_linger_requests
);
1241 static void __unregister_linger_request(struct ceph_osd_client
*osdc
,
1242 struct ceph_osd_request
*req
)
1244 WARN_ON(!req
->r_linger
);
1246 if (list_empty(&req
->r_linger_item
)) {
1247 dout("%s %p tid %llu not registered\n", __func__
, req
,
1252 dout("%s %p tid %llu\n", __func__
, req
, req
->r_tid
);
1253 list_del_init(&req
->r_linger_item
);
1256 list_del_init(&req
->r_linger_osd_item
);
1257 maybe_move_osd_to_lru(req
->r_osd
);
1258 if (list_empty(&req
->r_osd_item
))
1261 ceph_osdc_put_request(req
);
1264 void ceph_osdc_set_request_linger(struct ceph_osd_client
*osdc
,
1265 struct ceph_osd_request
*req
)
1267 if (!req
->r_linger
) {
1268 dout("set_request_linger %p\n", req
);
1272 EXPORT_SYMBOL(ceph_osdc_set_request_linger
);
1274 static bool __pool_full(struct ceph_pg_pool_info
*pi
)
1276 return pi
->flags
& CEPH_POOL_FLAG_FULL
;
1279 static bool have_pool_full(struct ceph_osd_client
*osdc
)
1283 for (n
= rb_first(&osdc
->osdmap
->pg_pools
); n
; n
= rb_next(n
)) {
1284 struct ceph_pg_pool_info
*pi
=
1285 rb_entry(n
, struct ceph_pg_pool_info
, node
);
1287 if (__pool_full(pi
))
1295 * Returns whether a request should be blocked from being sent
1296 * based on the current osdmap and osd_client settings.
1298 * Caller should hold map_sem for read.
1300 static bool target_should_be_paused(struct ceph_osd_client
*osdc
,
1301 const struct ceph_osd_request_target
*t
,
1302 struct ceph_pg_pool_info
*pi
)
1304 bool pauserd
= ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_PAUSERD
);
1305 bool pausewr
= ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_PAUSEWR
) ||
1306 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) ||
1309 WARN_ON(pi
->id
!= t
->base_oloc
.pool
);
1310 return (t
->flags
& CEPH_OSD_FLAG_READ
&& pauserd
) ||
1311 (t
->flags
& CEPH_OSD_FLAG_WRITE
&& pausewr
);
1314 enum calc_target_result
{
1315 CALC_TARGET_NO_ACTION
= 0,
1316 CALC_TARGET_NEED_RESEND
,
1317 CALC_TARGET_POOL_DNE
,
1320 static enum calc_target_result
calc_target(struct ceph_osd_client
*osdc
,
1321 struct ceph_osd_request_target
*t
,
1322 u32
*last_force_resend
,
1325 struct ceph_pg_pool_info
*pi
;
1326 struct ceph_pg pgid
, last_pgid
;
1327 struct ceph_osds up
, acting
;
1328 bool force_resend
= false;
1329 bool need_check_tiering
= false;
1330 bool need_resend
= false;
1331 bool sort_bitwise
= ceph_osdmap_flag(osdc
->osdmap
,
1332 CEPH_OSDMAP_SORTBITWISE
);
1333 enum calc_target_result ct_res
;
1336 pi
= ceph_pg_pool_by_id(osdc
->osdmap
, t
->base_oloc
.pool
);
1338 t
->osd
= CEPH_HOMELESS_OSD
;
1339 ct_res
= CALC_TARGET_POOL_DNE
;
1343 if (osdc
->osdmap
->epoch
== pi
->last_force_request_resend
) {
1344 if (last_force_resend
&&
1345 *last_force_resend
< pi
->last_force_request_resend
) {
1346 *last_force_resend
= pi
->last_force_request_resend
;
1347 force_resend
= true;
1348 } else if (!last_force_resend
) {
1349 force_resend
= true;
1352 if (ceph_oid_empty(&t
->target_oid
) || force_resend
) {
1353 ceph_oid_copy(&t
->target_oid
, &t
->base_oid
);
1354 need_check_tiering
= true;
1356 if (ceph_oloc_empty(&t
->target_oloc
) || force_resend
) {
1357 ceph_oloc_copy(&t
->target_oloc
, &t
->base_oloc
);
1358 need_check_tiering
= true;
1361 if (need_check_tiering
&&
1362 (t
->flags
& CEPH_OSD_FLAG_IGNORE_OVERLAY
) == 0) {
1363 if (t
->flags
& CEPH_OSD_FLAG_READ
&& pi
->read_tier
>= 0)
1364 t
->target_oloc
.pool
= pi
->read_tier
;
1365 if (t
->flags
& CEPH_OSD_FLAG_WRITE
&& pi
->write_tier
>= 0)
1366 t
->target_oloc
.pool
= pi
->write_tier
;
1369 ret
= ceph_object_locator_to_pg(osdc
->osdmap
, &t
->target_oid
,
1370 &t
->target_oloc
, &pgid
);
1372 WARN_ON(ret
!= -ENOENT
);
1373 t
->osd
= CEPH_HOMELESS_OSD
;
1374 ct_res
= CALC_TARGET_POOL_DNE
;
1377 last_pgid
.pool
= pgid
.pool
;
1378 last_pgid
.seed
= ceph_stable_mod(pgid
.seed
, t
->pg_num
, t
->pg_num_mask
);
1380 ceph_pg_to_up_acting_osds(osdc
->osdmap
, &pgid
, &up
, &acting
);
1382 ceph_is_new_interval(&t
->acting
,
1395 force_resend
= true;
1397 if (t
->paused
&& !target_should_be_paused(osdc
, t
, pi
)) {
1402 if (ceph_pg_compare(&t
->pgid
, &pgid
) ||
1403 ceph_osds_changed(&t
->acting
, &acting
, any_change
) ||
1405 t
->pgid
= pgid
; /* struct */
1406 ceph_osds_copy(&t
->acting
, &acting
);
1407 ceph_osds_copy(&t
->up
, &up
);
1409 t
->min_size
= pi
->min_size
;
1410 t
->pg_num
= pi
->pg_num
;
1411 t
->pg_num_mask
= pi
->pg_num_mask
;
1412 t
->sort_bitwise
= sort_bitwise
;
1414 t
->osd
= acting
.primary
;
1418 ct_res
= need_resend
? CALC_TARGET_NEED_RESEND
: CALC_TARGET_NO_ACTION
;
1420 dout("%s t %p -> ct_res %d osd %d\n", __func__
, t
, ct_res
, t
->osd
);
1424 static void __enqueue_request(struct ceph_osd_request
*req
)
1426 struct ceph_osd_client
*osdc
= req
->r_osdc
;
1428 dout("%s %p tid %llu to osd%d\n", __func__
, req
, req
->r_tid
,
1429 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1432 __remove_osd_from_lru(req
->r_osd
);
1433 list_add_tail(&req
->r_osd_item
, &req
->r_osd
->o_requests
);
1434 list_move_tail(&req
->r_req_lru_item
, &osdc
->req_unsent
);
1436 list_move_tail(&req
->r_req_lru_item
, &osdc
->req_notarget
);
1441 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1442 * (as needed), and set the request r_osd appropriately. If there is
1443 * no up osd, set r_osd to NULL. Move the request to the appropriate list
1444 * (unsent, homeless) or leave on in-flight lru.
1446 * Return 0 if unchanged, 1 if changed, or negative on error.
1448 * Caller should hold map_sem for read and request_mutex.
1450 static int __map_request(struct ceph_osd_client
*osdc
,
1451 struct ceph_osd_request
*req
, int force_resend
)
1453 enum calc_target_result ct_res
;
1456 dout("map_request %p tid %lld\n", req
, req
->r_tid
);
1458 ct_res
= calc_target(osdc
, &req
->r_t
, NULL
, force_resend
);
1460 case CALC_TARGET_POOL_DNE
:
1461 list_move(&req
->r_req_lru_item
, &osdc
->req_notarget
);
1463 case CALC_TARGET_NO_ACTION
:
1464 return 0; /* no change */
1466 BUG_ON(ct_res
!= CALC_TARGET_NEED_RESEND
);
1469 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1470 req
->r_tid
, req
->r_t
.pgid
.pool
, req
->r_t
.pgid
.seed
, req
->r_t
.osd
,
1471 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1474 __cancel_request(req
);
1475 list_del_init(&req
->r_osd_item
);
1476 list_del_init(&req
->r_linger_osd_item
);
1480 req
->r_osd
= lookup_osd(&osdc
->osds
, req
->r_t
.osd
);
1481 if (!req
->r_osd
&& req
->r_t
.osd
>= 0) {
1483 req
->r_osd
= create_osd(osdc
, req
->r_t
.osd
);
1485 list_move(&req
->r_req_lru_item
, &osdc
->req_notarget
);
1489 dout("map_request osd %p is osd%d\n", req
->r_osd
,
1491 insert_osd(&osdc
->osds
, req
->r_osd
);
1493 ceph_con_open(&req
->r_osd
->o_con
,
1494 CEPH_ENTITY_TYPE_OSD
, req
->r_osd
->o_osd
,
1495 &osdc
->osdmap
->osd_addr
[req
->r_osd
->o_osd
]);
1498 __enqueue_request(req
);
1499 err
= 1; /* osd or pg changed */
1505 static void setup_request_data(struct ceph_osd_request
*req
,
1506 struct ceph_msg
*msg
)
1511 if (!list_empty(&msg
->data
))
1514 WARN_ON(msg
->data_length
);
1515 for (i
= 0; i
< req
->r_num_ops
; i
++) {
1516 struct ceph_osd_req_op
*op
= &req
->r_ops
[i
];
1520 case CEPH_OSD_OP_WRITE
:
1521 case CEPH_OSD_OP_WRITEFULL
:
1522 WARN_ON(op
->indata_len
!= op
->extent
.length
);
1523 ceph_osdc_msg_data_add(msg
, &op
->extent
.osd_data
);
1525 case CEPH_OSD_OP_SETXATTR
:
1526 case CEPH_OSD_OP_CMPXATTR
:
1527 WARN_ON(op
->indata_len
!= op
->xattr
.name_len
+
1528 op
->xattr
.value_len
);
1529 ceph_osdc_msg_data_add(msg
, &op
->xattr
.osd_data
);
1533 case CEPH_OSD_OP_STAT
:
1534 ceph_osdc_msg_data_add(req
->r_reply
,
1537 case CEPH_OSD_OP_READ
:
1538 ceph_osdc_msg_data_add(req
->r_reply
,
1539 &op
->extent
.osd_data
);
1543 case CEPH_OSD_OP_CALL
:
1544 WARN_ON(op
->indata_len
!= op
->cls
.class_len
+
1545 op
->cls
.method_len
+
1546 op
->cls
.indata_len
);
1547 ceph_osdc_msg_data_add(msg
, &op
->cls
.request_info
);
1548 /* optional, can be NONE */
1549 ceph_osdc_msg_data_add(msg
, &op
->cls
.request_data
);
1550 /* optional, can be NONE */
1551 ceph_osdc_msg_data_add(req
->r_reply
,
1552 &op
->cls
.response_data
);
1556 data_len
+= op
->indata_len
;
1559 WARN_ON(data_len
!= msg
->data_length
);
1562 static void encode_request(struct ceph_osd_request
*req
, struct ceph_msg
*msg
)
1564 void *p
= msg
->front
.iov_base
;
1565 void *const end
= p
+ msg
->front_alloc_len
;
1569 if (req
->r_flags
& CEPH_OSD_FLAG_WRITE
) {
1570 /* snapshots aren't writeable */
1571 WARN_ON(req
->r_snapid
!= CEPH_NOSNAP
);
1573 WARN_ON(req
->r_mtime
.tv_sec
|| req
->r_mtime
.tv_nsec
||
1574 req
->r_data_offset
|| req
->r_snapc
);
1577 setup_request_data(req
, msg
);
1579 ceph_encode_32(&p
, 1); /* client_inc, always 1 */
1580 ceph_encode_32(&p
, req
->r_osdc
->osdmap
->epoch
);
1581 ceph_encode_32(&p
, req
->r_flags
);
1582 ceph_encode_timespec(p
, &req
->r_mtime
);
1583 p
+= sizeof(struct ceph_timespec
);
1584 /* aka reassert_version */
1585 memcpy(p
, &req
->r_replay_version
, sizeof(req
->r_replay_version
));
1586 p
+= sizeof(req
->r_replay_version
);
1589 ceph_encode_8(&p
, 4);
1590 ceph_encode_8(&p
, 4);
1591 ceph_encode_32(&p
, 8 + 4 + 4);
1592 ceph_encode_64(&p
, req
->r_t
.target_oloc
.pool
);
1593 ceph_encode_32(&p
, -1); /* preferred */
1594 ceph_encode_32(&p
, 0); /* key len */
1597 ceph_encode_8(&p
, 1);
1598 ceph_encode_64(&p
, req
->r_t
.pgid
.pool
);
1599 ceph_encode_32(&p
, req
->r_t
.pgid
.seed
);
1600 ceph_encode_32(&p
, -1); /* preferred */
1603 ceph_encode_32(&p
, req
->r_t
.target_oid
.name_len
);
1604 memcpy(p
, req
->r_t
.target_oid
.name
, req
->r_t
.target_oid
.name_len
);
1605 p
+= req
->r_t
.target_oid
.name_len
;
1607 /* ops, can imply data */
1608 ceph_encode_16(&p
, req
->r_num_ops
);
1609 for (i
= 0; i
< req
->r_num_ops
; i
++) {
1610 data_len
+= osd_req_encode_op(p
, &req
->r_ops
[i
]);
1611 p
+= sizeof(struct ceph_osd_op
);
1614 ceph_encode_64(&p
, req
->r_snapid
); /* snapid */
1616 ceph_encode_64(&p
, req
->r_snapc
->seq
);
1617 ceph_encode_32(&p
, req
->r_snapc
->num_snaps
);
1618 for (i
= 0; i
< req
->r_snapc
->num_snaps
; i
++)
1619 ceph_encode_64(&p
, req
->r_snapc
->snaps
[i
]);
1621 ceph_encode_64(&p
, 0); /* snap_seq */
1622 ceph_encode_32(&p
, 0); /* snaps len */
1625 ceph_encode_32(&p
, req
->r_attempts
); /* retry_attempt */
1628 msg
->front
.iov_len
= p
- msg
->front
.iov_base
;
1629 msg
->hdr
.version
= cpu_to_le16(4); /* MOSDOp v4 */
1630 msg
->hdr
.front_len
= cpu_to_le32(msg
->front
.iov_len
);
1631 msg
->hdr
.data_len
= cpu_to_le32(data_len
);
1633 * The header "data_off" is a hint to the receiver allowing it
1634 * to align received data into its buffers such that there's no
1635 * need to re-copy it before writing it to disk (direct I/O).
1637 msg
->hdr
.data_off
= cpu_to_le16(req
->r_data_offset
);
1639 dout("%s req %p oid %*pE oid_len %d front %zu data %u\n", __func__
,
1640 req
, req
->r_t
.target_oid
.name_len
, req
->r_t
.target_oid
.name
,
1641 req
->r_t
.target_oid
.name_len
, msg
->front
.iov_len
, data_len
);
1645 * @req has to be assigned a tid and registered.
1647 static void send_request(struct ceph_osd_request
*req
)
1649 struct ceph_osd
*osd
= req
->r_osd
;
1651 WARN_ON(osd
->o_osd
!= req
->r_t
.osd
);
1653 req
->r_flags
|= CEPH_OSD_FLAG_KNOWN_REDIR
;
1654 if (req
->r_attempts
)
1655 req
->r_flags
|= CEPH_OSD_FLAG_RETRY
;
1657 WARN_ON(req
->r_flags
& CEPH_OSD_FLAG_RETRY
);
1659 encode_request(req
, req
->r_request
);
1661 dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n",
1662 __func__
, req
, req
->r_tid
, req
->r_t
.pgid
.pool
, req
->r_t
.pgid
.seed
,
1663 req
->r_t
.osd
, req
->r_flags
, req
->r_attempts
);
1665 req
->r_t
.paused
= false;
1666 req
->r_stamp
= jiffies
;
1669 req
->r_sent
= osd
->o_incarnation
;
1670 req
->r_request
->hdr
.tid
= cpu_to_le64(req
->r_tid
);
1671 ceph_con_send(&osd
->o_con
, ceph_msg_get(req
->r_request
));
1675 * Send any requests in the queue (req_unsent).
1677 static void __send_queued(struct ceph_osd_client
*osdc
)
1679 struct ceph_osd_request
*req
, *tmp
;
1681 dout("__send_queued\n");
1682 list_for_each_entry_safe(req
, tmp
, &osdc
->req_unsent
, r_req_lru_item
) {
1683 list_move_tail(&req
->r_req_lru_item
, &osdc
->req_lru
);
1688 static void maybe_request_map(struct ceph_osd_client
*osdc
)
1690 bool continuous
= false;
1692 WARN_ON(!osdc
->osdmap
->epoch
);
1694 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) ||
1695 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_PAUSERD
) ||
1696 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_PAUSEWR
)) {
1697 dout("%s osdc %p continuous\n", __func__
, osdc
);
1700 dout("%s osdc %p onetime\n", __func__
, osdc
);
1703 if (ceph_monc_want_map(&osdc
->client
->monc
, CEPH_SUB_OSDMAP
,
1704 osdc
->osdmap
->epoch
+ 1, continuous
))
1705 ceph_monc_renew_subs(&osdc
->client
->monc
);
1709 * Caller should hold map_sem for read and request_mutex.
1711 static int __ceph_osdc_start_request(struct ceph_osd_client
*osdc
,
1712 struct ceph_osd_request
*req
,
1717 __register_request(osdc
, req
);
1719 req
->r_got_reply
= 0;
1720 rc
= __map_request(osdc
, req
, 0);
1723 dout("osdc_start_request failed map, "
1724 " will retry %lld\n", req
->r_tid
);
1727 __unregister_request(osdc
, req
);
1732 if (req
->r_osd
== NULL
) {
1733 dout("send_request %p no up osds in pg\n", req
);
1734 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1736 __send_queued(osdc
);
1742 static void __complete_request(struct ceph_osd_request
*req
)
1744 if (req
->r_callback
)
1745 req
->r_callback(req
);
1747 complete_all(&req
->r_completion
);
1751 * Timeout callback, called every N seconds. When 1 or more OSD
1752 * requests has been active for more than N seconds, we send a keepalive
1753 * (tag + timestamp) to its OSD to ensure any communications channel
1754 * reset is detected.
1756 static void handle_timeout(struct work_struct
*work
)
1758 struct ceph_osd_client
*osdc
=
1759 container_of(work
, struct ceph_osd_client
, timeout_work
.work
);
1760 struct ceph_options
*opts
= osdc
->client
->options
;
1761 struct ceph_osd_request
*req
;
1762 struct ceph_osd
*osd
;
1763 struct list_head slow_osds
;
1765 down_read(&osdc
->map_sem
);
1767 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1769 mutex_lock(&osdc
->request_mutex
);
1772 * ping osds that are a bit slow. this ensures that if there
1773 * is a break in the TCP connection we will notice, and reopen
1774 * a connection with that osd (from the fault callback).
1776 INIT_LIST_HEAD(&slow_osds
);
1777 list_for_each_entry(req
, &osdc
->req_lru
, r_req_lru_item
) {
1778 if (time_before(jiffies
,
1779 req
->r_stamp
+ opts
->osd_keepalive_timeout
))
1784 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1785 req
->r_tid
, osd
->o_osd
);
1786 list_move_tail(&osd
->o_keepalive_item
, &slow_osds
);
1788 while (!list_empty(&slow_osds
)) {
1789 osd
= list_entry(slow_osds
.next
, struct ceph_osd
,
1791 list_del_init(&osd
->o_keepalive_item
);
1792 ceph_con_keepalive(&osd
->o_con
);
1795 __send_queued(osdc
);
1796 mutex_unlock(&osdc
->request_mutex
);
1797 up_read(&osdc
->map_sem
);
1799 schedule_delayed_work(&osdc
->timeout_work
,
1800 osdc
->client
->options
->osd_keepalive_timeout
);
1803 static void handle_osds_timeout(struct work_struct
*work
)
1805 struct ceph_osd_client
*osdc
=
1806 container_of(work
, struct ceph_osd_client
,
1807 osds_timeout_work
.work
);
1808 unsigned long delay
= osdc
->client
->options
->osd_idle_ttl
/ 4;
1809 struct ceph_osd
*osd
, *nosd
;
1811 dout("%s osdc %p\n", __func__
, osdc
);
1812 down_read(&osdc
->map_sem
);
1813 mutex_lock(&osdc
->request_mutex
);
1815 list_for_each_entry_safe(osd
, nosd
, &osdc
->osd_lru
, o_osd_lru
) {
1816 if (time_before(jiffies
, osd
->lru_ttl
))
1819 remove_osd(osdc
, osd
);
1822 mutex_unlock(&osdc
->request_mutex
);
1823 up_read(&osdc
->map_sem
);
1824 schedule_delayed_work(&osdc
->osds_timeout_work
,
1825 round_jiffies_relative(delay
));
1828 static int ceph_oloc_decode(void **p
, void *end
,
1829 struct ceph_object_locator
*oloc
)
1831 u8 struct_v
, struct_cv
;
1836 ceph_decode_need(p
, end
, 1 + 1 + 4, e_inval
);
1837 struct_v
= ceph_decode_8(p
);
1838 struct_cv
= ceph_decode_8(p
);
1840 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1841 struct_v
, struct_cv
);
1844 if (struct_cv
> 6) {
1845 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1846 struct_v
, struct_cv
);
1849 len
= ceph_decode_32(p
);
1850 ceph_decode_need(p
, end
, len
, e_inval
);
1851 struct_end
= *p
+ len
;
1853 oloc
->pool
= ceph_decode_64(p
);
1854 *p
+= 4; /* skip preferred */
1856 len
= ceph_decode_32(p
);
1858 pr_warn("ceph_object_locator::key is set\n");
1862 if (struct_v
>= 5) {
1863 len
= ceph_decode_32(p
);
1865 pr_warn("ceph_object_locator::nspace is set\n");
1870 if (struct_v
>= 6) {
1871 s64 hash
= ceph_decode_64(p
);
1873 pr_warn("ceph_object_locator::hash is set\n");
1888 static int ceph_redirect_decode(void **p
, void *end
,
1889 struct ceph_request_redirect
*redir
)
1891 u8 struct_v
, struct_cv
;
1896 ceph_decode_need(p
, end
, 1 + 1 + 4, e_inval
);
1897 struct_v
= ceph_decode_8(p
);
1898 struct_cv
= ceph_decode_8(p
);
1899 if (struct_cv
> 1) {
1900 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1901 struct_v
, struct_cv
);
1904 len
= ceph_decode_32(p
);
1905 ceph_decode_need(p
, end
, len
, e_inval
);
1906 struct_end
= *p
+ len
;
1908 ret
= ceph_oloc_decode(p
, end
, &redir
->oloc
);
1912 len
= ceph_decode_32(p
);
1914 pr_warn("ceph_request_redirect::object_name is set\n");
1918 len
= ceph_decode_32(p
);
1919 *p
+= len
; /* skip osd_instructions */
1931 struct MOSDOpReply
{
1932 struct ceph_pg pgid
;
1937 u32 outdata_len
[CEPH_OSD_MAX_OPS
];
1938 s32 rval
[CEPH_OSD_MAX_OPS
];
1940 struct ceph_eversion replay_version
;
1942 struct ceph_request_redirect redirect
;
1945 static int decode_MOSDOpReply(const struct ceph_msg
*msg
, struct MOSDOpReply
*m
)
1947 void *p
= msg
->front
.iov_base
;
1948 void *const end
= p
+ msg
->front
.iov_len
;
1949 u16 version
= le16_to_cpu(msg
->hdr
.version
);
1950 struct ceph_eversion bad_replay_version
;
1956 ceph_decode_32_safe(&p
, end
, len
, e_inval
);
1957 ceph_decode_need(&p
, end
, len
, e_inval
);
1958 p
+= len
; /* skip oid */
1960 ret
= ceph_decode_pgid(&p
, end
, &m
->pgid
);
1964 ceph_decode_64_safe(&p
, end
, m
->flags
, e_inval
);
1965 ceph_decode_32_safe(&p
, end
, m
->result
, e_inval
);
1966 ceph_decode_need(&p
, end
, sizeof(bad_replay_version
), e_inval
);
1967 memcpy(&bad_replay_version
, p
, sizeof(bad_replay_version
));
1968 p
+= sizeof(bad_replay_version
);
1969 ceph_decode_32_safe(&p
, end
, m
->epoch
, e_inval
);
1971 ceph_decode_32_safe(&p
, end
, m
->num_ops
, e_inval
);
1972 if (m
->num_ops
> ARRAY_SIZE(m
->outdata_len
))
1975 ceph_decode_need(&p
, end
, m
->num_ops
* sizeof(struct ceph_osd_op
),
1977 for (i
= 0; i
< m
->num_ops
; i
++) {
1978 struct ceph_osd_op
*op
= p
;
1980 m
->outdata_len
[i
] = le32_to_cpu(op
->payload_len
);
1984 ceph_decode_32_safe(&p
, end
, m
->retry_attempt
, e_inval
);
1985 for (i
= 0; i
< m
->num_ops
; i
++)
1986 ceph_decode_32_safe(&p
, end
, m
->rval
[i
], e_inval
);
1989 ceph_decode_need(&p
, end
, sizeof(m
->replay_version
), e_inval
);
1990 memcpy(&m
->replay_version
, p
, sizeof(m
->replay_version
));
1991 p
+= sizeof(m
->replay_version
);
1992 ceph_decode_64_safe(&p
, end
, m
->user_version
, e_inval
);
1994 m
->replay_version
= bad_replay_version
; /* struct */
1995 m
->user_version
= le64_to_cpu(m
->replay_version
.version
);
2000 ceph_decode_8_safe(&p
, end
, decode_redir
, e_inval
);
2008 ret
= ceph_redirect_decode(&p
, end
, &m
->redirect
);
2012 ceph_oloc_init(&m
->redirect
.oloc
);
2022 * We are done with @req if
2023 * - @m is a safe reply, or
2024 * - @m is an unsafe reply and we didn't want a safe one
2026 static bool done_request(const struct ceph_osd_request
*req
,
2027 const struct MOSDOpReply
*m
)
2029 return (m
->result
< 0 ||
2030 (m
->flags
& CEPH_OSD_FLAG_ONDISK
) ||
2031 !(req
->r_flags
& CEPH_OSD_FLAG_ONDISK
));
2035 * handle osd op reply. either call the callback if it is specified,
2036 * or do the completion to wake up the waiting thread.
2038 * ->r_unsafe_callback is set? yes no
2040 * first reply is OK (needed r_cb/r_completion, r_cb/r_completion,
2041 * any or needed/got safe) r_safe_completion r_safe_completion
2043 * first reply is unsafe r_unsafe_cb(true) (nothing)
2045 * when we get the safe reply r_unsafe_cb(false), r_cb/r_completion,
2046 * r_safe_completion r_safe_completion
2048 static void handle_reply(struct ceph_osd_client
*osdc
, struct ceph_msg
*msg
)
2050 struct ceph_osd_request
*req
;
2051 struct MOSDOpReply m
;
2052 u64 tid
= le64_to_cpu(msg
->hdr
.tid
);
2058 dout("%s msg %p tid %llu\n", __func__
, msg
, tid
);
2060 down_read(&osdc
->map_sem
);
2061 mutex_lock(&osdc
->request_mutex
);
2062 req
= lookup_request(&osdc
->requests
, tid
);
2064 dout("%s no tid %llu\n", __func__
, tid
);
2067 ceph_osdc_get_request(req
);
2069 ret
= decode_MOSDOpReply(msg
, &m
);
2071 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2076 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2077 __func__
, req
, req
->r_tid
, m
.flags
, m
.pgid
.pool
, m
.pgid
.seed
,
2078 m
.epoch
, m
.retry_attempt
, le32_to_cpu(m
.replay_version
.epoch
),
2079 le64_to_cpu(m
.replay_version
.version
), m
.user_version
);
2081 if (m
.retry_attempt
>= 0) {
2082 if (m
.retry_attempt
!= req
->r_attempts
- 1) {
2083 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2084 req
, req
->r_tid
, m
.retry_attempt
,
2085 req
->r_attempts
- 1);
2089 WARN_ON(1); /* MOSDOpReply v4 is assumed */
2092 if (!ceph_oloc_empty(&m
.redirect
.oloc
)) {
2093 dout("req %p tid %llu redirect pool %lld\n", req
, req
->r_tid
,
2094 m
.redirect
.oloc
.pool
);
2095 __unregister_request(osdc
, req
);
2097 ceph_oloc_copy(&req
->r_t
.target_oloc
, &m
.redirect
.oloc
);
2100 * Start redirect requests with nofail=true. If
2101 * mapping fails, request will end up on the notarget
2102 * list, waiting for the new osdmap (which can take
2103 * a while), even though the original request mapped
2104 * successfully. In the future we might want to follow
2105 * original request's nofail setting here.
2107 ret
= __ceph_osdc_start_request(osdc
, req
, true);
2113 if (m
.num_ops
!= req
->r_num_ops
) {
2114 pr_err("num_ops %d != %d for tid %llu\n", m
.num_ops
,
2115 req
->r_num_ops
, req
->r_tid
);
2118 for (i
= 0; i
< req
->r_num_ops
; i
++) {
2119 dout(" req %p tid %llu op %d rval %d len %u\n", req
,
2120 req
->r_tid
, i
, m
.rval
[i
], m
.outdata_len
[i
]);
2121 req
->r_ops
[i
].rval
= m
.rval
[i
];
2122 req
->r_ops
[i
].outdata_len
= m
.outdata_len
[i
];
2123 data_len
+= m
.outdata_len
[i
];
2125 if (data_len
!= le32_to_cpu(msg
->hdr
.data_len
)) {
2126 pr_err("sum of lens %u != %u for tid %llu\n", data_len
,
2127 le32_to_cpu(msg
->hdr
.data_len
), req
->r_tid
);
2130 dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__
,
2131 req
, req
->r_tid
, req
->r_got_reply
, m
.result
, data_len
);
2133 already_acked
= req
->r_got_reply
;
2134 if (!already_acked
) {
2135 req
->r_result
= m
.result
?: data_len
;
2136 req
->r_replay_version
= m
.replay_version
; /* struct */
2137 req
->r_got_reply
= true;
2138 } else if (!(m
.flags
& CEPH_OSD_FLAG_ONDISK
)) {
2139 dout("req %p tid %llu dup ack\n", req
, req
->r_tid
);
2143 if (done_request(req
, &m
)) {
2144 __unregister_request(osdc
, req
);
2145 if (req
->r_linger
) {
2146 WARN_ON(req
->r_unsafe_callback
);
2147 __register_linger_request(osdc
, req
);
2151 mutex_unlock(&osdc
->request_mutex
);
2152 up_read(&osdc
->map_sem
);
2154 if (done_request(req
, &m
)) {
2155 if (already_acked
&& req
->r_unsafe_callback
) {
2156 dout("req %p tid %llu safe-cb\n", req
, req
->r_tid
);
2157 req
->r_unsafe_callback(req
, false);
2159 dout("req %p tid %llu cb\n", req
, req
->r_tid
);
2160 __complete_request(req
);
2163 if (req
->r_unsafe_callback
) {
2164 dout("req %p tid %llu unsafe-cb\n", req
, req
->r_tid
);
2165 req
->r_unsafe_callback(req
, true);
2170 if (m
.flags
& CEPH_OSD_FLAG_ONDISK
)
2171 complete_all(&req
->r_safe_completion
);
2173 ceph_osdc_put_request(req
);
2177 req
->r_result
= -EIO
;
2178 __unregister_request(osdc
, req
);
2179 __complete_request(req
);
2180 complete_all(&req
->r_safe_completion
);
2182 ceph_osdc_put_request(req
);
2184 mutex_unlock(&osdc
->request_mutex
);
2185 up_read(&osdc
->map_sem
);
2188 static void set_pool_was_full(struct ceph_osd_client
*osdc
)
2192 for (n
= rb_first(&osdc
->osdmap
->pg_pools
); n
; n
= rb_next(n
)) {
2193 struct ceph_pg_pool_info
*pi
=
2194 rb_entry(n
, struct ceph_pg_pool_info
, node
);
2196 pi
->was_full
= __pool_full(pi
);
2200 static void reset_changed_osds(struct ceph_osd_client
*osdc
)
2202 struct rb_node
*p
, *n
;
2204 dout("%s %p\n", __func__
, osdc
);
2205 for (p
= rb_first(&osdc
->osds
); p
; p
= n
) {
2206 struct ceph_osd
*osd
= rb_entry(p
, struct ceph_osd
, o_node
);
2209 if (!ceph_osd_is_up(osdc
->osdmap
, osd
->o_osd
) ||
2210 memcmp(&osd
->o_con
.peer_addr
,
2211 ceph_osd_addr(osdc
->osdmap
,
2213 sizeof(struct ceph_entity_addr
)) != 0)
2214 __reset_osd(osdc
, osd
);
2219 * Requeue requests whose mapping to an OSD has changed. If requests map to
2220 * no osd, request a new map.
2222 * Caller should hold map_sem for read.
2224 static void kick_requests(struct ceph_osd_client
*osdc
, bool force_resend
,
2225 bool force_resend_writes
)
2227 struct ceph_osd_request
*req
, *nreq
;
2231 bool force_resend_req
;
2233 dout("kick_requests %s %s\n", force_resend
? " (force resend)" : "",
2234 force_resend_writes
? " (force resend writes)" : "");
2235 mutex_lock(&osdc
->request_mutex
);
2236 for (p
= rb_first(&osdc
->requests
); p
; ) {
2237 req
= rb_entry(p
, struct ceph_osd_request
, r_node
);
2241 * For linger requests that have not yet been
2242 * registered, move them to the linger list; they'll
2243 * be sent to the osd in the loop below. Unregister
2244 * the request before re-registering it as a linger
2245 * request to ensure the __map_request() below
2246 * will decide it needs to be sent.
2248 if (req
->r_linger
&& list_empty(&req
->r_linger_item
)) {
2249 dout("%p tid %llu restart on osd%d\n",
2251 req
->r_osd
? req
->r_osd
->o_osd
: -1);
2252 ceph_osdc_get_request(req
);
2253 __unregister_request(osdc
, req
);
2254 __register_linger_request(osdc
, req
);
2255 ceph_osdc_put_request(req
);
2259 force_resend_req
= force_resend
||
2260 (force_resend_writes
&&
2261 req
->r_flags
& CEPH_OSD_FLAG_WRITE
);
2262 err
= __map_request(osdc
, req
, force_resend_req
);
2264 continue; /* error */
2265 if (req
->r_osd
== NULL
) {
2266 dout("%p tid %llu maps to no osd\n", req
, req
->r_tid
);
2267 needmap
++; /* request a newer map */
2268 } else if (err
> 0) {
2269 if (!req
->r_linger
) {
2270 dout("%p tid %llu requeued on osd%d\n", req
,
2272 req
->r_osd
? req
->r_osd
->o_osd
: -1);
2273 req
->r_flags
|= CEPH_OSD_FLAG_RETRY
;
2278 list_for_each_entry_safe(req
, nreq
, &osdc
->req_linger
,
2280 dout("linger req=%p req->r_osd=%p\n", req
, req
->r_osd
);
2282 err
= __map_request(osdc
, req
,
2283 force_resend
|| force_resend_writes
);
2284 dout("__map_request returned %d\n", err
);
2286 continue; /* hrm! */
2287 if (req
->r_osd
== NULL
|| err
> 0) {
2288 if (req
->r_osd
== NULL
) {
2289 dout("lingering %p tid %llu maps to no osd\n",
2292 * A homeless lingering request makes
2293 * no sense, as it's job is to keep
2294 * a particular OSD connection open.
2295 * Request a newer map and kick the
2296 * request, knowing that it won't be
2297 * resent until we actually get a map
2298 * that can tell us where to send it.
2303 dout("kicking lingering %p tid %llu osd%d\n", req
,
2304 req
->r_tid
, req
->r_osd
? req
->r_osd
->o_osd
: -1);
2305 __register_request(osdc
, req
);
2306 __unregister_linger_request(osdc
, req
);
2309 reset_changed_osds(osdc
);
2310 mutex_unlock(&osdc
->request_mutex
);
2313 dout("%d requests for down osds, need new map\n", needmap
);
2314 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
2318 static int handle_one_map(struct ceph_osd_client
*osdc
,
2319 void *p
, void *end
, bool incremental
)
2321 struct ceph_osdmap
*newmap
;
2323 bool skipped_map
= false;
2326 was_full
= ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
);
2327 set_pool_was_full(osdc
);
2330 newmap
= osdmap_apply_incremental(&p
, end
, osdc
->osdmap
);
2332 newmap
= ceph_osdmap_decode(&p
, end
);
2334 return PTR_ERR(newmap
);
2336 if (newmap
!= osdc
->osdmap
) {
2338 * Preserve ->was_full before destroying the old map.
2339 * For pools that weren't in the old map, ->was_full
2342 for (n
= rb_first(&newmap
->pg_pools
); n
; n
= rb_next(n
)) {
2343 struct ceph_pg_pool_info
*pi
=
2344 rb_entry(n
, struct ceph_pg_pool_info
, node
);
2345 struct ceph_pg_pool_info
*old_pi
;
2347 old_pi
= ceph_pg_pool_by_id(osdc
->osdmap
, pi
->id
);
2349 pi
->was_full
= old_pi
->was_full
;
2351 WARN_ON(pi
->was_full
);
2354 if (osdc
->osdmap
->epoch
&&
2355 osdc
->osdmap
->epoch
+ 1 < newmap
->epoch
) {
2356 WARN_ON(incremental
);
2360 ceph_osdmap_destroy(osdc
->osdmap
);
2361 osdc
->osdmap
= newmap
;
2364 was_full
&= !ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
);
2365 kick_requests(osdc
, skipped_map
, was_full
);
2371 * Process updated osd map.
2373 * The message contains any number of incremental and full maps, normally
2374 * indicating some sort of topology change in the cluster. Kick requests
2375 * off to different OSDs as needed.
2377 void ceph_osdc_handle_map(struct ceph_osd_client
*osdc
, struct ceph_msg
*msg
)
2379 void *p
= msg
->front
.iov_base
;
2380 void *const end
= p
+ msg
->front
.iov_len
;
2381 u32 nr_maps
, maplen
;
2383 struct ceph_fsid fsid
;
2384 bool handled_incremental
= false;
2385 bool was_pauserd
, was_pausewr
;
2386 bool pauserd
, pausewr
;
2389 dout("%s have %u\n", __func__
, osdc
->osdmap
->epoch
);
2390 down_write(&osdc
->map_sem
);
2393 ceph_decode_need(&p
, end
, sizeof(fsid
), bad
);
2394 ceph_decode_copy(&p
, &fsid
, sizeof(fsid
));
2395 if (ceph_check_fsid(osdc
->client
, &fsid
) < 0)
2398 was_pauserd
= ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_PAUSERD
);
2399 was_pausewr
= ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_PAUSEWR
) ||
2400 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) ||
2401 have_pool_full(osdc
);
2403 /* incremental maps */
2404 ceph_decode_32_safe(&p
, end
, nr_maps
, bad
);
2405 dout(" %d inc maps\n", nr_maps
);
2406 while (nr_maps
> 0) {
2407 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
2408 epoch
= ceph_decode_32(&p
);
2409 maplen
= ceph_decode_32(&p
);
2410 ceph_decode_need(&p
, end
, maplen
, bad
);
2411 if (osdc
->osdmap
->epoch
&&
2412 osdc
->osdmap
->epoch
+ 1 == epoch
) {
2413 dout("applying incremental map %u len %d\n",
2415 err
= handle_one_map(osdc
, p
, p
+ maplen
, true);
2418 handled_incremental
= true;
2420 dout("ignoring incremental map %u len %d\n",
2426 if (handled_incremental
)
2430 ceph_decode_32_safe(&p
, end
, nr_maps
, bad
);
2431 dout(" %d full maps\n", nr_maps
);
2433 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
2434 epoch
= ceph_decode_32(&p
);
2435 maplen
= ceph_decode_32(&p
);
2436 ceph_decode_need(&p
, end
, maplen
, bad
);
2438 dout("skipping non-latest full map %u len %d\n",
2440 } else if (osdc
->osdmap
->epoch
>= epoch
) {
2441 dout("skipping full map %u len %d, "
2442 "older than our %u\n", epoch
, maplen
,
2443 osdc
->osdmap
->epoch
);
2445 dout("taking full map %u len %d\n", epoch
, maplen
);
2446 err
= handle_one_map(osdc
, p
, p
+ maplen
, false);
2456 * subscribe to subsequent osdmap updates if full to ensure
2457 * we find out when we are no longer full and stop returning
2460 pauserd
= ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_PAUSERD
);
2461 pausewr
= ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_PAUSEWR
) ||
2462 ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
) ||
2463 have_pool_full(osdc
);
2464 if (was_pauserd
|| was_pausewr
|| pauserd
|| pausewr
)
2465 maybe_request_map(osdc
);
2467 mutex_lock(&osdc
->request_mutex
);
2468 __send_queued(osdc
);
2469 mutex_unlock(&osdc
->request_mutex
);
2471 ceph_monc_got_map(&osdc
->client
->monc
, CEPH_SUB_OSDMAP
,
2472 osdc
->osdmap
->epoch
);
2473 up_write(&osdc
->map_sem
);
2474 wake_up_all(&osdc
->client
->auth_wq
);
2478 pr_err("osdc handle_map corrupt msg\n");
2480 up_write(&osdc
->map_sem
);
2484 * watch/notify callback event infrastructure
2486 * These callbacks are used both for watch and notify operations.
2488 static void __release_event(struct kref
*kref
)
2490 struct ceph_osd_event
*event
=
2491 container_of(kref
, struct ceph_osd_event
, kref
);
2493 dout("__release_event %p\n", event
);
2497 static void get_event(struct ceph_osd_event
*event
)
2499 kref_get(&event
->kref
);
2502 void ceph_osdc_put_event(struct ceph_osd_event
*event
)
2504 kref_put(&event
->kref
, __release_event
);
2506 EXPORT_SYMBOL(ceph_osdc_put_event
);
2508 static void __insert_event(struct ceph_osd_client
*osdc
,
2509 struct ceph_osd_event
*new)
2511 struct rb_node
**p
= &osdc
->event_tree
.rb_node
;
2512 struct rb_node
*parent
= NULL
;
2513 struct ceph_osd_event
*event
= NULL
;
2517 event
= rb_entry(parent
, struct ceph_osd_event
, node
);
2518 if (new->cookie
< event
->cookie
)
2520 else if (new->cookie
> event
->cookie
)
2521 p
= &(*p
)->rb_right
;
2526 rb_link_node(&new->node
, parent
, p
);
2527 rb_insert_color(&new->node
, &osdc
->event_tree
);
2530 static struct ceph_osd_event
*__find_event(struct ceph_osd_client
*osdc
,
2533 struct rb_node
**p
= &osdc
->event_tree
.rb_node
;
2534 struct rb_node
*parent
= NULL
;
2535 struct ceph_osd_event
*event
= NULL
;
2539 event
= rb_entry(parent
, struct ceph_osd_event
, node
);
2540 if (cookie
< event
->cookie
)
2542 else if (cookie
> event
->cookie
)
2543 p
= &(*p
)->rb_right
;
2550 static void __remove_event(struct ceph_osd_event
*event
)
2552 struct ceph_osd_client
*osdc
= event
->osdc
;
2554 if (!RB_EMPTY_NODE(&event
->node
)) {
2555 dout("__remove_event removed %p\n", event
);
2556 rb_erase(&event
->node
, &osdc
->event_tree
);
2557 ceph_osdc_put_event(event
);
2559 dout("__remove_event didn't remove %p\n", event
);
2563 int ceph_osdc_create_event(struct ceph_osd_client
*osdc
,
2564 void (*event_cb
)(u64
, u64
, u8
, void *),
2565 void *data
, struct ceph_osd_event
**pevent
)
2567 struct ceph_osd_event
*event
;
2569 event
= kmalloc(sizeof(*event
), GFP_NOIO
);
2573 dout("create_event %p\n", event
);
2574 event
->cb
= event_cb
;
2575 event
->one_shot
= 0;
2578 INIT_LIST_HEAD(&event
->osd_node
);
2579 RB_CLEAR_NODE(&event
->node
);
2580 kref_init(&event
->kref
); /* one ref for us */
2581 kref_get(&event
->kref
); /* one ref for the caller */
2583 spin_lock(&osdc
->event_lock
);
2584 event
->cookie
= ++osdc
->event_count
;
2585 __insert_event(osdc
, event
);
2586 spin_unlock(&osdc
->event_lock
);
2591 EXPORT_SYMBOL(ceph_osdc_create_event
);
2593 void ceph_osdc_cancel_event(struct ceph_osd_event
*event
)
2595 struct ceph_osd_client
*osdc
= event
->osdc
;
2597 dout("cancel_event %p\n", event
);
2598 spin_lock(&osdc
->event_lock
);
2599 __remove_event(event
);
2600 spin_unlock(&osdc
->event_lock
);
2601 ceph_osdc_put_event(event
); /* caller's */
2603 EXPORT_SYMBOL(ceph_osdc_cancel_event
);
2606 static void do_event_work(struct work_struct
*work
)
2608 struct ceph_osd_event_work
*event_work
=
2609 container_of(work
, struct ceph_osd_event_work
, work
);
2610 struct ceph_osd_event
*event
= event_work
->event
;
2611 u64 ver
= event_work
->ver
;
2612 u64 notify_id
= event_work
->notify_id
;
2613 u8 opcode
= event_work
->opcode
;
2615 dout("do_event_work completing %p\n", event
);
2616 event
->cb(ver
, notify_id
, opcode
, event
->data
);
2617 dout("do_event_work completed %p\n", event
);
2618 ceph_osdc_put_event(event
);
2624 * Process osd watch notifications
2626 static void handle_watch_notify(struct ceph_osd_client
*osdc
,
2627 struct ceph_msg
*msg
)
2631 u64 cookie
, ver
, notify_id
;
2633 struct ceph_osd_event
*event
;
2634 struct ceph_osd_event_work
*event_work
;
2636 p
= msg
->front
.iov_base
;
2637 end
= p
+ msg
->front
.iov_len
;
2639 ceph_decode_8_safe(&p
, end
, proto_ver
, bad
);
2640 ceph_decode_8_safe(&p
, end
, opcode
, bad
);
2641 ceph_decode_64_safe(&p
, end
, cookie
, bad
);
2642 ceph_decode_64_safe(&p
, end
, ver
, bad
);
2643 ceph_decode_64_safe(&p
, end
, notify_id
, bad
);
2645 spin_lock(&osdc
->event_lock
);
2646 event
= __find_event(osdc
, cookie
);
2648 BUG_ON(event
->one_shot
);
2651 spin_unlock(&osdc
->event_lock
);
2652 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2653 cookie
, ver
, event
);
2655 event_work
= kmalloc(sizeof(*event_work
), GFP_NOIO
);
2657 pr_err("couldn't allocate event_work\n");
2658 ceph_osdc_put_event(event
);
2661 INIT_WORK(&event_work
->work
, do_event_work
);
2662 event_work
->event
= event
;
2663 event_work
->ver
= ver
;
2664 event_work
->notify_id
= notify_id
;
2665 event_work
->opcode
= opcode
;
2667 queue_work(osdc
->notify_wq
, &event_work
->work
);
2673 pr_err("osdc handle_watch_notify corrupt msg\n");
2677 * Register request, send initial attempt.
2679 int ceph_osdc_start_request(struct ceph_osd_client
*osdc
,
2680 struct ceph_osd_request
*req
,
2685 down_read(&osdc
->map_sem
);
2686 mutex_lock(&osdc
->request_mutex
);
2688 rc
= __ceph_osdc_start_request(osdc
, req
, nofail
);
2690 mutex_unlock(&osdc
->request_mutex
);
2691 up_read(&osdc
->map_sem
);
2695 EXPORT_SYMBOL(ceph_osdc_start_request
);
2698 * Unregister a registered request. The request is not completed (i.e.
2699 * no callbacks or wakeups) - higher layers are supposed to know what
2700 * they are canceling.
2702 void ceph_osdc_cancel_request(struct ceph_osd_request
*req
)
2704 struct ceph_osd_client
*osdc
= req
->r_osdc
;
2706 mutex_lock(&osdc
->request_mutex
);
2708 __unregister_linger_request(osdc
, req
);
2709 __unregister_request(osdc
, req
);
2710 mutex_unlock(&osdc
->request_mutex
);
2712 dout("%s %p tid %llu canceled\n", __func__
, req
, req
->r_tid
);
2714 EXPORT_SYMBOL(ceph_osdc_cancel_request
);
2717 * wait for a request to complete
2719 int ceph_osdc_wait_request(struct ceph_osd_client
*osdc
,
2720 struct ceph_osd_request
*req
)
2724 dout("%s %p tid %llu\n", __func__
, req
, req
->r_tid
);
2726 rc
= wait_for_completion_interruptible(&req
->r_completion
);
2728 dout("%s %p tid %llu interrupted\n", __func__
, req
, req
->r_tid
);
2729 ceph_osdc_cancel_request(req
);
2731 /* kludge - need to to wake ceph_osdc_sync() */
2732 complete_all(&req
->r_safe_completion
);
2736 dout("%s %p tid %llu result %d\n", __func__
, req
, req
->r_tid
,
2738 return req
->r_result
;
2740 EXPORT_SYMBOL(ceph_osdc_wait_request
);
2743 * sync - wait for all in-flight requests to flush. avoid starvation.
2745 void ceph_osdc_sync(struct ceph_osd_client
*osdc
)
2747 struct ceph_osd_request
*req
;
2748 u64 last_tid
, next_tid
= 0;
2750 mutex_lock(&osdc
->request_mutex
);
2751 last_tid
= osdc
->last_tid
;
2753 req
= __lookup_request_ge(osdc
, next_tid
);
2756 if (req
->r_tid
> last_tid
)
2759 next_tid
= req
->r_tid
+ 1;
2760 if ((req
->r_flags
& CEPH_OSD_FLAG_WRITE
) == 0)
2763 ceph_osdc_get_request(req
);
2764 mutex_unlock(&osdc
->request_mutex
);
2765 dout("sync waiting on tid %llu (last is %llu)\n",
2766 req
->r_tid
, last_tid
);
2767 wait_for_completion(&req
->r_safe_completion
);
2768 mutex_lock(&osdc
->request_mutex
);
2769 ceph_osdc_put_request(req
);
2771 mutex_unlock(&osdc
->request_mutex
);
2772 dout("sync done (thru tid %llu)\n", last_tid
);
2774 EXPORT_SYMBOL(ceph_osdc_sync
);
2777 * Call all pending notify callbacks - for use after a watch is
2778 * unregistered, to make sure no more callbacks for it will be invoked
2780 void ceph_osdc_flush_notifies(struct ceph_osd_client
*osdc
)
2782 flush_workqueue(osdc
->notify_wq
);
2784 EXPORT_SYMBOL(ceph_osdc_flush_notifies
);
2790 int ceph_osdc_init(struct ceph_osd_client
*osdc
, struct ceph_client
*client
)
2795 osdc
->client
= client
;
2796 init_rwsem(&osdc
->map_sem
);
2797 mutex_init(&osdc
->request_mutex
);
2799 osdc
->osds
= RB_ROOT
;
2800 INIT_LIST_HEAD(&osdc
->osd_lru
);
2801 spin_lock_init(&osdc
->osd_lru_lock
);
2802 osdc
->requests
= RB_ROOT
;
2803 INIT_LIST_HEAD(&osdc
->req_lru
);
2804 INIT_LIST_HEAD(&osdc
->req_unsent
);
2805 INIT_LIST_HEAD(&osdc
->req_notarget
);
2806 INIT_LIST_HEAD(&osdc
->req_linger
);
2807 osdc
->num_requests
= 0;
2808 INIT_DELAYED_WORK(&osdc
->timeout_work
, handle_timeout
);
2809 INIT_DELAYED_WORK(&osdc
->osds_timeout_work
, handle_osds_timeout
);
2810 spin_lock_init(&osdc
->event_lock
);
2811 osdc
->event_tree
= RB_ROOT
;
2812 osdc
->event_count
= 0;
2815 osdc
->osdmap
= ceph_osdmap_alloc();
2819 osdc
->req_mempool
= mempool_create_slab_pool(10,
2820 ceph_osd_request_cache
);
2821 if (!osdc
->req_mempool
)
2824 err
= ceph_msgpool_init(&osdc
->msgpool_op
, CEPH_MSG_OSD_OP
,
2825 PAGE_SIZE
, 10, true, "osd_op");
2828 err
= ceph_msgpool_init(&osdc
->msgpool_op_reply
, CEPH_MSG_OSD_OPREPLY
,
2829 PAGE_SIZE
, 10, true, "osd_op_reply");
2834 osdc
->notify_wq
= create_singlethread_workqueue("ceph-watch-notify");
2835 if (!osdc
->notify_wq
)
2836 goto out_msgpool_reply
;
2838 schedule_delayed_work(&osdc
->timeout_work
,
2839 osdc
->client
->options
->osd_keepalive_timeout
);
2840 schedule_delayed_work(&osdc
->osds_timeout_work
,
2841 round_jiffies_relative(osdc
->client
->options
->osd_idle_ttl
));
2846 ceph_msgpool_destroy(&osdc
->msgpool_op_reply
);
2848 ceph_msgpool_destroy(&osdc
->msgpool_op
);
2850 mempool_destroy(osdc
->req_mempool
);
2852 ceph_osdmap_destroy(osdc
->osdmap
);
2857 void ceph_osdc_stop(struct ceph_osd_client
*osdc
)
2859 flush_workqueue(osdc
->notify_wq
);
2860 destroy_workqueue(osdc
->notify_wq
);
2861 cancel_delayed_work_sync(&osdc
->timeout_work
);
2862 cancel_delayed_work_sync(&osdc
->osds_timeout_work
);
2864 mutex_lock(&osdc
->request_mutex
);
2865 while (!RB_EMPTY_ROOT(&osdc
->osds
)) {
2866 struct ceph_osd
*osd
= rb_entry(rb_first(&osdc
->osds
),
2867 struct ceph_osd
, o_node
);
2868 remove_osd(osdc
, osd
);
2870 mutex_unlock(&osdc
->request_mutex
);
2872 ceph_osdmap_destroy(osdc
->osdmap
);
2873 mempool_destroy(osdc
->req_mempool
);
2874 ceph_msgpool_destroy(&osdc
->msgpool_op
);
2875 ceph_msgpool_destroy(&osdc
->msgpool_op_reply
);
2879 * Read some contiguous pages. If we cross a stripe boundary, shorten
2880 * *plen. Return number of bytes read, or error.
2882 int ceph_osdc_readpages(struct ceph_osd_client
*osdc
,
2883 struct ceph_vino vino
, struct ceph_file_layout
*layout
,
2885 u32 truncate_seq
, u64 truncate_size
,
2886 struct page
**pages
, int num_pages
, int page_align
)
2888 struct ceph_osd_request
*req
;
2891 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino
.ino
,
2892 vino
.snap
, off
, *plen
);
2893 req
= ceph_osdc_new_request(osdc
, layout
, vino
, off
, plen
, 0, 1,
2894 CEPH_OSD_OP_READ
, CEPH_OSD_FLAG_READ
,
2895 NULL
, truncate_seq
, truncate_size
,
2898 return PTR_ERR(req
);
2900 /* it may be a short read due to an object boundary */
2901 osd_req_op_extent_osd_data_pages(req
, 0,
2902 pages
, *plen
, page_align
, false, false);
2904 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
2905 off
, *plen
, *plen
, page_align
);
2907 rc
= ceph_osdc_start_request(osdc
, req
, false);
2909 rc
= ceph_osdc_wait_request(osdc
, req
);
2911 ceph_osdc_put_request(req
);
2912 dout("readpages result %d\n", rc
);
2915 EXPORT_SYMBOL(ceph_osdc_readpages
);
2918 * do a synchronous write on N pages
2920 int ceph_osdc_writepages(struct ceph_osd_client
*osdc
, struct ceph_vino vino
,
2921 struct ceph_file_layout
*layout
,
2922 struct ceph_snap_context
*snapc
,
2924 u32 truncate_seq
, u64 truncate_size
,
2925 struct timespec
*mtime
,
2926 struct page
**pages
, int num_pages
)
2928 struct ceph_osd_request
*req
;
2930 int page_align
= off
& ~PAGE_MASK
;
2932 req
= ceph_osdc_new_request(osdc
, layout
, vino
, off
, &len
, 0, 1,
2934 CEPH_OSD_FLAG_ONDISK
| CEPH_OSD_FLAG_WRITE
,
2935 snapc
, truncate_seq
, truncate_size
,
2938 return PTR_ERR(req
);
2940 /* it may be a short write due to an object boundary */
2941 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, page_align
,
2943 dout("writepages %llu~%llu (%llu bytes)\n", off
, len
, len
);
2945 req
->r_mtime
= *mtime
;
2946 rc
= ceph_osdc_start_request(osdc
, req
, true);
2948 rc
= ceph_osdc_wait_request(osdc
, req
);
2950 ceph_osdc_put_request(req
);
2953 dout("writepages result %d\n", rc
);
2956 EXPORT_SYMBOL(ceph_osdc_writepages
);
2958 int ceph_osdc_setup(void)
2960 size_t size
= sizeof(struct ceph_osd_request
) +
2961 CEPH_OSD_SLAB_OPS
* sizeof(struct ceph_osd_req_op
);
2963 BUG_ON(ceph_osd_request_cache
);
2964 ceph_osd_request_cache
= kmem_cache_create("ceph_osd_request", size
,
2967 return ceph_osd_request_cache
? 0 : -ENOMEM
;
2969 EXPORT_SYMBOL(ceph_osdc_setup
);
2971 void ceph_osdc_cleanup(void)
2973 BUG_ON(!ceph_osd_request_cache
);
2974 kmem_cache_destroy(ceph_osd_request_cache
);
2975 ceph_osd_request_cache
= NULL
;
2977 EXPORT_SYMBOL(ceph_osdc_cleanup
);
2980 * handle incoming message
2982 static void dispatch(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2984 struct ceph_osd
*osd
= con
->private;
2985 struct ceph_osd_client
*osdc
;
2986 int type
= le16_to_cpu(msg
->hdr
.type
);
2993 case CEPH_MSG_OSD_MAP
:
2994 ceph_osdc_handle_map(osdc
, msg
);
2996 case CEPH_MSG_OSD_OPREPLY
:
2997 handle_reply(osdc
, msg
);
2999 case CEPH_MSG_WATCH_NOTIFY
:
3000 handle_watch_notify(osdc
, msg
);
3004 pr_err("received unknown message type %d %s\n", type
,
3005 ceph_msg_type_name(type
));
3012 * Lookup and return message for incoming reply. Don't try to do
3013 * anything about a larger than preallocated data portion of the
3014 * message at the moment - for now, just skip the message.
3016 static struct ceph_msg
*get_reply(struct ceph_connection
*con
,
3017 struct ceph_msg_header
*hdr
,
3020 struct ceph_osd
*osd
= con
->private;
3021 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
3023 struct ceph_osd_request
*req
;
3024 int front_len
= le32_to_cpu(hdr
->front_len
);
3025 int data_len
= le32_to_cpu(hdr
->data_len
);
3028 tid
= le64_to_cpu(hdr
->tid
);
3029 mutex_lock(&osdc
->request_mutex
);
3030 req
= lookup_request(&osdc
->requests
, tid
);
3032 dout("%s osd%d tid %llu unknown, skipping\n", __func__
,
3039 ceph_msg_revoke_incoming(req
->r_reply
);
3041 if (front_len
> req
->r_reply
->front_alloc_len
) {
3042 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
3043 __func__
, osd
->o_osd
, req
->r_tid
, front_len
,
3044 req
->r_reply
->front_alloc_len
);
3045 m
= ceph_msg_new(CEPH_MSG_OSD_OPREPLY
, front_len
, GFP_NOFS
,
3049 ceph_msg_put(req
->r_reply
);
3053 if (data_len
> req
->r_reply
->data_length
) {
3054 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
3055 __func__
, osd
->o_osd
, req
->r_tid
, data_len
,
3056 req
->r_reply
->data_length
);
3062 m
= ceph_msg_get(req
->r_reply
);
3063 dout("get_reply tid %lld %p\n", tid
, m
);
3066 mutex_unlock(&osdc
->request_mutex
);
3070 static struct ceph_msg
*alloc_msg(struct ceph_connection
*con
,
3071 struct ceph_msg_header
*hdr
,
3074 struct ceph_osd
*osd
= con
->private;
3075 int type
= le16_to_cpu(hdr
->type
);
3076 int front
= le32_to_cpu(hdr
->front_len
);
3080 case CEPH_MSG_OSD_MAP
:
3081 case CEPH_MSG_WATCH_NOTIFY
:
3082 return ceph_msg_new(type
, front
, GFP_NOFS
, false);
3083 case CEPH_MSG_OSD_OPREPLY
:
3084 return get_reply(con
, hdr
, skip
);
3086 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type
,
3094 * Wrappers to refcount containing ceph_osd struct
3096 static struct ceph_connection
*get_osd_con(struct ceph_connection
*con
)
3098 struct ceph_osd
*osd
= con
->private;
3104 static void put_osd_con(struct ceph_connection
*con
)
3106 struct ceph_osd
*osd
= con
->private;
3114 * Note: returned pointer is the address of a structure that's
3115 * managed separately. Caller must *not* attempt to free it.
3117 static struct ceph_auth_handshake
*get_authorizer(struct ceph_connection
*con
,
3118 int *proto
, int force_new
)
3120 struct ceph_osd
*o
= con
->private;
3121 struct ceph_osd_client
*osdc
= o
->o_osdc
;
3122 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
3123 struct ceph_auth_handshake
*auth
= &o
->o_auth
;
3125 if (force_new
&& auth
->authorizer
) {
3126 ceph_auth_destroy_authorizer(auth
->authorizer
);
3127 auth
->authorizer
= NULL
;
3129 if (!auth
->authorizer
) {
3130 int ret
= ceph_auth_create_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
,
3133 return ERR_PTR(ret
);
3135 int ret
= ceph_auth_update_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
,
3138 return ERR_PTR(ret
);
3140 *proto
= ac
->protocol
;
3146 static int verify_authorizer_reply(struct ceph_connection
*con
, int len
)
3148 struct ceph_osd
*o
= con
->private;
3149 struct ceph_osd_client
*osdc
= o
->o_osdc
;
3150 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
3152 return ceph_auth_verify_authorizer_reply(ac
, o
->o_auth
.authorizer
, len
);
3155 static int invalidate_authorizer(struct ceph_connection
*con
)
3157 struct ceph_osd
*o
= con
->private;
3158 struct ceph_osd_client
*osdc
= o
->o_osdc
;
3159 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
3161 ceph_auth_invalidate_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
);
3162 return ceph_monc_validate_auth(&osdc
->client
->monc
);
3165 static int osd_sign_message(struct ceph_msg
*msg
)
3167 struct ceph_osd
*o
= msg
->con
->private;
3168 struct ceph_auth_handshake
*auth
= &o
->o_auth
;
3170 return ceph_auth_sign_message(auth
, msg
);
3173 static int osd_check_message_signature(struct ceph_msg
*msg
)
3175 struct ceph_osd
*o
= msg
->con
->private;
3176 struct ceph_auth_handshake
*auth
= &o
->o_auth
;
3178 return ceph_auth_check_message_signature(auth
, msg
);
3181 static const struct ceph_connection_operations osd_con_ops
= {
3184 .dispatch
= dispatch
,
3185 .get_authorizer
= get_authorizer
,
3186 .verify_authorizer_reply
= verify_authorizer_reply
,
3187 .invalidate_authorizer
= invalidate_authorizer
,
3188 .alloc_msg
= alloc_msg
,
3189 .sign_message
= osd_sign_message
,
3190 .check_message_signature
= osd_check_message_signature
,