1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
5 #include <linux/highmem.h>
7 #include <linux/pagemap.h>
8 #include <linux/slab.h>
9 #include <linux/uaccess.h>
11 #include <linux/bio.h>
14 #include <linux/ceph/libceph.h>
15 #include <linux/ceph/osd_client.h>
16 #include <linux/ceph/messenger.h>
17 #include <linux/ceph/decode.h>
18 #include <linux/ceph/auth.h>
19 #include <linux/ceph/pagelist.h>
21 #define OSD_OP_FRONT_LEN 4096
22 #define OSD_OPREPLY_FRONT_LEN 512
24 static const struct ceph_connection_operations osd_con_ops
;
26 static void send_queued(struct ceph_osd_client
*osdc
);
27 static int __reset_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
);
28 static void __register_request(struct ceph_osd_client
*osdc
,
29 struct ceph_osd_request
*req
);
30 static void __unregister_linger_request(struct ceph_osd_client
*osdc
,
31 struct ceph_osd_request
*req
);
32 static void __send_request(struct ceph_osd_client
*osdc
,
33 struct ceph_osd_request
*req
);
35 static int op_needs_trail(int op
)
38 case CEPH_OSD_OP_GETXATTR
:
39 case CEPH_OSD_OP_SETXATTR
:
40 case CEPH_OSD_OP_CMPXATTR
:
41 case CEPH_OSD_OP_CALL
:
42 case CEPH_OSD_OP_NOTIFY
:
49 static int op_has_extent(int op
)
51 return (op
== CEPH_OSD_OP_READ
||
52 op
== CEPH_OSD_OP_WRITE
);
55 int ceph_calc_raw_layout(struct ceph_osd_client
*osdc
,
56 struct ceph_file_layout
*layout
,
58 u64 off
, u64
*plen
, u64
*bno
,
59 struct ceph_osd_request
*req
,
60 struct ceph_osd_req_op
*op
)
62 struct ceph_osd_request_head
*reqhead
= req
->r_request
->front
.iov_base
;
64 u64 objoff
, objlen
; /* extent in object */
67 reqhead
->snapid
= cpu_to_le64(snapid
);
70 r
= ceph_calc_file_object_mapping(layout
, off
, plen
, bno
,
75 dout(" skipping last %llu, final file extent %llu~%llu\n",
76 orig_len
- *plen
, off
, *plen
);
78 if (op_has_extent(op
->op
)) {
79 op
->extent
.offset
= objoff
;
80 op
->extent
.length
= objlen
;
82 req
->r_num_pages
= calc_pages_for(off
, *plen
);
83 req
->r_page_alignment
= off
& ~PAGE_MASK
;
84 if (op
->op
== CEPH_OSD_OP_WRITE
)
85 op
->payload_len
= *plen
;
87 dout("calc_layout bno=%llx %llu~%llu (%d pages)\n",
88 *bno
, objoff
, objlen
, req
->r_num_pages
);
91 EXPORT_SYMBOL(ceph_calc_raw_layout
);
94 * Implement client access to distributed object storage cluster.
96 * All data objects are stored within a cluster/cloud of OSDs, or
97 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
98 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
99 * remote daemons serving up and coordinating consistent and safe
102 * Cluster membership and the mapping of data objects onto storage devices
103 * are described by the osd map.
105 * We keep track of pending OSD requests (read, write), resubmit
106 * requests to different OSDs when the cluster topology/data layout
107 * change, or retry the affected requests when the communications
108 * channel with an OSD is reset.
112 * calculate the mapping of a file extent onto an object, and fill out the
113 * request accordingly. shorten extent as necessary if it crosses an
116 * fill osd op in request message.
118 static int calc_layout(struct ceph_osd_client
*osdc
,
119 struct ceph_vino vino
,
120 struct ceph_file_layout
*layout
,
122 struct ceph_osd_request
*req
,
123 struct ceph_osd_req_op
*op
)
128 r
= ceph_calc_raw_layout(osdc
, layout
, vino
.snap
, off
,
129 plen
, &bno
, req
, op
);
133 snprintf(req
->r_oid
, sizeof(req
->r_oid
), "%llx.%08llx", vino
.ino
, bno
);
134 req
->r_oid_len
= strlen(req
->r_oid
);
142 void ceph_osdc_release_request(struct kref
*kref
)
144 struct ceph_osd_request
*req
= container_of(kref
,
145 struct ceph_osd_request
,
149 ceph_msg_put(req
->r_request
);
150 if (req
->r_con_filling_msg
) {
151 dout("%s revoking pages %p from con %p\n", __func__
,
152 req
->r_pages
, req
->r_con_filling_msg
);
153 ceph_msg_revoke_incoming(req
->r_reply
);
154 req
->r_con_filling_msg
->ops
->put(req
->r_con_filling_msg
);
157 ceph_msg_put(req
->r_reply
);
158 if (req
->r_own_pages
)
159 ceph_release_page_vector(req
->r_pages
,
165 ceph_put_snap_context(req
->r_snapc
);
167 ceph_pagelist_release(req
->r_trail
);
171 mempool_free(req
, req
->r_osdc
->req_mempool
);
175 EXPORT_SYMBOL(ceph_osdc_release_request
);
177 static int get_num_ops(struct ceph_osd_req_op
*ops
, int *needs_trail
)
184 if (needs_trail
&& op_needs_trail(ops
[i
].op
))
192 struct ceph_osd_request
*ceph_osdc_alloc_request(struct ceph_osd_client
*osdc
,
194 struct ceph_snap_context
*snapc
,
195 struct ceph_osd_req_op
*ops
,
201 struct ceph_osd_request
*req
;
202 struct ceph_msg
*msg
;
204 int num_op
= get_num_ops(ops
, &needs_trail
);
205 size_t msg_size
= sizeof(struct ceph_osd_request_head
);
207 msg_size
+= num_op
*sizeof(struct ceph_osd_op
);
210 req
= mempool_alloc(osdc
->req_mempool
, gfp_flags
);
211 memset(req
, 0, sizeof(*req
));
213 req
= kzalloc(sizeof(*req
), gfp_flags
);
219 req
->r_mempool
= use_mempool
;
221 kref_init(&req
->r_kref
);
222 init_completion(&req
->r_completion
);
223 init_completion(&req
->r_safe_completion
);
224 rb_init_node(&req
->r_node
);
225 INIT_LIST_HEAD(&req
->r_unsafe_item
);
226 INIT_LIST_HEAD(&req
->r_linger_item
);
227 INIT_LIST_HEAD(&req
->r_linger_osd
);
228 INIT_LIST_HEAD(&req
->r_req_lru_item
);
229 INIT_LIST_HEAD(&req
->r_osd_item
);
231 req
->r_flags
= flags
;
233 WARN_ON((flags
& (CEPH_OSD_FLAG_READ
|CEPH_OSD_FLAG_WRITE
)) == 0);
235 /* create reply message */
237 msg
= ceph_msgpool_get(&osdc
->msgpool_op_reply
, 0);
239 msg
= ceph_msg_new(CEPH_MSG_OSD_OPREPLY
,
240 OSD_OPREPLY_FRONT_LEN
, gfp_flags
, true);
242 ceph_osdc_put_request(req
);
247 /* allocate space for the trailing data */
249 req
->r_trail
= kmalloc(sizeof(struct ceph_pagelist
), gfp_flags
);
251 ceph_osdc_put_request(req
);
254 ceph_pagelist_init(req
->r_trail
);
257 /* create request message; allow space for oid */
258 msg_size
+= MAX_OBJ_NAME_SIZE
;
260 msg_size
+= sizeof(u64
) * snapc
->num_snaps
;
262 msg
= ceph_msgpool_get(&osdc
->msgpool_op
, 0);
264 msg
= ceph_msg_new(CEPH_MSG_OSD_OP
, msg_size
, gfp_flags
, true);
266 ceph_osdc_put_request(req
);
270 memset(msg
->front
.iov_base
, 0, msg
->front
.iov_len
);
272 req
->r_request
= msg
;
273 req
->r_pages
= pages
;
283 EXPORT_SYMBOL(ceph_osdc_alloc_request
);
285 static void osd_req_encode_op(struct ceph_osd_request
*req
,
286 struct ceph_osd_op
*dst
,
287 struct ceph_osd_req_op
*src
)
289 dst
->op
= cpu_to_le16(src
->op
);
292 case CEPH_OSD_OP_READ
:
293 case CEPH_OSD_OP_WRITE
:
295 cpu_to_le64(src
->extent
.offset
);
297 cpu_to_le64(src
->extent
.length
);
298 dst
->extent
.truncate_size
=
299 cpu_to_le64(src
->extent
.truncate_size
);
300 dst
->extent
.truncate_seq
=
301 cpu_to_le32(src
->extent
.truncate_seq
);
304 case CEPH_OSD_OP_GETXATTR
:
305 case CEPH_OSD_OP_SETXATTR
:
306 case CEPH_OSD_OP_CMPXATTR
:
307 BUG_ON(!req
->r_trail
);
309 dst
->xattr
.name_len
= cpu_to_le32(src
->xattr
.name_len
);
310 dst
->xattr
.value_len
= cpu_to_le32(src
->xattr
.value_len
);
311 dst
->xattr
.cmp_op
= src
->xattr
.cmp_op
;
312 dst
->xattr
.cmp_mode
= src
->xattr
.cmp_mode
;
313 ceph_pagelist_append(req
->r_trail
, src
->xattr
.name
,
314 src
->xattr
.name_len
);
315 ceph_pagelist_append(req
->r_trail
, src
->xattr
.val
,
316 src
->xattr
.value_len
);
318 case CEPH_OSD_OP_CALL
:
319 BUG_ON(!req
->r_trail
);
321 dst
->cls
.class_len
= src
->cls
.class_len
;
322 dst
->cls
.method_len
= src
->cls
.method_len
;
323 dst
->cls
.indata_len
= cpu_to_le32(src
->cls
.indata_len
);
325 ceph_pagelist_append(req
->r_trail
, src
->cls
.class_name
,
327 ceph_pagelist_append(req
->r_trail
, src
->cls
.method_name
,
328 src
->cls
.method_len
);
329 ceph_pagelist_append(req
->r_trail
, src
->cls
.indata
,
330 src
->cls
.indata_len
);
332 case CEPH_OSD_OP_ROLLBACK
:
333 dst
->snap
.snapid
= cpu_to_le64(src
->snap
.snapid
);
335 case CEPH_OSD_OP_STARTSYNC
:
337 case CEPH_OSD_OP_NOTIFY
:
339 __le32 prot_ver
= cpu_to_le32(src
->watch
.prot_ver
);
340 __le32 timeout
= cpu_to_le32(src
->watch
.timeout
);
342 BUG_ON(!req
->r_trail
);
344 ceph_pagelist_append(req
->r_trail
,
345 &prot_ver
, sizeof(prot_ver
));
346 ceph_pagelist_append(req
->r_trail
,
347 &timeout
, sizeof(timeout
));
349 case CEPH_OSD_OP_NOTIFY_ACK
:
350 case CEPH_OSD_OP_WATCH
:
351 dst
->watch
.cookie
= cpu_to_le64(src
->watch
.cookie
);
352 dst
->watch
.ver
= cpu_to_le64(src
->watch
.ver
);
353 dst
->watch
.flag
= src
->watch
.flag
;
356 pr_err("unrecognized osd opcode %d\n", dst
->op
);
360 dst
->payload_len
= cpu_to_le32(src
->payload_len
);
364 * build new request AND message
367 void ceph_osdc_build_request(struct ceph_osd_request
*req
,
369 struct ceph_osd_req_op
*src_ops
,
370 struct ceph_snap_context
*snapc
,
371 struct timespec
*mtime
,
375 struct ceph_msg
*msg
= req
->r_request
;
376 struct ceph_osd_request_head
*head
;
377 struct ceph_osd_req_op
*src_op
;
378 struct ceph_osd_op
*op
;
380 int num_op
= get_num_ops(src_ops
, NULL
);
381 size_t msg_size
= sizeof(*head
) + num_op
*sizeof(*op
);
382 int flags
= req
->r_flags
;
386 head
= msg
->front
.iov_base
;
387 op
= (void *)(head
+ 1);
388 p
= (void *)(op
+ num_op
);
390 req
->r_snapc
= ceph_get_snap_context(snapc
);
392 head
->client_inc
= cpu_to_le32(1); /* always, for now. */
393 head
->flags
= cpu_to_le32(flags
);
394 if (flags
& CEPH_OSD_FLAG_WRITE
)
395 ceph_encode_timespec(&head
->mtime
, mtime
);
396 head
->num_ops
= cpu_to_le16(num_op
);
400 head
->object_len
= cpu_to_le32(oid_len
);
401 memcpy(p
, oid
, oid_len
);
406 osd_req_encode_op(req
, op
, src_op
);
412 data_len
+= req
->r_trail
->length
;
415 head
->snap_seq
= cpu_to_le64(snapc
->seq
);
416 head
->num_snaps
= cpu_to_le32(snapc
->num_snaps
);
417 for (i
= 0; i
< snapc
->num_snaps
; i
++) {
418 put_unaligned_le64(snapc
->snaps
[i
], p
);
423 if (flags
& CEPH_OSD_FLAG_WRITE
) {
424 req
->r_request
->hdr
.data_off
= cpu_to_le16(off
);
425 req
->r_request
->hdr
.data_len
= cpu_to_le32(*plen
+ data_len
);
426 } else if (data_len
) {
427 req
->r_request
->hdr
.data_off
= 0;
428 req
->r_request
->hdr
.data_len
= cpu_to_le32(data_len
);
431 req
->r_request
->page_alignment
= req
->r_page_alignment
;
433 BUG_ON(p
> msg
->front
.iov_base
+ msg
->front
.iov_len
);
434 msg_size
= p
- msg
->front
.iov_base
;
435 msg
->front
.iov_len
= msg_size
;
436 msg
->hdr
.front_len
= cpu_to_le32(msg_size
);
439 EXPORT_SYMBOL(ceph_osdc_build_request
);
442 * build new request AND message, calculate layout, and adjust file
445 * if the file was recently truncated, we include information about its
446 * old and new size so that the object can be updated appropriately. (we
447 * avoid synchronously deleting truncated objects because it's slow.)
449 * if @do_sync, include a 'startsync' command so that the osd will flush
452 struct ceph_osd_request
*ceph_osdc_new_request(struct ceph_osd_client
*osdc
,
453 struct ceph_file_layout
*layout
,
454 struct ceph_vino vino
,
456 int opcode
, int flags
,
457 struct ceph_snap_context
*snapc
,
461 struct timespec
*mtime
,
462 bool use_mempool
, int num_reply
,
465 struct ceph_osd_req_op ops
[3];
466 struct ceph_osd_request
*req
;
470 ops
[0].extent
.truncate_seq
= truncate_seq
;
471 ops
[0].extent
.truncate_size
= truncate_size
;
472 ops
[0].payload_len
= 0;
475 ops
[1].op
= CEPH_OSD_OP_STARTSYNC
;
476 ops
[1].payload_len
= 0;
481 req
= ceph_osdc_alloc_request(osdc
, flags
,
484 GFP_NOFS
, NULL
, NULL
);
486 return ERR_PTR(-ENOMEM
);
488 /* calculate max write size */
489 r
= calc_layout(osdc
, vino
, layout
, off
, plen
, req
, ops
);
492 req
->r_file_layout
= *layout
; /* keep a copy */
494 /* in case it differs from natural (file) alignment that
495 calc_layout filled in for us */
496 req
->r_num_pages
= calc_pages_for(page_align
, *plen
);
497 req
->r_page_alignment
= page_align
;
499 ceph_osdc_build_request(req
, off
, plen
, ops
,
502 req
->r_oid
, req
->r_oid_len
);
506 EXPORT_SYMBOL(ceph_osdc_new_request
);
509 * We keep osd requests in an rbtree, sorted by ->r_tid.
511 static void __insert_request(struct ceph_osd_client
*osdc
,
512 struct ceph_osd_request
*new)
514 struct rb_node
**p
= &osdc
->requests
.rb_node
;
515 struct rb_node
*parent
= NULL
;
516 struct ceph_osd_request
*req
= NULL
;
520 req
= rb_entry(parent
, struct ceph_osd_request
, r_node
);
521 if (new->r_tid
< req
->r_tid
)
523 else if (new->r_tid
> req
->r_tid
)
529 rb_link_node(&new->r_node
, parent
, p
);
530 rb_insert_color(&new->r_node
, &osdc
->requests
);
533 static struct ceph_osd_request
*__lookup_request(struct ceph_osd_client
*osdc
,
536 struct ceph_osd_request
*req
;
537 struct rb_node
*n
= osdc
->requests
.rb_node
;
540 req
= rb_entry(n
, struct ceph_osd_request
, r_node
);
541 if (tid
< req
->r_tid
)
543 else if (tid
> req
->r_tid
)
551 static struct ceph_osd_request
*
552 __lookup_request_ge(struct ceph_osd_client
*osdc
,
555 struct ceph_osd_request
*req
;
556 struct rb_node
*n
= osdc
->requests
.rb_node
;
559 req
= rb_entry(n
, struct ceph_osd_request
, r_node
);
560 if (tid
< req
->r_tid
) {
564 } else if (tid
> req
->r_tid
) {
574 * Resubmit requests pending on the given osd.
576 static void __kick_osd_requests(struct ceph_osd_client
*osdc
,
577 struct ceph_osd
*osd
)
579 struct ceph_osd_request
*req
, *nreq
;
582 dout("__kick_osd_requests osd%d\n", osd
->o_osd
);
583 err
= __reset_osd(osdc
, osd
);
587 list_for_each_entry(req
, &osd
->o_requests
, r_osd_item
) {
588 list_move(&req
->r_req_lru_item
, &osdc
->req_unsent
);
589 dout("requeued %p tid %llu osd%d\n", req
, req
->r_tid
,
592 req
->r_flags
|= CEPH_OSD_FLAG_RETRY
;
595 list_for_each_entry_safe(req
, nreq
, &osd
->o_linger_requests
,
598 * reregister request prior to unregistering linger so
599 * that r_osd is preserved.
601 BUG_ON(!list_empty(&req
->r_req_lru_item
));
602 __register_request(osdc
, req
);
603 list_add(&req
->r_req_lru_item
, &osdc
->req_unsent
);
604 list_add(&req
->r_osd_item
, &req
->r_osd
->o_requests
);
605 __unregister_linger_request(osdc
, req
);
606 dout("requeued lingering %p tid %llu osd%d\n", req
, req
->r_tid
,
612 * If the osd connection drops, we need to resubmit all requests.
614 static void osd_reset(struct ceph_connection
*con
)
616 struct ceph_osd
*osd
= con
->private;
617 struct ceph_osd_client
*osdc
;
621 dout("osd_reset osd%d\n", osd
->o_osd
);
623 down_read(&osdc
->map_sem
);
624 mutex_lock(&osdc
->request_mutex
);
625 __kick_osd_requests(osdc
, osd
);
626 mutex_unlock(&osdc
->request_mutex
);
628 up_read(&osdc
->map_sem
);
632 * Track open sessions with osds.
634 static struct ceph_osd
*create_osd(struct ceph_osd_client
*osdc
, int onum
)
636 struct ceph_osd
*osd
;
638 osd
= kzalloc(sizeof(*osd
), GFP_NOFS
);
642 atomic_set(&osd
->o_ref
, 1);
645 INIT_LIST_HEAD(&osd
->o_requests
);
646 INIT_LIST_HEAD(&osd
->o_linger_requests
);
647 INIT_LIST_HEAD(&osd
->o_osd_lru
);
648 osd
->o_incarnation
= 1;
650 ceph_con_init(&osd
->o_con
, osd
, &osd_con_ops
, &osdc
->client
->msgr
);
652 INIT_LIST_HEAD(&osd
->o_keepalive_item
);
656 static struct ceph_osd
*get_osd(struct ceph_osd
*osd
)
658 if (atomic_inc_not_zero(&osd
->o_ref
)) {
659 dout("get_osd %p %d -> %d\n", osd
, atomic_read(&osd
->o_ref
)-1,
660 atomic_read(&osd
->o_ref
));
663 dout("get_osd %p FAIL\n", osd
);
668 static void put_osd(struct ceph_osd
*osd
)
670 dout("put_osd %p %d -> %d\n", osd
, atomic_read(&osd
->o_ref
),
671 atomic_read(&osd
->o_ref
) - 1);
672 if (atomic_dec_and_test(&osd
->o_ref
) && osd
->o_auth
.authorizer
) {
673 struct ceph_auth_client
*ac
= osd
->o_osdc
->client
->monc
.auth
;
675 if (ac
->ops
&& ac
->ops
->destroy_authorizer
)
676 ac
->ops
->destroy_authorizer(ac
, osd
->o_auth
.authorizer
);
682 * remove an osd from our map
684 static void __remove_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
)
686 dout("__remove_osd %p\n", osd
);
687 BUG_ON(!list_empty(&osd
->o_requests
));
688 rb_erase(&osd
->o_node
, &osdc
->osds
);
689 list_del_init(&osd
->o_osd_lru
);
690 ceph_con_close(&osd
->o_con
);
694 static void remove_all_osds(struct ceph_osd_client
*osdc
)
696 dout("%s %p\n", __func__
, osdc
);
697 mutex_lock(&osdc
->request_mutex
);
698 while (!RB_EMPTY_ROOT(&osdc
->osds
)) {
699 struct ceph_osd
*osd
= rb_entry(rb_first(&osdc
->osds
),
700 struct ceph_osd
, o_node
);
701 __remove_osd(osdc
, osd
);
703 mutex_unlock(&osdc
->request_mutex
);
706 static void __move_osd_to_lru(struct ceph_osd_client
*osdc
,
707 struct ceph_osd
*osd
)
709 dout("__move_osd_to_lru %p\n", osd
);
710 BUG_ON(!list_empty(&osd
->o_osd_lru
));
711 list_add_tail(&osd
->o_osd_lru
, &osdc
->osd_lru
);
712 osd
->lru_ttl
= jiffies
+ osdc
->client
->options
->osd_idle_ttl
* HZ
;
715 static void __remove_osd_from_lru(struct ceph_osd
*osd
)
717 dout("__remove_osd_from_lru %p\n", osd
);
718 if (!list_empty(&osd
->o_osd_lru
))
719 list_del_init(&osd
->o_osd_lru
);
722 static void remove_old_osds(struct ceph_osd_client
*osdc
)
724 struct ceph_osd
*osd
, *nosd
;
726 dout("__remove_old_osds %p\n", osdc
);
727 mutex_lock(&osdc
->request_mutex
);
728 list_for_each_entry_safe(osd
, nosd
, &osdc
->osd_lru
, o_osd_lru
) {
729 if (time_before(jiffies
, osd
->lru_ttl
))
731 __remove_osd(osdc
, osd
);
733 mutex_unlock(&osdc
->request_mutex
);
739 static int __reset_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*osd
)
741 struct ceph_osd_request
*req
;
744 dout("__reset_osd %p osd%d\n", osd
, osd
->o_osd
);
745 if (list_empty(&osd
->o_requests
) &&
746 list_empty(&osd
->o_linger_requests
)) {
747 __remove_osd(osdc
, osd
);
749 } else if (memcmp(&osdc
->osdmap
->osd_addr
[osd
->o_osd
],
750 &osd
->o_con
.peer_addr
,
751 sizeof(osd
->o_con
.peer_addr
)) == 0 &&
752 !ceph_con_opened(&osd
->o_con
)) {
753 dout(" osd addr hasn't changed and connection never opened,"
754 " letting msgr retry");
755 /* touch each r_stamp for handle_timeout()'s benfit */
756 list_for_each_entry(req
, &osd
->o_requests
, r_osd_item
)
757 req
->r_stamp
= jiffies
;
760 ceph_con_close(&osd
->o_con
);
761 ceph_con_open(&osd
->o_con
, CEPH_ENTITY_TYPE_OSD
, osd
->o_osd
,
762 &osdc
->osdmap
->osd_addr
[osd
->o_osd
]);
763 osd
->o_incarnation
++;
768 static void __insert_osd(struct ceph_osd_client
*osdc
, struct ceph_osd
*new)
770 struct rb_node
**p
= &osdc
->osds
.rb_node
;
771 struct rb_node
*parent
= NULL
;
772 struct ceph_osd
*osd
= NULL
;
774 dout("__insert_osd %p osd%d\n", new, new->o_osd
);
777 osd
= rb_entry(parent
, struct ceph_osd
, o_node
);
778 if (new->o_osd
< osd
->o_osd
)
780 else if (new->o_osd
> osd
->o_osd
)
786 rb_link_node(&new->o_node
, parent
, p
);
787 rb_insert_color(&new->o_node
, &osdc
->osds
);
790 static struct ceph_osd
*__lookup_osd(struct ceph_osd_client
*osdc
, int o
)
792 struct ceph_osd
*osd
;
793 struct rb_node
*n
= osdc
->osds
.rb_node
;
796 osd
= rb_entry(n
, struct ceph_osd
, o_node
);
799 else if (o
> osd
->o_osd
)
807 static void __schedule_osd_timeout(struct ceph_osd_client
*osdc
)
809 schedule_delayed_work(&osdc
->timeout_work
,
810 osdc
->client
->options
->osd_keepalive_timeout
* HZ
);
813 static void __cancel_osd_timeout(struct ceph_osd_client
*osdc
)
815 cancel_delayed_work(&osdc
->timeout_work
);
819 * Register request, assign tid. If this is the first request, set up
822 static void __register_request(struct ceph_osd_client
*osdc
,
823 struct ceph_osd_request
*req
)
825 req
->r_tid
= ++osdc
->last_tid
;
826 req
->r_request
->hdr
.tid
= cpu_to_le64(req
->r_tid
);
827 dout("__register_request %p tid %lld\n", req
, req
->r_tid
);
828 __insert_request(osdc
, req
);
829 ceph_osdc_get_request(req
);
830 osdc
->num_requests
++;
831 if (osdc
->num_requests
== 1) {
832 dout(" first request, scheduling timeout\n");
833 __schedule_osd_timeout(osdc
);
837 static void register_request(struct ceph_osd_client
*osdc
,
838 struct ceph_osd_request
*req
)
840 mutex_lock(&osdc
->request_mutex
);
841 __register_request(osdc
, req
);
842 mutex_unlock(&osdc
->request_mutex
);
846 * called under osdc->request_mutex
848 static void __unregister_request(struct ceph_osd_client
*osdc
,
849 struct ceph_osd_request
*req
)
851 if (RB_EMPTY_NODE(&req
->r_node
)) {
852 dout("__unregister_request %p tid %lld not registered\n",
857 dout("__unregister_request %p tid %lld\n", req
, req
->r_tid
);
858 rb_erase(&req
->r_node
, &osdc
->requests
);
859 osdc
->num_requests
--;
862 /* make sure the original request isn't in flight. */
863 ceph_msg_revoke(req
->r_request
);
865 list_del_init(&req
->r_osd_item
);
866 if (list_empty(&req
->r_osd
->o_requests
) &&
867 list_empty(&req
->r_osd
->o_linger_requests
)) {
868 dout("moving osd to %p lru\n", req
->r_osd
);
869 __move_osd_to_lru(osdc
, req
->r_osd
);
871 if (list_empty(&req
->r_linger_item
))
875 list_del_init(&req
->r_req_lru_item
);
876 ceph_osdc_put_request(req
);
878 if (osdc
->num_requests
== 0) {
879 dout(" no requests, canceling timeout\n");
880 __cancel_osd_timeout(osdc
);
885 * Cancel a previously queued request message
887 static void __cancel_request(struct ceph_osd_request
*req
)
889 if (req
->r_sent
&& req
->r_osd
) {
890 ceph_msg_revoke(req
->r_request
);
895 static void __register_linger_request(struct ceph_osd_client
*osdc
,
896 struct ceph_osd_request
*req
)
898 dout("__register_linger_request %p\n", req
);
899 list_add_tail(&req
->r_linger_item
, &osdc
->req_linger
);
901 list_add_tail(&req
->r_linger_osd
,
902 &req
->r_osd
->o_linger_requests
);
905 static void __unregister_linger_request(struct ceph_osd_client
*osdc
,
906 struct ceph_osd_request
*req
)
908 dout("__unregister_linger_request %p\n", req
);
909 list_del_init(&req
->r_linger_item
);
911 list_del_init(&req
->r_linger_osd
);
913 if (list_empty(&req
->r_osd
->o_requests
) &&
914 list_empty(&req
->r_osd
->o_linger_requests
)) {
915 dout("moving osd to %p lru\n", req
->r_osd
);
916 __move_osd_to_lru(osdc
, req
->r_osd
);
918 if (list_empty(&req
->r_osd_item
))
923 void ceph_osdc_unregister_linger_request(struct ceph_osd_client
*osdc
,
924 struct ceph_osd_request
*req
)
926 mutex_lock(&osdc
->request_mutex
);
928 __unregister_linger_request(osdc
, req
);
929 ceph_osdc_put_request(req
);
931 mutex_unlock(&osdc
->request_mutex
);
933 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request
);
935 void ceph_osdc_set_request_linger(struct ceph_osd_client
*osdc
,
936 struct ceph_osd_request
*req
)
938 if (!req
->r_linger
) {
939 dout("set_request_linger %p\n", req
);
942 * caller is now responsible for calling
943 * unregister_linger_request
945 ceph_osdc_get_request(req
);
948 EXPORT_SYMBOL(ceph_osdc_set_request_linger
);
951 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
952 * (as needed), and set the request r_osd appropriately. If there is
953 * no up osd, set r_osd to NULL. Move the request to the appropriate list
954 * (unsent, homeless) or leave on in-flight lru.
956 * Return 0 if unchanged, 1 if changed, or negative on error.
958 * Caller should hold map_sem for read and request_mutex.
960 static int __map_request(struct ceph_osd_client
*osdc
,
961 struct ceph_osd_request
*req
, int force_resend
)
963 struct ceph_osd_request_head
*reqhead
= req
->r_request
->front
.iov_base
;
965 int acting
[CEPH_PG_MAX_SIZE
];
969 dout("map_request %p tid %lld\n", req
, req
->r_tid
);
970 err
= ceph_calc_object_layout(&reqhead
->layout
, req
->r_oid
,
971 &req
->r_file_layout
, osdc
->osdmap
);
973 list_move(&req
->r_req_lru_item
, &osdc
->req_notarget
);
976 pgid
= reqhead
->layout
.ol_pgid
;
979 err
= ceph_calc_pg_acting(osdc
->osdmap
, pgid
, acting
);
985 if ((!force_resend
&&
986 req
->r_osd
&& req
->r_osd
->o_osd
== o
&&
987 req
->r_sent
>= req
->r_osd
->o_incarnation
&&
988 req
->r_num_pg_osds
== num
&&
989 memcmp(req
->r_pg_osds
, acting
, sizeof(acting
[0])*num
) == 0) ||
990 (req
->r_osd
== NULL
&& o
== -1))
991 return 0; /* no change */
993 dout("map_request tid %llu pgid %d.%x osd%d (was osd%d)\n",
994 req
->r_tid
, le32_to_cpu(pgid
.pool
), le16_to_cpu(pgid
.ps
), o
,
995 req
->r_osd
? req
->r_osd
->o_osd
: -1);
997 /* record full pg acting set */
998 memcpy(req
->r_pg_osds
, acting
, sizeof(acting
[0]) * num
);
999 req
->r_num_pg_osds
= num
;
1002 __cancel_request(req
);
1003 list_del_init(&req
->r_osd_item
);
1007 req
->r_osd
= __lookup_osd(osdc
, o
);
1008 if (!req
->r_osd
&& o
>= 0) {
1010 req
->r_osd
= create_osd(osdc
, o
);
1012 list_move(&req
->r_req_lru_item
, &osdc
->req_notarget
);
1016 dout("map_request osd %p is osd%d\n", req
->r_osd
, o
);
1017 __insert_osd(osdc
, req
->r_osd
);
1019 ceph_con_open(&req
->r_osd
->o_con
,
1020 CEPH_ENTITY_TYPE_OSD
, o
,
1021 &osdc
->osdmap
->osd_addr
[o
]);
1025 __remove_osd_from_lru(req
->r_osd
);
1026 list_add(&req
->r_osd_item
, &req
->r_osd
->o_requests
);
1027 list_move(&req
->r_req_lru_item
, &osdc
->req_unsent
);
1029 list_move(&req
->r_req_lru_item
, &osdc
->req_notarget
);
1031 err
= 1; /* osd or pg changed */
1038 * caller should hold map_sem (for read) and request_mutex
1040 static void __send_request(struct ceph_osd_client
*osdc
,
1041 struct ceph_osd_request
*req
)
1043 struct ceph_osd_request_head
*reqhead
;
1045 dout("send_request %p tid %llu to osd%d flags %d\n",
1046 req
, req
->r_tid
, req
->r_osd
->o_osd
, req
->r_flags
);
1048 reqhead
= req
->r_request
->front
.iov_base
;
1049 reqhead
->osdmap_epoch
= cpu_to_le32(osdc
->osdmap
->epoch
);
1050 reqhead
->flags
|= cpu_to_le32(req
->r_flags
); /* e.g., RETRY */
1051 reqhead
->reassert_version
= req
->r_reassert_version
;
1053 req
->r_stamp
= jiffies
;
1054 list_move_tail(&req
->r_req_lru_item
, &osdc
->req_lru
);
1056 ceph_msg_get(req
->r_request
); /* send consumes a ref */
1057 ceph_con_send(&req
->r_osd
->o_con
, req
->r_request
);
1058 req
->r_sent
= req
->r_osd
->o_incarnation
;
1062 * Send any requests in the queue (req_unsent).
1064 static void send_queued(struct ceph_osd_client
*osdc
)
1066 struct ceph_osd_request
*req
, *tmp
;
1068 dout("send_queued\n");
1069 mutex_lock(&osdc
->request_mutex
);
1070 list_for_each_entry_safe(req
, tmp
, &osdc
->req_unsent
, r_req_lru_item
) {
1071 __send_request(osdc
, req
);
1073 mutex_unlock(&osdc
->request_mutex
);
1077 * Timeout callback, called every N seconds when 1 or more osd
1078 * requests has been active for more than N seconds. When this
1079 * happens, we ping all OSDs with requests who have timed out to
1080 * ensure any communications channel reset is detected. Reset the
1081 * request timeouts another N seconds in the future as we go.
1082 * Reschedule the timeout event another N seconds in future (unless
1083 * there are no open requests).
1085 static void handle_timeout(struct work_struct
*work
)
1087 struct ceph_osd_client
*osdc
=
1088 container_of(work
, struct ceph_osd_client
, timeout_work
.work
);
1089 struct ceph_osd_request
*req
;
1090 struct ceph_osd
*osd
;
1091 unsigned long keepalive
=
1092 osdc
->client
->options
->osd_keepalive_timeout
* HZ
;
1093 struct list_head slow_osds
;
1095 down_read(&osdc
->map_sem
);
1097 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1099 mutex_lock(&osdc
->request_mutex
);
1102 * ping osds that are a bit slow. this ensures that if there
1103 * is a break in the TCP connection we will notice, and reopen
1104 * a connection with that osd (from the fault callback).
1106 INIT_LIST_HEAD(&slow_osds
);
1107 list_for_each_entry(req
, &osdc
->req_lru
, r_req_lru_item
) {
1108 if (time_before(jiffies
, req
->r_stamp
+ keepalive
))
1113 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1114 req
->r_tid
, osd
->o_osd
);
1115 list_move_tail(&osd
->o_keepalive_item
, &slow_osds
);
1117 while (!list_empty(&slow_osds
)) {
1118 osd
= list_entry(slow_osds
.next
, struct ceph_osd
,
1120 list_del_init(&osd
->o_keepalive_item
);
1121 ceph_con_keepalive(&osd
->o_con
);
1124 __schedule_osd_timeout(osdc
);
1125 mutex_unlock(&osdc
->request_mutex
);
1127 up_read(&osdc
->map_sem
);
1130 static void handle_osds_timeout(struct work_struct
*work
)
1132 struct ceph_osd_client
*osdc
=
1133 container_of(work
, struct ceph_osd_client
,
1134 osds_timeout_work
.work
);
1135 unsigned long delay
=
1136 osdc
->client
->options
->osd_idle_ttl
* HZ
>> 2;
1138 dout("osds timeout\n");
1139 down_read(&osdc
->map_sem
);
1140 remove_old_osds(osdc
);
1141 up_read(&osdc
->map_sem
);
1143 schedule_delayed_work(&osdc
->osds_timeout_work
,
1144 round_jiffies_relative(delay
));
1147 static void complete_request(struct ceph_osd_request
*req
)
1149 if (req
->r_safe_callback
)
1150 req
->r_safe_callback(req
, NULL
);
1151 complete_all(&req
->r_safe_completion
); /* fsync waiter */
1155 * handle osd op reply. either call the callback if it is specified,
1156 * or do the completion to wake up the waiting thread.
1158 static void handle_reply(struct ceph_osd_client
*osdc
, struct ceph_msg
*msg
,
1159 struct ceph_connection
*con
)
1161 struct ceph_osd_reply_head
*rhead
= msg
->front
.iov_base
;
1162 struct ceph_osd_request
*req
;
1164 int numops
, object_len
, flags
;
1167 tid
= le64_to_cpu(msg
->hdr
.tid
);
1168 if (msg
->front
.iov_len
< sizeof(*rhead
))
1170 numops
= le32_to_cpu(rhead
->num_ops
);
1171 object_len
= le32_to_cpu(rhead
->object_len
);
1172 result
= le32_to_cpu(rhead
->result
);
1173 if (msg
->front
.iov_len
!= sizeof(*rhead
) + object_len
+
1174 numops
* sizeof(struct ceph_osd_op
))
1176 dout("handle_reply %p tid %llu result %d\n", msg
, tid
, (int)result
);
1178 mutex_lock(&osdc
->request_mutex
);
1179 req
= __lookup_request(osdc
, tid
);
1181 dout("handle_reply tid %llu dne\n", tid
);
1182 mutex_unlock(&osdc
->request_mutex
);
1185 ceph_osdc_get_request(req
);
1186 flags
= le32_to_cpu(rhead
->flags
);
1189 * if this connection filled our message, drop our reference now, to
1190 * avoid a (safe but slower) revoke later.
1192 if (req
->r_con_filling_msg
== con
&& req
->r_reply
== msg
) {
1193 dout(" dropping con_filling_msg ref %p\n", con
);
1194 req
->r_con_filling_msg
= NULL
;
1198 if (!req
->r_got_reply
) {
1201 req
->r_result
= le32_to_cpu(rhead
->result
);
1202 bytes
= le32_to_cpu(msg
->hdr
.data_len
);
1203 dout("handle_reply result %d bytes %d\n", req
->r_result
,
1205 if (req
->r_result
== 0)
1206 req
->r_result
= bytes
;
1208 /* in case this is a write and we need to replay, */
1209 req
->r_reassert_version
= rhead
->reassert_version
;
1211 req
->r_got_reply
= 1;
1212 } else if ((flags
& CEPH_OSD_FLAG_ONDISK
) == 0) {
1213 dout("handle_reply tid %llu dup ack\n", tid
);
1214 mutex_unlock(&osdc
->request_mutex
);
1218 dout("handle_reply tid %llu flags %d\n", tid
, flags
);
1220 if (req
->r_linger
&& (flags
& CEPH_OSD_FLAG_ONDISK
))
1221 __register_linger_request(osdc
, req
);
1223 /* either this is a read, or we got the safe response */
1225 (flags
& CEPH_OSD_FLAG_ONDISK
) ||
1226 ((flags
& CEPH_OSD_FLAG_WRITE
) == 0))
1227 __unregister_request(osdc
, req
);
1229 mutex_unlock(&osdc
->request_mutex
);
1231 if (req
->r_callback
)
1232 req
->r_callback(req
, msg
);
1234 complete_all(&req
->r_completion
);
1236 if (flags
& CEPH_OSD_FLAG_ONDISK
)
1237 complete_request(req
);
1240 dout("req=%p req->r_linger=%d\n", req
, req
->r_linger
);
1241 ceph_osdc_put_request(req
);
1245 pr_err("corrupt osd_op_reply got %d %d expected %d\n",
1246 (int)msg
->front
.iov_len
, le32_to_cpu(msg
->hdr
.front_len
),
1247 (int)sizeof(*rhead
));
1251 static void reset_changed_osds(struct ceph_osd_client
*osdc
)
1253 struct rb_node
*p
, *n
;
1255 for (p
= rb_first(&osdc
->osds
); p
; p
= n
) {
1256 struct ceph_osd
*osd
= rb_entry(p
, struct ceph_osd
, o_node
);
1259 if (!ceph_osd_is_up(osdc
->osdmap
, osd
->o_osd
) ||
1260 memcmp(&osd
->o_con
.peer_addr
,
1261 ceph_osd_addr(osdc
->osdmap
,
1263 sizeof(struct ceph_entity_addr
)) != 0)
1264 __reset_osd(osdc
, osd
);
1269 * Requeue requests whose mapping to an OSD has changed. If requests map to
1270 * no osd, request a new map.
1272 * Caller should hold map_sem for read and request_mutex.
1274 static void kick_requests(struct ceph_osd_client
*osdc
, int force_resend
)
1276 struct ceph_osd_request
*req
, *nreq
;
1281 dout("kick_requests %s\n", force_resend
? " (force resend)" : "");
1282 mutex_lock(&osdc
->request_mutex
);
1283 for (p
= rb_first(&osdc
->requests
); p
; ) {
1284 req
= rb_entry(p
, struct ceph_osd_request
, r_node
);
1286 err
= __map_request(osdc
, req
, force_resend
);
1288 continue; /* error */
1289 if (req
->r_osd
== NULL
) {
1290 dout("%p tid %llu maps to no osd\n", req
, req
->r_tid
);
1291 needmap
++; /* request a newer map */
1292 } else if (err
> 0) {
1293 if (!req
->r_linger
) {
1294 dout("%p tid %llu requeued on osd%d\n", req
,
1296 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1297 req
->r_flags
|= CEPH_OSD_FLAG_RETRY
;
1300 if (req
->r_linger
&& list_empty(&req
->r_linger_item
)) {
1302 * register as a linger so that we will
1303 * re-submit below and get a new tid
1305 dout("%p tid %llu restart on osd%d\n",
1307 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1308 __register_linger_request(osdc
, req
);
1309 __unregister_request(osdc
, req
);
1313 list_for_each_entry_safe(req
, nreq
, &osdc
->req_linger
,
1315 dout("linger req=%p req->r_osd=%p\n", req
, req
->r_osd
);
1317 err
= __map_request(osdc
, req
, force_resend
);
1319 continue; /* no change and no osd was specified */
1321 continue; /* hrm! */
1322 if (req
->r_osd
== NULL
) {
1323 dout("tid %llu maps to no valid osd\n", req
->r_tid
);
1324 needmap
++; /* request a newer map */
1328 dout("kicking lingering %p tid %llu osd%d\n", req
, req
->r_tid
,
1329 req
->r_osd
? req
->r_osd
->o_osd
: -1);
1330 __unregister_linger_request(osdc
, req
);
1331 __register_request(osdc
, req
);
1333 mutex_unlock(&osdc
->request_mutex
);
1336 dout("%d requests for down osds, need new map\n", needmap
);
1337 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1343 * Process updated osd map.
1345 * The message contains any number of incremental and full maps, normally
1346 * indicating some sort of topology change in the cluster. Kick requests
1347 * off to different OSDs as needed.
1349 void ceph_osdc_handle_map(struct ceph_osd_client
*osdc
, struct ceph_msg
*msg
)
1351 void *p
, *end
, *next
;
1352 u32 nr_maps
, maplen
;
1354 struct ceph_osdmap
*newmap
= NULL
, *oldmap
;
1356 struct ceph_fsid fsid
;
1358 dout("handle_map have %u\n", osdc
->osdmap
? osdc
->osdmap
->epoch
: 0);
1359 p
= msg
->front
.iov_base
;
1360 end
= p
+ msg
->front
.iov_len
;
1363 ceph_decode_need(&p
, end
, sizeof(fsid
), bad
);
1364 ceph_decode_copy(&p
, &fsid
, sizeof(fsid
));
1365 if (ceph_check_fsid(osdc
->client
, &fsid
) < 0)
1368 down_write(&osdc
->map_sem
);
1370 /* incremental maps */
1371 ceph_decode_32_safe(&p
, end
, nr_maps
, bad
);
1372 dout(" %d inc maps\n", nr_maps
);
1373 while (nr_maps
> 0) {
1374 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
1375 epoch
= ceph_decode_32(&p
);
1376 maplen
= ceph_decode_32(&p
);
1377 ceph_decode_need(&p
, end
, maplen
, bad
);
1379 if (osdc
->osdmap
&& osdc
->osdmap
->epoch
+1 == epoch
) {
1380 dout("applying incremental map %u len %d\n",
1382 newmap
= osdmap_apply_incremental(&p
, next
,
1384 &osdc
->client
->msgr
);
1385 if (IS_ERR(newmap
)) {
1386 err
= PTR_ERR(newmap
);
1390 if (newmap
!= osdc
->osdmap
) {
1391 ceph_osdmap_destroy(osdc
->osdmap
);
1392 osdc
->osdmap
= newmap
;
1394 kick_requests(osdc
, 0);
1395 reset_changed_osds(osdc
);
1397 dout("ignoring incremental map %u len %d\n",
1407 ceph_decode_32_safe(&p
, end
, nr_maps
, bad
);
1408 dout(" %d full maps\n", nr_maps
);
1410 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
1411 epoch
= ceph_decode_32(&p
);
1412 maplen
= ceph_decode_32(&p
);
1413 ceph_decode_need(&p
, end
, maplen
, bad
);
1415 dout("skipping non-latest full map %u len %d\n",
1417 } else if (osdc
->osdmap
&& osdc
->osdmap
->epoch
>= epoch
) {
1418 dout("skipping full map %u len %d, "
1419 "older than our %u\n", epoch
, maplen
,
1420 osdc
->osdmap
->epoch
);
1422 int skipped_map
= 0;
1424 dout("taking full map %u len %d\n", epoch
, maplen
);
1425 newmap
= osdmap_decode(&p
, p
+maplen
);
1426 if (IS_ERR(newmap
)) {
1427 err
= PTR_ERR(newmap
);
1431 oldmap
= osdc
->osdmap
;
1432 osdc
->osdmap
= newmap
;
1434 if (oldmap
->epoch
+ 1 < newmap
->epoch
)
1436 ceph_osdmap_destroy(oldmap
);
1438 kick_requests(osdc
, skipped_map
);
1445 downgrade_write(&osdc
->map_sem
);
1446 ceph_monc_got_osdmap(&osdc
->client
->monc
, osdc
->osdmap
->epoch
);
1449 * subscribe to subsequent osdmap updates if full to ensure
1450 * we find out when we are no longer full and stop returning
1453 if (ceph_osdmap_flag(osdc
->osdmap
, CEPH_OSDMAP_FULL
))
1454 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1457 up_read(&osdc
->map_sem
);
1458 wake_up_all(&osdc
->client
->auth_wq
);
1462 pr_err("osdc handle_map corrupt msg\n");
1464 up_write(&osdc
->map_sem
);
1469 * watch/notify callback event infrastructure
1471 * These callbacks are used both for watch and notify operations.
1473 static void __release_event(struct kref
*kref
)
1475 struct ceph_osd_event
*event
=
1476 container_of(kref
, struct ceph_osd_event
, kref
);
1478 dout("__release_event %p\n", event
);
1482 static void get_event(struct ceph_osd_event
*event
)
1484 kref_get(&event
->kref
);
1487 void ceph_osdc_put_event(struct ceph_osd_event
*event
)
1489 kref_put(&event
->kref
, __release_event
);
1491 EXPORT_SYMBOL(ceph_osdc_put_event
);
1493 static void __insert_event(struct ceph_osd_client
*osdc
,
1494 struct ceph_osd_event
*new)
1496 struct rb_node
**p
= &osdc
->event_tree
.rb_node
;
1497 struct rb_node
*parent
= NULL
;
1498 struct ceph_osd_event
*event
= NULL
;
1502 event
= rb_entry(parent
, struct ceph_osd_event
, node
);
1503 if (new->cookie
< event
->cookie
)
1505 else if (new->cookie
> event
->cookie
)
1506 p
= &(*p
)->rb_right
;
1511 rb_link_node(&new->node
, parent
, p
);
1512 rb_insert_color(&new->node
, &osdc
->event_tree
);
1515 static struct ceph_osd_event
*__find_event(struct ceph_osd_client
*osdc
,
1518 struct rb_node
**p
= &osdc
->event_tree
.rb_node
;
1519 struct rb_node
*parent
= NULL
;
1520 struct ceph_osd_event
*event
= NULL
;
1524 event
= rb_entry(parent
, struct ceph_osd_event
, node
);
1525 if (cookie
< event
->cookie
)
1527 else if (cookie
> event
->cookie
)
1528 p
= &(*p
)->rb_right
;
1535 static void __remove_event(struct ceph_osd_event
*event
)
1537 struct ceph_osd_client
*osdc
= event
->osdc
;
1539 if (!RB_EMPTY_NODE(&event
->node
)) {
1540 dout("__remove_event removed %p\n", event
);
1541 rb_erase(&event
->node
, &osdc
->event_tree
);
1542 ceph_osdc_put_event(event
);
1544 dout("__remove_event didn't remove %p\n", event
);
1548 int ceph_osdc_create_event(struct ceph_osd_client
*osdc
,
1549 void (*event_cb
)(u64
, u64
, u8
, void *),
1550 int one_shot
, void *data
,
1551 struct ceph_osd_event
**pevent
)
1553 struct ceph_osd_event
*event
;
1555 event
= kmalloc(sizeof(*event
), GFP_NOIO
);
1559 dout("create_event %p\n", event
);
1560 event
->cb
= event_cb
;
1561 event
->one_shot
= one_shot
;
1564 INIT_LIST_HEAD(&event
->osd_node
);
1565 kref_init(&event
->kref
); /* one ref for us */
1566 kref_get(&event
->kref
); /* one ref for the caller */
1567 init_completion(&event
->completion
);
1569 spin_lock(&osdc
->event_lock
);
1570 event
->cookie
= ++osdc
->event_count
;
1571 __insert_event(osdc
, event
);
1572 spin_unlock(&osdc
->event_lock
);
1577 EXPORT_SYMBOL(ceph_osdc_create_event
);
1579 void ceph_osdc_cancel_event(struct ceph_osd_event
*event
)
1581 struct ceph_osd_client
*osdc
= event
->osdc
;
1583 dout("cancel_event %p\n", event
);
1584 spin_lock(&osdc
->event_lock
);
1585 __remove_event(event
);
1586 spin_unlock(&osdc
->event_lock
);
1587 ceph_osdc_put_event(event
); /* caller's */
1589 EXPORT_SYMBOL(ceph_osdc_cancel_event
);
1592 static void do_event_work(struct work_struct
*work
)
1594 struct ceph_osd_event_work
*event_work
=
1595 container_of(work
, struct ceph_osd_event_work
, work
);
1596 struct ceph_osd_event
*event
= event_work
->event
;
1597 u64 ver
= event_work
->ver
;
1598 u64 notify_id
= event_work
->notify_id
;
1599 u8 opcode
= event_work
->opcode
;
1601 dout("do_event_work completing %p\n", event
);
1602 event
->cb(ver
, notify_id
, opcode
, event
->data
);
1603 complete(&event
->completion
);
1604 dout("do_event_work completed %p\n", event
);
1605 ceph_osdc_put_event(event
);
1611 * Process osd watch notifications
1613 void handle_watch_notify(struct ceph_osd_client
*osdc
, struct ceph_msg
*msg
)
1617 u64 cookie
, ver
, notify_id
;
1619 struct ceph_osd_event
*event
;
1620 struct ceph_osd_event_work
*event_work
;
1622 p
= msg
->front
.iov_base
;
1623 end
= p
+ msg
->front
.iov_len
;
1625 ceph_decode_8_safe(&p
, end
, proto_ver
, bad
);
1626 ceph_decode_8_safe(&p
, end
, opcode
, bad
);
1627 ceph_decode_64_safe(&p
, end
, cookie
, bad
);
1628 ceph_decode_64_safe(&p
, end
, ver
, bad
);
1629 ceph_decode_64_safe(&p
, end
, notify_id
, bad
);
1631 spin_lock(&osdc
->event_lock
);
1632 event
= __find_event(osdc
, cookie
);
1635 if (event
->one_shot
)
1636 __remove_event(event
);
1638 spin_unlock(&osdc
->event_lock
);
1639 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1640 cookie
, ver
, event
);
1642 event_work
= kmalloc(sizeof(*event_work
), GFP_NOIO
);
1644 dout("ERROR: could not allocate event_work\n");
1647 INIT_WORK(&event_work
->work
, do_event_work
);
1648 event_work
->event
= event
;
1649 event_work
->ver
= ver
;
1650 event_work
->notify_id
= notify_id
;
1651 event_work
->opcode
= opcode
;
1652 if (!queue_work(osdc
->notify_wq
, &event_work
->work
)) {
1653 dout("WARNING: failed to queue notify event work\n");
1661 complete(&event
->completion
);
1662 ceph_osdc_put_event(event
);
1666 pr_err("osdc handle_watch_notify corrupt msg\n");
1670 int ceph_osdc_wait_event(struct ceph_osd_event
*event
, unsigned long timeout
)
1674 dout("wait_event %p\n", event
);
1675 err
= wait_for_completion_interruptible_timeout(&event
->completion
,
1677 ceph_osdc_put_event(event
);
1680 dout("wait_event %p returns %d\n", event
, err
);
1683 EXPORT_SYMBOL(ceph_osdc_wait_event
);
1686 * Register request, send initial attempt.
1688 int ceph_osdc_start_request(struct ceph_osd_client
*osdc
,
1689 struct ceph_osd_request
*req
,
1694 req
->r_request
->pages
= req
->r_pages
;
1695 req
->r_request
->nr_pages
= req
->r_num_pages
;
1697 req
->r_request
->bio
= req
->r_bio
;
1699 req
->r_request
->trail
= req
->r_trail
;
1701 register_request(osdc
, req
);
1703 down_read(&osdc
->map_sem
);
1704 mutex_lock(&osdc
->request_mutex
);
1706 * a racing kick_requests() may have sent the message for us
1707 * while we dropped request_mutex above, so only send now if
1708 * the request still han't been touched yet.
1710 if (req
->r_sent
== 0) {
1711 rc
= __map_request(osdc
, req
, 0);
1714 dout("osdc_start_request failed map, "
1715 " will retry %lld\n", req
->r_tid
);
1720 if (req
->r_osd
== NULL
) {
1721 dout("send_request %p no up osds in pg\n", req
);
1722 ceph_monc_request_next_osdmap(&osdc
->client
->monc
);
1724 __send_request(osdc
, req
);
1730 mutex_unlock(&osdc
->request_mutex
);
1731 up_read(&osdc
->map_sem
);
1734 EXPORT_SYMBOL(ceph_osdc_start_request
);
1737 * wait for a request to complete
1739 int ceph_osdc_wait_request(struct ceph_osd_client
*osdc
,
1740 struct ceph_osd_request
*req
)
1744 rc
= wait_for_completion_interruptible(&req
->r_completion
);
1746 mutex_lock(&osdc
->request_mutex
);
1747 __cancel_request(req
);
1748 __unregister_request(osdc
, req
);
1749 mutex_unlock(&osdc
->request_mutex
);
1750 complete_request(req
);
1751 dout("wait_request tid %llu canceled/timed out\n", req
->r_tid
);
1755 dout("wait_request tid %llu result %d\n", req
->r_tid
, req
->r_result
);
1756 return req
->r_result
;
1758 EXPORT_SYMBOL(ceph_osdc_wait_request
);
1761 * sync - wait for all in-flight requests to flush. avoid starvation.
1763 void ceph_osdc_sync(struct ceph_osd_client
*osdc
)
1765 struct ceph_osd_request
*req
;
1766 u64 last_tid
, next_tid
= 0;
1768 mutex_lock(&osdc
->request_mutex
);
1769 last_tid
= osdc
->last_tid
;
1771 req
= __lookup_request_ge(osdc
, next_tid
);
1774 if (req
->r_tid
> last_tid
)
1777 next_tid
= req
->r_tid
+ 1;
1778 if ((req
->r_flags
& CEPH_OSD_FLAG_WRITE
) == 0)
1781 ceph_osdc_get_request(req
);
1782 mutex_unlock(&osdc
->request_mutex
);
1783 dout("sync waiting on tid %llu (last is %llu)\n",
1784 req
->r_tid
, last_tid
);
1785 wait_for_completion(&req
->r_safe_completion
);
1786 mutex_lock(&osdc
->request_mutex
);
1787 ceph_osdc_put_request(req
);
1789 mutex_unlock(&osdc
->request_mutex
);
1790 dout("sync done (thru tid %llu)\n", last_tid
);
1792 EXPORT_SYMBOL(ceph_osdc_sync
);
1797 int ceph_osdc_init(struct ceph_osd_client
*osdc
, struct ceph_client
*client
)
1802 osdc
->client
= client
;
1803 osdc
->osdmap
= NULL
;
1804 init_rwsem(&osdc
->map_sem
);
1805 init_completion(&osdc
->map_waiters
);
1806 osdc
->last_requested_map
= 0;
1807 mutex_init(&osdc
->request_mutex
);
1809 osdc
->osds
= RB_ROOT
;
1810 INIT_LIST_HEAD(&osdc
->osd_lru
);
1811 osdc
->requests
= RB_ROOT
;
1812 INIT_LIST_HEAD(&osdc
->req_lru
);
1813 INIT_LIST_HEAD(&osdc
->req_unsent
);
1814 INIT_LIST_HEAD(&osdc
->req_notarget
);
1815 INIT_LIST_HEAD(&osdc
->req_linger
);
1816 osdc
->num_requests
= 0;
1817 INIT_DELAYED_WORK(&osdc
->timeout_work
, handle_timeout
);
1818 INIT_DELAYED_WORK(&osdc
->osds_timeout_work
, handle_osds_timeout
);
1819 spin_lock_init(&osdc
->event_lock
);
1820 osdc
->event_tree
= RB_ROOT
;
1821 osdc
->event_count
= 0;
1823 schedule_delayed_work(&osdc
->osds_timeout_work
,
1824 round_jiffies_relative(osdc
->client
->options
->osd_idle_ttl
* HZ
));
1827 osdc
->req_mempool
= mempool_create_kmalloc_pool(10,
1828 sizeof(struct ceph_osd_request
));
1829 if (!osdc
->req_mempool
)
1832 err
= ceph_msgpool_init(&osdc
->msgpool_op
, CEPH_MSG_OSD_OP
,
1833 OSD_OP_FRONT_LEN
, 10, true,
1837 err
= ceph_msgpool_init(&osdc
->msgpool_op_reply
, CEPH_MSG_OSD_OPREPLY
,
1838 OSD_OPREPLY_FRONT_LEN
, 10, true,
1843 osdc
->notify_wq
= create_singlethread_workqueue("ceph-watch-notify");
1844 if (IS_ERR(osdc
->notify_wq
)) {
1845 err
= PTR_ERR(osdc
->notify_wq
);
1846 osdc
->notify_wq
= NULL
;
1852 ceph_msgpool_destroy(&osdc
->msgpool_op
);
1854 mempool_destroy(osdc
->req_mempool
);
1858 EXPORT_SYMBOL(ceph_osdc_init
);
1860 void ceph_osdc_stop(struct ceph_osd_client
*osdc
)
1862 flush_workqueue(osdc
->notify_wq
);
1863 destroy_workqueue(osdc
->notify_wq
);
1864 cancel_delayed_work_sync(&osdc
->timeout_work
);
1865 cancel_delayed_work_sync(&osdc
->osds_timeout_work
);
1867 ceph_osdmap_destroy(osdc
->osdmap
);
1868 osdc
->osdmap
= NULL
;
1870 remove_all_osds(osdc
);
1871 mempool_destroy(osdc
->req_mempool
);
1872 ceph_msgpool_destroy(&osdc
->msgpool_op
);
1873 ceph_msgpool_destroy(&osdc
->msgpool_op_reply
);
1875 EXPORT_SYMBOL(ceph_osdc_stop
);
1878 * Read some contiguous pages. If we cross a stripe boundary, shorten
1879 * *plen. Return number of bytes read, or error.
1881 int ceph_osdc_readpages(struct ceph_osd_client
*osdc
,
1882 struct ceph_vino vino
, struct ceph_file_layout
*layout
,
1884 u32 truncate_seq
, u64 truncate_size
,
1885 struct page
**pages
, int num_pages
, int page_align
)
1887 struct ceph_osd_request
*req
;
1890 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino
.ino
,
1891 vino
.snap
, off
, *plen
);
1892 req
= ceph_osdc_new_request(osdc
, layout
, vino
, off
, plen
,
1893 CEPH_OSD_OP_READ
, CEPH_OSD_FLAG_READ
,
1894 NULL
, 0, truncate_seq
, truncate_size
, NULL
,
1895 false, 1, page_align
);
1897 return PTR_ERR(req
);
1899 /* it may be a short read due to an object boundary */
1900 req
->r_pages
= pages
;
1902 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1903 off
, *plen
, req
->r_num_pages
, page_align
);
1905 rc
= ceph_osdc_start_request(osdc
, req
, false);
1907 rc
= ceph_osdc_wait_request(osdc
, req
);
1909 ceph_osdc_put_request(req
);
1910 dout("readpages result %d\n", rc
);
1913 EXPORT_SYMBOL(ceph_osdc_readpages
);
1916 * do a synchronous write on N pages
1918 int ceph_osdc_writepages(struct ceph_osd_client
*osdc
, struct ceph_vino vino
,
1919 struct ceph_file_layout
*layout
,
1920 struct ceph_snap_context
*snapc
,
1922 u32 truncate_seq
, u64 truncate_size
,
1923 struct timespec
*mtime
,
1924 struct page
**pages
, int num_pages
,
1925 int flags
, int do_sync
, bool nofail
)
1927 struct ceph_osd_request
*req
;
1929 int page_align
= off
& ~PAGE_MASK
;
1931 BUG_ON(vino
.snap
!= CEPH_NOSNAP
);
1932 req
= ceph_osdc_new_request(osdc
, layout
, vino
, off
, &len
,
1934 flags
| CEPH_OSD_FLAG_ONDISK
|
1935 CEPH_OSD_FLAG_WRITE
,
1937 truncate_seq
, truncate_size
, mtime
,
1938 nofail
, 1, page_align
);
1940 return PTR_ERR(req
);
1942 /* it may be a short write due to an object boundary */
1943 req
->r_pages
= pages
;
1944 dout("writepages %llu~%llu (%d pages)\n", off
, len
,
1947 rc
= ceph_osdc_start_request(osdc
, req
, nofail
);
1949 rc
= ceph_osdc_wait_request(osdc
, req
);
1951 ceph_osdc_put_request(req
);
1954 dout("writepages result %d\n", rc
);
1957 EXPORT_SYMBOL(ceph_osdc_writepages
);
1960 * handle incoming message
1962 static void dispatch(struct ceph_connection
*con
, struct ceph_msg
*msg
)
1964 struct ceph_osd
*osd
= con
->private;
1965 struct ceph_osd_client
*osdc
;
1966 int type
= le16_to_cpu(msg
->hdr
.type
);
1973 case CEPH_MSG_OSD_MAP
:
1974 ceph_osdc_handle_map(osdc
, msg
);
1976 case CEPH_MSG_OSD_OPREPLY
:
1977 handle_reply(osdc
, msg
, con
);
1979 case CEPH_MSG_WATCH_NOTIFY
:
1980 handle_watch_notify(osdc
, msg
);
1984 pr_err("received unknown message type %d %s\n", type
,
1985 ceph_msg_type_name(type
));
1992 * lookup and return message for incoming reply. set up reply message
1995 static struct ceph_msg
*get_reply(struct ceph_connection
*con
,
1996 struct ceph_msg_header
*hdr
,
1999 struct ceph_osd
*osd
= con
->private;
2000 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
2002 struct ceph_osd_request
*req
;
2003 int front
= le32_to_cpu(hdr
->front_len
);
2004 int data_len
= le32_to_cpu(hdr
->data_len
);
2007 tid
= le64_to_cpu(hdr
->tid
);
2008 mutex_lock(&osdc
->request_mutex
);
2009 req
= __lookup_request(osdc
, tid
);
2013 dout("get_reply unknown tid %llu from osd%d\n", tid
,
2018 if (req
->r_con_filling_msg
) {
2019 dout("%s revoking msg %p from old con %p\n", __func__
,
2020 req
->r_reply
, req
->r_con_filling_msg
);
2021 ceph_msg_revoke_incoming(req
->r_reply
);
2022 req
->r_con_filling_msg
->ops
->put(req
->r_con_filling_msg
);
2023 req
->r_con_filling_msg
= NULL
;
2026 if (front
> req
->r_reply
->front
.iov_len
) {
2027 pr_warning("get_reply front %d > preallocated %d\n",
2028 front
, (int)req
->r_reply
->front
.iov_len
);
2029 m
= ceph_msg_new(CEPH_MSG_OSD_OPREPLY
, front
, GFP_NOFS
, false);
2032 ceph_msg_put(req
->r_reply
);
2035 m
= ceph_msg_get(req
->r_reply
);
2038 int want
= calc_pages_for(req
->r_page_alignment
, data_len
);
2040 if (unlikely(req
->r_num_pages
< want
)) {
2041 pr_warning("tid %lld reply has %d bytes %d pages, we"
2042 " had only %d pages ready\n", tid
, data_len
,
2043 want
, req
->r_num_pages
);
2049 m
->pages
= req
->r_pages
;
2050 m
->nr_pages
= req
->r_num_pages
;
2051 m
->page_alignment
= req
->r_page_alignment
;
2053 m
->bio
= req
->r_bio
;
2057 req
->r_con_filling_msg
= con
->ops
->get(con
);
2058 dout("get_reply tid %lld %p\n", tid
, m
);
2061 mutex_unlock(&osdc
->request_mutex
);
2066 static struct ceph_msg
*alloc_msg(struct ceph_connection
*con
,
2067 struct ceph_msg_header
*hdr
,
2070 struct ceph_osd
*osd
= con
->private;
2071 int type
= le16_to_cpu(hdr
->type
);
2072 int front
= le32_to_cpu(hdr
->front_len
);
2076 case CEPH_MSG_OSD_MAP
:
2077 case CEPH_MSG_WATCH_NOTIFY
:
2078 return ceph_msg_new(type
, front
, GFP_NOFS
, false);
2079 case CEPH_MSG_OSD_OPREPLY
:
2080 return get_reply(con
, hdr
, skip
);
2082 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type
,
2090 * Wrappers to refcount containing ceph_osd struct
2092 static struct ceph_connection
*get_osd_con(struct ceph_connection
*con
)
2094 struct ceph_osd
*osd
= con
->private;
2100 static void put_osd_con(struct ceph_connection
*con
)
2102 struct ceph_osd
*osd
= con
->private;
2110 * Note: returned pointer is the address of a structure that's
2111 * managed separately. Caller must *not* attempt to free it.
2113 static struct ceph_auth_handshake
*get_authorizer(struct ceph_connection
*con
,
2114 int *proto
, int force_new
)
2116 struct ceph_osd
*o
= con
->private;
2117 struct ceph_osd_client
*osdc
= o
->o_osdc
;
2118 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
2119 struct ceph_auth_handshake
*auth
= &o
->o_auth
;
2121 if (force_new
&& auth
->authorizer
) {
2122 if (ac
->ops
&& ac
->ops
->destroy_authorizer
)
2123 ac
->ops
->destroy_authorizer(ac
, auth
->authorizer
);
2124 auth
->authorizer
= NULL
;
2126 if (!auth
->authorizer
&& ac
->ops
&& ac
->ops
->create_authorizer
) {
2127 int ret
= ac
->ops
->create_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
,
2130 return ERR_PTR(ret
);
2132 *proto
= ac
->protocol
;
2138 static int verify_authorizer_reply(struct ceph_connection
*con
, int len
)
2140 struct ceph_osd
*o
= con
->private;
2141 struct ceph_osd_client
*osdc
= o
->o_osdc
;
2142 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
2145 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2146 * XXX which do we do: succeed or fail?
2148 return ac
->ops
->verify_authorizer_reply(ac
, o
->o_auth
.authorizer
, len
);
2151 static int invalidate_authorizer(struct ceph_connection
*con
)
2153 struct ceph_osd
*o
= con
->private;
2154 struct ceph_osd_client
*osdc
= o
->o_osdc
;
2155 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
2157 if (ac
->ops
&& ac
->ops
->invalidate_authorizer
)
2158 ac
->ops
->invalidate_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
);
2160 return ceph_monc_validate_auth(&osdc
->client
->monc
);
2163 static const struct ceph_connection_operations osd_con_ops
= {
2166 .dispatch
= dispatch
,
2167 .get_authorizer
= get_authorizer
,
2168 .verify_authorizer_reply
= verify_authorizer_reply
,
2169 .invalidate_authorizer
= invalidate_authorizer
,
2170 .alloc_msg
= alloc_msg
,