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
;
29 * Implement client access to distributed object storage cluster.
31 * All data objects are stored within a cluster/cloud of OSDs, or
32 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
33 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
34 * remote daemons serving up and coordinating consistent and safe
37 * Cluster membership and the mapping of data objects onto storage devices
38 * are described by the osd map.
40 * We keep track of pending OSD requests (read, write), resubmit
41 * requests to different OSDs when the cluster topology/data layout
42 * change, or retry the affected requests when the communications
43 * channel with an OSD is reset.
46 static void link_request(struct ceph_osd
*osd
, struct ceph_osd_request
*req
);
47 static void unlink_request(struct ceph_osd
*osd
, struct ceph_osd_request
*req
);
48 static void link_linger(struct ceph_osd
*osd
,
49 struct ceph_osd_linger_request
*lreq
);
50 static void unlink_linger(struct ceph_osd
*osd
,
51 struct ceph_osd_linger_request
*lreq
);
54 static inline bool rwsem_is_wrlocked(struct rw_semaphore
*sem
)
58 if (unlikely(down_read_trylock(sem
))) {
65 static inline void verify_osdc_locked(struct ceph_osd_client
*osdc
)
67 WARN_ON(!rwsem_is_locked(&osdc
->lock
));
69 static inline void verify_osdc_wrlocked(struct ceph_osd_client
*osdc
)
71 WARN_ON(!rwsem_is_wrlocked(&osdc
->lock
));
73 static inline void verify_osd_locked(struct ceph_osd
*osd
)
75 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
77 WARN_ON(!(mutex_is_locked(&osd
->lock
) &&
78 rwsem_is_locked(&osdc
->lock
)) &&
79 !rwsem_is_wrlocked(&osdc
->lock
));
81 static inline void verify_lreq_locked(struct ceph_osd_linger_request
*lreq
)
83 WARN_ON(!mutex_is_locked(&lreq
->lock
));
86 static inline void verify_osdc_locked(struct ceph_osd_client
*osdc
) { }
87 static inline void verify_osdc_wrlocked(struct ceph_osd_client
*osdc
) { }
88 static inline void verify_osd_locked(struct ceph_osd
*osd
) { }
89 static inline void verify_lreq_locked(struct ceph_osd_linger_request
*lreq
) { }
93 * calculate the mapping of a file extent onto an object, and fill out the
94 * request accordingly. shorten extent as necessary if it crosses an
97 * fill osd op in request message.
99 static int calc_layout(struct ceph_file_layout
*layout
, u64 off
, u64
*plen
,
100 u64
*objnum
, u64
*objoff
, u64
*objlen
)
102 u64 orig_len
= *plen
;
106 r
= ceph_calc_file_object_mapping(layout
, off
, orig_len
, objnum
,
110 if (*objlen
< orig_len
) {
112 dout(" skipping last %llu, final file extent %llu~%llu\n",
113 orig_len
- *plen
, off
, *plen
);
116 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum
, *objoff
, *objlen
);
121 static void ceph_osd_data_init(struct ceph_osd_data
*osd_data
)
123 memset(osd_data
, 0, sizeof (*osd_data
));
124 osd_data
->type
= CEPH_OSD_DATA_TYPE_NONE
;
127 static void ceph_osd_data_pages_init(struct ceph_osd_data
*osd_data
,
128 struct page
**pages
, u64 length
, u32 alignment
,
129 bool pages_from_pool
, bool own_pages
)
131 osd_data
->type
= CEPH_OSD_DATA_TYPE_PAGES
;
132 osd_data
->pages
= pages
;
133 osd_data
->length
= length
;
134 osd_data
->alignment
= alignment
;
135 osd_data
->pages_from_pool
= pages_from_pool
;
136 osd_data
->own_pages
= own_pages
;
139 static void ceph_osd_data_pagelist_init(struct ceph_osd_data
*osd_data
,
140 struct ceph_pagelist
*pagelist
)
142 osd_data
->type
= CEPH_OSD_DATA_TYPE_PAGELIST
;
143 osd_data
->pagelist
= pagelist
;
147 static void ceph_osd_data_bio_init(struct ceph_osd_data
*osd_data
,
148 struct bio
*bio
, size_t bio_length
)
150 osd_data
->type
= CEPH_OSD_DATA_TYPE_BIO
;
152 osd_data
->bio_length
= bio_length
;
154 #endif /* CONFIG_BLOCK */
156 #define osd_req_op_data(oreq, whch, typ, fld) \
158 struct ceph_osd_request *__oreq = (oreq); \
159 unsigned int __whch = (whch); \
160 BUG_ON(__whch >= __oreq->r_num_ops); \
161 &__oreq->r_ops[__whch].typ.fld; \
164 static struct ceph_osd_data
*
165 osd_req_op_raw_data_in(struct ceph_osd_request
*osd_req
, unsigned int which
)
167 BUG_ON(which
>= osd_req
->r_num_ops
);
169 return &osd_req
->r_ops
[which
].raw_data_in
;
172 struct ceph_osd_data
*
173 osd_req_op_extent_osd_data(struct ceph_osd_request
*osd_req
,
176 return osd_req_op_data(osd_req
, which
, extent
, osd_data
);
178 EXPORT_SYMBOL(osd_req_op_extent_osd_data
);
180 void osd_req_op_raw_data_in_pages(struct ceph_osd_request
*osd_req
,
181 unsigned int which
, struct page
**pages
,
182 u64 length
, u32 alignment
,
183 bool pages_from_pool
, bool own_pages
)
185 struct ceph_osd_data
*osd_data
;
187 osd_data
= osd_req_op_raw_data_in(osd_req
, which
);
188 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
189 pages_from_pool
, own_pages
);
191 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages
);
193 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request
*osd_req
,
194 unsigned int which
, struct page
**pages
,
195 u64 length
, u32 alignment
,
196 bool pages_from_pool
, bool own_pages
)
198 struct ceph_osd_data
*osd_data
;
200 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
201 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
202 pages_from_pool
, own_pages
);
204 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages
);
206 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request
*osd_req
,
207 unsigned int which
, struct ceph_pagelist
*pagelist
)
209 struct ceph_osd_data
*osd_data
;
211 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
212 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
214 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist
);
217 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request
*osd_req
,
218 unsigned int which
, struct bio
*bio
, size_t bio_length
)
220 struct ceph_osd_data
*osd_data
;
222 osd_data
= osd_req_op_data(osd_req
, which
, extent
, osd_data
);
223 ceph_osd_data_bio_init(osd_data
, bio
, bio_length
);
225 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio
);
226 #endif /* CONFIG_BLOCK */
228 static void osd_req_op_cls_request_info_pagelist(
229 struct ceph_osd_request
*osd_req
,
230 unsigned int which
, struct ceph_pagelist
*pagelist
)
232 struct ceph_osd_data
*osd_data
;
234 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_info
);
235 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
238 void osd_req_op_cls_request_data_pagelist(
239 struct ceph_osd_request
*osd_req
,
240 unsigned int which
, struct ceph_pagelist
*pagelist
)
242 struct ceph_osd_data
*osd_data
;
244 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_data
);
245 ceph_osd_data_pagelist_init(osd_data
, pagelist
);
246 osd_req
->r_ops
[which
].cls
.indata_len
+= pagelist
->length
;
247 osd_req
->r_ops
[which
].indata_len
+= pagelist
->length
;
249 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist
);
251 void osd_req_op_cls_request_data_pages(struct ceph_osd_request
*osd_req
,
252 unsigned int which
, struct page
**pages
, u64 length
,
253 u32 alignment
, bool pages_from_pool
, bool own_pages
)
255 struct ceph_osd_data
*osd_data
;
257 osd_data
= osd_req_op_data(osd_req
, which
, cls
, request_data
);
258 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
259 pages_from_pool
, own_pages
);
260 osd_req
->r_ops
[which
].cls
.indata_len
+= length
;
261 osd_req
->r_ops
[which
].indata_len
+= length
;
263 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages
);
265 void osd_req_op_cls_response_data_pages(struct ceph_osd_request
*osd_req
,
266 unsigned int which
, struct page
**pages
, u64 length
,
267 u32 alignment
, bool pages_from_pool
, bool own_pages
)
269 struct ceph_osd_data
*osd_data
;
271 osd_data
= osd_req_op_data(osd_req
, which
, cls
, response_data
);
272 ceph_osd_data_pages_init(osd_data
, pages
, length
, alignment
,
273 pages_from_pool
, own_pages
);
275 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages
);
277 static u64
ceph_osd_data_length(struct ceph_osd_data
*osd_data
)
279 switch (osd_data
->type
) {
280 case CEPH_OSD_DATA_TYPE_NONE
:
282 case CEPH_OSD_DATA_TYPE_PAGES
:
283 return osd_data
->length
;
284 case CEPH_OSD_DATA_TYPE_PAGELIST
:
285 return (u64
)osd_data
->pagelist
->length
;
287 case CEPH_OSD_DATA_TYPE_BIO
:
288 return (u64
)osd_data
->bio_length
;
289 #endif /* CONFIG_BLOCK */
291 WARN(true, "unrecognized data type %d\n", (int)osd_data
->type
);
296 static void ceph_osd_data_release(struct ceph_osd_data
*osd_data
)
298 if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGES
&& osd_data
->own_pages
) {
301 num_pages
= calc_pages_for((u64
)osd_data
->alignment
,
302 (u64
)osd_data
->length
);
303 ceph_release_page_vector(osd_data
->pages
, num_pages
);
305 ceph_osd_data_init(osd_data
);
308 static void osd_req_op_data_release(struct ceph_osd_request
*osd_req
,
311 struct ceph_osd_req_op
*op
;
313 BUG_ON(which
>= osd_req
->r_num_ops
);
314 op
= &osd_req
->r_ops
[which
];
317 case CEPH_OSD_OP_READ
:
318 case CEPH_OSD_OP_WRITE
:
319 case CEPH_OSD_OP_WRITEFULL
:
320 ceph_osd_data_release(&op
->extent
.osd_data
);
322 case CEPH_OSD_OP_CALL
:
323 ceph_osd_data_release(&op
->cls
.request_info
);
324 ceph_osd_data_release(&op
->cls
.request_data
);
325 ceph_osd_data_release(&op
->cls
.response_data
);
327 case CEPH_OSD_OP_SETXATTR
:
328 case CEPH_OSD_OP_CMPXATTR
:
329 ceph_osd_data_release(&op
->xattr
.osd_data
);
331 case CEPH_OSD_OP_STAT
:
332 ceph_osd_data_release(&op
->raw_data_in
);
334 case CEPH_OSD_OP_NOTIFY_ACK
:
335 ceph_osd_data_release(&op
->notify_ack
.request_data
);
337 case CEPH_OSD_OP_NOTIFY
:
338 ceph_osd_data_release(&op
->notify
.request_data
);
339 ceph_osd_data_release(&op
->notify
.response_data
);
341 case CEPH_OSD_OP_LIST_WATCHERS
:
342 ceph_osd_data_release(&op
->list_watchers
.response_data
);
350 * Assumes @t is zero-initialized.
352 static void target_init(struct ceph_osd_request_target
*t
)
354 ceph_oid_init(&t
->base_oid
);
355 ceph_oloc_init(&t
->base_oloc
);
356 ceph_oid_init(&t
->target_oid
);
357 ceph_oloc_init(&t
->target_oloc
);
359 ceph_osds_init(&t
->acting
);
360 ceph_osds_init(&t
->up
);
364 t
->osd
= CEPH_HOMELESS_OSD
;
367 static void target_copy(struct ceph_osd_request_target
*dest
,
368 const struct ceph_osd_request_target
*src
)
370 ceph_oid_copy(&dest
->base_oid
, &src
->base_oid
);
371 ceph_oloc_copy(&dest
->base_oloc
, &src
->base_oloc
);
372 ceph_oid_copy(&dest
->target_oid
, &src
->target_oid
);
373 ceph_oloc_copy(&dest
->target_oloc
, &src
->target_oloc
);
375 dest
->pgid
= src
->pgid
; /* struct */
376 dest
->pg_num
= src
->pg_num
;
377 dest
->pg_num_mask
= src
->pg_num_mask
;
378 ceph_osds_copy(&dest
->acting
, &src
->acting
);
379 ceph_osds_copy(&dest
->up
, &src
->up
);
380 dest
->size
= src
->size
;
381 dest
->min_size
= src
->min_size
;
382 dest
->sort_bitwise
= src
->sort_bitwise
;
384 dest
->flags
= src
->flags
;
385 dest
->paused
= src
->paused
;
387 dest
->osd
= src
->osd
;
390 static void target_destroy(struct ceph_osd_request_target
*t
)
392 ceph_oid_destroy(&t
->base_oid
);
393 ceph_oloc_destroy(&t
->base_oloc
);
394 ceph_oid_destroy(&t
->target_oid
);
395 ceph_oloc_destroy(&t
->target_oloc
);
401 static void request_release_checks(struct ceph_osd_request
*req
)
403 WARN_ON(!RB_EMPTY_NODE(&req
->r_node
));
404 WARN_ON(!RB_EMPTY_NODE(&req
->r_mc_node
));
405 WARN_ON(!list_empty(&req
->r_unsafe_item
));
409 static void ceph_osdc_release_request(struct kref
*kref
)
411 struct ceph_osd_request
*req
= container_of(kref
,
412 struct ceph_osd_request
, r_kref
);
415 dout("%s %p (r_request %p r_reply %p)\n", __func__
, req
,
416 req
->r_request
, req
->r_reply
);
417 request_release_checks(req
);
420 ceph_msg_put(req
->r_request
);
422 ceph_msg_put(req
->r_reply
);
424 for (which
= 0; which
< req
->r_num_ops
; which
++)
425 osd_req_op_data_release(req
, which
);
427 target_destroy(&req
->r_t
);
428 ceph_put_snap_context(req
->r_snapc
);
431 mempool_free(req
, req
->r_osdc
->req_mempool
);
432 else if (req
->r_num_ops
<= CEPH_OSD_SLAB_OPS
)
433 kmem_cache_free(ceph_osd_request_cache
, req
);
438 void ceph_osdc_get_request(struct ceph_osd_request
*req
)
440 dout("%s %p (was %d)\n", __func__
, req
,
441 atomic_read(&req
->r_kref
.refcount
));
442 kref_get(&req
->r_kref
);
444 EXPORT_SYMBOL(ceph_osdc_get_request
);
446 void ceph_osdc_put_request(struct ceph_osd_request
*req
)
449 dout("%s %p (was %d)\n", __func__
, req
,
450 atomic_read(&req
->r_kref
.refcount
));
451 kref_put(&req
->r_kref
, ceph_osdc_release_request
);
454 EXPORT_SYMBOL(ceph_osdc_put_request
);
456 static void request_init(struct ceph_osd_request
*req
)
458 /* req only, each op is zeroed in _osd_req_op_init() */
459 memset(req
, 0, sizeof(*req
));
461 kref_init(&req
->r_kref
);
462 init_completion(&req
->r_completion
);
463 init_completion(&req
->r_safe_completion
);
464 RB_CLEAR_NODE(&req
->r_node
);
465 RB_CLEAR_NODE(&req
->r_mc_node
);
466 INIT_LIST_HEAD(&req
->r_unsafe_item
);
468 target_init(&req
->r_t
);
472 * This is ugly, but it allows us to reuse linger registration and ping
473 * requests, keeping the structure of the code around send_linger{_ping}()
474 * reasonable. Setting up a min_nr=2 mempool for each linger request
475 * and dealing with copying ops (this blasts req only, watch op remains
476 * intact) isn't any better.
478 static void request_reinit(struct ceph_osd_request
*req
)
480 struct ceph_osd_client
*osdc
= req
->r_osdc
;
481 bool mempool
= req
->r_mempool
;
482 unsigned int num_ops
= req
->r_num_ops
;
483 u64 snapid
= req
->r_snapid
;
484 struct ceph_snap_context
*snapc
= req
->r_snapc
;
485 bool linger
= req
->r_linger
;
486 struct ceph_msg
*request_msg
= req
->r_request
;
487 struct ceph_msg
*reply_msg
= req
->r_reply
;
489 dout("%s req %p\n", __func__
, req
);
490 WARN_ON(atomic_read(&req
->r_kref
.refcount
) != 1);
491 request_release_checks(req
);
493 WARN_ON(atomic_read(&request_msg
->kref
.refcount
) != 1);
494 WARN_ON(atomic_read(&reply_msg
->kref
.refcount
) != 1);
495 target_destroy(&req
->r_t
);
499 req
->r_mempool
= mempool
;
500 req
->r_num_ops
= num_ops
;
501 req
->r_snapid
= snapid
;
502 req
->r_snapc
= snapc
;
503 req
->r_linger
= linger
;
504 req
->r_request
= request_msg
;
505 req
->r_reply
= reply_msg
;
508 struct ceph_osd_request
*ceph_osdc_alloc_request(struct ceph_osd_client
*osdc
,
509 struct ceph_snap_context
*snapc
,
510 unsigned int num_ops
,
514 struct ceph_osd_request
*req
;
517 BUG_ON(num_ops
> CEPH_OSD_SLAB_OPS
);
518 req
= mempool_alloc(osdc
->req_mempool
, gfp_flags
);
519 } else if (num_ops
<= CEPH_OSD_SLAB_OPS
) {
520 req
= kmem_cache_alloc(ceph_osd_request_cache
, gfp_flags
);
522 BUG_ON(num_ops
> CEPH_OSD_MAX_OPS
);
523 req
= kmalloc(sizeof(*req
) + num_ops
* sizeof(req
->r_ops
[0]),
531 req
->r_mempool
= use_mempool
;
532 req
->r_num_ops
= num_ops
;
533 req
->r_snapid
= CEPH_NOSNAP
;
534 req
->r_snapc
= ceph_get_snap_context(snapc
);
536 dout("%s req %p\n", __func__
, req
);
539 EXPORT_SYMBOL(ceph_osdc_alloc_request
);
541 static int ceph_oloc_encoding_size(struct ceph_object_locator
*oloc
)
543 return 8 + 4 + 4 + 4 + (oloc
->pool_ns
? oloc
->pool_ns
->len
: 0);
546 int ceph_osdc_alloc_messages(struct ceph_osd_request
*req
, gfp_t gfp
)
548 struct ceph_osd_client
*osdc
= req
->r_osdc
;
549 struct ceph_msg
*msg
;
552 WARN_ON(ceph_oid_empty(&req
->r_base_oid
));
553 WARN_ON(ceph_oloc_empty(&req
->r_base_oloc
));
555 /* create request message */
556 msg_size
= 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */
557 msg_size
+= 4 + 4 + 4 + 8; /* mtime, reassert_version */
558 msg_size
+= CEPH_ENCODING_START_BLK_LEN
+
559 ceph_oloc_encoding_size(&req
->r_base_oloc
); /* oloc */
560 msg_size
+= 1 + 8 + 4 + 4; /* pgid */
561 msg_size
+= 4 + req
->r_base_oid
.name_len
; /* oid */
562 msg_size
+= 2 + req
->r_num_ops
* sizeof(struct ceph_osd_op
);
563 msg_size
+= 8; /* snapid */
564 msg_size
+= 8; /* snap_seq */
565 msg_size
+= 4 + 8 * (req
->r_snapc
? req
->r_snapc
->num_snaps
: 0);
566 msg_size
+= 4; /* retry_attempt */
569 msg
= ceph_msgpool_get(&osdc
->msgpool_op
, 0);
571 msg
= ceph_msg_new(CEPH_MSG_OSD_OP
, msg_size
, gfp
, true);
575 memset(msg
->front
.iov_base
, 0, msg
->front
.iov_len
);
576 req
->r_request
= msg
;
578 /* create reply message */
579 msg_size
= OSD_OPREPLY_FRONT_LEN
;
580 msg_size
+= req
->r_base_oid
.name_len
;
581 msg_size
+= req
->r_num_ops
* sizeof(struct ceph_osd_op
);
584 msg
= ceph_msgpool_get(&osdc
->msgpool_op_reply
, 0);
586 msg
= ceph_msg_new(CEPH_MSG_OSD_OPREPLY
, msg_size
, gfp
, true);
594 EXPORT_SYMBOL(ceph_osdc_alloc_messages
);
596 static bool osd_req_opcode_valid(u16 opcode
)
599 #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
600 __CEPH_FORALL_OSD_OPS(GENERATE_CASE
)
608 * This is an osd op init function for opcodes that have no data or
609 * other information associated with them. It also serves as a
610 * common init routine for all the other init functions, below.
612 static struct ceph_osd_req_op
*
613 _osd_req_op_init(struct ceph_osd_request
*osd_req
, unsigned int which
,
614 u16 opcode
, u32 flags
)
616 struct ceph_osd_req_op
*op
;
618 BUG_ON(which
>= osd_req
->r_num_ops
);
619 BUG_ON(!osd_req_opcode_valid(opcode
));
621 op
= &osd_req
->r_ops
[which
];
622 memset(op
, 0, sizeof (*op
));
629 void osd_req_op_init(struct ceph_osd_request
*osd_req
,
630 unsigned int which
, u16 opcode
, u32 flags
)
632 (void)_osd_req_op_init(osd_req
, which
, opcode
, flags
);
634 EXPORT_SYMBOL(osd_req_op_init
);
636 void osd_req_op_extent_init(struct ceph_osd_request
*osd_req
,
637 unsigned int which
, u16 opcode
,
638 u64 offset
, u64 length
,
639 u64 truncate_size
, u32 truncate_seq
)
641 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
643 size_t payload_len
= 0;
645 BUG_ON(opcode
!= CEPH_OSD_OP_READ
&& opcode
!= CEPH_OSD_OP_WRITE
&&
646 opcode
!= CEPH_OSD_OP_WRITEFULL
&& opcode
!= CEPH_OSD_OP_ZERO
&&
647 opcode
!= CEPH_OSD_OP_TRUNCATE
);
649 op
->extent
.offset
= offset
;
650 op
->extent
.length
= length
;
651 op
->extent
.truncate_size
= truncate_size
;
652 op
->extent
.truncate_seq
= truncate_seq
;
653 if (opcode
== CEPH_OSD_OP_WRITE
|| opcode
== CEPH_OSD_OP_WRITEFULL
)
654 payload_len
+= length
;
656 op
->indata_len
= payload_len
;
658 EXPORT_SYMBOL(osd_req_op_extent_init
);
660 void osd_req_op_extent_update(struct ceph_osd_request
*osd_req
,
661 unsigned int which
, u64 length
)
663 struct ceph_osd_req_op
*op
;
666 BUG_ON(which
>= osd_req
->r_num_ops
);
667 op
= &osd_req
->r_ops
[which
];
668 previous
= op
->extent
.length
;
670 if (length
== previous
)
671 return; /* Nothing to do */
672 BUG_ON(length
> previous
);
674 op
->extent
.length
= length
;
675 op
->indata_len
-= previous
- length
;
677 EXPORT_SYMBOL(osd_req_op_extent_update
);
679 void osd_req_op_extent_dup_last(struct ceph_osd_request
*osd_req
,
680 unsigned int which
, u64 offset_inc
)
682 struct ceph_osd_req_op
*op
, *prev_op
;
684 BUG_ON(which
+ 1 >= osd_req
->r_num_ops
);
686 prev_op
= &osd_req
->r_ops
[which
];
687 op
= _osd_req_op_init(osd_req
, which
+ 1, prev_op
->op
, prev_op
->flags
);
688 /* dup previous one */
689 op
->indata_len
= prev_op
->indata_len
;
690 op
->outdata_len
= prev_op
->outdata_len
;
691 op
->extent
= prev_op
->extent
;
693 op
->extent
.offset
+= offset_inc
;
694 op
->extent
.length
-= offset_inc
;
696 if (op
->op
== CEPH_OSD_OP_WRITE
|| op
->op
== CEPH_OSD_OP_WRITEFULL
)
697 op
->indata_len
-= offset_inc
;
699 EXPORT_SYMBOL(osd_req_op_extent_dup_last
);
701 void osd_req_op_cls_init(struct ceph_osd_request
*osd_req
, unsigned int which
,
702 u16 opcode
, const char *class, const char *method
)
704 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
706 struct ceph_pagelist
*pagelist
;
707 size_t payload_len
= 0;
710 BUG_ON(opcode
!= CEPH_OSD_OP_CALL
);
712 pagelist
= kmalloc(sizeof (*pagelist
), GFP_NOFS
);
714 ceph_pagelist_init(pagelist
);
716 op
->cls
.class_name
= class;
717 size
= strlen(class);
718 BUG_ON(size
> (size_t) U8_MAX
);
719 op
->cls
.class_len
= size
;
720 ceph_pagelist_append(pagelist
, class, size
);
723 op
->cls
.method_name
= method
;
724 size
= strlen(method
);
725 BUG_ON(size
> (size_t) U8_MAX
);
726 op
->cls
.method_len
= size
;
727 ceph_pagelist_append(pagelist
, method
, size
);
730 osd_req_op_cls_request_info_pagelist(osd_req
, which
, pagelist
);
732 op
->indata_len
= payload_len
;
734 EXPORT_SYMBOL(osd_req_op_cls_init
);
736 int osd_req_op_xattr_init(struct ceph_osd_request
*osd_req
, unsigned int which
,
737 u16 opcode
, const char *name
, const void *value
,
738 size_t size
, u8 cmp_op
, u8 cmp_mode
)
740 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
742 struct ceph_pagelist
*pagelist
;
745 BUG_ON(opcode
!= CEPH_OSD_OP_SETXATTR
&& opcode
!= CEPH_OSD_OP_CMPXATTR
);
747 pagelist
= kmalloc(sizeof(*pagelist
), GFP_NOFS
);
751 ceph_pagelist_init(pagelist
);
753 payload_len
= strlen(name
);
754 op
->xattr
.name_len
= payload_len
;
755 ceph_pagelist_append(pagelist
, name
, payload_len
);
757 op
->xattr
.value_len
= size
;
758 ceph_pagelist_append(pagelist
, value
, size
);
761 op
->xattr
.cmp_op
= cmp_op
;
762 op
->xattr
.cmp_mode
= cmp_mode
;
764 ceph_osd_data_pagelist_init(&op
->xattr
.osd_data
, pagelist
);
765 op
->indata_len
= payload_len
;
768 EXPORT_SYMBOL(osd_req_op_xattr_init
);
771 * @watch_opcode: CEPH_OSD_WATCH_OP_*
773 static void osd_req_op_watch_init(struct ceph_osd_request
*req
, int which
,
774 u64 cookie
, u8 watch_opcode
)
776 struct ceph_osd_req_op
*op
;
778 op
= _osd_req_op_init(req
, which
, CEPH_OSD_OP_WATCH
, 0);
779 op
->watch
.cookie
= cookie
;
780 op
->watch
.op
= watch_opcode
;
784 void osd_req_op_alloc_hint_init(struct ceph_osd_request
*osd_req
,
786 u64 expected_object_size
,
787 u64 expected_write_size
)
789 struct ceph_osd_req_op
*op
= _osd_req_op_init(osd_req
, which
,
790 CEPH_OSD_OP_SETALLOCHINT
,
793 op
->alloc_hint
.expected_object_size
= expected_object_size
;
794 op
->alloc_hint
.expected_write_size
= expected_write_size
;
797 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
798 * not worth a feature bit. Set FAILOK per-op flag to make
799 * sure older osds don't trip over an unsupported opcode.
801 op
->flags
|= CEPH_OSD_OP_FLAG_FAILOK
;
803 EXPORT_SYMBOL(osd_req_op_alloc_hint_init
);
805 static void ceph_osdc_msg_data_add(struct ceph_msg
*msg
,
806 struct ceph_osd_data
*osd_data
)
808 u64 length
= ceph_osd_data_length(osd_data
);
810 if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGES
) {
811 BUG_ON(length
> (u64
) SIZE_MAX
);
813 ceph_msg_data_add_pages(msg
, osd_data
->pages
,
814 length
, osd_data
->alignment
);
815 } else if (osd_data
->type
== CEPH_OSD_DATA_TYPE_PAGELIST
) {
817 ceph_msg_data_add_pagelist(msg
, osd_data
->pagelist
);
819 } else if (osd_data
->type
== CEPH_OSD_DATA_TYPE_BIO
) {
820 ceph_msg_data_add_bio(msg
, osd_data
->bio
, length
);
823 BUG_ON(osd_data
->type
!= CEPH_OSD_DATA_TYPE_NONE
);
827 static u32
osd_req_encode_op(struct ceph_osd_op
*dst
,
828 const struct ceph_osd_req_op
*src
)
830 if (WARN_ON(!osd_req_opcode_valid(src
->op
))) {
831 pr_err("unrecognized osd opcode %d\n", src
->op
);
837 case CEPH_OSD_OP_STAT
:
839 case CEPH_OSD_OP_READ
:
840 case CEPH_OSD_OP_WRITE
:
841 case CEPH_OSD_OP_WRITEFULL
:
842 case CEPH_OSD_OP_ZERO
:
843 case CEPH_OSD_OP_TRUNCATE
:
844 dst
->extent
.offset
= cpu_to_le64(src
->extent
.offset
);
845 dst
->extent
.length
= cpu_to_le64(src
->extent
.length
);
846 dst
->extent
.truncate_size
=
847 cpu_to_le64(src
->extent
.truncate_size
);
848 dst
->extent
.truncate_seq
=
849 cpu_to_le32(src
->extent
.truncate_seq
);
851 case CEPH_OSD_OP_CALL
:
852 dst
->cls
.class_len
= src
->cls
.class_len
;
853 dst
->cls
.method_len
= src
->cls
.method_len
;
854 dst
->cls
.indata_len
= cpu_to_le32(src
->cls
.indata_len
);
856 case CEPH_OSD_OP_STARTSYNC
:
858 case CEPH_OSD_OP_WATCH
:
859 dst
->watch
.cookie
= cpu_to_le64(src
->watch
.cookie
);
860 dst
->watch
.ver
= cpu_to_le64(0);
861 dst
->watch
.op
= src
->watch
.op
;
862 dst
->watch
.gen
= cpu_to_le32(src
->watch
.gen
);
864 case CEPH_OSD_OP_NOTIFY_ACK
:
866 case CEPH_OSD_OP_NOTIFY
:
867 dst
->notify
.cookie
= cpu_to_le64(src
->notify
.cookie
);
869 case CEPH_OSD_OP_LIST_WATCHERS
:
871 case CEPH_OSD_OP_SETALLOCHINT
:
872 dst
->alloc_hint
.expected_object_size
=
873 cpu_to_le64(src
->alloc_hint
.expected_object_size
);
874 dst
->alloc_hint
.expected_write_size
=
875 cpu_to_le64(src
->alloc_hint
.expected_write_size
);
877 case CEPH_OSD_OP_SETXATTR
:
878 case CEPH_OSD_OP_CMPXATTR
:
879 dst
->xattr
.name_len
= cpu_to_le32(src
->xattr
.name_len
);
880 dst
->xattr
.value_len
= cpu_to_le32(src
->xattr
.value_len
);
881 dst
->xattr
.cmp_op
= src
->xattr
.cmp_op
;
882 dst
->xattr
.cmp_mode
= src
->xattr
.cmp_mode
;
884 case CEPH_OSD_OP_CREATE
:
885 case CEPH_OSD_OP_DELETE
:
888 pr_err("unsupported osd opcode %s\n",
889 ceph_osd_op_name(src
->op
));
895 dst
->op
= cpu_to_le16(src
->op
);
896 dst
->flags
= cpu_to_le32(src
->flags
);
897 dst
->payload_len
= cpu_to_le32(src
->indata_len
);
899 return src
->indata_len
;
903 * build new request AND message, calculate layout, and adjust file
906 * if the file was recently truncated, we include information about its
907 * old and new size so that the object can be updated appropriately. (we
908 * avoid synchronously deleting truncated objects because it's slow.)
910 * if @do_sync, include a 'startsync' command so that the osd will flush
913 struct ceph_osd_request
*ceph_osdc_new_request(struct ceph_osd_client
*osdc
,
914 struct ceph_file_layout
*layout
,
915 struct ceph_vino vino
,
917 unsigned int which
, int num_ops
,
918 int opcode
, int flags
,
919 struct ceph_snap_context
*snapc
,
924 struct ceph_osd_request
*req
;
930 BUG_ON(opcode
!= CEPH_OSD_OP_READ
&& opcode
!= CEPH_OSD_OP_WRITE
&&
931 opcode
!= CEPH_OSD_OP_ZERO
&& opcode
!= CEPH_OSD_OP_TRUNCATE
&&
932 opcode
!= CEPH_OSD_OP_CREATE
&& opcode
!= CEPH_OSD_OP_DELETE
);
934 req
= ceph_osdc_alloc_request(osdc
, snapc
, num_ops
, use_mempool
,
941 /* calculate max write size */
942 r
= calc_layout(layout
, off
, plen
, &objnum
, &objoff
, &objlen
);
946 if (opcode
== CEPH_OSD_OP_CREATE
|| opcode
== CEPH_OSD_OP_DELETE
) {
947 osd_req_op_init(req
, which
, opcode
, 0);
949 u32 object_size
= layout
->object_size
;
950 u32 object_base
= off
- objoff
;
951 if (!(truncate_seq
== 1 && truncate_size
== -1ULL)) {
952 if (truncate_size
<= object_base
) {
955 truncate_size
-= object_base
;
956 if (truncate_size
> object_size
)
957 truncate_size
= object_size
;
960 osd_req_op_extent_init(req
, which
, opcode
, objoff
, objlen
,
961 truncate_size
, truncate_seq
);
964 req
->r_flags
= flags
;
965 req
->r_base_oloc
.pool
= layout
->pool_id
;
966 req
->r_base_oloc
.pool_ns
= ceph_try_get_string(layout
->pool_ns
);
967 ceph_oid_printf(&req
->r_base_oid
, "%llx.%08llx", vino
.ino
, objnum
);
969 req
->r_snapid
= vino
.snap
;
970 if (flags
& CEPH_OSD_FLAG_WRITE
)
971 req
->r_data_offset
= off
;
973 r
= ceph_osdc_alloc_messages(req
, GFP_NOFS
);
980 ceph_osdc_put_request(req
);
983 EXPORT_SYMBOL(ceph_osdc_new_request
);
986 * We keep osd requests in an rbtree, sorted by ->r_tid.
988 DEFINE_RB_FUNCS(request
, struct ceph_osd_request
, r_tid
, r_node
)
989 DEFINE_RB_FUNCS(request_mc
, struct ceph_osd_request
, r_tid
, r_mc_node
)
991 static bool osd_homeless(struct ceph_osd
*osd
)
993 return osd
->o_osd
== CEPH_HOMELESS_OSD
;
996 static bool osd_registered(struct ceph_osd
*osd
)
998 verify_osdc_locked(osd
->o_osdc
);
1000 return !RB_EMPTY_NODE(&osd
->o_node
);
1004 * Assumes @osd is zero-initialized.
1006 static void osd_init(struct ceph_osd
*osd
)
1008 atomic_set(&osd
->o_ref
, 1);
1009 RB_CLEAR_NODE(&osd
->o_node
);
1010 osd
->o_requests
= RB_ROOT
;
1011 osd
->o_linger_requests
= RB_ROOT
;
1012 INIT_LIST_HEAD(&osd
->o_osd_lru
);
1013 INIT_LIST_HEAD(&osd
->o_keepalive_item
);
1014 osd
->o_incarnation
= 1;
1015 mutex_init(&osd
->lock
);
1018 static void osd_cleanup(struct ceph_osd
*osd
)
1020 WARN_ON(!RB_EMPTY_NODE(&osd
->o_node
));
1021 WARN_ON(!RB_EMPTY_ROOT(&osd
->o_requests
));
1022 WARN_ON(!RB_EMPTY_ROOT(&osd
->o_linger_requests
));
1023 WARN_ON(!list_empty(&osd
->o_osd_lru
));
1024 WARN_ON(!list_empty(&osd
->o_keepalive_item
));
1026 if (osd
->o_auth
.authorizer
) {
1027 WARN_ON(osd_homeless(osd
));
1028 ceph_auth_destroy_authorizer(osd
->o_auth
.authorizer
);
1033 * Track open sessions with osds.
1035 static struct ceph_osd
*create_osd(struct ceph_osd_client
*osdc
, int onum
)
1037 struct ceph_osd
*osd
;
1039 WARN_ON(onum
== CEPH_HOMELESS_OSD
);
1041 osd
= kzalloc(sizeof(*osd
), GFP_NOIO
| __GFP_NOFAIL
);
1046 ceph_con_init(&osd
->o_con
, osd
, &osd_con_ops
, &osdc
->client
->msgr
);
1051 static struct ceph_osd
*get_osd(struct ceph_osd
*osd
)
1053 if (atomic_inc_not_zero(&osd
->o_ref
)) {
1054 dout("get_osd %p %d -> %d\n", osd
, atomic_read(&osd
->o_ref
)-1,
1055 atomic_read(&osd
->o_ref
));
1058 dout("get_osd %p FAIL\n", osd
);
1063 static void put_osd(struct ceph_osd
*osd
)
1065 dout("put_osd %p %d -> %d\n", osd
, atomic_read(&osd
->o_ref
),
1066 atomic_read(&osd
->o_ref
) - 1);
1067 if (atomic_dec_and_test(&osd
->o_ref
)) {
1073 DEFINE_RB_FUNCS(osd
, struct ceph_osd
, o_osd
, o_node
)
1075 static void __move_osd_to_lru(struct ceph_osd
*osd
)
1077 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
1079 dout("%s osd %p osd%d\n", __func__
, osd
, osd
->o_osd
);
1080 BUG_ON(!list_empty(&osd
->o_osd_lru
));
1082 spin_lock(&osdc
->osd_lru_lock
);
1083 list_add_tail(&osd
->o_osd_lru
, &osdc
->osd_lru
);
1084 spin_unlock(&osdc
->osd_lru_lock
);
1086 osd
->lru_ttl
= jiffies
+ osdc
->client
->options
->osd_idle_ttl
;
1089 static void maybe_move_osd_to_lru(struct ceph_osd
*osd
)
1091 if (RB_EMPTY_ROOT(&osd
->o_requests
) &&
1092 RB_EMPTY_ROOT(&osd
->o_linger_requests
))
1093 __move_osd_to_lru(osd
);
1096 static void __remove_osd_from_lru(struct ceph_osd
*osd
)
1098 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
1100 dout("%s osd %p osd%d\n", __func__
, osd
, osd
->o_osd
);
1102 spin_lock(&osdc
->osd_lru_lock
);
1103 if (!list_empty(&osd
->o_osd_lru
))
1104 list_del_init(&osd
->o_osd_lru
);
1105 spin_unlock(&osdc
->osd_lru_lock
);
1109 * Close the connection and assign any leftover requests to the
1112 static void close_osd(struct ceph_osd
*osd
)
1114 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
1117 verify_osdc_wrlocked(osdc
);
1118 dout("%s osd %p osd%d\n", __func__
, osd
, osd
->o_osd
);
1120 ceph_con_close(&osd
->o_con
);
1122 for (n
= rb_first(&osd
->o_requests
); n
; ) {
1123 struct ceph_osd_request
*req
=
1124 rb_entry(n
, struct ceph_osd_request
, r_node
);
1126 n
= rb_next(n
); /* unlink_request() */
1128 dout(" reassigning req %p tid %llu\n", req
, req
->r_tid
);
1129 unlink_request(osd
, req
);
1130 link_request(&osdc
->homeless_osd
, req
);
1132 for (n
= rb_first(&osd
->o_linger_requests
); n
; ) {
1133 struct ceph_osd_linger_request
*lreq
=
1134 rb_entry(n
, struct ceph_osd_linger_request
, node
);
1136 n
= rb_next(n
); /* unlink_linger() */
1138 dout(" reassigning lreq %p linger_id %llu\n", lreq
,
1140 unlink_linger(osd
, lreq
);
1141 link_linger(&osdc
->homeless_osd
, lreq
);
1144 __remove_osd_from_lru(osd
);
1145 erase_osd(&osdc
->osds
, osd
);
1152 static int reopen_osd(struct ceph_osd
*osd
)
1154 struct ceph_entity_addr
*peer_addr
;
1156 dout("%s osd %p osd%d\n", __func__
, osd
, osd
->o_osd
);
1158 if (RB_EMPTY_ROOT(&osd
->o_requests
) &&
1159 RB_EMPTY_ROOT(&osd
->o_linger_requests
)) {
1164 peer_addr
= &osd
->o_osdc
->osdmap
->osd_addr
[osd
->o_osd
];
1165 if (!memcmp(peer_addr
, &osd
->o_con
.peer_addr
, sizeof (*peer_addr
)) &&
1166 !ceph_con_opened(&osd
->o_con
)) {
1169 dout("osd addr hasn't changed and connection never opened, "
1170 "letting msgr retry\n");
1171 /* touch each r_stamp for handle_timeout()'s benfit */
1172 for (n
= rb_first(&osd
->o_requests
); n
; n
= rb_next(n
)) {
1173 struct ceph_osd_request
*req
=
1174 rb_entry(n
, struct ceph_osd_request
, r_node
);
1175 req
->r_stamp
= jiffies
;
1181 ceph_con_close(&osd
->o_con
);
1182 ceph_con_open(&osd
->o_con
, CEPH_ENTITY_TYPE_OSD
, osd
->o_osd
, peer_addr
);
1183 osd
->o_incarnation
++;
1188 static struct ceph_osd
*lookup_create_osd(struct ceph_osd_client
*osdc
, int o
,
1191 struct ceph_osd
*osd
;
1194 verify_osdc_wrlocked(osdc
);
1196 verify_osdc_locked(osdc
);
1198 if (o
!= CEPH_HOMELESS_OSD
)
1199 osd
= lookup_osd(&osdc
->osds
, o
);
1201 osd
= &osdc
->homeless_osd
;
1204 return ERR_PTR(-EAGAIN
);
1206 osd
= create_osd(osdc
, o
);
1207 insert_osd(&osdc
->osds
, osd
);
1208 ceph_con_open(&osd
->o_con
, CEPH_ENTITY_TYPE_OSD
, osd
->o_osd
,
1209 &osdc
->osdmap
->osd_addr
[osd
->o_osd
]);
1212 dout("%s osdc %p osd%d -> osd %p\n", __func__
, osdc
, o
, osd
);
1217 * Create request <-> OSD session relation.
1219 * @req has to be assigned a tid, @osd may be homeless.
1221 static void link_request(struct ceph_osd
*osd
, struct ceph_osd_request
*req
)
1223 verify_osd_locked(osd
);
1224 WARN_ON(!req
->r_tid
|| req
->r_osd
);
1225 dout("%s osd %p osd%d req %p tid %llu\n", __func__
, osd
, osd
->o_osd
,
1228 if (!osd_homeless(osd
))
1229 __remove_osd_from_lru(osd
);
1231 atomic_inc(&osd
->o_osdc
->num_homeless
);
1234 insert_request(&osd
->o_requests
, req
);
1238 static void unlink_request(struct ceph_osd
*osd
, struct ceph_osd_request
*req
)
1240 verify_osd_locked(osd
);
1241 WARN_ON(req
->r_osd
!= osd
);
1242 dout("%s osd %p osd%d req %p tid %llu\n", __func__
, osd
, osd
->o_osd
,
1246 erase_request(&osd
->o_requests
, req
);
1249 if (!osd_homeless(osd
))
1250 maybe_move_osd_to_lru(osd
);
1252 atomic_dec(&osd
->o_osdc
->num_homeless
);
1255 static bool __pool_full(struct ceph_pg_pool_info
*pi
)
1257 return pi
->flags
& CEPH_POOL_FLAG_FULL
;
1260 static bool have_pool_full(struct ceph_osd_client
*osdc
)
1264 for (n
= rb_first(&osdc
->osdmap
->pg_pools
); n
; n
= rb_next(n
)) {
1265 struct ceph_pg_pool_info
*pi
=
1266 rb_entry(n
, struct ceph_pg_pool_info
, node
);
1268 if (__pool_full(pi
))
1275 static bool pool_full(struct ceph_osd_client
*osdc
, s64 pool_id
)
1277 struct ceph_pg_pool_info
*pi
;
1279 pi
= ceph_pg_pool_by_id(osdc
->osdmap
, pool_id
);
1283 return __pool_full(pi
);
1287 * Returns whether a request should be blocked from being sent
1288 * based on the current osdmap and osd_client settings.
1290 static bool target_should_be_paused(struct ceph_osd_client
*osdc
,
1291 const struct ceph_osd_request_target
*t
,
1292 struct ceph_pg_pool_info
*pi
)
1294 bool pauserd
= ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSERD
);
1295 bool pausewr
= ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSEWR
) ||
1296 ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
) ||
1299 WARN_ON(pi
->id
!= t
->base_oloc
.pool
);
1300 return (t
->flags
& CEPH_OSD_FLAG_READ
&& pauserd
) ||
1301 (t
->flags
& CEPH_OSD_FLAG_WRITE
&& pausewr
);
1304 enum calc_target_result
{
1305 CALC_TARGET_NO_ACTION
= 0,
1306 CALC_TARGET_NEED_RESEND
,
1307 CALC_TARGET_POOL_DNE
,
1310 static enum calc_target_result
calc_target(struct ceph_osd_client
*osdc
,
1311 struct ceph_osd_request_target
*t
,
1312 u32
*last_force_resend
,
1315 struct ceph_pg_pool_info
*pi
;
1316 struct ceph_pg pgid
, last_pgid
;
1317 struct ceph_osds up
, acting
;
1318 bool force_resend
= false;
1319 bool need_check_tiering
= false;
1320 bool need_resend
= false;
1321 bool sort_bitwise
= ceph_osdmap_flag(osdc
, CEPH_OSDMAP_SORTBITWISE
);
1322 enum calc_target_result ct_res
;
1325 pi
= ceph_pg_pool_by_id(osdc
->osdmap
, t
->base_oloc
.pool
);
1327 t
->osd
= CEPH_HOMELESS_OSD
;
1328 ct_res
= CALC_TARGET_POOL_DNE
;
1332 if (osdc
->osdmap
->epoch
== pi
->last_force_request_resend
) {
1333 if (last_force_resend
&&
1334 *last_force_resend
< pi
->last_force_request_resend
) {
1335 *last_force_resend
= pi
->last_force_request_resend
;
1336 force_resend
= true;
1337 } else if (!last_force_resend
) {
1338 force_resend
= true;
1341 if (ceph_oid_empty(&t
->target_oid
) || force_resend
) {
1342 ceph_oid_copy(&t
->target_oid
, &t
->base_oid
);
1343 need_check_tiering
= true;
1345 if (ceph_oloc_empty(&t
->target_oloc
) || force_resend
) {
1346 ceph_oloc_copy(&t
->target_oloc
, &t
->base_oloc
);
1347 need_check_tiering
= true;
1350 if (need_check_tiering
&&
1351 (t
->flags
& CEPH_OSD_FLAG_IGNORE_OVERLAY
) == 0) {
1352 if (t
->flags
& CEPH_OSD_FLAG_READ
&& pi
->read_tier
>= 0)
1353 t
->target_oloc
.pool
= pi
->read_tier
;
1354 if (t
->flags
& CEPH_OSD_FLAG_WRITE
&& pi
->write_tier
>= 0)
1355 t
->target_oloc
.pool
= pi
->write_tier
;
1358 ret
= ceph_object_locator_to_pg(osdc
->osdmap
, &t
->target_oid
,
1359 &t
->target_oloc
, &pgid
);
1361 WARN_ON(ret
!= -ENOENT
);
1362 t
->osd
= CEPH_HOMELESS_OSD
;
1363 ct_res
= CALC_TARGET_POOL_DNE
;
1366 last_pgid
.pool
= pgid
.pool
;
1367 last_pgid
.seed
= ceph_stable_mod(pgid
.seed
, t
->pg_num
, t
->pg_num_mask
);
1369 ceph_pg_to_up_acting_osds(osdc
->osdmap
, &pgid
, &up
, &acting
);
1371 ceph_is_new_interval(&t
->acting
,
1384 force_resend
= true;
1386 if (t
->paused
&& !target_should_be_paused(osdc
, t
, pi
)) {
1391 if (ceph_pg_compare(&t
->pgid
, &pgid
) ||
1392 ceph_osds_changed(&t
->acting
, &acting
, any_change
) ||
1394 t
->pgid
= pgid
; /* struct */
1395 ceph_osds_copy(&t
->acting
, &acting
);
1396 ceph_osds_copy(&t
->up
, &up
);
1398 t
->min_size
= pi
->min_size
;
1399 t
->pg_num
= pi
->pg_num
;
1400 t
->pg_num_mask
= pi
->pg_num_mask
;
1401 t
->sort_bitwise
= sort_bitwise
;
1403 t
->osd
= acting
.primary
;
1407 ct_res
= need_resend
? CALC_TARGET_NEED_RESEND
: CALC_TARGET_NO_ACTION
;
1409 dout("%s t %p -> ct_res %d osd %d\n", __func__
, t
, ct_res
, t
->osd
);
1413 static void setup_request_data(struct ceph_osd_request
*req
,
1414 struct ceph_msg
*msg
)
1419 if (!list_empty(&msg
->data
))
1422 WARN_ON(msg
->data_length
);
1423 for (i
= 0; i
< req
->r_num_ops
; i
++) {
1424 struct ceph_osd_req_op
*op
= &req
->r_ops
[i
];
1428 case CEPH_OSD_OP_WRITE
:
1429 case CEPH_OSD_OP_WRITEFULL
:
1430 WARN_ON(op
->indata_len
!= op
->extent
.length
);
1431 ceph_osdc_msg_data_add(msg
, &op
->extent
.osd_data
);
1433 case CEPH_OSD_OP_SETXATTR
:
1434 case CEPH_OSD_OP_CMPXATTR
:
1435 WARN_ON(op
->indata_len
!= op
->xattr
.name_len
+
1436 op
->xattr
.value_len
);
1437 ceph_osdc_msg_data_add(msg
, &op
->xattr
.osd_data
);
1439 case CEPH_OSD_OP_NOTIFY_ACK
:
1440 ceph_osdc_msg_data_add(msg
,
1441 &op
->notify_ack
.request_data
);
1445 case CEPH_OSD_OP_STAT
:
1446 ceph_osdc_msg_data_add(req
->r_reply
,
1449 case CEPH_OSD_OP_READ
:
1450 ceph_osdc_msg_data_add(req
->r_reply
,
1451 &op
->extent
.osd_data
);
1453 case CEPH_OSD_OP_LIST_WATCHERS
:
1454 ceph_osdc_msg_data_add(req
->r_reply
,
1455 &op
->list_watchers
.response_data
);
1459 case CEPH_OSD_OP_CALL
:
1460 WARN_ON(op
->indata_len
!= op
->cls
.class_len
+
1461 op
->cls
.method_len
+
1462 op
->cls
.indata_len
);
1463 ceph_osdc_msg_data_add(msg
, &op
->cls
.request_info
);
1464 /* optional, can be NONE */
1465 ceph_osdc_msg_data_add(msg
, &op
->cls
.request_data
);
1466 /* optional, can be NONE */
1467 ceph_osdc_msg_data_add(req
->r_reply
,
1468 &op
->cls
.response_data
);
1470 case CEPH_OSD_OP_NOTIFY
:
1471 ceph_osdc_msg_data_add(msg
,
1472 &op
->notify
.request_data
);
1473 ceph_osdc_msg_data_add(req
->r_reply
,
1474 &op
->notify
.response_data
);
1478 data_len
+= op
->indata_len
;
1481 WARN_ON(data_len
!= msg
->data_length
);
1484 static void encode_request(struct ceph_osd_request
*req
, struct ceph_msg
*msg
)
1486 void *p
= msg
->front
.iov_base
;
1487 void *const end
= p
+ msg
->front_alloc_len
;
1491 if (req
->r_flags
& CEPH_OSD_FLAG_WRITE
) {
1492 /* snapshots aren't writeable */
1493 WARN_ON(req
->r_snapid
!= CEPH_NOSNAP
);
1495 WARN_ON(req
->r_mtime
.tv_sec
|| req
->r_mtime
.tv_nsec
||
1496 req
->r_data_offset
|| req
->r_snapc
);
1499 setup_request_data(req
, msg
);
1501 ceph_encode_32(&p
, 1); /* client_inc, always 1 */
1502 ceph_encode_32(&p
, req
->r_osdc
->osdmap
->epoch
);
1503 ceph_encode_32(&p
, req
->r_flags
);
1504 ceph_encode_timespec(p
, &req
->r_mtime
);
1505 p
+= sizeof(struct ceph_timespec
);
1506 /* aka reassert_version */
1507 memcpy(p
, &req
->r_replay_version
, sizeof(req
->r_replay_version
));
1508 p
+= sizeof(req
->r_replay_version
);
1511 ceph_start_encoding(&p
, 5, 4,
1512 ceph_oloc_encoding_size(&req
->r_t
.target_oloc
));
1513 ceph_encode_64(&p
, req
->r_t
.target_oloc
.pool
);
1514 ceph_encode_32(&p
, -1); /* preferred */
1515 ceph_encode_32(&p
, 0); /* key len */
1516 if (req
->r_t
.target_oloc
.pool_ns
)
1517 ceph_encode_string(&p
, end
, req
->r_t
.target_oloc
.pool_ns
->str
,
1518 req
->r_t
.target_oloc
.pool_ns
->len
);
1520 ceph_encode_32(&p
, 0);
1523 ceph_encode_8(&p
, 1);
1524 ceph_encode_64(&p
, req
->r_t
.pgid
.pool
);
1525 ceph_encode_32(&p
, req
->r_t
.pgid
.seed
);
1526 ceph_encode_32(&p
, -1); /* preferred */
1529 ceph_encode_32(&p
, req
->r_t
.target_oid
.name_len
);
1530 memcpy(p
, req
->r_t
.target_oid
.name
, req
->r_t
.target_oid
.name_len
);
1531 p
+= req
->r_t
.target_oid
.name_len
;
1533 /* ops, can imply data */
1534 ceph_encode_16(&p
, req
->r_num_ops
);
1535 for (i
= 0; i
< req
->r_num_ops
; i
++) {
1536 data_len
+= osd_req_encode_op(p
, &req
->r_ops
[i
]);
1537 p
+= sizeof(struct ceph_osd_op
);
1540 ceph_encode_64(&p
, req
->r_snapid
); /* snapid */
1542 ceph_encode_64(&p
, req
->r_snapc
->seq
);
1543 ceph_encode_32(&p
, req
->r_snapc
->num_snaps
);
1544 for (i
= 0; i
< req
->r_snapc
->num_snaps
; i
++)
1545 ceph_encode_64(&p
, req
->r_snapc
->snaps
[i
]);
1547 ceph_encode_64(&p
, 0); /* snap_seq */
1548 ceph_encode_32(&p
, 0); /* snaps len */
1551 ceph_encode_32(&p
, req
->r_attempts
); /* retry_attempt */
1554 msg
->front
.iov_len
= p
- msg
->front
.iov_base
;
1555 msg
->hdr
.version
= cpu_to_le16(4); /* MOSDOp v4 */
1556 msg
->hdr
.front_len
= cpu_to_le32(msg
->front
.iov_len
);
1557 msg
->hdr
.data_len
= cpu_to_le32(data_len
);
1559 * The header "data_off" is a hint to the receiver allowing it
1560 * to align received data into its buffers such that there's no
1561 * need to re-copy it before writing it to disk (direct I/O).
1563 msg
->hdr
.data_off
= cpu_to_le16(req
->r_data_offset
);
1565 dout("%s req %p oid %s oid_len %d front %zu data %u\n", __func__
,
1566 req
, req
->r_t
.target_oid
.name
, req
->r_t
.target_oid
.name_len
,
1567 msg
->front
.iov_len
, data_len
);
1571 * @req has to be assigned a tid and registered.
1573 static void send_request(struct ceph_osd_request
*req
)
1575 struct ceph_osd
*osd
= req
->r_osd
;
1577 verify_osd_locked(osd
);
1578 WARN_ON(osd
->o_osd
!= req
->r_t
.osd
);
1581 * We may have a previously queued request message hanging
1582 * around. Cancel it to avoid corrupting the msgr.
1585 ceph_msg_revoke(req
->r_request
);
1587 req
->r_flags
|= CEPH_OSD_FLAG_KNOWN_REDIR
;
1588 if (req
->r_attempts
)
1589 req
->r_flags
|= CEPH_OSD_FLAG_RETRY
;
1591 WARN_ON(req
->r_flags
& CEPH_OSD_FLAG_RETRY
);
1593 encode_request(req
, req
->r_request
);
1595 dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n",
1596 __func__
, req
, req
->r_tid
, req
->r_t
.pgid
.pool
, req
->r_t
.pgid
.seed
,
1597 req
->r_t
.osd
, req
->r_flags
, req
->r_attempts
);
1599 req
->r_t
.paused
= false;
1600 req
->r_stamp
= jiffies
;
1603 req
->r_sent
= osd
->o_incarnation
;
1604 req
->r_request
->hdr
.tid
= cpu_to_le64(req
->r_tid
);
1605 ceph_con_send(&osd
->o_con
, ceph_msg_get(req
->r_request
));
1608 static void maybe_request_map(struct ceph_osd_client
*osdc
)
1610 bool continuous
= false;
1612 verify_osdc_locked(osdc
);
1613 WARN_ON(!osdc
->osdmap
->epoch
);
1615 if (ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
) ||
1616 ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSERD
) ||
1617 ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSEWR
)) {
1618 dout("%s osdc %p continuous\n", __func__
, osdc
);
1621 dout("%s osdc %p onetime\n", __func__
, osdc
);
1624 if (ceph_monc_want_map(&osdc
->client
->monc
, CEPH_SUB_OSDMAP
,
1625 osdc
->osdmap
->epoch
+ 1, continuous
))
1626 ceph_monc_renew_subs(&osdc
->client
->monc
);
1629 static void send_map_check(struct ceph_osd_request
*req
);
1631 static void __submit_request(struct ceph_osd_request
*req
, bool wrlocked
)
1633 struct ceph_osd_client
*osdc
= req
->r_osdc
;
1634 struct ceph_osd
*osd
;
1635 enum calc_target_result ct_res
;
1636 bool need_send
= false;
1637 bool promoted
= false;
1639 WARN_ON(req
->r_tid
|| req
->r_got_reply
);
1640 dout("%s req %p wrlocked %d\n", __func__
, req
, wrlocked
);
1643 ct_res
= calc_target(osdc
, &req
->r_t
, &req
->r_last_force_resend
, false);
1644 if (ct_res
== CALC_TARGET_POOL_DNE
&& !wrlocked
)
1647 osd
= lookup_create_osd(osdc
, req
->r_t
.osd
, wrlocked
);
1649 WARN_ON(PTR_ERR(osd
) != -EAGAIN
|| wrlocked
);
1653 if ((req
->r_flags
& CEPH_OSD_FLAG_WRITE
) &&
1654 ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSEWR
)) {
1655 dout("req %p pausewr\n", req
);
1656 req
->r_t
.paused
= true;
1657 maybe_request_map(osdc
);
1658 } else if ((req
->r_flags
& CEPH_OSD_FLAG_READ
) &&
1659 ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSERD
)) {
1660 dout("req %p pauserd\n", req
);
1661 req
->r_t
.paused
= true;
1662 maybe_request_map(osdc
);
1663 } else if ((req
->r_flags
& CEPH_OSD_FLAG_WRITE
) &&
1664 !(req
->r_flags
& (CEPH_OSD_FLAG_FULL_TRY
|
1665 CEPH_OSD_FLAG_FULL_FORCE
)) &&
1666 (ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
) ||
1667 pool_full(osdc
, req
->r_t
.base_oloc
.pool
))) {
1668 dout("req %p full/pool_full\n", req
);
1669 pr_warn_ratelimited("FULL or reached pool quota\n");
1670 req
->r_t
.paused
= true;
1671 maybe_request_map(osdc
);
1672 } else if (!osd_homeless(osd
)) {
1675 maybe_request_map(osdc
);
1678 mutex_lock(&osd
->lock
);
1680 * Assign the tid atomically with send_request() to protect
1681 * multiple writes to the same object from racing with each
1682 * other, resulting in out of order ops on the OSDs.
1684 req
->r_tid
= atomic64_inc_return(&osdc
->last_tid
);
1685 link_request(osd
, req
);
1688 mutex_unlock(&osd
->lock
);
1690 if (ct_res
== CALC_TARGET_POOL_DNE
)
1691 send_map_check(req
);
1694 downgrade_write(&osdc
->lock
);
1698 up_read(&osdc
->lock
);
1699 down_write(&osdc
->lock
);
1705 static void account_request(struct ceph_osd_request
*req
)
1707 unsigned int mask
= CEPH_OSD_FLAG_ACK
| CEPH_OSD_FLAG_ONDISK
;
1709 if (req
->r_flags
& CEPH_OSD_FLAG_READ
) {
1710 WARN_ON(req
->r_flags
& mask
);
1711 req
->r_flags
|= CEPH_OSD_FLAG_ACK
;
1712 } else if (req
->r_flags
& CEPH_OSD_FLAG_WRITE
)
1713 WARN_ON(!(req
->r_flags
& mask
));
1717 WARN_ON(req
->r_unsafe_callback
&& (req
->r_flags
& mask
) != mask
);
1718 atomic_inc(&req
->r_osdc
->num_requests
);
1721 static void submit_request(struct ceph_osd_request
*req
, bool wrlocked
)
1723 ceph_osdc_get_request(req
);
1724 account_request(req
);
1725 __submit_request(req
, wrlocked
);
1728 static void __finish_request(struct ceph_osd_request
*req
)
1730 struct ceph_osd_client
*osdc
= req
->r_osdc
;
1731 struct ceph_osd
*osd
= req
->r_osd
;
1733 verify_osd_locked(osd
);
1734 dout("%s req %p tid %llu\n", __func__
, req
, req
->r_tid
);
1736 WARN_ON(lookup_request_mc(&osdc
->map_checks
, req
->r_tid
));
1737 unlink_request(osd
, req
);
1738 atomic_dec(&osdc
->num_requests
);
1741 * If an OSD has failed or returned and a request has been sent
1742 * twice, it's possible to get a reply and end up here while the
1743 * request message is queued for delivery. We will ignore the
1744 * reply, so not a big deal, but better to try and catch it.
1746 ceph_msg_revoke(req
->r_request
);
1747 ceph_msg_revoke_incoming(req
->r_reply
);
1750 static void finish_request(struct ceph_osd_request
*req
)
1752 __finish_request(req
);
1753 ceph_osdc_put_request(req
);
1756 static void __complete_request(struct ceph_osd_request
*req
)
1758 if (req
->r_callback
)
1759 req
->r_callback(req
);
1761 complete_all(&req
->r_completion
);
1765 * Note that this is open-coded in handle_reply(), which has to deal
1766 * with ack vs commit, dup acks, etc.
1768 static void complete_request(struct ceph_osd_request
*req
, int err
)
1770 dout("%s req %p tid %llu err %d\n", __func__
, req
, req
->r_tid
, err
);
1772 req
->r_result
= err
;
1773 __finish_request(req
);
1774 __complete_request(req
);
1775 complete_all(&req
->r_safe_completion
);
1776 ceph_osdc_put_request(req
);
1779 static void cancel_map_check(struct ceph_osd_request
*req
)
1781 struct ceph_osd_client
*osdc
= req
->r_osdc
;
1782 struct ceph_osd_request
*lookup_req
;
1784 verify_osdc_wrlocked(osdc
);
1786 lookup_req
= lookup_request_mc(&osdc
->map_checks
, req
->r_tid
);
1790 WARN_ON(lookup_req
!= req
);
1791 erase_request_mc(&osdc
->map_checks
, req
);
1792 ceph_osdc_put_request(req
);
1795 static void cancel_request(struct ceph_osd_request
*req
)
1797 dout("%s req %p tid %llu\n", __func__
, req
, req
->r_tid
);
1799 cancel_map_check(req
);
1800 finish_request(req
);
1803 static void check_pool_dne(struct ceph_osd_request
*req
)
1805 struct ceph_osd_client
*osdc
= req
->r_osdc
;
1806 struct ceph_osdmap
*map
= osdc
->osdmap
;
1808 verify_osdc_wrlocked(osdc
);
1809 WARN_ON(!map
->epoch
);
1811 if (req
->r_attempts
) {
1813 * We sent a request earlier, which means that
1814 * previously the pool existed, and now it does not
1815 * (i.e., it was deleted).
1817 req
->r_map_dne_bound
= map
->epoch
;
1818 dout("%s req %p tid %llu pool disappeared\n", __func__
, req
,
1821 dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__
,
1822 req
, req
->r_tid
, req
->r_map_dne_bound
, map
->epoch
);
1825 if (req
->r_map_dne_bound
) {
1826 if (map
->epoch
>= req
->r_map_dne_bound
) {
1827 /* we had a new enough map */
1828 pr_info_ratelimited("tid %llu pool does not exist\n",
1830 complete_request(req
, -ENOENT
);
1833 send_map_check(req
);
1837 static void map_check_cb(struct ceph_mon_generic_request
*greq
)
1839 struct ceph_osd_client
*osdc
= &greq
->monc
->client
->osdc
;
1840 struct ceph_osd_request
*req
;
1841 u64 tid
= greq
->private_data
;
1843 WARN_ON(greq
->result
|| !greq
->u
.newest
);
1845 down_write(&osdc
->lock
);
1846 req
= lookup_request_mc(&osdc
->map_checks
, tid
);
1848 dout("%s tid %llu dne\n", __func__
, tid
);
1852 dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__
,
1853 req
, req
->r_tid
, req
->r_map_dne_bound
, greq
->u
.newest
);
1854 if (!req
->r_map_dne_bound
)
1855 req
->r_map_dne_bound
= greq
->u
.newest
;
1856 erase_request_mc(&osdc
->map_checks
, req
);
1857 check_pool_dne(req
);
1859 ceph_osdc_put_request(req
);
1861 up_write(&osdc
->lock
);
1864 static void send_map_check(struct ceph_osd_request
*req
)
1866 struct ceph_osd_client
*osdc
= req
->r_osdc
;
1867 struct ceph_osd_request
*lookup_req
;
1870 verify_osdc_wrlocked(osdc
);
1872 lookup_req
= lookup_request_mc(&osdc
->map_checks
, req
->r_tid
);
1874 WARN_ON(lookup_req
!= req
);
1878 ceph_osdc_get_request(req
);
1879 insert_request_mc(&osdc
->map_checks
, req
);
1880 ret
= ceph_monc_get_version_async(&osdc
->client
->monc
, "osdmap",
1881 map_check_cb
, req
->r_tid
);
1886 * lingering requests, watch/notify v2 infrastructure
1888 static void linger_release(struct kref
*kref
)
1890 struct ceph_osd_linger_request
*lreq
=
1891 container_of(kref
, struct ceph_osd_linger_request
, kref
);
1893 dout("%s lreq %p reg_req %p ping_req %p\n", __func__
, lreq
,
1894 lreq
->reg_req
, lreq
->ping_req
);
1895 WARN_ON(!RB_EMPTY_NODE(&lreq
->node
));
1896 WARN_ON(!RB_EMPTY_NODE(&lreq
->osdc_node
));
1897 WARN_ON(!RB_EMPTY_NODE(&lreq
->mc_node
));
1898 WARN_ON(!list_empty(&lreq
->scan_item
));
1899 WARN_ON(!list_empty(&lreq
->pending_lworks
));
1903 ceph_osdc_put_request(lreq
->reg_req
);
1905 ceph_osdc_put_request(lreq
->ping_req
);
1906 target_destroy(&lreq
->t
);
1910 static void linger_put(struct ceph_osd_linger_request
*lreq
)
1913 kref_put(&lreq
->kref
, linger_release
);
1916 static struct ceph_osd_linger_request
*
1917 linger_get(struct ceph_osd_linger_request
*lreq
)
1919 kref_get(&lreq
->kref
);
1923 static struct ceph_osd_linger_request
*
1924 linger_alloc(struct ceph_osd_client
*osdc
)
1926 struct ceph_osd_linger_request
*lreq
;
1928 lreq
= kzalloc(sizeof(*lreq
), GFP_NOIO
);
1932 kref_init(&lreq
->kref
);
1933 mutex_init(&lreq
->lock
);
1934 RB_CLEAR_NODE(&lreq
->node
);
1935 RB_CLEAR_NODE(&lreq
->osdc_node
);
1936 RB_CLEAR_NODE(&lreq
->mc_node
);
1937 INIT_LIST_HEAD(&lreq
->scan_item
);
1938 INIT_LIST_HEAD(&lreq
->pending_lworks
);
1939 init_completion(&lreq
->reg_commit_wait
);
1940 init_completion(&lreq
->notify_finish_wait
);
1943 target_init(&lreq
->t
);
1945 dout("%s lreq %p\n", __func__
, lreq
);
1949 DEFINE_RB_INSDEL_FUNCS(linger
, struct ceph_osd_linger_request
, linger_id
, node
)
1950 DEFINE_RB_FUNCS(linger_osdc
, struct ceph_osd_linger_request
, linger_id
, osdc_node
)
1951 DEFINE_RB_FUNCS(linger_mc
, struct ceph_osd_linger_request
, linger_id
, mc_node
)
1954 * Create linger request <-> OSD session relation.
1956 * @lreq has to be registered, @osd may be homeless.
1958 static void link_linger(struct ceph_osd
*osd
,
1959 struct ceph_osd_linger_request
*lreq
)
1961 verify_osd_locked(osd
);
1962 WARN_ON(!lreq
->linger_id
|| lreq
->osd
);
1963 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__
, osd
,
1964 osd
->o_osd
, lreq
, lreq
->linger_id
);
1966 if (!osd_homeless(osd
))
1967 __remove_osd_from_lru(osd
);
1969 atomic_inc(&osd
->o_osdc
->num_homeless
);
1972 insert_linger(&osd
->o_linger_requests
, lreq
);
1976 static void unlink_linger(struct ceph_osd
*osd
,
1977 struct ceph_osd_linger_request
*lreq
)
1979 verify_osd_locked(osd
);
1980 WARN_ON(lreq
->osd
!= osd
);
1981 dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__
, osd
,
1982 osd
->o_osd
, lreq
, lreq
->linger_id
);
1985 erase_linger(&osd
->o_linger_requests
, lreq
);
1988 if (!osd_homeless(osd
))
1989 maybe_move_osd_to_lru(osd
);
1991 atomic_dec(&osd
->o_osdc
->num_homeless
);
1994 static bool __linger_registered(struct ceph_osd_linger_request
*lreq
)
1996 verify_osdc_locked(lreq
->osdc
);
1998 return !RB_EMPTY_NODE(&lreq
->osdc_node
);
2001 static bool linger_registered(struct ceph_osd_linger_request
*lreq
)
2003 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2006 down_read(&osdc
->lock
);
2007 registered
= __linger_registered(lreq
);
2008 up_read(&osdc
->lock
);
2013 static void linger_register(struct ceph_osd_linger_request
*lreq
)
2015 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2017 verify_osdc_wrlocked(osdc
);
2018 WARN_ON(lreq
->linger_id
);
2021 lreq
->linger_id
= ++osdc
->last_linger_id
;
2022 insert_linger_osdc(&osdc
->linger_requests
, lreq
);
2025 static void linger_unregister(struct ceph_osd_linger_request
*lreq
)
2027 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2029 verify_osdc_wrlocked(osdc
);
2031 erase_linger_osdc(&osdc
->linger_requests
, lreq
);
2035 static void cancel_linger_request(struct ceph_osd_request
*req
)
2037 struct ceph_osd_linger_request
*lreq
= req
->r_priv
;
2039 WARN_ON(!req
->r_linger
);
2040 cancel_request(req
);
2044 struct linger_work
{
2045 struct work_struct work
;
2046 struct ceph_osd_linger_request
*lreq
;
2047 struct list_head pending_item
;
2048 unsigned long queued_stamp
;
2054 void *payload
; /* points into @msg front */
2057 struct ceph_msg
*msg
; /* for ceph_msg_put() */
2065 static struct linger_work
*lwork_alloc(struct ceph_osd_linger_request
*lreq
,
2068 struct linger_work
*lwork
;
2070 lwork
= kzalloc(sizeof(*lwork
), GFP_NOIO
);
2074 INIT_WORK(&lwork
->work
, workfn
);
2075 INIT_LIST_HEAD(&lwork
->pending_item
);
2076 lwork
->lreq
= linger_get(lreq
);
2081 static void lwork_free(struct linger_work
*lwork
)
2083 struct ceph_osd_linger_request
*lreq
= lwork
->lreq
;
2085 mutex_lock(&lreq
->lock
);
2086 list_del(&lwork
->pending_item
);
2087 mutex_unlock(&lreq
->lock
);
2093 static void lwork_queue(struct linger_work
*lwork
)
2095 struct ceph_osd_linger_request
*lreq
= lwork
->lreq
;
2096 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2098 verify_lreq_locked(lreq
);
2099 WARN_ON(!list_empty(&lwork
->pending_item
));
2101 lwork
->queued_stamp
= jiffies
;
2102 list_add_tail(&lwork
->pending_item
, &lreq
->pending_lworks
);
2103 queue_work(osdc
->notify_wq
, &lwork
->work
);
2106 static void do_watch_notify(struct work_struct
*w
)
2108 struct linger_work
*lwork
= container_of(w
, struct linger_work
, work
);
2109 struct ceph_osd_linger_request
*lreq
= lwork
->lreq
;
2111 if (!linger_registered(lreq
)) {
2112 dout("%s lreq %p not registered\n", __func__
, lreq
);
2116 WARN_ON(!lreq
->is_watch
);
2117 dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n",
2118 __func__
, lreq
, lwork
->notify
.notify_id
, lwork
->notify
.notifier_id
,
2119 lwork
->notify
.payload_len
);
2120 lreq
->wcb(lreq
->data
, lwork
->notify
.notify_id
, lreq
->linger_id
,
2121 lwork
->notify
.notifier_id
, lwork
->notify
.payload
,
2122 lwork
->notify
.payload_len
);
2125 ceph_msg_put(lwork
->notify
.msg
);
2129 static void do_watch_error(struct work_struct
*w
)
2131 struct linger_work
*lwork
= container_of(w
, struct linger_work
, work
);
2132 struct ceph_osd_linger_request
*lreq
= lwork
->lreq
;
2134 if (!linger_registered(lreq
)) {
2135 dout("%s lreq %p not registered\n", __func__
, lreq
);
2139 dout("%s lreq %p err %d\n", __func__
, lreq
, lwork
->error
.err
);
2140 lreq
->errcb(lreq
->data
, lreq
->linger_id
, lwork
->error
.err
);
2146 static void queue_watch_error(struct ceph_osd_linger_request
*lreq
)
2148 struct linger_work
*lwork
;
2150 lwork
= lwork_alloc(lreq
, do_watch_error
);
2152 pr_err("failed to allocate error-lwork\n");
2156 lwork
->error
.err
= lreq
->last_error
;
2160 static void linger_reg_commit_complete(struct ceph_osd_linger_request
*lreq
,
2163 if (!completion_done(&lreq
->reg_commit_wait
)) {
2164 lreq
->reg_commit_error
= (result
<= 0 ? result
: 0);
2165 complete_all(&lreq
->reg_commit_wait
);
2169 static void linger_commit_cb(struct ceph_osd_request
*req
)
2171 struct ceph_osd_linger_request
*lreq
= req
->r_priv
;
2173 mutex_lock(&lreq
->lock
);
2174 dout("%s lreq %p linger_id %llu result %d\n", __func__
, lreq
,
2175 lreq
->linger_id
, req
->r_result
);
2176 WARN_ON(!__linger_registered(lreq
));
2177 linger_reg_commit_complete(lreq
, req
->r_result
);
2178 lreq
->committed
= true;
2180 if (!lreq
->is_watch
) {
2181 struct ceph_osd_data
*osd_data
=
2182 osd_req_op_data(req
, 0, notify
, response_data
);
2183 void *p
= page_address(osd_data
->pages
[0]);
2185 WARN_ON(req
->r_ops
[0].op
!= CEPH_OSD_OP_NOTIFY
||
2186 osd_data
->type
!= CEPH_OSD_DATA_TYPE_PAGES
);
2188 /* make note of the notify_id */
2189 if (req
->r_ops
[0].outdata_len
>= sizeof(u64
)) {
2190 lreq
->notify_id
= ceph_decode_64(&p
);
2191 dout("lreq %p notify_id %llu\n", lreq
,
2194 dout("lreq %p no notify_id\n", lreq
);
2198 mutex_unlock(&lreq
->lock
);
2202 static int normalize_watch_error(int err
)
2205 * Translate ENOENT -> ENOTCONN so that a delete->disconnection
2206 * notification and a failure to reconnect because we raced with
2207 * the delete appear the same to the user.
2215 static void linger_reconnect_cb(struct ceph_osd_request
*req
)
2217 struct ceph_osd_linger_request
*lreq
= req
->r_priv
;
2219 mutex_lock(&lreq
->lock
);
2220 dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__
,
2221 lreq
, lreq
->linger_id
, req
->r_result
, lreq
->last_error
);
2222 if (req
->r_result
< 0) {
2223 if (!lreq
->last_error
) {
2224 lreq
->last_error
= normalize_watch_error(req
->r_result
);
2225 queue_watch_error(lreq
);
2229 mutex_unlock(&lreq
->lock
);
2233 static void send_linger(struct ceph_osd_linger_request
*lreq
)
2235 struct ceph_osd_request
*req
= lreq
->reg_req
;
2236 struct ceph_osd_req_op
*op
= &req
->r_ops
[0];
2238 verify_osdc_wrlocked(req
->r_osdc
);
2239 dout("%s lreq %p linger_id %llu\n", __func__
, lreq
, lreq
->linger_id
);
2242 cancel_linger_request(req
);
2244 request_reinit(req
);
2245 ceph_oid_copy(&req
->r_base_oid
, &lreq
->t
.base_oid
);
2246 ceph_oloc_copy(&req
->r_base_oloc
, &lreq
->t
.base_oloc
);
2247 req
->r_flags
= lreq
->t
.flags
;
2248 req
->r_mtime
= lreq
->mtime
;
2250 mutex_lock(&lreq
->lock
);
2251 if (lreq
->is_watch
&& lreq
->committed
) {
2252 WARN_ON(op
->op
!= CEPH_OSD_OP_WATCH
||
2253 op
->watch
.cookie
!= lreq
->linger_id
);
2254 op
->watch
.op
= CEPH_OSD_WATCH_OP_RECONNECT
;
2255 op
->watch
.gen
= ++lreq
->register_gen
;
2256 dout("lreq %p reconnect register_gen %u\n", lreq
,
2258 req
->r_callback
= linger_reconnect_cb
;
2260 if (!lreq
->is_watch
)
2261 lreq
->notify_id
= 0;
2263 WARN_ON(op
->watch
.op
!= CEPH_OSD_WATCH_OP_WATCH
);
2264 dout("lreq %p register\n", lreq
);
2265 req
->r_callback
= linger_commit_cb
;
2267 mutex_unlock(&lreq
->lock
);
2269 req
->r_priv
= linger_get(lreq
);
2270 req
->r_linger
= true;
2272 submit_request(req
, true);
2275 static void linger_ping_cb(struct ceph_osd_request
*req
)
2277 struct ceph_osd_linger_request
*lreq
= req
->r_priv
;
2279 mutex_lock(&lreq
->lock
);
2280 dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n",
2281 __func__
, lreq
, lreq
->linger_id
, req
->r_result
, lreq
->ping_sent
,
2283 if (lreq
->register_gen
== req
->r_ops
[0].watch
.gen
) {
2284 if (!req
->r_result
) {
2285 lreq
->watch_valid_thru
= lreq
->ping_sent
;
2286 } else if (!lreq
->last_error
) {
2287 lreq
->last_error
= normalize_watch_error(req
->r_result
);
2288 queue_watch_error(lreq
);
2291 dout("lreq %p register_gen %u ignoring old pong %u\n", lreq
,
2292 lreq
->register_gen
, req
->r_ops
[0].watch
.gen
);
2295 mutex_unlock(&lreq
->lock
);
2299 static void send_linger_ping(struct ceph_osd_linger_request
*lreq
)
2301 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2302 struct ceph_osd_request
*req
= lreq
->ping_req
;
2303 struct ceph_osd_req_op
*op
= &req
->r_ops
[0];
2305 if (ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSERD
)) {
2306 dout("%s PAUSERD\n", __func__
);
2310 lreq
->ping_sent
= jiffies
;
2311 dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n",
2312 __func__
, lreq
, lreq
->linger_id
, lreq
->ping_sent
,
2313 lreq
->register_gen
);
2316 cancel_linger_request(req
);
2318 request_reinit(req
);
2319 target_copy(&req
->r_t
, &lreq
->t
);
2321 WARN_ON(op
->op
!= CEPH_OSD_OP_WATCH
||
2322 op
->watch
.cookie
!= lreq
->linger_id
||
2323 op
->watch
.op
!= CEPH_OSD_WATCH_OP_PING
);
2324 op
->watch
.gen
= lreq
->register_gen
;
2325 req
->r_callback
= linger_ping_cb
;
2326 req
->r_priv
= linger_get(lreq
);
2327 req
->r_linger
= true;
2329 ceph_osdc_get_request(req
);
2330 account_request(req
);
2331 req
->r_tid
= atomic64_inc_return(&osdc
->last_tid
);
2332 link_request(lreq
->osd
, req
);
2336 static void linger_submit(struct ceph_osd_linger_request
*lreq
)
2338 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2339 struct ceph_osd
*osd
;
2341 calc_target(osdc
, &lreq
->t
, &lreq
->last_force_resend
, false);
2342 osd
= lookup_create_osd(osdc
, lreq
->t
.osd
, true);
2343 link_linger(osd
, lreq
);
2348 static void cancel_linger_map_check(struct ceph_osd_linger_request
*lreq
)
2350 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2351 struct ceph_osd_linger_request
*lookup_lreq
;
2353 verify_osdc_wrlocked(osdc
);
2355 lookup_lreq
= lookup_linger_mc(&osdc
->linger_map_checks
,
2360 WARN_ON(lookup_lreq
!= lreq
);
2361 erase_linger_mc(&osdc
->linger_map_checks
, lreq
);
2366 * @lreq has to be both registered and linked.
2368 static void __linger_cancel(struct ceph_osd_linger_request
*lreq
)
2370 if (lreq
->is_watch
&& lreq
->ping_req
->r_osd
)
2371 cancel_linger_request(lreq
->ping_req
);
2372 if (lreq
->reg_req
->r_osd
)
2373 cancel_linger_request(lreq
->reg_req
);
2374 cancel_linger_map_check(lreq
);
2375 unlink_linger(lreq
->osd
, lreq
);
2376 linger_unregister(lreq
);
2379 static void linger_cancel(struct ceph_osd_linger_request
*lreq
)
2381 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2383 down_write(&osdc
->lock
);
2384 if (__linger_registered(lreq
))
2385 __linger_cancel(lreq
);
2386 up_write(&osdc
->lock
);
2389 static void send_linger_map_check(struct ceph_osd_linger_request
*lreq
);
2391 static void check_linger_pool_dne(struct ceph_osd_linger_request
*lreq
)
2393 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2394 struct ceph_osdmap
*map
= osdc
->osdmap
;
2396 verify_osdc_wrlocked(osdc
);
2397 WARN_ON(!map
->epoch
);
2399 if (lreq
->register_gen
) {
2400 lreq
->map_dne_bound
= map
->epoch
;
2401 dout("%s lreq %p linger_id %llu pool disappeared\n", __func__
,
2402 lreq
, lreq
->linger_id
);
2404 dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n",
2405 __func__
, lreq
, lreq
->linger_id
, lreq
->map_dne_bound
,
2409 if (lreq
->map_dne_bound
) {
2410 if (map
->epoch
>= lreq
->map_dne_bound
) {
2411 /* we had a new enough map */
2412 pr_info("linger_id %llu pool does not exist\n",
2414 linger_reg_commit_complete(lreq
, -ENOENT
);
2415 __linger_cancel(lreq
);
2418 send_linger_map_check(lreq
);
2422 static void linger_map_check_cb(struct ceph_mon_generic_request
*greq
)
2424 struct ceph_osd_client
*osdc
= &greq
->monc
->client
->osdc
;
2425 struct ceph_osd_linger_request
*lreq
;
2426 u64 linger_id
= greq
->private_data
;
2428 WARN_ON(greq
->result
|| !greq
->u
.newest
);
2430 down_write(&osdc
->lock
);
2431 lreq
= lookup_linger_mc(&osdc
->linger_map_checks
, linger_id
);
2433 dout("%s linger_id %llu dne\n", __func__
, linger_id
);
2437 dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n",
2438 __func__
, lreq
, lreq
->linger_id
, lreq
->map_dne_bound
,
2440 if (!lreq
->map_dne_bound
)
2441 lreq
->map_dne_bound
= greq
->u
.newest
;
2442 erase_linger_mc(&osdc
->linger_map_checks
, lreq
);
2443 check_linger_pool_dne(lreq
);
2447 up_write(&osdc
->lock
);
2450 static void send_linger_map_check(struct ceph_osd_linger_request
*lreq
)
2452 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2453 struct ceph_osd_linger_request
*lookup_lreq
;
2456 verify_osdc_wrlocked(osdc
);
2458 lookup_lreq
= lookup_linger_mc(&osdc
->linger_map_checks
,
2461 WARN_ON(lookup_lreq
!= lreq
);
2466 insert_linger_mc(&osdc
->linger_map_checks
, lreq
);
2467 ret
= ceph_monc_get_version_async(&osdc
->client
->monc
, "osdmap",
2468 linger_map_check_cb
, lreq
->linger_id
);
2472 static int linger_reg_commit_wait(struct ceph_osd_linger_request
*lreq
)
2476 dout("%s lreq %p linger_id %llu\n", __func__
, lreq
, lreq
->linger_id
);
2477 ret
= wait_for_completion_interruptible(&lreq
->reg_commit_wait
);
2478 return ret
?: lreq
->reg_commit_error
;
2481 static int linger_notify_finish_wait(struct ceph_osd_linger_request
*lreq
)
2485 dout("%s lreq %p linger_id %llu\n", __func__
, lreq
, lreq
->linger_id
);
2486 ret
= wait_for_completion_interruptible(&lreq
->notify_finish_wait
);
2487 return ret
?: lreq
->notify_finish_error
;
2491 * Timeout callback, called every N seconds. When 1 or more OSD
2492 * requests has been active for more than N seconds, we send a keepalive
2493 * (tag + timestamp) to its OSD to ensure any communications channel
2494 * reset is detected.
2496 static void handle_timeout(struct work_struct
*work
)
2498 struct ceph_osd_client
*osdc
=
2499 container_of(work
, struct ceph_osd_client
, timeout_work
.work
);
2500 struct ceph_options
*opts
= osdc
->client
->options
;
2501 unsigned long cutoff
= jiffies
- opts
->osd_keepalive_timeout
;
2502 LIST_HEAD(slow_osds
);
2503 struct rb_node
*n
, *p
;
2505 dout("%s osdc %p\n", __func__
, osdc
);
2506 down_write(&osdc
->lock
);
2509 * ping osds that are a bit slow. this ensures that if there
2510 * is a break in the TCP connection we will notice, and reopen
2511 * a connection with that osd (from the fault callback).
2513 for (n
= rb_first(&osdc
->osds
); n
; n
= rb_next(n
)) {
2514 struct ceph_osd
*osd
= rb_entry(n
, struct ceph_osd
, o_node
);
2517 for (p
= rb_first(&osd
->o_requests
); p
; p
= rb_next(p
)) {
2518 struct ceph_osd_request
*req
=
2519 rb_entry(p
, struct ceph_osd_request
, r_node
);
2521 if (time_before(req
->r_stamp
, cutoff
)) {
2522 dout(" req %p tid %llu on osd%d is laggy\n",
2523 req
, req
->r_tid
, osd
->o_osd
);
2527 for (p
= rb_first(&osd
->o_linger_requests
); p
; p
= rb_next(p
)) {
2528 struct ceph_osd_linger_request
*lreq
=
2529 rb_entry(p
, struct ceph_osd_linger_request
, node
);
2531 dout(" lreq %p linger_id %llu is served by osd%d\n",
2532 lreq
, lreq
->linger_id
, osd
->o_osd
);
2535 mutex_lock(&lreq
->lock
);
2536 if (lreq
->is_watch
&& lreq
->committed
&& !lreq
->last_error
)
2537 send_linger_ping(lreq
);
2538 mutex_unlock(&lreq
->lock
);
2542 list_move_tail(&osd
->o_keepalive_item
, &slow_osds
);
2545 if (atomic_read(&osdc
->num_homeless
) || !list_empty(&slow_osds
))
2546 maybe_request_map(osdc
);
2548 while (!list_empty(&slow_osds
)) {
2549 struct ceph_osd
*osd
= list_first_entry(&slow_osds
,
2552 list_del_init(&osd
->o_keepalive_item
);
2553 ceph_con_keepalive(&osd
->o_con
);
2556 up_write(&osdc
->lock
);
2557 schedule_delayed_work(&osdc
->timeout_work
,
2558 osdc
->client
->options
->osd_keepalive_timeout
);
2561 static void handle_osds_timeout(struct work_struct
*work
)
2563 struct ceph_osd_client
*osdc
=
2564 container_of(work
, struct ceph_osd_client
,
2565 osds_timeout_work
.work
);
2566 unsigned long delay
= osdc
->client
->options
->osd_idle_ttl
/ 4;
2567 struct ceph_osd
*osd
, *nosd
;
2569 dout("%s osdc %p\n", __func__
, osdc
);
2570 down_write(&osdc
->lock
);
2571 list_for_each_entry_safe(osd
, nosd
, &osdc
->osd_lru
, o_osd_lru
) {
2572 if (time_before(jiffies
, osd
->lru_ttl
))
2575 WARN_ON(!RB_EMPTY_ROOT(&osd
->o_requests
));
2576 WARN_ON(!RB_EMPTY_ROOT(&osd
->o_linger_requests
));
2580 up_write(&osdc
->lock
);
2581 schedule_delayed_work(&osdc
->osds_timeout_work
,
2582 round_jiffies_relative(delay
));
2585 static int ceph_oloc_decode(void **p
, void *end
,
2586 struct ceph_object_locator
*oloc
)
2588 u8 struct_v
, struct_cv
;
2593 ceph_decode_need(p
, end
, 1 + 1 + 4, e_inval
);
2594 struct_v
= ceph_decode_8(p
);
2595 struct_cv
= ceph_decode_8(p
);
2597 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
2598 struct_v
, struct_cv
);
2601 if (struct_cv
> 6) {
2602 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
2603 struct_v
, struct_cv
);
2606 len
= ceph_decode_32(p
);
2607 ceph_decode_need(p
, end
, len
, e_inval
);
2608 struct_end
= *p
+ len
;
2610 oloc
->pool
= ceph_decode_64(p
);
2611 *p
+= 4; /* skip preferred */
2613 len
= ceph_decode_32(p
);
2615 pr_warn("ceph_object_locator::key is set\n");
2619 if (struct_v
>= 5) {
2620 bool changed
= false;
2622 len
= ceph_decode_32(p
);
2624 ceph_decode_need(p
, end
, len
, e_inval
);
2625 if (!oloc
->pool_ns
||
2626 ceph_compare_string(oloc
->pool_ns
, *p
, len
))
2634 /* redirect changes namespace */
2635 pr_warn("ceph_object_locator::nspace is changed\n");
2640 if (struct_v
>= 6) {
2641 s64 hash
= ceph_decode_64(p
);
2643 pr_warn("ceph_object_locator::hash is set\n");
2658 static int ceph_redirect_decode(void **p
, void *end
,
2659 struct ceph_request_redirect
*redir
)
2661 u8 struct_v
, struct_cv
;
2666 ceph_decode_need(p
, end
, 1 + 1 + 4, e_inval
);
2667 struct_v
= ceph_decode_8(p
);
2668 struct_cv
= ceph_decode_8(p
);
2669 if (struct_cv
> 1) {
2670 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
2671 struct_v
, struct_cv
);
2674 len
= ceph_decode_32(p
);
2675 ceph_decode_need(p
, end
, len
, e_inval
);
2676 struct_end
= *p
+ len
;
2678 ret
= ceph_oloc_decode(p
, end
, &redir
->oloc
);
2682 len
= ceph_decode_32(p
);
2684 pr_warn("ceph_request_redirect::object_name is set\n");
2688 len
= ceph_decode_32(p
);
2689 *p
+= len
; /* skip osd_instructions */
2701 struct MOSDOpReply
{
2702 struct ceph_pg pgid
;
2707 u32 outdata_len
[CEPH_OSD_MAX_OPS
];
2708 s32 rval
[CEPH_OSD_MAX_OPS
];
2710 struct ceph_eversion replay_version
;
2712 struct ceph_request_redirect redirect
;
2715 static int decode_MOSDOpReply(const struct ceph_msg
*msg
, struct MOSDOpReply
*m
)
2717 void *p
= msg
->front
.iov_base
;
2718 void *const end
= p
+ msg
->front
.iov_len
;
2719 u16 version
= le16_to_cpu(msg
->hdr
.version
);
2720 struct ceph_eversion bad_replay_version
;
2726 ceph_decode_32_safe(&p
, end
, len
, e_inval
);
2727 ceph_decode_need(&p
, end
, len
, e_inval
);
2728 p
+= len
; /* skip oid */
2730 ret
= ceph_decode_pgid(&p
, end
, &m
->pgid
);
2734 ceph_decode_64_safe(&p
, end
, m
->flags
, e_inval
);
2735 ceph_decode_32_safe(&p
, end
, m
->result
, e_inval
);
2736 ceph_decode_need(&p
, end
, sizeof(bad_replay_version
), e_inval
);
2737 memcpy(&bad_replay_version
, p
, sizeof(bad_replay_version
));
2738 p
+= sizeof(bad_replay_version
);
2739 ceph_decode_32_safe(&p
, end
, m
->epoch
, e_inval
);
2741 ceph_decode_32_safe(&p
, end
, m
->num_ops
, e_inval
);
2742 if (m
->num_ops
> ARRAY_SIZE(m
->outdata_len
))
2745 ceph_decode_need(&p
, end
, m
->num_ops
* sizeof(struct ceph_osd_op
),
2747 for (i
= 0; i
< m
->num_ops
; i
++) {
2748 struct ceph_osd_op
*op
= p
;
2750 m
->outdata_len
[i
] = le32_to_cpu(op
->payload_len
);
2754 ceph_decode_32_safe(&p
, end
, m
->retry_attempt
, e_inval
);
2755 for (i
= 0; i
< m
->num_ops
; i
++)
2756 ceph_decode_32_safe(&p
, end
, m
->rval
[i
], e_inval
);
2759 ceph_decode_need(&p
, end
, sizeof(m
->replay_version
), e_inval
);
2760 memcpy(&m
->replay_version
, p
, sizeof(m
->replay_version
));
2761 p
+= sizeof(m
->replay_version
);
2762 ceph_decode_64_safe(&p
, end
, m
->user_version
, e_inval
);
2764 m
->replay_version
= bad_replay_version
; /* struct */
2765 m
->user_version
= le64_to_cpu(m
->replay_version
.version
);
2770 ceph_decode_8_safe(&p
, end
, decode_redir
, e_inval
);
2778 ret
= ceph_redirect_decode(&p
, end
, &m
->redirect
);
2782 ceph_oloc_init(&m
->redirect
.oloc
);
2792 * We are done with @req if
2793 * - @m is a safe reply, or
2794 * - @m is an unsafe reply and we didn't want a safe one
2796 static bool done_request(const struct ceph_osd_request
*req
,
2797 const struct MOSDOpReply
*m
)
2799 return (m
->result
< 0 ||
2800 (m
->flags
& CEPH_OSD_FLAG_ONDISK
) ||
2801 !(req
->r_flags
& CEPH_OSD_FLAG_ONDISK
));
2805 * handle osd op reply. either call the callback if it is specified,
2806 * or do the completion to wake up the waiting thread.
2808 * ->r_unsafe_callback is set? yes no
2810 * first reply is OK (needed r_cb/r_completion, r_cb/r_completion,
2811 * any or needed/got safe) r_safe_completion r_safe_completion
2813 * first reply is unsafe r_unsafe_cb(true) (nothing)
2815 * when we get the safe reply r_unsafe_cb(false), r_cb/r_completion,
2816 * r_safe_completion r_safe_completion
2818 static void handle_reply(struct ceph_osd
*osd
, struct ceph_msg
*msg
)
2820 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
2821 struct ceph_osd_request
*req
;
2822 struct MOSDOpReply m
;
2823 u64 tid
= le64_to_cpu(msg
->hdr
.tid
);
2829 dout("%s msg %p tid %llu\n", __func__
, msg
, tid
);
2831 down_read(&osdc
->lock
);
2832 if (!osd_registered(osd
)) {
2833 dout("%s osd%d unknown\n", __func__
, osd
->o_osd
);
2834 goto out_unlock_osdc
;
2836 WARN_ON(osd
->o_osd
!= le64_to_cpu(msg
->hdr
.src
.num
));
2838 mutex_lock(&osd
->lock
);
2839 req
= lookup_request(&osd
->o_requests
, tid
);
2841 dout("%s osd%d tid %llu unknown\n", __func__
, osd
->o_osd
, tid
);
2842 goto out_unlock_session
;
2845 m
.redirect
.oloc
.pool_ns
= req
->r_t
.target_oloc
.pool_ns
;
2846 ret
= decode_MOSDOpReply(msg
, &m
);
2847 m
.redirect
.oloc
.pool_ns
= NULL
;
2849 pr_err("failed to decode MOSDOpReply for tid %llu: %d\n",
2854 dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n",
2855 __func__
, req
, req
->r_tid
, m
.flags
, m
.pgid
.pool
, m
.pgid
.seed
,
2856 m
.epoch
, m
.retry_attempt
, le32_to_cpu(m
.replay_version
.epoch
),
2857 le64_to_cpu(m
.replay_version
.version
), m
.user_version
);
2859 if (m
.retry_attempt
>= 0) {
2860 if (m
.retry_attempt
!= req
->r_attempts
- 1) {
2861 dout("req %p tid %llu retry_attempt %d != %d, ignoring\n",
2862 req
, req
->r_tid
, m
.retry_attempt
,
2863 req
->r_attempts
- 1);
2864 goto out_unlock_session
;
2867 WARN_ON(1); /* MOSDOpReply v4 is assumed */
2870 if (!ceph_oloc_empty(&m
.redirect
.oloc
)) {
2871 dout("req %p tid %llu redirect pool %lld\n", req
, req
->r_tid
,
2872 m
.redirect
.oloc
.pool
);
2873 unlink_request(osd
, req
);
2874 mutex_unlock(&osd
->lock
);
2877 * Not ceph_oloc_copy() - changing pool_ns is not
2880 req
->r_t
.target_oloc
.pool
= m
.redirect
.oloc
.pool
;
2881 req
->r_flags
|= CEPH_OSD_FLAG_REDIRECTED
;
2883 __submit_request(req
, false);
2884 goto out_unlock_osdc
;
2887 if (m
.num_ops
!= req
->r_num_ops
) {
2888 pr_err("num_ops %d != %d for tid %llu\n", m
.num_ops
,
2889 req
->r_num_ops
, req
->r_tid
);
2892 for (i
= 0; i
< req
->r_num_ops
; i
++) {
2893 dout(" req %p tid %llu op %d rval %d len %u\n", req
,
2894 req
->r_tid
, i
, m
.rval
[i
], m
.outdata_len
[i
]);
2895 req
->r_ops
[i
].rval
= m
.rval
[i
];
2896 req
->r_ops
[i
].outdata_len
= m
.outdata_len
[i
];
2897 data_len
+= m
.outdata_len
[i
];
2899 if (data_len
!= le32_to_cpu(msg
->hdr
.data_len
)) {
2900 pr_err("sum of lens %u != %u for tid %llu\n", data_len
,
2901 le32_to_cpu(msg
->hdr
.data_len
), req
->r_tid
);
2904 dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__
,
2905 req
, req
->r_tid
, req
->r_got_reply
, m
.result
, data_len
);
2907 already_acked
= req
->r_got_reply
;
2908 if (!already_acked
) {
2909 req
->r_result
= m
.result
?: data_len
;
2910 req
->r_replay_version
= m
.replay_version
; /* struct */
2911 req
->r_got_reply
= true;
2912 } else if (!(m
.flags
& CEPH_OSD_FLAG_ONDISK
)) {
2913 dout("req %p tid %llu dup ack\n", req
, req
->r_tid
);
2914 goto out_unlock_session
;
2917 if (done_request(req
, &m
)) {
2918 __finish_request(req
);
2919 if (req
->r_linger
) {
2920 WARN_ON(req
->r_unsafe_callback
);
2921 dout("req %p tid %llu cb (locked)\n", req
, req
->r_tid
);
2922 __complete_request(req
);
2926 mutex_unlock(&osd
->lock
);
2927 up_read(&osdc
->lock
);
2929 if (done_request(req
, &m
)) {
2930 if (already_acked
&& req
->r_unsafe_callback
) {
2931 dout("req %p tid %llu safe-cb\n", req
, req
->r_tid
);
2932 req
->r_unsafe_callback(req
, false);
2933 } else if (!req
->r_linger
) {
2934 dout("req %p tid %llu cb\n", req
, req
->r_tid
);
2935 __complete_request(req
);
2937 if (m
.flags
& CEPH_OSD_FLAG_ONDISK
)
2938 complete_all(&req
->r_safe_completion
);
2939 ceph_osdc_put_request(req
);
2941 if (req
->r_unsafe_callback
) {
2942 dout("req %p tid %llu unsafe-cb\n", req
, req
->r_tid
);
2943 req
->r_unsafe_callback(req
, true);
2952 complete_request(req
, -EIO
);
2954 mutex_unlock(&osd
->lock
);
2956 up_read(&osdc
->lock
);
2959 static void set_pool_was_full(struct ceph_osd_client
*osdc
)
2963 for (n
= rb_first(&osdc
->osdmap
->pg_pools
); n
; n
= rb_next(n
)) {
2964 struct ceph_pg_pool_info
*pi
=
2965 rb_entry(n
, struct ceph_pg_pool_info
, node
);
2967 pi
->was_full
= __pool_full(pi
);
2971 static bool pool_cleared_full(struct ceph_osd_client
*osdc
, s64 pool_id
)
2973 struct ceph_pg_pool_info
*pi
;
2975 pi
= ceph_pg_pool_by_id(osdc
->osdmap
, pool_id
);
2979 return pi
->was_full
&& !__pool_full(pi
);
2982 static enum calc_target_result
2983 recalc_linger_target(struct ceph_osd_linger_request
*lreq
)
2985 struct ceph_osd_client
*osdc
= lreq
->osdc
;
2986 enum calc_target_result ct_res
;
2988 ct_res
= calc_target(osdc
, &lreq
->t
, &lreq
->last_force_resend
, true);
2989 if (ct_res
== CALC_TARGET_NEED_RESEND
) {
2990 struct ceph_osd
*osd
;
2992 osd
= lookup_create_osd(osdc
, lreq
->t
.osd
, true);
2993 if (osd
!= lreq
->osd
) {
2994 unlink_linger(lreq
->osd
, lreq
);
2995 link_linger(osd
, lreq
);
3003 * Requeue requests whose mapping to an OSD has changed.
3005 static void scan_requests(struct ceph_osd
*osd
,
3008 bool check_pool_cleared_full
,
3009 struct rb_root
*need_resend
,
3010 struct list_head
*need_resend_linger
)
3012 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
3014 bool force_resend_writes
;
3016 for (n
= rb_first(&osd
->o_linger_requests
); n
; ) {
3017 struct ceph_osd_linger_request
*lreq
=
3018 rb_entry(n
, struct ceph_osd_linger_request
, node
);
3019 enum calc_target_result ct_res
;
3021 n
= rb_next(n
); /* recalc_linger_target() */
3023 dout("%s lreq %p linger_id %llu\n", __func__
, lreq
,
3025 ct_res
= recalc_linger_target(lreq
);
3027 case CALC_TARGET_NO_ACTION
:
3028 force_resend_writes
= cleared_full
||
3029 (check_pool_cleared_full
&&
3030 pool_cleared_full(osdc
, lreq
->t
.base_oloc
.pool
));
3031 if (!force_resend
&& !force_resend_writes
)
3035 case CALC_TARGET_NEED_RESEND
:
3036 cancel_linger_map_check(lreq
);
3038 * scan_requests() for the previous epoch(s)
3039 * may have already added it to the list, since
3040 * it's not unlinked here.
3042 if (list_empty(&lreq
->scan_item
))
3043 list_add_tail(&lreq
->scan_item
, need_resend_linger
);
3045 case CALC_TARGET_POOL_DNE
:
3046 check_linger_pool_dne(lreq
);
3051 for (n
= rb_first(&osd
->o_requests
); n
; ) {
3052 struct ceph_osd_request
*req
=
3053 rb_entry(n
, struct ceph_osd_request
, r_node
);
3054 enum calc_target_result ct_res
;
3056 n
= rb_next(n
); /* unlink_request(), check_pool_dne() */
3058 dout("%s req %p tid %llu\n", __func__
, req
, req
->r_tid
);
3059 ct_res
= calc_target(osdc
, &req
->r_t
,
3060 &req
->r_last_force_resend
, false);
3062 case CALC_TARGET_NO_ACTION
:
3063 force_resend_writes
= cleared_full
||
3064 (check_pool_cleared_full
&&
3065 pool_cleared_full(osdc
, req
->r_t
.base_oloc
.pool
));
3066 if (!force_resend
&&
3067 (!(req
->r_flags
& CEPH_OSD_FLAG_WRITE
) ||
3068 !force_resend_writes
))
3072 case CALC_TARGET_NEED_RESEND
:
3073 cancel_map_check(req
);
3074 unlink_request(osd
, req
);
3075 insert_request(need_resend
, req
);
3077 case CALC_TARGET_POOL_DNE
:
3078 check_pool_dne(req
);
3084 static int handle_one_map(struct ceph_osd_client
*osdc
,
3085 void *p
, void *end
, bool incremental
,
3086 struct rb_root
*need_resend
,
3087 struct list_head
*need_resend_linger
)
3089 struct ceph_osdmap
*newmap
;
3091 bool skipped_map
= false;
3094 was_full
= ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
);
3095 set_pool_was_full(osdc
);
3098 newmap
= osdmap_apply_incremental(&p
, end
, osdc
->osdmap
);
3100 newmap
= ceph_osdmap_decode(&p
, end
);
3102 return PTR_ERR(newmap
);
3104 if (newmap
!= osdc
->osdmap
) {
3106 * Preserve ->was_full before destroying the old map.
3107 * For pools that weren't in the old map, ->was_full
3110 for (n
= rb_first(&newmap
->pg_pools
); n
; n
= rb_next(n
)) {
3111 struct ceph_pg_pool_info
*pi
=
3112 rb_entry(n
, struct ceph_pg_pool_info
, node
);
3113 struct ceph_pg_pool_info
*old_pi
;
3115 old_pi
= ceph_pg_pool_by_id(osdc
->osdmap
, pi
->id
);
3117 pi
->was_full
= old_pi
->was_full
;
3119 WARN_ON(pi
->was_full
);
3122 if (osdc
->osdmap
->epoch
&&
3123 osdc
->osdmap
->epoch
+ 1 < newmap
->epoch
) {
3124 WARN_ON(incremental
);
3128 ceph_osdmap_destroy(osdc
->osdmap
);
3129 osdc
->osdmap
= newmap
;
3132 was_full
&= !ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
);
3133 scan_requests(&osdc
->homeless_osd
, skipped_map
, was_full
, true,
3134 need_resend
, need_resend_linger
);
3136 for (n
= rb_first(&osdc
->osds
); n
; ) {
3137 struct ceph_osd
*osd
= rb_entry(n
, struct ceph_osd
, o_node
);
3139 n
= rb_next(n
); /* close_osd() */
3141 scan_requests(osd
, skipped_map
, was_full
, true, need_resend
,
3142 need_resend_linger
);
3143 if (!ceph_osd_is_up(osdc
->osdmap
, osd
->o_osd
) ||
3144 memcmp(&osd
->o_con
.peer_addr
,
3145 ceph_osd_addr(osdc
->osdmap
, osd
->o_osd
),
3146 sizeof(struct ceph_entity_addr
)))
3153 static void kick_requests(struct ceph_osd_client
*osdc
,
3154 struct rb_root
*need_resend
,
3155 struct list_head
*need_resend_linger
)
3157 struct ceph_osd_linger_request
*lreq
, *nlreq
;
3160 for (n
= rb_first(need_resend
); n
; ) {
3161 struct ceph_osd_request
*req
=
3162 rb_entry(n
, struct ceph_osd_request
, r_node
);
3163 struct ceph_osd
*osd
;
3166 erase_request(need_resend
, req
); /* before link_request() */
3168 WARN_ON(req
->r_osd
);
3169 calc_target(osdc
, &req
->r_t
, NULL
, false);
3170 osd
= lookup_create_osd(osdc
, req
->r_t
.osd
, true);
3171 link_request(osd
, req
);
3172 if (!req
->r_linger
) {
3173 if (!osd_homeless(osd
) && !req
->r_t
.paused
)
3176 cancel_linger_request(req
);
3180 list_for_each_entry_safe(lreq
, nlreq
, need_resend_linger
, scan_item
) {
3181 if (!osd_homeless(lreq
->osd
))
3184 list_del_init(&lreq
->scan_item
);
3189 * Process updated osd map.
3191 * The message contains any number of incremental and full maps, normally
3192 * indicating some sort of topology change in the cluster. Kick requests
3193 * off to different OSDs as needed.
3195 void ceph_osdc_handle_map(struct ceph_osd_client
*osdc
, struct ceph_msg
*msg
)
3197 void *p
= msg
->front
.iov_base
;
3198 void *const end
= p
+ msg
->front
.iov_len
;
3199 u32 nr_maps
, maplen
;
3201 struct ceph_fsid fsid
;
3202 struct rb_root need_resend
= RB_ROOT
;
3203 LIST_HEAD(need_resend_linger
);
3204 bool handled_incremental
= false;
3205 bool was_pauserd
, was_pausewr
;
3206 bool pauserd
, pausewr
;
3209 dout("%s have %u\n", __func__
, osdc
->osdmap
->epoch
);
3210 down_write(&osdc
->lock
);
3213 ceph_decode_need(&p
, end
, sizeof(fsid
), bad
);
3214 ceph_decode_copy(&p
, &fsid
, sizeof(fsid
));
3215 if (ceph_check_fsid(osdc
->client
, &fsid
) < 0)
3218 was_pauserd
= ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSERD
);
3219 was_pausewr
= ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSEWR
) ||
3220 ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
) ||
3221 have_pool_full(osdc
);
3223 /* incremental maps */
3224 ceph_decode_32_safe(&p
, end
, nr_maps
, bad
);
3225 dout(" %d inc maps\n", nr_maps
);
3226 while (nr_maps
> 0) {
3227 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
3228 epoch
= ceph_decode_32(&p
);
3229 maplen
= ceph_decode_32(&p
);
3230 ceph_decode_need(&p
, end
, maplen
, bad
);
3231 if (osdc
->osdmap
->epoch
&&
3232 osdc
->osdmap
->epoch
+ 1 == epoch
) {
3233 dout("applying incremental map %u len %d\n",
3235 err
= handle_one_map(osdc
, p
, p
+ maplen
, true,
3236 &need_resend
, &need_resend_linger
);
3239 handled_incremental
= true;
3241 dout("ignoring incremental map %u len %d\n",
3247 if (handled_incremental
)
3251 ceph_decode_32_safe(&p
, end
, nr_maps
, bad
);
3252 dout(" %d full maps\n", nr_maps
);
3254 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
3255 epoch
= ceph_decode_32(&p
);
3256 maplen
= ceph_decode_32(&p
);
3257 ceph_decode_need(&p
, end
, maplen
, bad
);
3259 dout("skipping non-latest full map %u len %d\n",
3261 } else if (osdc
->osdmap
->epoch
>= epoch
) {
3262 dout("skipping full map %u len %d, "
3263 "older than our %u\n", epoch
, maplen
,
3264 osdc
->osdmap
->epoch
);
3266 dout("taking full map %u len %d\n", epoch
, maplen
);
3267 err
= handle_one_map(osdc
, p
, p
+ maplen
, false,
3268 &need_resend
, &need_resend_linger
);
3278 * subscribe to subsequent osdmap updates if full to ensure
3279 * we find out when we are no longer full and stop returning
3282 pauserd
= ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSERD
);
3283 pausewr
= ceph_osdmap_flag(osdc
, CEPH_OSDMAP_PAUSEWR
) ||
3284 ceph_osdmap_flag(osdc
, CEPH_OSDMAP_FULL
) ||
3285 have_pool_full(osdc
);
3286 if (was_pauserd
|| was_pausewr
|| pauserd
|| pausewr
)
3287 maybe_request_map(osdc
);
3289 kick_requests(osdc
, &need_resend
, &need_resend_linger
);
3291 ceph_monc_got_map(&osdc
->client
->monc
, CEPH_SUB_OSDMAP
,
3292 osdc
->osdmap
->epoch
);
3293 up_write(&osdc
->lock
);
3294 wake_up_all(&osdc
->client
->auth_wq
);
3298 pr_err("osdc handle_map corrupt msg\n");
3300 up_write(&osdc
->lock
);
3304 * Resubmit requests pending on the given osd.
3306 static void kick_osd_requests(struct ceph_osd
*osd
)
3310 for (n
= rb_first(&osd
->o_requests
); n
; ) {
3311 struct ceph_osd_request
*req
=
3312 rb_entry(n
, struct ceph_osd_request
, r_node
);
3314 n
= rb_next(n
); /* cancel_linger_request() */
3316 if (!req
->r_linger
) {
3317 if (!req
->r_t
.paused
)
3320 cancel_linger_request(req
);
3323 for (n
= rb_first(&osd
->o_linger_requests
); n
; n
= rb_next(n
)) {
3324 struct ceph_osd_linger_request
*lreq
=
3325 rb_entry(n
, struct ceph_osd_linger_request
, node
);
3332 * If the osd connection drops, we need to resubmit all requests.
3334 static void osd_fault(struct ceph_connection
*con
)
3336 struct ceph_osd
*osd
= con
->private;
3337 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
3339 dout("%s osd %p osd%d\n", __func__
, osd
, osd
->o_osd
);
3341 down_write(&osdc
->lock
);
3342 if (!osd_registered(osd
)) {
3343 dout("%s osd%d unknown\n", __func__
, osd
->o_osd
);
3347 if (!reopen_osd(osd
))
3348 kick_osd_requests(osd
);
3349 maybe_request_map(osdc
);
3352 up_write(&osdc
->lock
);
3356 * Process osd watch notifications
3358 static void handle_watch_notify(struct ceph_osd_client
*osdc
,
3359 struct ceph_msg
*msg
)
3361 void *p
= msg
->front
.iov_base
;
3362 void *const end
= p
+ msg
->front
.iov_len
;
3363 struct ceph_osd_linger_request
*lreq
;
3364 struct linger_work
*lwork
;
3365 u8 proto_ver
, opcode
;
3366 u64 cookie
, notify_id
;
3367 u64 notifier_id
= 0;
3368 s32 return_code
= 0;
3369 void *payload
= NULL
;
3370 u32 payload_len
= 0;
3372 ceph_decode_8_safe(&p
, end
, proto_ver
, bad
);
3373 ceph_decode_8_safe(&p
, end
, opcode
, bad
);
3374 ceph_decode_64_safe(&p
, end
, cookie
, bad
);
3375 p
+= 8; /* skip ver */
3376 ceph_decode_64_safe(&p
, end
, notify_id
, bad
);
3378 if (proto_ver
>= 1) {
3379 ceph_decode_32_safe(&p
, end
, payload_len
, bad
);
3380 ceph_decode_need(&p
, end
, payload_len
, bad
);
3385 if (le16_to_cpu(msg
->hdr
.version
) >= 2)
3386 ceph_decode_32_safe(&p
, end
, return_code
, bad
);
3388 if (le16_to_cpu(msg
->hdr
.version
) >= 3)
3389 ceph_decode_64_safe(&p
, end
, notifier_id
, bad
);
3391 down_read(&osdc
->lock
);
3392 lreq
= lookup_linger_osdc(&osdc
->linger_requests
, cookie
);
3394 dout("%s opcode %d cookie %llu dne\n", __func__
, opcode
,
3396 goto out_unlock_osdc
;
3399 mutex_lock(&lreq
->lock
);
3400 dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__
,
3401 opcode
, cookie
, lreq
, lreq
->is_watch
);
3402 if (opcode
== CEPH_WATCH_EVENT_DISCONNECT
) {
3403 if (!lreq
->last_error
) {
3404 lreq
->last_error
= -ENOTCONN
;
3405 queue_watch_error(lreq
);
3407 } else if (!lreq
->is_watch
) {
3408 /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */
3409 if (lreq
->notify_id
&& lreq
->notify_id
!= notify_id
) {
3410 dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq
,
3411 lreq
->notify_id
, notify_id
);
3412 } else if (!completion_done(&lreq
->notify_finish_wait
)) {
3413 struct ceph_msg_data
*data
=
3414 list_first_entry_or_null(&msg
->data
,
3415 struct ceph_msg_data
,
3419 if (lreq
->preply_pages
) {
3420 WARN_ON(data
->type
!=
3421 CEPH_MSG_DATA_PAGES
);
3422 *lreq
->preply_pages
= data
->pages
;
3423 *lreq
->preply_len
= data
->length
;
3425 ceph_release_page_vector(data
->pages
,
3426 calc_pages_for(0, data
->length
));
3429 lreq
->notify_finish_error
= return_code
;
3430 complete_all(&lreq
->notify_finish_wait
);
3433 /* CEPH_WATCH_EVENT_NOTIFY */
3434 lwork
= lwork_alloc(lreq
, do_watch_notify
);
3436 pr_err("failed to allocate notify-lwork\n");
3437 goto out_unlock_lreq
;
3440 lwork
->notify
.notify_id
= notify_id
;
3441 lwork
->notify
.notifier_id
= notifier_id
;
3442 lwork
->notify
.payload
= payload
;
3443 lwork
->notify
.payload_len
= payload_len
;
3444 lwork
->notify
.msg
= ceph_msg_get(msg
);
3449 mutex_unlock(&lreq
->lock
);
3451 up_read(&osdc
->lock
);
3455 pr_err("osdc handle_watch_notify corrupt msg\n");
3459 * Register request, send initial attempt.
3461 int ceph_osdc_start_request(struct ceph_osd_client
*osdc
,
3462 struct ceph_osd_request
*req
,
3465 down_read(&osdc
->lock
);
3466 submit_request(req
, false);
3467 up_read(&osdc
->lock
);
3471 EXPORT_SYMBOL(ceph_osdc_start_request
);
3474 * Unregister a registered request. The request is not completed (i.e.
3475 * no callbacks or wakeups) - higher layers are supposed to know what
3476 * they are canceling.
3478 void ceph_osdc_cancel_request(struct ceph_osd_request
*req
)
3480 struct ceph_osd_client
*osdc
= req
->r_osdc
;
3482 down_write(&osdc
->lock
);
3484 cancel_request(req
);
3485 up_write(&osdc
->lock
);
3487 EXPORT_SYMBOL(ceph_osdc_cancel_request
);
3490 * @timeout: in jiffies, 0 means "wait forever"
3492 static int wait_request_timeout(struct ceph_osd_request
*req
,
3493 unsigned long timeout
)
3497 dout("%s req %p tid %llu\n", __func__
, req
, req
->r_tid
);
3498 left
= wait_for_completion_killable_timeout(&req
->r_completion
,
3499 ceph_timeout_jiffies(timeout
));
3501 left
= left
?: -ETIMEDOUT
;
3502 ceph_osdc_cancel_request(req
);
3504 /* kludge - need to to wake ceph_osdc_sync() */
3505 complete_all(&req
->r_safe_completion
);
3507 left
= req
->r_result
; /* completed */
3514 * wait for a request to complete
3516 int ceph_osdc_wait_request(struct ceph_osd_client
*osdc
,
3517 struct ceph_osd_request
*req
)
3519 return wait_request_timeout(req
, 0);
3521 EXPORT_SYMBOL(ceph_osdc_wait_request
);
3524 * sync - wait for all in-flight requests to flush. avoid starvation.
3526 void ceph_osdc_sync(struct ceph_osd_client
*osdc
)
3528 struct rb_node
*n
, *p
;
3529 u64 last_tid
= atomic64_read(&osdc
->last_tid
);
3532 down_read(&osdc
->lock
);
3533 for (n
= rb_first(&osdc
->osds
); n
; n
= rb_next(n
)) {
3534 struct ceph_osd
*osd
= rb_entry(n
, struct ceph_osd
, o_node
);
3536 mutex_lock(&osd
->lock
);
3537 for (p
= rb_first(&osd
->o_requests
); p
; p
= rb_next(p
)) {
3538 struct ceph_osd_request
*req
=
3539 rb_entry(p
, struct ceph_osd_request
, r_node
);
3541 if (req
->r_tid
> last_tid
)
3544 if (!(req
->r_flags
& CEPH_OSD_FLAG_WRITE
))
3547 ceph_osdc_get_request(req
);
3548 mutex_unlock(&osd
->lock
);
3549 up_read(&osdc
->lock
);
3550 dout("%s waiting on req %p tid %llu last_tid %llu\n",
3551 __func__
, req
, req
->r_tid
, last_tid
);
3552 wait_for_completion(&req
->r_safe_completion
);
3553 ceph_osdc_put_request(req
);
3557 mutex_unlock(&osd
->lock
);
3560 up_read(&osdc
->lock
);
3561 dout("%s done last_tid %llu\n", __func__
, last_tid
);
3563 EXPORT_SYMBOL(ceph_osdc_sync
);
3565 static struct ceph_osd_request
*
3566 alloc_linger_request(struct ceph_osd_linger_request
*lreq
)
3568 struct ceph_osd_request
*req
;
3570 req
= ceph_osdc_alloc_request(lreq
->osdc
, NULL
, 1, false, GFP_NOIO
);
3574 ceph_oid_copy(&req
->r_base_oid
, &lreq
->t
.base_oid
);
3575 ceph_oloc_copy(&req
->r_base_oloc
, &lreq
->t
.base_oloc
);
3577 if (ceph_osdc_alloc_messages(req
, GFP_NOIO
)) {
3578 ceph_osdc_put_request(req
);
3586 * Returns a handle, caller owns a ref.
3588 struct ceph_osd_linger_request
*
3589 ceph_osdc_watch(struct ceph_osd_client
*osdc
,
3590 struct ceph_object_id
*oid
,
3591 struct ceph_object_locator
*oloc
,
3592 rados_watchcb2_t wcb
,
3593 rados_watcherrcb_t errcb
,
3596 struct ceph_osd_linger_request
*lreq
;
3599 lreq
= linger_alloc(osdc
);
3601 return ERR_PTR(-ENOMEM
);
3603 lreq
->is_watch
= true;
3605 lreq
->errcb
= errcb
;
3607 lreq
->watch_valid_thru
= jiffies
;
3609 ceph_oid_copy(&lreq
->t
.base_oid
, oid
);
3610 ceph_oloc_copy(&lreq
->t
.base_oloc
, oloc
);
3611 lreq
->t
.flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
3612 lreq
->mtime
= CURRENT_TIME
;
3614 lreq
->reg_req
= alloc_linger_request(lreq
);
3615 if (!lreq
->reg_req
) {
3620 lreq
->ping_req
= alloc_linger_request(lreq
);
3621 if (!lreq
->ping_req
) {
3626 down_write(&osdc
->lock
);
3627 linger_register(lreq
); /* before osd_req_op_* */
3628 osd_req_op_watch_init(lreq
->reg_req
, 0, lreq
->linger_id
,
3629 CEPH_OSD_WATCH_OP_WATCH
);
3630 osd_req_op_watch_init(lreq
->ping_req
, 0, lreq
->linger_id
,
3631 CEPH_OSD_WATCH_OP_PING
);
3632 linger_submit(lreq
);
3633 up_write(&osdc
->lock
);
3635 ret
= linger_reg_commit_wait(lreq
);
3637 linger_cancel(lreq
);
3645 return ERR_PTR(ret
);
3647 EXPORT_SYMBOL(ceph_osdc_watch
);
3652 * Times out after mount_timeout to preserve rbd unmap behaviour
3653 * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap
3654 * with mount_timeout").
3656 int ceph_osdc_unwatch(struct ceph_osd_client
*osdc
,
3657 struct ceph_osd_linger_request
*lreq
)
3659 struct ceph_options
*opts
= osdc
->client
->options
;
3660 struct ceph_osd_request
*req
;
3663 req
= ceph_osdc_alloc_request(osdc
, NULL
, 1, false, GFP_NOIO
);
3667 ceph_oid_copy(&req
->r_base_oid
, &lreq
->t
.base_oid
);
3668 ceph_oloc_copy(&req
->r_base_oloc
, &lreq
->t
.base_oloc
);
3669 req
->r_flags
= CEPH_OSD_FLAG_WRITE
| CEPH_OSD_FLAG_ONDISK
;
3670 req
->r_mtime
= CURRENT_TIME
;
3671 osd_req_op_watch_init(req
, 0, lreq
->linger_id
,
3672 CEPH_OSD_WATCH_OP_UNWATCH
);
3674 ret
= ceph_osdc_alloc_messages(req
, GFP_NOIO
);
3678 ceph_osdc_start_request(osdc
, req
, false);
3679 linger_cancel(lreq
);
3681 ret
= wait_request_timeout(req
, opts
->mount_timeout
);
3684 ceph_osdc_put_request(req
);
3687 EXPORT_SYMBOL(ceph_osdc_unwatch
);
3689 static int osd_req_op_notify_ack_init(struct ceph_osd_request
*req
, int which
,
3690 u64 notify_id
, u64 cookie
, void *payload
,
3693 struct ceph_osd_req_op
*op
;
3694 struct ceph_pagelist
*pl
;
3697 op
= _osd_req_op_init(req
, which
, CEPH_OSD_OP_NOTIFY_ACK
, 0);
3699 pl
= kmalloc(sizeof(*pl
), GFP_NOIO
);
3703 ceph_pagelist_init(pl
);
3704 ret
= ceph_pagelist_encode_64(pl
, notify_id
);
3705 ret
|= ceph_pagelist_encode_64(pl
, cookie
);
3707 ret
|= ceph_pagelist_encode_32(pl
, payload_len
);
3708 ret
|= ceph_pagelist_append(pl
, payload
, payload_len
);
3710 ret
|= ceph_pagelist_encode_32(pl
, 0);
3713 ceph_pagelist_release(pl
);
3717 ceph_osd_data_pagelist_init(&op
->notify_ack
.request_data
, pl
);
3718 op
->indata_len
= pl
->length
;
3722 int ceph_osdc_notify_ack(struct ceph_osd_client
*osdc
,
3723 struct ceph_object_id
*oid
,
3724 struct ceph_object_locator
*oloc
,
3730 struct ceph_osd_request
*req
;
3733 req
= ceph_osdc_alloc_request(osdc
, NULL
, 1, false, GFP_NOIO
);
3737 ceph_oid_copy(&req
->r_base_oid
, oid
);
3738 ceph_oloc_copy(&req
->r_base_oloc
, oloc
);
3739 req
->r_flags
= CEPH_OSD_FLAG_READ
;
3741 ret
= ceph_osdc_alloc_messages(req
, GFP_NOIO
);
3745 ret
= osd_req_op_notify_ack_init(req
, 0, notify_id
, cookie
, payload
,
3750 ceph_osdc_start_request(osdc
, req
, false);
3751 ret
= ceph_osdc_wait_request(osdc
, req
);
3754 ceph_osdc_put_request(req
);
3757 EXPORT_SYMBOL(ceph_osdc_notify_ack
);
3759 static int osd_req_op_notify_init(struct ceph_osd_request
*req
, int which
,
3760 u64 cookie
, u32 prot_ver
, u32 timeout
,
3761 void *payload
, size_t payload_len
)
3763 struct ceph_osd_req_op
*op
;
3764 struct ceph_pagelist
*pl
;
3767 op
= _osd_req_op_init(req
, which
, CEPH_OSD_OP_NOTIFY
, 0);
3768 op
->notify
.cookie
= cookie
;
3770 pl
= kmalloc(sizeof(*pl
), GFP_NOIO
);
3774 ceph_pagelist_init(pl
);
3775 ret
= ceph_pagelist_encode_32(pl
, 1); /* prot_ver */
3776 ret
|= ceph_pagelist_encode_32(pl
, timeout
);
3777 ret
|= ceph_pagelist_encode_32(pl
, payload_len
);
3778 ret
|= ceph_pagelist_append(pl
, payload
, payload_len
);
3780 ceph_pagelist_release(pl
);
3784 ceph_osd_data_pagelist_init(&op
->notify
.request_data
, pl
);
3785 op
->indata_len
= pl
->length
;
3790 * @timeout: in seconds
3792 * @preply_{pages,len} are initialized both on success and error.
3793 * The caller is responsible for:
3795 * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len))
3797 int ceph_osdc_notify(struct ceph_osd_client
*osdc
,
3798 struct ceph_object_id
*oid
,
3799 struct ceph_object_locator
*oloc
,
3803 struct page
***preply_pages
,
3806 struct ceph_osd_linger_request
*lreq
;
3807 struct page
**pages
;
3812 *preply_pages
= NULL
;
3816 lreq
= linger_alloc(osdc
);
3820 lreq
->preply_pages
= preply_pages
;
3821 lreq
->preply_len
= preply_len
;
3823 ceph_oid_copy(&lreq
->t
.base_oid
, oid
);
3824 ceph_oloc_copy(&lreq
->t
.base_oloc
, oloc
);
3825 lreq
->t
.flags
= CEPH_OSD_FLAG_READ
;
3827 lreq
->reg_req
= alloc_linger_request(lreq
);
3828 if (!lreq
->reg_req
) {
3834 pages
= ceph_alloc_page_vector(1, GFP_NOIO
);
3835 if (IS_ERR(pages
)) {
3836 ret
= PTR_ERR(pages
);
3840 down_write(&osdc
->lock
);
3841 linger_register(lreq
); /* before osd_req_op_* */
3842 ret
= osd_req_op_notify_init(lreq
->reg_req
, 0, lreq
->linger_id
, 1,
3843 timeout
, payload
, payload_len
);
3845 linger_unregister(lreq
);
3846 up_write(&osdc
->lock
);
3847 ceph_release_page_vector(pages
, 1);
3850 ceph_osd_data_pages_init(osd_req_op_data(lreq
->reg_req
, 0, notify
,
3852 pages
, PAGE_SIZE
, 0, false, true);
3853 linger_submit(lreq
);
3854 up_write(&osdc
->lock
);
3856 ret
= linger_reg_commit_wait(lreq
);
3858 ret
= linger_notify_finish_wait(lreq
);
3860 dout("lreq %p failed to initiate notify %d\n", lreq
, ret
);
3862 linger_cancel(lreq
);
3867 EXPORT_SYMBOL(ceph_osdc_notify
);
3870 * Return the number of milliseconds since the watch was last
3871 * confirmed, or an error. If there is an error, the watch is no
3872 * longer valid, and should be destroyed with ceph_osdc_unwatch().
3874 int ceph_osdc_watch_check(struct ceph_osd_client
*osdc
,
3875 struct ceph_osd_linger_request
*lreq
)
3877 unsigned long stamp
, age
;
3880 down_read(&osdc
->lock
);
3881 mutex_lock(&lreq
->lock
);
3882 stamp
= lreq
->watch_valid_thru
;
3883 if (!list_empty(&lreq
->pending_lworks
)) {
3884 struct linger_work
*lwork
=
3885 list_first_entry(&lreq
->pending_lworks
,
3889 if (time_before(lwork
->queued_stamp
, stamp
))
3890 stamp
= lwork
->queued_stamp
;
3892 age
= jiffies
- stamp
;
3893 dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__
,
3894 lreq
, lreq
->linger_id
, age
, lreq
->last_error
);
3895 /* we are truncating to msecs, so return a safe upper bound */
3896 ret
= lreq
->last_error
?: 1 + jiffies_to_msecs(age
);
3898 mutex_unlock(&lreq
->lock
);
3899 up_read(&osdc
->lock
);
3903 static int decode_watcher(void **p
, void *end
, struct ceph_watch_item
*item
)
3909 ret
= ceph_start_decoding(p
, end
, 2, "watch_item_t",
3910 &struct_v
, &struct_len
);
3914 ceph_decode_copy(p
, &item
->name
, sizeof(item
->name
));
3915 item
->cookie
= ceph_decode_64(p
);
3916 *p
+= 4; /* skip timeout_seconds */
3917 if (struct_v
>= 2) {
3918 ceph_decode_copy(p
, &item
->addr
, sizeof(item
->addr
));
3919 ceph_decode_addr(&item
->addr
);
3922 dout("%s %s%llu cookie %llu addr %s\n", __func__
,
3923 ENTITY_NAME(item
->name
), item
->cookie
,
3924 ceph_pr_addr(&item
->addr
.in_addr
));
3928 static int decode_watchers(void **p
, void *end
,
3929 struct ceph_watch_item
**watchers
,
3937 ret
= ceph_start_decoding(p
, end
, 1, "obj_list_watch_response_t",
3938 &struct_v
, &struct_len
);
3942 *num_watchers
= ceph_decode_32(p
);
3943 *watchers
= kcalloc(*num_watchers
, sizeof(**watchers
), GFP_NOIO
);
3947 for (i
= 0; i
< *num_watchers
; i
++) {
3948 ret
= decode_watcher(p
, end
, *watchers
+ i
);
3959 * On success, the caller is responsible for:
3963 int ceph_osdc_list_watchers(struct ceph_osd_client
*osdc
,
3964 struct ceph_object_id
*oid
,
3965 struct ceph_object_locator
*oloc
,
3966 struct ceph_watch_item
**watchers
,
3969 struct ceph_osd_request
*req
;
3970 struct page
**pages
;
3973 req
= ceph_osdc_alloc_request(osdc
, NULL
, 1, false, GFP_NOIO
);
3977 ceph_oid_copy(&req
->r_base_oid
, oid
);
3978 ceph_oloc_copy(&req
->r_base_oloc
, oloc
);
3979 req
->r_flags
= CEPH_OSD_FLAG_READ
;
3981 ret
= ceph_osdc_alloc_messages(req
, GFP_NOIO
);
3985 pages
= ceph_alloc_page_vector(1, GFP_NOIO
);
3986 if (IS_ERR(pages
)) {
3987 ret
= PTR_ERR(pages
);
3991 osd_req_op_init(req
, 0, CEPH_OSD_OP_LIST_WATCHERS
, 0);
3992 ceph_osd_data_pages_init(osd_req_op_data(req
, 0, list_watchers
,
3994 pages
, PAGE_SIZE
, 0, false, true);
3996 ceph_osdc_start_request(osdc
, req
, false);
3997 ret
= ceph_osdc_wait_request(osdc
, req
);
3999 void *p
= page_address(pages
[0]);
4000 void *const end
= p
+ req
->r_ops
[0].outdata_len
;
4002 ret
= decode_watchers(&p
, end
, watchers
, num_watchers
);
4006 ceph_osdc_put_request(req
);
4009 EXPORT_SYMBOL(ceph_osdc_list_watchers
);
4012 * Call all pending notify callbacks - for use after a watch is
4013 * unregistered, to make sure no more callbacks for it will be invoked
4015 void ceph_osdc_flush_notifies(struct ceph_osd_client
*osdc
)
4017 dout("%s osdc %p\n", __func__
, osdc
);
4018 flush_workqueue(osdc
->notify_wq
);
4020 EXPORT_SYMBOL(ceph_osdc_flush_notifies
);
4022 void ceph_osdc_maybe_request_map(struct ceph_osd_client
*osdc
)
4024 down_read(&osdc
->lock
);
4025 maybe_request_map(osdc
);
4026 up_read(&osdc
->lock
);
4028 EXPORT_SYMBOL(ceph_osdc_maybe_request_map
);
4031 * Execute an OSD class method on an object.
4033 * @flags: CEPH_OSD_FLAG_*
4034 * @resp_len: out param for reply length
4036 int ceph_osdc_call(struct ceph_osd_client
*osdc
,
4037 struct ceph_object_id
*oid
,
4038 struct ceph_object_locator
*oloc
,
4039 const char *class, const char *method
,
4041 struct page
*req_page
, size_t req_len
,
4042 struct page
*resp_page
, size_t *resp_len
)
4044 struct ceph_osd_request
*req
;
4047 req
= ceph_osdc_alloc_request(osdc
, NULL
, 1, false, GFP_NOIO
);
4051 ceph_oid_copy(&req
->r_base_oid
, oid
);
4052 ceph_oloc_copy(&req
->r_base_oloc
, oloc
);
4053 req
->r_flags
= flags
;
4055 ret
= ceph_osdc_alloc_messages(req
, GFP_NOIO
);
4059 osd_req_op_cls_init(req
, 0, CEPH_OSD_OP_CALL
, class, method
);
4061 osd_req_op_cls_request_data_pages(req
, 0, &req_page
, req_len
,
4064 osd_req_op_cls_response_data_pages(req
, 0, &resp_page
,
4065 PAGE_SIZE
, 0, false, false);
4067 ceph_osdc_start_request(osdc
, req
, false);
4068 ret
= ceph_osdc_wait_request(osdc
, req
);
4070 ret
= req
->r_ops
[0].rval
;
4072 *resp_len
= req
->r_ops
[0].outdata_len
;
4076 ceph_osdc_put_request(req
);
4079 EXPORT_SYMBOL(ceph_osdc_call
);
4084 int ceph_osdc_init(struct ceph_osd_client
*osdc
, struct ceph_client
*client
)
4089 osdc
->client
= client
;
4090 init_rwsem(&osdc
->lock
);
4091 osdc
->osds
= RB_ROOT
;
4092 INIT_LIST_HEAD(&osdc
->osd_lru
);
4093 spin_lock_init(&osdc
->osd_lru_lock
);
4094 osd_init(&osdc
->homeless_osd
);
4095 osdc
->homeless_osd
.o_osdc
= osdc
;
4096 osdc
->homeless_osd
.o_osd
= CEPH_HOMELESS_OSD
;
4097 osdc
->linger_requests
= RB_ROOT
;
4098 osdc
->map_checks
= RB_ROOT
;
4099 osdc
->linger_map_checks
= RB_ROOT
;
4100 INIT_DELAYED_WORK(&osdc
->timeout_work
, handle_timeout
);
4101 INIT_DELAYED_WORK(&osdc
->osds_timeout_work
, handle_osds_timeout
);
4104 osdc
->osdmap
= ceph_osdmap_alloc();
4108 osdc
->req_mempool
= mempool_create_slab_pool(10,
4109 ceph_osd_request_cache
);
4110 if (!osdc
->req_mempool
)
4113 err
= ceph_msgpool_init(&osdc
->msgpool_op
, CEPH_MSG_OSD_OP
,
4114 PAGE_SIZE
, 10, true, "osd_op");
4117 err
= ceph_msgpool_init(&osdc
->msgpool_op_reply
, CEPH_MSG_OSD_OPREPLY
,
4118 PAGE_SIZE
, 10, true, "osd_op_reply");
4123 osdc
->notify_wq
= create_singlethread_workqueue("ceph-watch-notify");
4124 if (!osdc
->notify_wq
)
4125 goto out_msgpool_reply
;
4127 schedule_delayed_work(&osdc
->timeout_work
,
4128 osdc
->client
->options
->osd_keepalive_timeout
);
4129 schedule_delayed_work(&osdc
->osds_timeout_work
,
4130 round_jiffies_relative(osdc
->client
->options
->osd_idle_ttl
));
4135 ceph_msgpool_destroy(&osdc
->msgpool_op_reply
);
4137 ceph_msgpool_destroy(&osdc
->msgpool_op
);
4139 mempool_destroy(osdc
->req_mempool
);
4141 ceph_osdmap_destroy(osdc
->osdmap
);
4146 void ceph_osdc_stop(struct ceph_osd_client
*osdc
)
4148 flush_workqueue(osdc
->notify_wq
);
4149 destroy_workqueue(osdc
->notify_wq
);
4150 cancel_delayed_work_sync(&osdc
->timeout_work
);
4151 cancel_delayed_work_sync(&osdc
->osds_timeout_work
);
4153 down_write(&osdc
->lock
);
4154 while (!RB_EMPTY_ROOT(&osdc
->osds
)) {
4155 struct ceph_osd
*osd
= rb_entry(rb_first(&osdc
->osds
),
4156 struct ceph_osd
, o_node
);
4159 up_write(&osdc
->lock
);
4160 WARN_ON(atomic_read(&osdc
->homeless_osd
.o_ref
) != 1);
4161 osd_cleanup(&osdc
->homeless_osd
);
4163 WARN_ON(!list_empty(&osdc
->osd_lru
));
4164 WARN_ON(!RB_EMPTY_ROOT(&osdc
->linger_requests
));
4165 WARN_ON(!RB_EMPTY_ROOT(&osdc
->map_checks
));
4166 WARN_ON(!RB_EMPTY_ROOT(&osdc
->linger_map_checks
));
4167 WARN_ON(atomic_read(&osdc
->num_requests
));
4168 WARN_ON(atomic_read(&osdc
->num_homeless
));
4170 ceph_osdmap_destroy(osdc
->osdmap
);
4171 mempool_destroy(osdc
->req_mempool
);
4172 ceph_msgpool_destroy(&osdc
->msgpool_op
);
4173 ceph_msgpool_destroy(&osdc
->msgpool_op_reply
);
4177 * Read some contiguous pages. If we cross a stripe boundary, shorten
4178 * *plen. Return number of bytes read, or error.
4180 int ceph_osdc_readpages(struct ceph_osd_client
*osdc
,
4181 struct ceph_vino vino
, struct ceph_file_layout
*layout
,
4183 u32 truncate_seq
, u64 truncate_size
,
4184 struct page
**pages
, int num_pages
, int page_align
)
4186 struct ceph_osd_request
*req
;
4189 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino
.ino
,
4190 vino
.snap
, off
, *plen
);
4191 req
= ceph_osdc_new_request(osdc
, layout
, vino
, off
, plen
, 0, 1,
4192 CEPH_OSD_OP_READ
, CEPH_OSD_FLAG_READ
,
4193 NULL
, truncate_seq
, truncate_size
,
4196 return PTR_ERR(req
);
4198 /* it may be a short read due to an object boundary */
4199 osd_req_op_extent_osd_data_pages(req
, 0,
4200 pages
, *plen
, page_align
, false, false);
4202 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
4203 off
, *plen
, *plen
, page_align
);
4205 rc
= ceph_osdc_start_request(osdc
, req
, false);
4207 rc
= ceph_osdc_wait_request(osdc
, req
);
4209 ceph_osdc_put_request(req
);
4210 dout("readpages result %d\n", rc
);
4213 EXPORT_SYMBOL(ceph_osdc_readpages
);
4216 * do a synchronous write on N pages
4218 int ceph_osdc_writepages(struct ceph_osd_client
*osdc
, struct ceph_vino vino
,
4219 struct ceph_file_layout
*layout
,
4220 struct ceph_snap_context
*snapc
,
4222 u32 truncate_seq
, u64 truncate_size
,
4223 struct timespec
*mtime
,
4224 struct page
**pages
, int num_pages
)
4226 struct ceph_osd_request
*req
;
4228 int page_align
= off
& ~PAGE_MASK
;
4230 req
= ceph_osdc_new_request(osdc
, layout
, vino
, off
, &len
, 0, 1,
4232 CEPH_OSD_FLAG_ONDISK
| CEPH_OSD_FLAG_WRITE
,
4233 snapc
, truncate_seq
, truncate_size
,
4236 return PTR_ERR(req
);
4238 /* it may be a short write due to an object boundary */
4239 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, page_align
,
4241 dout("writepages %llu~%llu (%llu bytes)\n", off
, len
, len
);
4243 req
->r_mtime
= *mtime
;
4244 rc
= ceph_osdc_start_request(osdc
, req
, true);
4246 rc
= ceph_osdc_wait_request(osdc
, req
);
4248 ceph_osdc_put_request(req
);
4251 dout("writepages result %d\n", rc
);
4254 EXPORT_SYMBOL(ceph_osdc_writepages
);
4256 int ceph_osdc_setup(void)
4258 size_t size
= sizeof(struct ceph_osd_request
) +
4259 CEPH_OSD_SLAB_OPS
* sizeof(struct ceph_osd_req_op
);
4261 BUG_ON(ceph_osd_request_cache
);
4262 ceph_osd_request_cache
= kmem_cache_create("ceph_osd_request", size
,
4265 return ceph_osd_request_cache
? 0 : -ENOMEM
;
4267 EXPORT_SYMBOL(ceph_osdc_setup
);
4269 void ceph_osdc_cleanup(void)
4271 BUG_ON(!ceph_osd_request_cache
);
4272 kmem_cache_destroy(ceph_osd_request_cache
);
4273 ceph_osd_request_cache
= NULL
;
4275 EXPORT_SYMBOL(ceph_osdc_cleanup
);
4278 * handle incoming message
4280 static void dispatch(struct ceph_connection
*con
, struct ceph_msg
*msg
)
4282 struct ceph_osd
*osd
= con
->private;
4283 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
4284 int type
= le16_to_cpu(msg
->hdr
.type
);
4287 case CEPH_MSG_OSD_MAP
:
4288 ceph_osdc_handle_map(osdc
, msg
);
4290 case CEPH_MSG_OSD_OPREPLY
:
4291 handle_reply(osd
, msg
);
4293 case CEPH_MSG_WATCH_NOTIFY
:
4294 handle_watch_notify(osdc
, msg
);
4298 pr_err("received unknown message type %d %s\n", type
,
4299 ceph_msg_type_name(type
));
4306 * Lookup and return message for incoming reply. Don't try to do
4307 * anything about a larger than preallocated data portion of the
4308 * message at the moment - for now, just skip the message.
4310 static struct ceph_msg
*get_reply(struct ceph_connection
*con
,
4311 struct ceph_msg_header
*hdr
,
4314 struct ceph_osd
*osd
= con
->private;
4315 struct ceph_osd_client
*osdc
= osd
->o_osdc
;
4316 struct ceph_msg
*m
= NULL
;
4317 struct ceph_osd_request
*req
;
4318 int front_len
= le32_to_cpu(hdr
->front_len
);
4319 int data_len
= le32_to_cpu(hdr
->data_len
);
4320 u64 tid
= le64_to_cpu(hdr
->tid
);
4322 down_read(&osdc
->lock
);
4323 if (!osd_registered(osd
)) {
4324 dout("%s osd%d unknown, skipping\n", __func__
, osd
->o_osd
);
4326 goto out_unlock_osdc
;
4328 WARN_ON(osd
->o_osd
!= le64_to_cpu(hdr
->src
.num
));
4330 mutex_lock(&osd
->lock
);
4331 req
= lookup_request(&osd
->o_requests
, tid
);
4333 dout("%s osd%d tid %llu unknown, skipping\n", __func__
,
4336 goto out_unlock_session
;
4339 ceph_msg_revoke_incoming(req
->r_reply
);
4341 if (front_len
> req
->r_reply
->front_alloc_len
) {
4342 pr_warn("%s osd%d tid %llu front %d > preallocated %d\n",
4343 __func__
, osd
->o_osd
, req
->r_tid
, front_len
,
4344 req
->r_reply
->front_alloc_len
);
4345 m
= ceph_msg_new(CEPH_MSG_OSD_OPREPLY
, front_len
, GFP_NOFS
,
4348 goto out_unlock_session
;
4349 ceph_msg_put(req
->r_reply
);
4353 if (data_len
> req
->r_reply
->data_length
) {
4354 pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n",
4355 __func__
, osd
->o_osd
, req
->r_tid
, data_len
,
4356 req
->r_reply
->data_length
);
4359 goto out_unlock_session
;
4362 m
= ceph_msg_get(req
->r_reply
);
4363 dout("get_reply tid %lld %p\n", tid
, m
);
4366 mutex_unlock(&osd
->lock
);
4368 up_read(&osdc
->lock
);
4373 * TODO: switch to a msg-owned pagelist
4375 static struct ceph_msg
*alloc_msg_with_page_vector(struct ceph_msg_header
*hdr
)
4378 int type
= le16_to_cpu(hdr
->type
);
4379 u32 front_len
= le32_to_cpu(hdr
->front_len
);
4380 u32 data_len
= le32_to_cpu(hdr
->data_len
);
4382 m
= ceph_msg_new(type
, front_len
, GFP_NOIO
, false);
4387 struct page
**pages
;
4388 struct ceph_osd_data osd_data
;
4390 pages
= ceph_alloc_page_vector(calc_pages_for(0, data_len
),
4392 if (IS_ERR(pages
)) {
4397 ceph_osd_data_pages_init(&osd_data
, pages
, data_len
, 0, false,
4399 ceph_osdc_msg_data_add(m
, &osd_data
);
4405 static struct ceph_msg
*alloc_msg(struct ceph_connection
*con
,
4406 struct ceph_msg_header
*hdr
,
4409 struct ceph_osd
*osd
= con
->private;
4410 int type
= le16_to_cpu(hdr
->type
);
4414 case CEPH_MSG_OSD_MAP
:
4415 case CEPH_MSG_WATCH_NOTIFY
:
4416 return alloc_msg_with_page_vector(hdr
);
4417 case CEPH_MSG_OSD_OPREPLY
:
4418 return get_reply(con
, hdr
, skip
);
4420 pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__
,
4428 * Wrappers to refcount containing ceph_osd struct
4430 static struct ceph_connection
*get_osd_con(struct ceph_connection
*con
)
4432 struct ceph_osd
*osd
= con
->private;
4438 static void put_osd_con(struct ceph_connection
*con
)
4440 struct ceph_osd
*osd
= con
->private;
4448 * Note: returned pointer is the address of a structure that's
4449 * managed separately. Caller must *not* attempt to free it.
4451 static struct ceph_auth_handshake
*get_authorizer(struct ceph_connection
*con
,
4452 int *proto
, int force_new
)
4454 struct ceph_osd
*o
= con
->private;
4455 struct ceph_osd_client
*osdc
= o
->o_osdc
;
4456 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
4457 struct ceph_auth_handshake
*auth
= &o
->o_auth
;
4459 if (force_new
&& auth
->authorizer
) {
4460 ceph_auth_destroy_authorizer(auth
->authorizer
);
4461 auth
->authorizer
= NULL
;
4463 if (!auth
->authorizer
) {
4464 int ret
= ceph_auth_create_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
,
4467 return ERR_PTR(ret
);
4469 int ret
= ceph_auth_update_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
,
4472 return ERR_PTR(ret
);
4474 *proto
= ac
->protocol
;
4480 static int verify_authorizer_reply(struct ceph_connection
*con
, int len
)
4482 struct ceph_osd
*o
= con
->private;
4483 struct ceph_osd_client
*osdc
= o
->o_osdc
;
4484 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
4486 return ceph_auth_verify_authorizer_reply(ac
, o
->o_auth
.authorizer
, len
);
4489 static int invalidate_authorizer(struct ceph_connection
*con
)
4491 struct ceph_osd
*o
= con
->private;
4492 struct ceph_osd_client
*osdc
= o
->o_osdc
;
4493 struct ceph_auth_client
*ac
= osdc
->client
->monc
.auth
;
4495 ceph_auth_invalidate_authorizer(ac
, CEPH_ENTITY_TYPE_OSD
);
4496 return ceph_monc_validate_auth(&osdc
->client
->monc
);
4499 static int osd_sign_message(struct ceph_msg
*msg
)
4501 struct ceph_osd
*o
= msg
->con
->private;
4502 struct ceph_auth_handshake
*auth
= &o
->o_auth
;
4504 return ceph_auth_sign_message(auth
, msg
);
4507 static int osd_check_message_signature(struct ceph_msg
*msg
)
4509 struct ceph_osd
*o
= msg
->con
->private;
4510 struct ceph_auth_handshake
*auth
= &o
->o_auth
;
4512 return ceph_auth_check_message_signature(auth
, msg
);
4515 static const struct ceph_connection_operations osd_con_ops
= {
4518 .dispatch
= dispatch
,
4519 .get_authorizer
= get_authorizer
,
4520 .verify_authorizer_reply
= verify_authorizer_reply
,
4521 .invalidate_authorizer
= invalidate_authorizer
,
4522 .alloc_msg
= alloc_msg
,
4523 .sign_message
= osd_sign_message
,
4524 .check_message_signature
= osd_check_message_signature
,