blob: 61147fe701813e7e0c0d485af3217a61eb320e0b [file] [log] [blame]
Alex Eldera4ce40a2013-04-05 01:27:12 -05001
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002#include <linux/ceph/ceph_debug.h>
Sage Weilf24e9982009-10-06 11:31:10 -07003
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07004#include <linux/module.h>
Sage Weilf24e9982009-10-06 11:31:10 -07005#include <linux/err.h>
6#include <linux/highmem.h>
7#include <linux/mm.h>
8#include <linux/pagemap.h>
9#include <linux/slab.h>
10#include <linux/uaccess.h>
Yehuda Sadeh68b44762010-04-06 15:01:27 -070011#ifdef CONFIG_BLOCK
12#include <linux/bio.h>
13#endif
Sage Weilf24e9982009-10-06 11:31:10 -070014
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070015#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>
Sage Weilf24e9982009-10-06 11:31:10 -070021
Sage Weilc16e7862010-03-01 13:02:00 -080022#define OSD_OP_FRONT_LEN 4096
23#define OSD_OPREPLY_FRONT_LEN 512
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -080024
Alex Elder5522ae02013-05-01 12:43:04 -050025static struct kmem_cache *ceph_osd_request_cache;
26
Tobias Klauser9e327892010-05-20 10:40:19 +020027static const struct ceph_connection_operations osd_con_ops;
Sage Weilf24e9982009-10-06 11:31:10 -070028
Alex Elderf9d25192013-02-15 11:42:29 -060029static void __send_queued(struct ceph_osd_client *osdc);
Sage Weil6f6c7002011-01-17 20:34:08 -080030static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -070031static void __register_request(struct ceph_osd_client *osdc,
32 struct ceph_osd_request *req);
33static void __unregister_linger_request(struct ceph_osd_client *osdc,
34 struct ceph_osd_request *req);
Sage Weil56e925b2012-01-03 12:34:34 -080035static void __send_request(struct ceph_osd_client *osdc,
36 struct ceph_osd_request *req);
Sage Weilf24e9982009-10-06 11:31:10 -070037
38/*
39 * Implement client access to distributed object storage cluster.
40 *
41 * All data objects are stored within a cluster/cloud of OSDs, or
42 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
43 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
44 * remote daemons serving up and coordinating consistent and safe
45 * access to storage.
46 *
47 * Cluster membership and the mapping of data objects onto storage devices
48 * are described by the osd map.
49 *
50 * We keep track of pending OSD requests (read, write), resubmit
51 * requests to different OSDs when the cluster topology/data layout
52 * change, or retry the affected requests when the communications
53 * channel with an OSD is reset.
54 */
55
56/*
57 * calculate the mapping of a file extent onto an object, and fill out the
58 * request accordingly. shorten extent as necessary if it crosses an
59 * object boundary.
60 *
61 * fill osd op in request message.
62 */
Alex Elderdbe0fc42013-02-15 22:10:17 -060063static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
Alex Eldera19dadf2013-03-13 20:50:01 -050064 u64 *objnum, u64 *objoff, u64 *objlen)
Sage Weilf24e9982009-10-06 11:31:10 -070065{
Alex Elder60e56f12013-02-15 11:42:29 -060066 u64 orig_len = *plen;
Sage Weild63b77f2012-09-24 20:59:48 -070067 int r;
Sage Weilf24e9982009-10-06 11:31:10 -070068
Alex Elder60e56f12013-02-15 11:42:29 -060069 /* object extent? */
Alex Elder75d1c942013-03-13 20:50:00 -050070 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
71 objoff, objlen);
Sage Weild63b77f2012-09-24 20:59:48 -070072 if (r < 0)
73 return r;
Alex Elder75d1c942013-03-13 20:50:00 -050074 if (*objlen < orig_len) {
75 *plen = *objlen;
Alex Elder60e56f12013-02-15 11:42:29 -060076 dout(" skipping last %llu, final file extent %llu~%llu\n",
77 orig_len - *plen, off, *plen);
78 }
79
Alex Elder75d1c942013-03-13 20:50:00 -050080 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
Sage Weilf24e9982009-10-06 11:31:10 -070081
Alex Elder3ff5f382013-02-15 22:10:17 -060082 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -070083}
84
Alex Elderc54d47b2013-04-03 01:28:57 -050085static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
86{
87 memset(osd_data, 0, sizeof (*osd_data));
88 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
89}
90
Alex Eldera4ce40a2013-04-05 01:27:12 -050091static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -050092 struct page **pages, u64 length, u32 alignment,
93 bool pages_from_pool, bool own_pages)
94{
95 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
96 osd_data->pages = pages;
97 osd_data->length = length;
98 osd_data->alignment = alignment;
99 osd_data->pages_from_pool = pages_from_pool;
100 osd_data->own_pages = own_pages;
101}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500102
Alex Eldera4ce40a2013-04-05 01:27:12 -0500103static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -0500104 struct ceph_pagelist *pagelist)
105{
106 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
107 osd_data->pagelist = pagelist;
108}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500109
110#ifdef CONFIG_BLOCK
Alex Eldera4ce40a2013-04-05 01:27:12 -0500111static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -0500112 struct bio *bio, size_t bio_length)
113{
114 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
115 osd_data->bio = bio;
116 osd_data->bio_length = bio_length;
117}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500118#endif /* CONFIG_BLOCK */
119
Alex Elder863c7eb2013-04-15 14:50:36 -0500120#define osd_req_op_data(oreq, whch, typ, fld) \
121 ({ \
122 BUG_ON(whch >= (oreq)->r_num_ops); \
123 &(oreq)->r_ops[whch].typ.fld; \
124 })
125
Alex Elder49719772013-02-11 12:33:24 -0600126static struct ceph_osd_data *
127osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
128{
129 BUG_ON(which >= osd_req->r_num_ops);
130
131 return &osd_req->r_ops[which].raw_data_in;
132}
133
Alex Eldera4ce40a2013-04-05 01:27:12 -0500134struct ceph_osd_data *
135osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500136 unsigned int which)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500137{
Alex Elder863c7eb2013-04-15 14:50:36 -0500138 return osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500139}
140EXPORT_SYMBOL(osd_req_op_extent_osd_data);
141
142struct ceph_osd_data *
Alex Eldera4ce40a2013-04-05 01:27:12 -0500143osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
144 unsigned int which)
145{
Alex Elder863c7eb2013-04-15 14:50:36 -0500146 return osd_req_op_data(osd_req, which, cls, response_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500147}
148EXPORT_SYMBOL(osd_req_op_cls_response_data); /* ??? */
149
Alex Elder49719772013-02-11 12:33:24 -0600150void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
151 unsigned int which, struct page **pages,
152 u64 length, u32 alignment,
153 bool pages_from_pool, bool own_pages)
154{
155 struct ceph_osd_data *osd_data;
156
157 osd_data = osd_req_op_raw_data_in(osd_req, which);
158 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
159 pages_from_pool, own_pages);
160}
161EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
162
Alex Eldera4ce40a2013-04-05 01:27:12 -0500163void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500164 unsigned int which, struct page **pages,
165 u64 length, u32 alignment,
Alex Eldera4ce40a2013-04-05 01:27:12 -0500166 bool pages_from_pool, bool own_pages)
167{
168 struct ceph_osd_data *osd_data;
169
Alex Elder863c7eb2013-04-15 14:50:36 -0500170 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500171 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
172 pages_from_pool, own_pages);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500173}
174EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
175
176void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500177 unsigned int which, struct ceph_pagelist *pagelist)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500178{
179 struct ceph_osd_data *osd_data;
180
Alex Elder863c7eb2013-04-15 14:50:36 -0500181 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500182 ceph_osd_data_pagelist_init(osd_data, pagelist);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500183}
184EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
185
186#ifdef CONFIG_BLOCK
187void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500188 unsigned int which, struct bio *bio, size_t bio_length)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500189{
190 struct ceph_osd_data *osd_data;
Alex Elder863c7eb2013-04-15 14:50:36 -0500191
192 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500193 ceph_osd_data_bio_init(osd_data, bio, bio_length);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500194}
195EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
196#endif /* CONFIG_BLOCK */
197
198static void osd_req_op_cls_request_info_pagelist(
199 struct ceph_osd_request *osd_req,
200 unsigned int which, struct ceph_pagelist *pagelist)
201{
202 struct ceph_osd_data *osd_data;
203
Alex Elder863c7eb2013-04-15 14:50:36 -0500204 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500205 ceph_osd_data_pagelist_init(osd_data, pagelist);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500206}
207
Alex Elder04017e22013-04-05 14:46:02 -0500208void osd_req_op_cls_request_data_pagelist(
209 struct ceph_osd_request *osd_req,
210 unsigned int which, struct ceph_pagelist *pagelist)
211{
212 struct ceph_osd_data *osd_data;
213
Alex Elder863c7eb2013-04-15 14:50:36 -0500214 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
Alex Elder04017e22013-04-05 14:46:02 -0500215 ceph_osd_data_pagelist_init(osd_data, pagelist);
216}
217EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
218
Alex Elder6c57b552013-04-19 15:34:49 -0500219void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
220 unsigned int which, struct page **pages, u64 length,
221 u32 alignment, bool pages_from_pool, bool own_pages)
222{
223 struct ceph_osd_data *osd_data;
224
225 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
226 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
227 pages_from_pool, own_pages);
228}
229EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
230
Alex Eldera4ce40a2013-04-05 01:27:12 -0500231void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
232 unsigned int which, struct page **pages, u64 length,
233 u32 alignment, bool pages_from_pool, bool own_pages)
234{
235 struct ceph_osd_data *osd_data;
236
Alex Elder863c7eb2013-04-15 14:50:36 -0500237 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500238 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
239 pages_from_pool, own_pages);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500240}
241EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
242
Alex Elder23c08a9cb2013-04-03 01:28:58 -0500243static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
244{
245 switch (osd_data->type) {
246 case CEPH_OSD_DATA_TYPE_NONE:
247 return 0;
248 case CEPH_OSD_DATA_TYPE_PAGES:
249 return osd_data->length;
250 case CEPH_OSD_DATA_TYPE_PAGELIST:
251 return (u64)osd_data->pagelist->length;
252#ifdef CONFIG_BLOCK
253 case CEPH_OSD_DATA_TYPE_BIO:
254 return (u64)osd_data->bio_length;
255#endif /* CONFIG_BLOCK */
256 default:
257 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
258 return 0;
259 }
260}
261
Alex Elderc54d47b2013-04-03 01:28:57 -0500262static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
263{
Alex Elder54764922013-04-05 01:27:12 -0500264 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
Alex Elderc54d47b2013-04-03 01:28:57 -0500265 int num_pages;
266
267 num_pages = calc_pages_for((u64)osd_data->alignment,
268 (u64)osd_data->length);
269 ceph_release_page_vector(osd_data->pages, num_pages);
270 }
Alex Elder54764922013-04-05 01:27:12 -0500271 ceph_osd_data_init(osd_data);
272}
273
274static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
275 unsigned int which)
276{
277 struct ceph_osd_req_op *op;
278
279 BUG_ON(which >= osd_req->r_num_ops);
280 op = &osd_req->r_ops[which];
281
282 switch (op->op) {
283 case CEPH_OSD_OP_READ:
284 case CEPH_OSD_OP_WRITE:
285 ceph_osd_data_release(&op->extent.osd_data);
286 break;
287 case CEPH_OSD_OP_CALL:
288 ceph_osd_data_release(&op->cls.request_info);
Alex Elder04017e22013-04-05 14:46:02 -0500289 ceph_osd_data_release(&op->cls.request_data);
Alex Elder54764922013-04-05 01:27:12 -0500290 ceph_osd_data_release(&op->cls.response_data);
291 break;
292 default:
293 break;
294 }
Alex Elderc54d47b2013-04-03 01:28:57 -0500295}
296
Sage Weilf24e9982009-10-06 11:31:10 -0700297/*
298 * requests
299 */
Sage Weil415e49a2009-12-07 13:37:03 -0800300void ceph_osdc_release_request(struct kref *kref)
Sage Weilf24e9982009-10-06 11:31:10 -0700301{
Alex Elderc54d47b2013-04-03 01:28:57 -0500302 struct ceph_osd_request *req;
Alex Elder54764922013-04-05 01:27:12 -0500303 unsigned int which;
Sage Weil415e49a2009-12-07 13:37:03 -0800304
Alex Elderc54d47b2013-04-03 01:28:57 -0500305 req = container_of(kref, struct ceph_osd_request, r_kref);
Sage Weil415e49a2009-12-07 13:37:03 -0800306 if (req->r_request)
307 ceph_msg_put(req->r_request);
Alex Elderace6d3a2013-04-01 16:12:14 -0500308 if (req->r_reply) {
Alex Elder8921d112012-06-01 14:56:43 -0500309 ceph_msg_revoke_incoming(req->r_reply);
Alex Elderab8cb342012-06-04 14:43:32 -0500310 ceph_msg_put(req->r_reply);
Alex Elderace6d3a2013-04-01 16:12:14 -0500311 }
Alex Elder0fff87e2013-02-14 12:16:43 -0600312
Alex Elder54764922013-04-05 01:27:12 -0500313 for (which = 0; which < req->r_num_ops; which++)
314 osd_req_op_data_release(req, which);
Alex Elder0fff87e2013-02-14 12:16:43 -0600315
Sage Weil415e49a2009-12-07 13:37:03 -0800316 ceph_put_snap_context(req->r_snapc);
317 if (req->r_mempool)
318 mempool_free(req, req->r_osdc->req_mempool);
319 else
Alex Elder5522ae02013-05-01 12:43:04 -0500320 kmem_cache_free(ceph_osd_request_cache, req);
321
Sage Weilf24e9982009-10-06 11:31:10 -0700322}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700323EXPORT_SYMBOL(ceph_osdc_release_request);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700324
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700325struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700326 struct ceph_snap_context *snapc,
Sage Weil1b83bef2013-02-25 16:11:12 -0800327 unsigned int num_ops,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700328 bool use_mempool,
Alex Elder54a54002012-11-13 21:11:15 -0600329 gfp_t gfp_flags)
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700330{
331 struct ceph_osd_request *req;
332 struct ceph_msg *msg;
Sage Weil1b83bef2013-02-25 16:11:12 -0800333 size_t msg_size;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700334
Alex Elder79528732013-04-03 21:32:51 -0500335 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
336 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
337
Sage Weil1b83bef2013-02-25 16:11:12 -0800338 msg_size = 4 + 4 + 8 + 8 + 4+8;
339 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
340 msg_size += 1 + 8 + 4 + 4; /* pg_t */
341 msg_size += 4 + MAX_OBJ_NAME_SIZE;
342 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
343 msg_size += 8; /* snapid */
344 msg_size += 8; /* snap_seq */
345 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
346 msg_size += 4;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700347
348 if (use_mempool) {
349 req = mempool_alloc(osdc->req_mempool, gfp_flags);
350 memset(req, 0, sizeof(*req));
351 } else {
Alex Elder5522ae02013-05-01 12:43:04 -0500352 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700353 }
354 if (req == NULL)
355 return NULL;
356
357 req->r_osdc = osdc;
358 req->r_mempool = use_mempool;
Alex Elder79528732013-04-03 21:32:51 -0500359 req->r_num_ops = num_ops;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700360
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700361 kref_init(&req->r_kref);
362 init_completion(&req->r_completion);
363 init_completion(&req->r_safe_completion);
Alex Eldera978fa22012-12-17 12:23:48 -0600364 RB_CLEAR_NODE(&req->r_node);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700365 INIT_LIST_HEAD(&req->r_unsafe_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700366 INIT_LIST_HEAD(&req->r_linger_item);
367 INIT_LIST_HEAD(&req->r_linger_osd);
Sage Weil935b6392011-09-16 11:13:17 -0700368 INIT_LIST_HEAD(&req->r_req_lru_item);
Sage Weilcd430452012-07-09 14:31:41 -0700369 INIT_LIST_HEAD(&req->r_osd_item);
370
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700371 /* create reply message */
372 if (use_mempool)
373 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
374 else
375 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
Sage Weilb61c2762011-08-09 15:03:46 -0700376 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700377 if (!msg) {
378 ceph_osdc_put_request(req);
379 return NULL;
380 }
381 req->r_reply = msg;
382
383 /* create request message; allow space for oid */
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700384 if (use_mempool)
385 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
386 else
Sage Weilb61c2762011-08-09 15:03:46 -0700387 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700388 if (!msg) {
389 ceph_osdc_put_request(req);
390 return NULL;
391 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700392
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700393 memset(msg->front.iov_base, 0, msg->front.iov_len);
394
395 req->r_request = msg;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700396
397 return req;
398}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700399EXPORT_SYMBOL(ceph_osdc_alloc_request);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700400
Alex Eldera8dd0a32013-03-13 20:50:00 -0500401static bool osd_req_opcode_valid(u16 opcode)
402{
403 switch (opcode) {
404 case CEPH_OSD_OP_READ:
405 case CEPH_OSD_OP_STAT:
406 case CEPH_OSD_OP_MAPEXT:
407 case CEPH_OSD_OP_MASKTRUNC:
408 case CEPH_OSD_OP_SPARSE_READ:
409 case CEPH_OSD_OP_NOTIFY:
410 case CEPH_OSD_OP_NOTIFY_ACK:
411 case CEPH_OSD_OP_ASSERT_VER:
412 case CEPH_OSD_OP_WRITE:
413 case CEPH_OSD_OP_WRITEFULL:
414 case CEPH_OSD_OP_TRUNCATE:
415 case CEPH_OSD_OP_ZERO:
416 case CEPH_OSD_OP_DELETE:
417 case CEPH_OSD_OP_APPEND:
418 case CEPH_OSD_OP_STARTSYNC:
419 case CEPH_OSD_OP_SETTRUNC:
420 case CEPH_OSD_OP_TRIMTRUNC:
421 case CEPH_OSD_OP_TMAPUP:
422 case CEPH_OSD_OP_TMAPPUT:
423 case CEPH_OSD_OP_TMAPGET:
424 case CEPH_OSD_OP_CREATE:
425 case CEPH_OSD_OP_ROLLBACK:
426 case CEPH_OSD_OP_WATCH:
427 case CEPH_OSD_OP_OMAPGETKEYS:
428 case CEPH_OSD_OP_OMAPGETVALS:
429 case CEPH_OSD_OP_OMAPGETHEADER:
430 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
431 case CEPH_OSD_OP_OMAPSETVALS:
432 case CEPH_OSD_OP_OMAPSETHEADER:
433 case CEPH_OSD_OP_OMAPCLEAR:
434 case CEPH_OSD_OP_OMAPRMKEYS:
435 case CEPH_OSD_OP_OMAP_CMP:
436 case CEPH_OSD_OP_CLONERANGE:
437 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
438 case CEPH_OSD_OP_SRC_CMPXATTR:
439 case CEPH_OSD_OP_GETXATTR:
440 case CEPH_OSD_OP_GETXATTRS:
441 case CEPH_OSD_OP_CMPXATTR:
442 case CEPH_OSD_OP_SETXATTR:
443 case CEPH_OSD_OP_SETXATTRS:
444 case CEPH_OSD_OP_RESETXATTRS:
445 case CEPH_OSD_OP_RMXATTR:
446 case CEPH_OSD_OP_PULL:
447 case CEPH_OSD_OP_PUSH:
448 case CEPH_OSD_OP_BALANCEREADS:
449 case CEPH_OSD_OP_UNBALANCEREADS:
450 case CEPH_OSD_OP_SCRUB:
451 case CEPH_OSD_OP_SCRUB_RESERVE:
452 case CEPH_OSD_OP_SCRUB_UNRESERVE:
453 case CEPH_OSD_OP_SCRUB_STOP:
454 case CEPH_OSD_OP_SCRUB_MAP:
455 case CEPH_OSD_OP_WRLOCK:
456 case CEPH_OSD_OP_WRUNLOCK:
457 case CEPH_OSD_OP_RDLOCK:
458 case CEPH_OSD_OP_RDUNLOCK:
459 case CEPH_OSD_OP_UPLOCK:
460 case CEPH_OSD_OP_DNLOCK:
461 case CEPH_OSD_OP_CALL:
462 case CEPH_OSD_OP_PGLS:
463 case CEPH_OSD_OP_PGLS_FILTER:
464 return true;
465 default:
466 return false;
467 }
468}
469
Alex Elder33803f32013-03-13 20:50:00 -0500470/*
471 * This is an osd op init function for opcodes that have no data or
472 * other information associated with them. It also serves as a
473 * common init routine for all the other init functions, below.
474 */
Alex Elderc99d2d42013-04-05 01:27:11 -0500475static struct ceph_osd_req_op *
Alex Elder49719772013-02-11 12:33:24 -0600476_osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
Alex Elderc99d2d42013-04-05 01:27:11 -0500477 u16 opcode)
Alex Elder33803f32013-03-13 20:50:00 -0500478{
Alex Elderc99d2d42013-04-05 01:27:11 -0500479 struct ceph_osd_req_op *op;
480
481 BUG_ON(which >= osd_req->r_num_ops);
Alex Elder33803f32013-03-13 20:50:00 -0500482 BUG_ON(!osd_req_opcode_valid(opcode));
483
Alex Elderc99d2d42013-04-05 01:27:11 -0500484 op = &osd_req->r_ops[which];
Alex Elder33803f32013-03-13 20:50:00 -0500485 memset(op, 0, sizeof (*op));
Alex Elder33803f32013-03-13 20:50:00 -0500486 op->op = opcode;
Alex Elderc99d2d42013-04-05 01:27:11 -0500487
488 return op;
Alex Elder33803f32013-03-13 20:50:00 -0500489}
490
Alex Elder49719772013-02-11 12:33:24 -0600491void osd_req_op_init(struct ceph_osd_request *osd_req,
492 unsigned int which, u16 opcode)
493{
494 (void)_osd_req_op_init(osd_req, which, opcode);
495}
496EXPORT_SYMBOL(osd_req_op_init);
497
Alex Elderc99d2d42013-04-05 01:27:11 -0500498void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
499 unsigned int which, u16 opcode,
Alex Elder33803f32013-03-13 20:50:00 -0500500 u64 offset, u64 length,
501 u64 truncate_size, u32 truncate_seq)
502{
Alex Elder49719772013-02-11 12:33:24 -0600503 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder33803f32013-03-13 20:50:00 -0500504 size_t payload_len = 0;
505
506 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
507
Alex Elder33803f32013-03-13 20:50:00 -0500508 op->extent.offset = offset;
509 op->extent.length = length;
510 op->extent.truncate_size = truncate_size;
511 op->extent.truncate_seq = truncate_seq;
512 if (opcode == CEPH_OSD_OP_WRITE)
513 payload_len += length;
514
515 op->payload_len = payload_len;
516}
517EXPORT_SYMBOL(osd_req_op_extent_init);
518
Alex Elderc99d2d42013-04-05 01:27:11 -0500519void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
520 unsigned int which, u64 length)
Alex Eldere5975c72013-03-14 14:09:05 -0500521{
Alex Elderc99d2d42013-04-05 01:27:11 -0500522 struct ceph_osd_req_op *op;
523 u64 previous;
524
525 BUG_ON(which >= osd_req->r_num_ops);
526 op = &osd_req->r_ops[which];
527 previous = op->extent.length;
Alex Eldere5975c72013-03-14 14:09:05 -0500528
529 if (length == previous)
530 return; /* Nothing to do */
531 BUG_ON(length > previous);
532
533 op->extent.length = length;
534 op->payload_len -= previous - length;
535}
536EXPORT_SYMBOL(osd_req_op_extent_update);
537
Alex Elderc99d2d42013-04-05 01:27:11 -0500538void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
Alex Elder04017e22013-04-05 14:46:02 -0500539 u16 opcode, const char *class, const char *method)
Alex Elder33803f32013-03-13 20:50:00 -0500540{
Alex Elder49719772013-02-11 12:33:24 -0600541 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder5f562df2013-04-05 01:27:12 -0500542 struct ceph_pagelist *pagelist;
Alex Elder33803f32013-03-13 20:50:00 -0500543 size_t payload_len = 0;
544 size_t size;
545
546 BUG_ON(opcode != CEPH_OSD_OP_CALL);
547
Alex Elder5f562df2013-04-05 01:27:12 -0500548 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
549 BUG_ON(!pagelist);
550 ceph_pagelist_init(pagelist);
551
Alex Elder33803f32013-03-13 20:50:00 -0500552 op->cls.class_name = class;
553 size = strlen(class);
554 BUG_ON(size > (size_t) U8_MAX);
555 op->cls.class_len = size;
Alex Elder5f562df2013-04-05 01:27:12 -0500556 ceph_pagelist_append(pagelist, class, size);
Alex Elder33803f32013-03-13 20:50:00 -0500557 payload_len += size;
558
559 op->cls.method_name = method;
560 size = strlen(method);
561 BUG_ON(size > (size_t) U8_MAX);
562 op->cls.method_len = size;
Alex Elder5f562df2013-04-05 01:27:12 -0500563 ceph_pagelist_append(pagelist, method, size);
Alex Elder33803f32013-03-13 20:50:00 -0500564 payload_len += size;
565
Alex Eldera4ce40a2013-04-05 01:27:12 -0500566 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
Alex Elder5f562df2013-04-05 01:27:12 -0500567
Alex Elder33803f32013-03-13 20:50:00 -0500568 op->cls.argc = 0; /* currently unused */
569
570 op->payload_len = payload_len;
571}
572EXPORT_SYMBOL(osd_req_op_cls_init);
Alex Elder8c042b02013-04-03 01:28:58 -0500573
Alex Elderc99d2d42013-04-05 01:27:11 -0500574void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
575 unsigned int which, u16 opcode,
Alex Elder33803f32013-03-13 20:50:00 -0500576 u64 cookie, u64 version, int flag)
577{
Alex Elder49719772013-02-11 12:33:24 -0600578 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder33803f32013-03-13 20:50:00 -0500579
Alex Elderc99d2d42013-04-05 01:27:11 -0500580 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
Alex Elder33803f32013-03-13 20:50:00 -0500581
582 op->watch.cookie = cookie;
Alex Elder9ef1ee52013-04-21 16:51:50 -0500583 op->watch.ver = version;
Alex Elder33803f32013-03-13 20:50:00 -0500584 if (opcode == CEPH_OSD_OP_WATCH && flag)
Alex Elderc99d2d42013-04-05 01:27:11 -0500585 op->watch.flag = (u8)1;
Alex Elder33803f32013-03-13 20:50:00 -0500586}
587EXPORT_SYMBOL(osd_req_op_watch_init);
588
Alex Elder90af3602013-04-05 14:46:01 -0500589static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
Alex Elderec9123c2013-04-05 01:27:12 -0500590 struct ceph_osd_data *osd_data)
591{
592 u64 length = ceph_osd_data_length(osd_data);
593
594 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
595 BUG_ON(length > (u64) SIZE_MAX);
596 if (length)
Alex Elder90af3602013-04-05 14:46:01 -0500597 ceph_msg_data_add_pages(msg, osd_data->pages,
Alex Elderec9123c2013-04-05 01:27:12 -0500598 length, osd_data->alignment);
599 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
600 BUG_ON(!length);
Alex Elder90af3602013-04-05 14:46:01 -0500601 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
Alex Elderec9123c2013-04-05 01:27:12 -0500602#ifdef CONFIG_BLOCK
603 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
Alex Elder90af3602013-04-05 14:46:01 -0500604 ceph_msg_data_add_bio(msg, osd_data->bio, length);
Alex Elderec9123c2013-04-05 01:27:12 -0500605#endif
606 } else {
607 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
608 }
609}
610
Alex Elder175face2013-03-08 13:35:36 -0600611static u64 osd_req_encode_op(struct ceph_osd_request *req,
Alex Elder79528732013-04-03 21:32:51 -0500612 struct ceph_osd_op *dst, unsigned int which)
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700613{
Alex Elder79528732013-04-03 21:32:51 -0500614 struct ceph_osd_req_op *src;
Alex Elder04017e22013-04-05 14:46:02 -0500615 struct ceph_osd_data *osd_data;
Alex Elder54d50642013-04-03 01:28:58 -0500616 u64 request_data_len = 0;
Alex Elder04017e22013-04-05 14:46:02 -0500617 u64 data_length;
Alex Elder175face2013-03-08 13:35:36 -0600618
Alex Elder79528732013-04-03 21:32:51 -0500619 BUG_ON(which >= req->r_num_ops);
620 src = &req->r_ops[which];
Alex Eldera8dd0a32013-03-13 20:50:00 -0500621 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
622 pr_err("unrecognized osd opcode %d\n", src->op);
623
624 return 0;
625 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700626
Alex Elder065a68f2012-04-20 15:49:43 -0500627 switch (src->op) {
Alex Elderfbfab532013-02-08 09:55:48 -0600628 case CEPH_OSD_OP_STAT:
Alex Elder49719772013-02-11 12:33:24 -0600629 osd_data = &src->raw_data_in;
630 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Alex Elderfbfab532013-02-08 09:55:48 -0600631 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700632 case CEPH_OSD_OP_READ:
633 case CEPH_OSD_OP_WRITE:
Alex Elder175face2013-03-08 13:35:36 -0600634 if (src->op == CEPH_OSD_OP_WRITE)
Alex Elder54d50642013-04-03 01:28:58 -0500635 request_data_len = src->extent.length;
Alex Elder175face2013-03-08 13:35:36 -0600636 dst->extent.offset = cpu_to_le64(src->extent.offset);
637 dst->extent.length = cpu_to_le64(src->extent.length);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700638 dst->extent.truncate_size =
639 cpu_to_le64(src->extent.truncate_size);
640 dst->extent.truncate_seq =
641 cpu_to_le32(src->extent.truncate_seq);
Alex Elder04017e22013-04-05 14:46:02 -0500642 osd_data = &src->extent.osd_data;
Alex Elder54764922013-04-05 01:27:12 -0500643 if (src->op == CEPH_OSD_OP_WRITE)
Alex Elder04017e22013-04-05 14:46:02 -0500644 ceph_osdc_msg_data_add(req->r_request, osd_data);
Alex Elder54764922013-04-05 01:27:12 -0500645 else
Alex Elder04017e22013-04-05 14:46:02 -0500646 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700647 break;
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700648 case CEPH_OSD_OP_CALL:
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700649 dst->cls.class_len = src->cls.class_len;
650 dst->cls.method_len = src->cls.method_len;
Alex Elder04017e22013-04-05 14:46:02 -0500651 osd_data = &src->cls.request_info;
652 ceph_osdc_msg_data_add(req->r_request, osd_data);
653 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
654 request_data_len = osd_data->pagelist->length;
655
656 osd_data = &src->cls.request_data;
657 data_length = ceph_osd_data_length(osd_data);
658 if (data_length) {
659 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
660 dst->cls.indata_len = cpu_to_le32(data_length);
661 ceph_osdc_msg_data_add(req->r_request, osd_data);
662 src->payload_len += data_length;
663 request_data_len += data_length;
664 }
665 osd_data = &src->cls.response_data;
666 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700667 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700668 case CEPH_OSD_OP_STARTSYNC:
669 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700670 case CEPH_OSD_OP_NOTIFY_ACK:
671 case CEPH_OSD_OP_WATCH:
672 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
673 dst->watch.ver = cpu_to_le64(src->watch.ver);
674 dst->watch.flag = src->watch.flag;
675 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700676 default:
Alex Elder4c464592013-02-15 11:42:30 -0600677 pr_err("unsupported osd opcode %s\n",
Alex Elder8f63ca22013-03-04 11:08:29 -0600678 ceph_osd_op_name(src->op));
Alex Elder4c464592013-02-15 11:42:30 -0600679 WARN_ON(1);
Alex Eldera8dd0a32013-03-13 20:50:00 -0500680
681 return 0;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700682 }
Alex Eldera8dd0a32013-03-13 20:50:00 -0500683 dst->op = cpu_to_le16(src->op);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700684 dst->payload_len = cpu_to_le32(src->payload_len);
Alex Elder175face2013-03-08 13:35:36 -0600685
Alex Elder54d50642013-04-03 01:28:58 -0500686 return request_data_len;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700687}
688
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700689/*
Sage Weilf24e9982009-10-06 11:31:10 -0700690 * build new request AND message, calculate layout, and adjust file
691 * extent as needed.
692 *
693 * if the file was recently truncated, we include information about its
694 * old and new size so that the object can be updated appropriately. (we
695 * avoid synchronously deleting truncated objects because it's slow.)
696 *
697 * if @do_sync, include a 'startsync' command so that the osd will flush
698 * data quickly.
699 */
700struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
701 struct ceph_file_layout *layout,
702 struct ceph_vino vino,
Alex Elderacead002013-03-14 14:09:05 -0500703 u64 off, u64 *plen, int num_ops,
Sage Weilf24e9982009-10-06 11:31:10 -0700704 int opcode, int flags,
705 struct ceph_snap_context *snapc,
Sage Weilf24e9982009-10-06 11:31:10 -0700706 u32 truncate_seq,
707 u64 truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -0600708 bool use_mempool)
Sage Weilf24e9982009-10-06 11:31:10 -0700709{
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700710 struct ceph_osd_request *req;
Alex Elder75d1c942013-03-13 20:50:00 -0500711 u64 objnum = 0;
712 u64 objoff = 0;
713 u64 objlen = 0;
Alex Elderd18d1e22013-03-13 20:50:01 -0500714 u32 object_size;
715 u64 object_base;
Sage Weil68162822012-09-24 21:01:02 -0700716 int r;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700717
Alex Elderd18d1e22013-03-13 20:50:01 -0500718 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700719
Alex Elderacead002013-03-14 14:09:05 -0500720 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
Alex Elderae7ca4a32012-11-13 21:11:15 -0600721 GFP_NOFS);
Sage Weil4ad12622011-05-03 09:23:36 -0700722 if (!req)
Sage Weil68162822012-09-24 21:01:02 -0700723 return ERR_PTR(-ENOMEM);
Alex Elder79528732013-04-03 21:32:51 -0500724
Alex Elderd178a9e2012-11-13 21:11:15 -0600725 req->r_flags = flags;
Sage Weilf24e9982009-10-06 11:31:10 -0700726
727 /* calculate max write size */
Alex Eldera19dadf2013-03-13 20:50:01 -0500728 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
Alex Elder3ff5f382013-02-15 22:10:17 -0600729 if (r < 0) {
730 ceph_osdc_put_request(req);
Sage Weil68162822012-09-24 21:01:02 -0700731 return ERR_PTR(r);
Alex Elder3ff5f382013-02-15 22:10:17 -0600732 }
Alex Eldera19dadf2013-03-13 20:50:01 -0500733
Alex Elderd18d1e22013-03-13 20:50:01 -0500734 object_size = le32_to_cpu(layout->fl_object_size);
735 object_base = off - objoff;
736 if (truncate_size <= object_base) {
737 truncate_size = 0;
738 } else {
739 truncate_size -= object_base;
740 if (truncate_size > object_size)
741 truncate_size = object_size;
Alex Eldera19dadf2013-03-13 20:50:01 -0500742 }
Alex Elderd18d1e22013-03-13 20:50:01 -0500743
Alex Elderc99d2d42013-04-05 01:27:11 -0500744 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
Alex Elderb0270322013-03-13 20:50:01 -0500745 truncate_size, truncate_seq);
Alex Elder8c042b02013-04-03 01:28:58 -0500746
Alex Elderacead002013-03-14 14:09:05 -0500747 /*
748 * A second op in the ops array means the caller wants to
749 * also issue a include a 'startsync' command so that the
750 * osd will flush data quickly.
751 */
752 if (num_ops > 1)
Alex Elderc99d2d42013-04-05 01:27:11 -0500753 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
Alex Elderd18d1e22013-03-13 20:50:01 -0500754
Sage Weilf24e9982009-10-06 11:31:10 -0700755 req->r_file_layout = *layout; /* keep a copy */
756
Alex Elder75d1c942013-03-13 20:50:00 -0500757 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
758 vino.ino, objnum);
Alex Elderdbe0fc42013-02-15 22:10:17 -0600759 req->r_oid_len = strlen(req->r_oid);
760
Sage Weilf24e9982009-10-06 11:31:10 -0700761 return req;
762}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700763EXPORT_SYMBOL(ceph_osdc_new_request);
Sage Weilf24e9982009-10-06 11:31:10 -0700764
765/*
766 * We keep osd requests in an rbtree, sorted by ->r_tid.
767 */
768static void __insert_request(struct ceph_osd_client *osdc,
769 struct ceph_osd_request *new)
770{
771 struct rb_node **p = &osdc->requests.rb_node;
772 struct rb_node *parent = NULL;
773 struct ceph_osd_request *req = NULL;
774
775 while (*p) {
776 parent = *p;
777 req = rb_entry(parent, struct ceph_osd_request, r_node);
778 if (new->r_tid < req->r_tid)
779 p = &(*p)->rb_left;
780 else if (new->r_tid > req->r_tid)
781 p = &(*p)->rb_right;
782 else
783 BUG();
784 }
785
786 rb_link_node(&new->r_node, parent, p);
787 rb_insert_color(&new->r_node, &osdc->requests);
788}
789
790static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
791 u64 tid)
792{
793 struct ceph_osd_request *req;
794 struct rb_node *n = osdc->requests.rb_node;
795
796 while (n) {
797 req = rb_entry(n, struct ceph_osd_request, r_node);
798 if (tid < req->r_tid)
799 n = n->rb_left;
800 else if (tid > req->r_tid)
801 n = n->rb_right;
802 else
803 return req;
804 }
805 return NULL;
806}
807
808static struct ceph_osd_request *
809__lookup_request_ge(struct ceph_osd_client *osdc,
810 u64 tid)
811{
812 struct ceph_osd_request *req;
813 struct rb_node *n = osdc->requests.rb_node;
814
815 while (n) {
816 req = rb_entry(n, struct ceph_osd_request, r_node);
817 if (tid < req->r_tid) {
818 if (!n->rb_left)
819 return req;
820 n = n->rb_left;
821 } else if (tid > req->r_tid) {
822 n = n->rb_right;
823 } else {
824 return req;
825 }
826 }
827 return NULL;
828}
829
Sage Weil6f6c7002011-01-17 20:34:08 -0800830/*
831 * Resubmit requests pending on the given osd.
832 */
833static void __kick_osd_requests(struct ceph_osd_client *osdc,
834 struct ceph_osd *osd)
835{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700836 struct ceph_osd_request *req, *nreq;
Alex Eldere02493c2013-03-25 18:16:11 -0500837 LIST_HEAD(resend);
Sage Weil6f6c7002011-01-17 20:34:08 -0800838 int err;
839
840 dout("__kick_osd_requests osd%d\n", osd->o_osd);
841 err = __reset_osd(osdc, osd);
Alex Elder685a7552012-12-07 09:57:58 -0600842 if (err)
Sage Weil6f6c7002011-01-17 20:34:08 -0800843 return;
Alex Eldere02493c2013-03-25 18:16:11 -0500844 /*
845 * Build up a list of requests to resend by traversing the
846 * osd's list of requests. Requests for a given object are
847 * sent in tid order, and that is also the order they're
848 * kept on this list. Therefore all requests that are in
849 * flight will be found first, followed by all requests that
850 * have not yet been sent. And to resend requests while
851 * preserving this order we will want to put any sent
852 * requests back on the front of the osd client's unsent
853 * list.
854 *
855 * So we build a separate ordered list of already-sent
856 * requests for the affected osd and splice it onto the
857 * front of the osd client's unsent list. Once we've seen a
858 * request that has not yet been sent we're done. Those
859 * requests are already sitting right where they belong.
860 */
Sage Weil6f6c7002011-01-17 20:34:08 -0800861 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
Alex Eldere02493c2013-03-25 18:16:11 -0500862 if (!req->r_sent)
863 break;
864 list_move_tail(&req->r_req_lru_item, &resend);
865 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
Sage Weil6f6c7002011-01-17 20:34:08 -0800866 osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700867 if (!req->r_linger)
868 req->r_flags |= CEPH_OSD_FLAG_RETRY;
869 }
Alex Eldere02493c2013-03-25 18:16:11 -0500870 list_splice(&resend, &osdc->req_unsent);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700871
Alex Eldere02493c2013-03-25 18:16:11 -0500872 /*
873 * Linger requests are re-registered before sending, which
874 * sets up a new tid for each. We add them to the unsent
875 * list at the end to keep things in tid order.
876 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700877 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
878 r_linger_osd) {
Sage Weil77f38e02011-04-06 09:09:16 -0700879 /*
880 * reregister request prior to unregistering linger so
881 * that r_osd is preserved.
882 */
883 BUG_ON(!list_empty(&req->r_req_lru_item));
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700884 __register_request(osdc, req);
Alex Eldere02493c2013-03-25 18:16:11 -0500885 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
Alex Elderad885922013-03-25 18:16:11 -0500886 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
Sage Weil77f38e02011-04-06 09:09:16 -0700887 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700888 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
889 osd->o_osd);
Sage Weil6f6c7002011-01-17 20:34:08 -0800890 }
891}
892
Sage Weilf24e9982009-10-06 11:31:10 -0700893/*
Sage Weil81b024e2009-10-09 10:29:18 -0700894 * If the osd connection drops, we need to resubmit all requests.
Sage Weilf24e9982009-10-06 11:31:10 -0700895 */
896static void osd_reset(struct ceph_connection *con)
897{
898 struct ceph_osd *osd = con->private;
899 struct ceph_osd_client *osdc;
900
901 if (!osd)
902 return;
903 dout("osd_reset osd%d\n", osd->o_osd);
904 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -0700905 down_read(&osdc->map_sem);
Sage Weil83aff952012-11-28 12:28:24 -0800906 mutex_lock(&osdc->request_mutex);
907 __kick_osd_requests(osdc, osd);
Alex Elderf9d25192013-02-15 11:42:29 -0600908 __send_queued(osdc);
Sage Weil83aff952012-11-28 12:28:24 -0800909 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -0700910 up_read(&osdc->map_sem);
911}
912
913/*
914 * Track open sessions with osds.
915 */
Alex Eldere10006f2012-05-26 23:26:43 -0500916static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
Sage Weilf24e9982009-10-06 11:31:10 -0700917{
918 struct ceph_osd *osd;
919
920 osd = kzalloc(sizeof(*osd), GFP_NOFS);
921 if (!osd)
922 return NULL;
923
924 atomic_set(&osd->o_ref, 1);
925 osd->o_osdc = osdc;
Alex Eldere10006f2012-05-26 23:26:43 -0500926 osd->o_osd = onum;
Alex Elderf4077312012-12-06 07:22:04 -0600927 RB_CLEAR_NODE(&osd->o_node);
Sage Weilf24e9982009-10-06 11:31:10 -0700928 INIT_LIST_HEAD(&osd->o_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700929 INIT_LIST_HEAD(&osd->o_linger_requests);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800930 INIT_LIST_HEAD(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700931 osd->o_incarnation = 1;
932
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700933 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800934
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800935 INIT_LIST_HEAD(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -0700936 return osd;
937}
938
939static struct ceph_osd *get_osd(struct ceph_osd *osd)
940{
941 if (atomic_inc_not_zero(&osd->o_ref)) {
942 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
943 atomic_read(&osd->o_ref));
944 return osd;
945 } else {
946 dout("get_osd %p FAIL\n", osd);
947 return NULL;
948 }
949}
950
951static void put_osd(struct ceph_osd *osd)
952{
953 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
954 atomic_read(&osd->o_ref) - 1);
Alex Eldera2556512012-05-16 15:16:39 -0500955 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
Sage Weil79494d12010-05-27 14:15:49 -0700956 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
957
Sage Weil27859f92013-03-25 10:26:14 -0700958 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
Sage Weilf24e9982009-10-06 11:31:10 -0700959 kfree(osd);
Sage Weil79494d12010-05-27 14:15:49 -0700960 }
Sage Weilf24e9982009-10-06 11:31:10 -0700961}
962
963/*
964 * remove an osd from our map
965 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800966static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -0700967{
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800968 dout("__remove_osd %p\n", osd);
Sage Weilf24e9982009-10-06 11:31:10 -0700969 BUG_ON(!list_empty(&osd->o_requests));
970 rb_erase(&osd->o_node, &osdc->osds);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800971 list_del_init(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700972 ceph_con_close(&osd->o_con);
973 put_osd(osd);
974}
975
Sage Weilaca420b2011-08-31 14:45:53 -0700976static void remove_all_osds(struct ceph_osd_client *osdc)
977{
Jiaju Zhang048a9d22012-07-20 08:18:36 -0500978 dout("%s %p\n", __func__, osdc);
Sage Weilaca420b2011-08-31 14:45:53 -0700979 mutex_lock(&osdc->request_mutex);
980 while (!RB_EMPTY_ROOT(&osdc->osds)) {
981 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
982 struct ceph_osd, o_node);
983 __remove_osd(osdc, osd);
984 }
985 mutex_unlock(&osdc->request_mutex);
986}
987
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800988static void __move_osd_to_lru(struct ceph_osd_client *osdc,
989 struct ceph_osd *osd)
990{
991 dout("__move_osd_to_lru %p\n", osd);
992 BUG_ON(!list_empty(&osd->o_osd_lru));
993 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700994 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800995}
996
997static void __remove_osd_from_lru(struct ceph_osd *osd)
998{
999 dout("__remove_osd_from_lru %p\n", osd);
1000 if (!list_empty(&osd->o_osd_lru))
1001 list_del_init(&osd->o_osd_lru);
1002}
1003
Sage Weilaca420b2011-08-31 14:45:53 -07001004static void remove_old_osds(struct ceph_osd_client *osdc)
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001005{
1006 struct ceph_osd *osd, *nosd;
1007
1008 dout("__remove_old_osds %p\n", osdc);
1009 mutex_lock(&osdc->request_mutex);
1010 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
Sage Weilaca420b2011-08-31 14:45:53 -07001011 if (time_before(jiffies, osd->lru_ttl))
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001012 break;
1013 __remove_osd(osdc, osd);
1014 }
1015 mutex_unlock(&osdc->request_mutex);
1016}
1017
Sage Weilf24e9982009-10-06 11:31:10 -07001018/*
1019 * reset osd connect
1020 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001021static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -07001022{
Alex Elderc3acb182012-12-07 09:57:58 -06001023 struct ceph_entity_addr *peer_addr;
Sage Weilf24e9982009-10-06 11:31:10 -07001024
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001025 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001026 if (list_empty(&osd->o_requests) &&
1027 list_empty(&osd->o_linger_requests)) {
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001028 __remove_osd(osdc, osd);
Alex Elderc3acb182012-12-07 09:57:58 -06001029
1030 return -ENODEV;
1031 }
1032
1033 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1034 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1035 !ceph_con_opened(&osd->o_con)) {
1036 struct ceph_osd_request *req;
1037
Sage Weil87b315a2010-03-22 14:51:18 -07001038 dout(" osd addr hasn't changed and connection never opened,"
1039 " letting msgr retry");
1040 /* touch each r_stamp for handle_timeout()'s benfit */
1041 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1042 req->r_stamp = jiffies;
Alex Elderc3acb182012-12-07 09:57:58 -06001043
1044 return -EAGAIN;
Sage Weilf24e9982009-10-06 11:31:10 -07001045 }
Alex Elderc3acb182012-12-07 09:57:58 -06001046
1047 ceph_con_close(&osd->o_con);
1048 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1049 osd->o_incarnation++;
1050
1051 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001052}
1053
1054static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1055{
1056 struct rb_node **p = &osdc->osds.rb_node;
1057 struct rb_node *parent = NULL;
1058 struct ceph_osd *osd = NULL;
1059
Sage Weilaca420b2011-08-31 14:45:53 -07001060 dout("__insert_osd %p osd%d\n", new, new->o_osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001061 while (*p) {
1062 parent = *p;
1063 osd = rb_entry(parent, struct ceph_osd, o_node);
1064 if (new->o_osd < osd->o_osd)
1065 p = &(*p)->rb_left;
1066 else if (new->o_osd > osd->o_osd)
1067 p = &(*p)->rb_right;
1068 else
1069 BUG();
1070 }
1071
1072 rb_link_node(&new->o_node, parent, p);
1073 rb_insert_color(&new->o_node, &osdc->osds);
1074}
1075
1076static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1077{
1078 struct ceph_osd *osd;
1079 struct rb_node *n = osdc->osds.rb_node;
1080
1081 while (n) {
1082 osd = rb_entry(n, struct ceph_osd, o_node);
1083 if (o < osd->o_osd)
1084 n = n->rb_left;
1085 else if (o > osd->o_osd)
1086 n = n->rb_right;
1087 else
1088 return osd;
1089 }
1090 return NULL;
1091}
1092
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001093static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1094{
1095 schedule_delayed_work(&osdc->timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001096 osdc->client->options->osd_keepalive_timeout * HZ);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001097}
1098
1099static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1100{
1101 cancel_delayed_work(&osdc->timeout_work);
1102}
Sage Weilf24e9982009-10-06 11:31:10 -07001103
1104/*
1105 * Register request, assign tid. If this is the first request, set up
1106 * the timeout event.
1107 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001108static void __register_request(struct ceph_osd_client *osdc,
1109 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001110{
Sage Weilf24e9982009-10-06 11:31:10 -07001111 req->r_tid = ++osdc->last_tid;
Sage Weil6df058c2009-12-22 11:24:33 -08001112 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
Sage Weil77f38e02011-04-06 09:09:16 -07001113 dout("__register_request %p tid %lld\n", req, req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07001114 __insert_request(osdc, req);
1115 ceph_osdc_get_request(req);
1116 osdc->num_requests++;
Sage Weilf24e9982009-10-06 11:31:10 -07001117 if (osdc->num_requests == 1) {
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001118 dout(" first request, scheduling timeout\n");
1119 __schedule_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001120 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001121}
1122
Sage Weilf24e9982009-10-06 11:31:10 -07001123/*
1124 * called under osdc->request_mutex
1125 */
1126static void __unregister_request(struct ceph_osd_client *osdc,
1127 struct ceph_osd_request *req)
1128{
Sage Weil35f9f8a2012-05-16 15:16:38 -05001129 if (RB_EMPTY_NODE(&req->r_node)) {
1130 dout("__unregister_request %p tid %lld not registered\n",
1131 req, req->r_tid);
1132 return;
1133 }
1134
Sage Weilf24e9982009-10-06 11:31:10 -07001135 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1136 rb_erase(&req->r_node, &osdc->requests);
1137 osdc->num_requests--;
1138
Sage Weil0ba64782009-10-08 16:57:16 -07001139 if (req->r_osd) {
1140 /* make sure the original request isn't in flight. */
Alex Elder6740a842012-06-01 14:56:43 -05001141 ceph_msg_revoke(req->r_request);
Sage Weil0ba64782009-10-08 16:57:16 -07001142
1143 list_del_init(&req->r_osd_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001144 if (list_empty(&req->r_osd->o_requests) &&
1145 list_empty(&req->r_osd->o_linger_requests)) {
1146 dout("moving osd to %p lru\n", req->r_osd);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001147 __move_osd_to_lru(osdc, req->r_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001148 }
Sage Weilfbdb9192011-03-29 12:11:06 -07001149 if (list_empty(&req->r_linger_item))
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001150 req->r_osd = NULL;
Sage Weil0ba64782009-10-08 16:57:16 -07001151 }
Sage Weilf24e9982009-10-06 11:31:10 -07001152
Alex Elder7d5f2482012-11-29 08:37:03 -06001153 list_del_init(&req->r_req_lru_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001154 ceph_osdc_put_request(req);
1155
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001156 if (osdc->num_requests == 0) {
1157 dout(" no requests, canceling timeout\n");
1158 __cancel_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001159 }
1160}
1161
1162/*
1163 * Cancel a previously queued request message
1164 */
1165static void __cancel_request(struct ceph_osd_request *req)
1166{
Sage Weil6bc18872010-09-27 10:18:52 -07001167 if (req->r_sent && req->r_osd) {
Alex Elder6740a842012-06-01 14:56:43 -05001168 ceph_msg_revoke(req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001169 req->r_sent = 0;
1170 }
1171}
1172
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001173static void __register_linger_request(struct ceph_osd_client *osdc,
1174 struct ceph_osd_request *req)
1175{
1176 dout("__register_linger_request %p\n", req);
Alex Elder96e4dac2013-05-22 20:54:25 -05001177 ceph_osdc_get_request(req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001178 list_add_tail(&req->r_linger_item, &osdc->req_linger);
Sage Weil6194ea82012-07-30 16:19:28 -07001179 if (req->r_osd)
1180 list_add_tail(&req->r_linger_osd,
1181 &req->r_osd->o_linger_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001182}
1183
1184static void __unregister_linger_request(struct ceph_osd_client *osdc,
1185 struct ceph_osd_request *req)
1186{
1187 dout("__unregister_linger_request %p\n", req);
Alex Elder61c74032012-12-06 09:37:23 -06001188 list_del_init(&req->r_linger_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001189 if (req->r_osd) {
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001190 list_del_init(&req->r_linger_osd);
1191
1192 if (list_empty(&req->r_osd->o_requests) &&
1193 list_empty(&req->r_osd->o_linger_requests)) {
1194 dout("moving osd to %p lru\n", req->r_osd);
1195 __move_osd_to_lru(osdc, req->r_osd);
1196 }
Sage Weilfbdb9192011-03-29 12:11:06 -07001197 if (list_empty(&req->r_osd_item))
1198 req->r_osd = NULL;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001199 }
Alex Elder96e4dac2013-05-22 20:54:25 -05001200 ceph_osdc_put_request(req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001201}
1202
1203void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1204 struct ceph_osd_request *req)
1205{
1206 mutex_lock(&osdc->request_mutex);
1207 if (req->r_linger) {
Alex Elderc10ebbf2013-05-09 14:56:32 -05001208 req->r_linger = 0;
Alex Elder96e4dac2013-05-22 20:54:25 -05001209 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001210 }
1211 mutex_unlock(&osdc->request_mutex);
1212}
1213EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1214
1215void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1216 struct ceph_osd_request *req)
1217{
1218 if (!req->r_linger) {
1219 dout("set_request_linger %p\n", req);
1220 req->r_linger = 1;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001221 }
1222}
1223EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1224
Sage Weilf24e9982009-10-06 11:31:10 -07001225/*
1226 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1227 * (as needed), and set the request r_osd appropriately. If there is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001228 * no up osd, set r_osd to NULL. Move the request to the appropriate list
Sage Weil6f6c7002011-01-17 20:34:08 -08001229 * (unsent, homeless) or leave on in-flight lru.
Sage Weilf24e9982009-10-06 11:31:10 -07001230 *
1231 * Return 0 if unchanged, 1 if changed, or negative on error.
1232 *
1233 * Caller should hold map_sem for read and request_mutex.
1234 */
Sage Weil6f6c7002011-01-17 20:34:08 -08001235static int __map_request(struct ceph_osd_client *osdc,
Sage Weil38d64532011-10-14 13:33:55 -07001236 struct ceph_osd_request *req, int force_resend)
Sage Weilf24e9982009-10-06 11:31:10 -07001237{
Sage Weil5b191d92013-02-23 10:38:16 -08001238 struct ceph_pg pgid;
Sage Weild85b7052010-05-10 10:24:48 -07001239 int acting[CEPH_PG_MAX_SIZE];
1240 int o = -1, num = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001241 int err;
Sage Weilf24e9982009-10-06 11:31:10 -07001242
Sage Weil6f6c7002011-01-17 20:34:08 -08001243 dout("map_request %p tid %lld\n", req, req->r_tid);
Alex Elder41766f82013-03-01 18:00:15 -06001244 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1245 ceph_file_layout_pg_pool(req->r_file_layout));
Sage Weil6f6c7002011-01-17 20:34:08 -08001246 if (err) {
1247 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilf24e9982009-10-06 11:31:10 -07001248 return err;
Sage Weil6f6c7002011-01-17 20:34:08 -08001249 }
Sage Weil7740a422010-01-08 15:58:25 -08001250 req->r_pgid = pgid;
1251
Sage Weild85b7052010-05-10 10:24:48 -07001252 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1253 if (err > 0) {
1254 o = acting[0];
1255 num = err;
1256 }
Sage Weilf24e9982009-10-06 11:31:10 -07001257
Sage Weil38d64532011-10-14 13:33:55 -07001258 if ((!force_resend &&
1259 req->r_osd && req->r_osd->o_osd == o &&
Sage Weild85b7052010-05-10 10:24:48 -07001260 req->r_sent >= req->r_osd->o_incarnation &&
1261 req->r_num_pg_osds == num &&
1262 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001263 (req->r_osd == NULL && o == -1))
1264 return 0; /* no change */
1265
Sage Weil5b191d92013-02-23 10:38:16 -08001266 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1267 req->r_tid, pgid.pool, pgid.seed, o,
Sage Weilf24e9982009-10-06 11:31:10 -07001268 req->r_osd ? req->r_osd->o_osd : -1);
1269
Sage Weild85b7052010-05-10 10:24:48 -07001270 /* record full pg acting set */
1271 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1272 req->r_num_pg_osds = num;
1273
Sage Weilf24e9982009-10-06 11:31:10 -07001274 if (req->r_osd) {
1275 __cancel_request(req);
1276 list_del_init(&req->r_osd_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001277 req->r_osd = NULL;
1278 }
1279
1280 req->r_osd = __lookup_osd(osdc, o);
1281 if (!req->r_osd && o >= 0) {
Sage Weilc99eb1c2010-02-26 09:37:33 -08001282 err = -ENOMEM;
Alex Eldere10006f2012-05-26 23:26:43 -05001283 req->r_osd = create_osd(osdc, o);
Sage Weil6f6c7002011-01-17 20:34:08 -08001284 if (!req->r_osd) {
1285 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilc99eb1c2010-02-26 09:37:33 -08001286 goto out;
Sage Weil6f6c7002011-01-17 20:34:08 -08001287 }
Sage Weilf24e9982009-10-06 11:31:10 -07001288
Sage Weil6f6c7002011-01-17 20:34:08 -08001289 dout("map_request osd %p is osd%d\n", req->r_osd, o);
Sage Weilf24e9982009-10-06 11:31:10 -07001290 __insert_osd(osdc, req->r_osd);
1291
Sage Weilb7a9e5d2012-06-27 12:24:08 -07001292 ceph_con_open(&req->r_osd->o_con,
1293 CEPH_ENTITY_TYPE_OSD, o,
1294 &osdc->osdmap->osd_addr[o]);
Sage Weilf24e9982009-10-06 11:31:10 -07001295 }
1296
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001297 if (req->r_osd) {
1298 __remove_osd_from_lru(req->r_osd);
Alex Elderad885922013-03-25 18:16:11 -05001299 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1300 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
Sage Weil6f6c7002011-01-17 20:34:08 -08001301 } else {
Alex Elderad885922013-03-25 18:16:11 -05001302 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001303 }
Sage Weild85b7052010-05-10 10:24:48 -07001304 err = 1; /* osd or pg changed */
Sage Weilf24e9982009-10-06 11:31:10 -07001305
1306out:
Sage Weilf24e9982009-10-06 11:31:10 -07001307 return err;
1308}
1309
1310/*
1311 * caller should hold map_sem (for read) and request_mutex
1312 */
Sage Weil56e925b2012-01-03 12:34:34 -08001313static void __send_request(struct ceph_osd_client *osdc,
1314 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001315{
Sage Weil1b83bef2013-02-25 16:11:12 -08001316 void *p;
Sage Weilf24e9982009-10-06 11:31:10 -07001317
Sage Weil1b83bef2013-02-25 16:11:12 -08001318 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1319 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1320 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
Sage Weilf24e9982009-10-06 11:31:10 -07001321
Sage Weil1b83bef2013-02-25 16:11:12 -08001322 /* fill in message content that changes each time we send it */
1323 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1324 put_unaligned_le32(req->r_flags, req->r_request_flags);
1325 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1326 p = req->r_request_pgid;
1327 ceph_encode_64(&p, req->r_pgid.pool);
1328 ceph_encode_32(&p, req->r_pgid.seed);
1329 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1330 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1331 sizeof(req->r_reassert_version));
Sage Weil2169aea2013-02-25 16:13:08 -08001332
Sage Weil3dd72fc2010-03-22 14:42:30 -07001333 req->r_stamp = jiffies;
Henry C Chang07a27e22010-08-22 21:34:27 -07001334 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001335
1336 ceph_msg_get(req->r_request); /* send consumes a ref */
Alex Elder26be8802013-04-15 11:20:42 -05001337
1338 /* Mark the request unsafe if this is the first timet's being sent. */
1339
1340 if (!req->r_sent && req->r_unsafe_callback)
1341 req->r_unsafe_callback(req, true);
Sage Weilf24e9982009-10-06 11:31:10 -07001342 req->r_sent = req->r_osd->o_incarnation;
Alex Elder26be8802013-04-15 11:20:42 -05001343
1344 ceph_con_send(&req->r_osd->o_con, req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001345}
1346
1347/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001348 * Send any requests in the queue (req_unsent).
1349 */
Alex Elderf9d25192013-02-15 11:42:29 -06001350static void __send_queued(struct ceph_osd_client *osdc)
Sage Weil6f6c7002011-01-17 20:34:08 -08001351{
1352 struct ceph_osd_request *req, *tmp;
1353
Alex Elderf9d25192013-02-15 11:42:29 -06001354 dout("__send_queued\n");
1355 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
Sage Weil6f6c7002011-01-17 20:34:08 -08001356 __send_request(osdc, req);
Sage Weil6f6c7002011-01-17 20:34:08 -08001357}
1358
1359/*
Sage Weilf24e9982009-10-06 11:31:10 -07001360 * Timeout callback, called every N seconds when 1 or more osd
1361 * requests has been active for more than N seconds. When this
1362 * happens, we ping all OSDs with requests who have timed out to
1363 * ensure any communications channel reset is detected. Reset the
1364 * request timeouts another N seconds in the future as we go.
1365 * Reschedule the timeout event another N seconds in future (unless
1366 * there are no open requests).
1367 */
1368static void handle_timeout(struct work_struct *work)
1369{
1370 struct ceph_osd_client *osdc =
1371 container_of(work, struct ceph_osd_client, timeout_work.work);
Sage Weil83aff952012-11-28 12:28:24 -08001372 struct ceph_osd_request *req;
Sage Weilf24e9982009-10-06 11:31:10 -07001373 struct ceph_osd *osd;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001374 unsigned long keepalive =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001375 osdc->client->options->osd_keepalive_timeout * HZ;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001376 struct list_head slow_osds;
Sage Weilf24e9982009-10-06 11:31:10 -07001377 dout("timeout\n");
1378 down_read(&osdc->map_sem);
1379
1380 ceph_monc_request_next_osdmap(&osdc->client->monc);
1381
1382 mutex_lock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001383
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001384 /*
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001385 * ping osds that are a bit slow. this ensures that if there
1386 * is a break in the TCP connection we will notice, and reopen
1387 * a connection with that osd (from the fault callback).
1388 */
1389 INIT_LIST_HEAD(&slow_osds);
1390 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
Sage Weil3dd72fc2010-03-22 14:42:30 -07001391 if (time_before(jiffies, req->r_stamp + keepalive))
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001392 break;
1393
1394 osd = req->r_osd;
1395 BUG_ON(!osd);
1396 dout(" tid %llu is slow, will send keepalive on osd%d\n",
Sage Weilf24e9982009-10-06 11:31:10 -07001397 req->r_tid, osd->o_osd);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001398 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1399 }
1400 while (!list_empty(&slow_osds)) {
1401 osd = list_entry(slow_osds.next, struct ceph_osd,
1402 o_keepalive_item);
1403 list_del_init(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001404 ceph_con_keepalive(&osd->o_con);
1405 }
1406
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001407 __schedule_osd_timeout(osdc);
Alex Elderf9d25192013-02-15 11:42:29 -06001408 __send_queued(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001409 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001410 up_read(&osdc->map_sem);
1411}
1412
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001413static void handle_osds_timeout(struct work_struct *work)
1414{
1415 struct ceph_osd_client *osdc =
1416 container_of(work, struct ceph_osd_client,
1417 osds_timeout_work.work);
1418 unsigned long delay =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001419 osdc->client->options->osd_idle_ttl * HZ >> 2;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001420
1421 dout("osds timeout\n");
1422 down_read(&osdc->map_sem);
Sage Weilaca420b2011-08-31 14:45:53 -07001423 remove_old_osds(osdc);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001424 up_read(&osdc->map_sem);
1425
1426 schedule_delayed_work(&osdc->osds_timeout_work,
1427 round_jiffies_relative(delay));
1428}
1429
Sage Weil25845472011-06-03 09:37:09 -07001430static void complete_request(struct ceph_osd_request *req)
1431{
Alex Elder26be8802013-04-15 11:20:42 -05001432 if (req->r_unsafe_callback)
1433 req->r_unsafe_callback(req, false);
Sage Weil25845472011-06-03 09:37:09 -07001434 complete_all(&req->r_safe_completion); /* fsync waiter */
1435}
1436
Sage Weilf24e9982009-10-06 11:31:10 -07001437/*
1438 * handle osd op reply. either call the callback if it is specified,
1439 * or do the completion to wake up the waiting thread.
1440 */
Sage Weil350b1c32009-12-22 10:45:45 -08001441static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1442 struct ceph_connection *con)
Sage Weilf24e9982009-10-06 11:31:10 -07001443{
Sage Weil1b83bef2013-02-25 16:11:12 -08001444 void *p, *end;
Sage Weilf24e9982009-10-06 11:31:10 -07001445 struct ceph_osd_request *req;
1446 u64 tid;
Sage Weil1b83bef2013-02-25 16:11:12 -08001447 int object_len;
Alex Elder79528732013-04-03 21:32:51 -05001448 unsigned int numops;
1449 int payload_len, flags;
Sage Weil0ceed5d2010-05-11 09:53:18 -07001450 s32 result;
Sage Weil1b83bef2013-02-25 16:11:12 -08001451 s32 retry_attempt;
1452 struct ceph_pg pg;
1453 int err;
1454 u32 reassert_epoch;
1455 u64 reassert_version;
1456 u32 osdmap_epoch;
Alex Elder0d5af162013-02-27 10:26:25 -06001457 int already_completed;
Alex Elder9fc6e062013-04-03 01:28:57 -05001458 u32 bytes;
Alex Elder79528732013-04-03 21:32:51 -05001459 unsigned int i;
Sage Weilf24e9982009-10-06 11:31:10 -07001460
Sage Weil6df058c2009-12-22 11:24:33 -08001461 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil1b83bef2013-02-25 16:11:12 -08001462 dout("handle_reply %p tid %llu\n", msg, tid);
1463
1464 p = msg->front.iov_base;
1465 end = p + msg->front.iov_len;
1466
1467 ceph_decode_need(&p, end, 4, bad);
1468 object_len = ceph_decode_32(&p);
1469 ceph_decode_need(&p, end, object_len, bad);
1470 p += object_len;
1471
Alex Elderef4859d2013-04-01 18:58:26 -05001472 err = ceph_decode_pgid(&p, end, &pg);
Sage Weil1b83bef2013-02-25 16:11:12 -08001473 if (err)
Sage Weilf24e9982009-10-06 11:31:10 -07001474 goto bad;
Sage Weil1b83bef2013-02-25 16:11:12 -08001475
1476 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1477 flags = ceph_decode_64(&p);
1478 result = ceph_decode_32(&p);
1479 reassert_epoch = ceph_decode_32(&p);
1480 reassert_version = ceph_decode_64(&p);
1481 osdmap_epoch = ceph_decode_32(&p);
1482
Sage Weilf24e9982009-10-06 11:31:10 -07001483 /* lookup */
1484 mutex_lock(&osdc->request_mutex);
1485 req = __lookup_request(osdc, tid);
1486 if (req == NULL) {
1487 dout("handle_reply tid %llu dne\n", tid);
Alex Elder8058fd42013-04-01 18:58:26 -05001488 goto bad_mutex;
Sage Weilf24e9982009-10-06 11:31:10 -07001489 }
1490 ceph_osdc_get_request(req);
Sage Weil1b83bef2013-02-25 16:11:12 -08001491
1492 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1493 req, result);
1494
1495 ceph_decode_need(&p, end, 4, bad);
1496 numops = ceph_decode_32(&p);
1497 if (numops > CEPH_OSD_MAX_OP)
1498 goto bad_put;
1499 if (numops != req->r_num_ops)
1500 goto bad_put;
1501 payload_len = 0;
1502 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1503 for (i = 0; i < numops; i++) {
1504 struct ceph_osd_op *op = p;
1505 int len;
1506
1507 len = le32_to_cpu(op->payload_len);
1508 req->r_reply_op_len[i] = len;
1509 dout(" op %d has %d bytes\n", i, len);
1510 payload_len += len;
1511 p += sizeof(*op);
1512 }
Alex Elder9fc6e062013-04-03 01:28:57 -05001513 bytes = le32_to_cpu(msg->hdr.data_len);
1514 if (payload_len != bytes) {
Sage Weil1b83bef2013-02-25 16:11:12 -08001515 pr_warning("sum of op payload lens %d != data_len %d",
Alex Elder9fc6e062013-04-03 01:28:57 -05001516 payload_len, bytes);
Sage Weil1b83bef2013-02-25 16:11:12 -08001517 goto bad_put;
1518 }
1519
1520 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1521 retry_attempt = ceph_decode_32(&p);
1522 for (i = 0; i < numops; i++)
1523 req->r_reply_op_result[i] = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001524
Sage Weilf24e9982009-10-06 11:31:10 -07001525 if (!req->r_got_reply) {
Sage Weilf24e9982009-10-06 11:31:10 -07001526
Sage Weil1b83bef2013-02-25 16:11:12 -08001527 req->r_result = result;
Sage Weilf24e9982009-10-06 11:31:10 -07001528 dout("handle_reply result %d bytes %d\n", req->r_result,
1529 bytes);
1530 if (req->r_result == 0)
1531 req->r_result = bytes;
1532
1533 /* in case this is a write and we need to replay, */
Sage Weil1b83bef2013-02-25 16:11:12 -08001534 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1535 req->r_reassert_version.version = cpu_to_le64(reassert_version);
Sage Weilf24e9982009-10-06 11:31:10 -07001536
1537 req->r_got_reply = 1;
1538 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1539 dout("handle_reply tid %llu dup ack\n", tid);
Sage Weil34b43a52009-12-01 12:23:54 -08001540 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001541 goto done;
1542 }
1543
1544 dout("handle_reply tid %llu flags %d\n", tid, flags);
1545
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001546 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1547 __register_linger_request(osdc, req);
1548
Sage Weilf24e9982009-10-06 11:31:10 -07001549 /* either this is a read, or we got the safe response */
Sage Weil0ceed5d2010-05-11 09:53:18 -07001550 if (result < 0 ||
1551 (flags & CEPH_OSD_FLAG_ONDISK) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001552 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1553 __unregister_request(osdc, req);
1554
Alex Elder0d5af162013-02-27 10:26:25 -06001555 already_completed = req->r_completed;
1556 req->r_completed = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07001557 mutex_unlock(&osdc->request_mutex);
Alex Elder0d5af162013-02-27 10:26:25 -06001558 if (already_completed)
1559 goto done;
Sage Weilf24e9982009-10-06 11:31:10 -07001560
1561 if (req->r_callback)
1562 req->r_callback(req, msg);
1563 else
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001564 complete_all(&req->r_completion);
Sage Weilf24e9982009-10-06 11:31:10 -07001565
Sage Weil25845472011-06-03 09:37:09 -07001566 if (flags & CEPH_OSD_FLAG_ONDISK)
1567 complete_request(req);
Sage Weilf24e9982009-10-06 11:31:10 -07001568
1569done:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001570 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07001571 ceph_osdc_put_request(req);
1572 return;
1573
Sage Weil1b83bef2013-02-25 16:11:12 -08001574bad_put:
1575 ceph_osdc_put_request(req);
Alex Elder8058fd42013-04-01 18:58:26 -05001576bad_mutex:
1577 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001578bad:
Sage Weil1b83bef2013-02-25 16:11:12 -08001579 pr_err("corrupt osd_op_reply got %d %d\n",
1580 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
Sage Weil9ec7cab2009-12-14 15:13:47 -08001581 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001582}
1583
Sage Weil6f6c7002011-01-17 20:34:08 -08001584static void reset_changed_osds(struct ceph_osd_client *osdc)
Sage Weilf24e9982009-10-06 11:31:10 -07001585{
Sage Weilf24e9982009-10-06 11:31:10 -07001586 struct rb_node *p, *n;
Sage Weilf24e9982009-10-06 11:31:10 -07001587
Sage Weil6f6c7002011-01-17 20:34:08 -08001588 for (p = rb_first(&osdc->osds); p; p = n) {
1589 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
Sage Weilf24e9982009-10-06 11:31:10 -07001590
Sage Weil6f6c7002011-01-17 20:34:08 -08001591 n = rb_next(p);
1592 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1593 memcmp(&osd->o_con.peer_addr,
1594 ceph_osd_addr(osdc->osdmap,
1595 osd->o_osd),
1596 sizeof(struct ceph_entity_addr)) != 0)
1597 __reset_osd(osdc, osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001598 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001599}
1600
1601/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001602 * Requeue requests whose mapping to an OSD has changed. If requests map to
1603 * no osd, request a new map.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001604 *
Alex Eldere6d50f62012-12-26 14:31:40 -06001605 * Caller should hold map_sem for read.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001606 */
Sage Weil38d64532011-10-14 13:33:55 -07001607static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001608{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001609 struct ceph_osd_request *req, *nreq;
Sage Weil6f6c7002011-01-17 20:34:08 -08001610 struct rb_node *p;
1611 int needmap = 0;
1612 int err;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001613
Sage Weil38d64532011-10-14 13:33:55 -07001614 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001615 mutex_lock(&osdc->request_mutex);
Sage Weil6194ea82012-07-30 16:19:28 -07001616 for (p = rb_first(&osdc->requests); p; ) {
Sage Weil6f6c7002011-01-17 20:34:08 -08001617 req = rb_entry(p, struct ceph_osd_request, r_node);
Sage Weil6194ea82012-07-30 16:19:28 -07001618 p = rb_next(p);
Alex Elderab60b162012-12-19 15:52:36 -06001619
1620 /*
1621 * For linger requests that have not yet been
1622 * registered, move them to the linger list; they'll
1623 * be sent to the osd in the loop below. Unregister
1624 * the request before re-registering it as a linger
1625 * request to ensure the __map_request() below
1626 * will decide it needs to be sent.
1627 */
1628 if (req->r_linger && list_empty(&req->r_linger_item)) {
1629 dout("%p tid %llu restart on osd%d\n",
1630 req, req->r_tid,
1631 req->r_osd ? req->r_osd->o_osd : -1);
Alex Elder96e4dac2013-05-22 20:54:25 -05001632 ceph_osdc_get_request(req);
Alex Elderab60b162012-12-19 15:52:36 -06001633 __unregister_request(osdc, req);
1634 __register_linger_request(osdc, req);
Alex Elder96e4dac2013-05-22 20:54:25 -05001635 ceph_osdc_put_request(req);
Alex Elderab60b162012-12-19 15:52:36 -06001636 continue;
1637 }
1638
Sage Weil38d64532011-10-14 13:33:55 -07001639 err = __map_request(osdc, req, force_resend);
Sage Weil6f6c7002011-01-17 20:34:08 -08001640 if (err < 0)
1641 continue; /* error */
1642 if (req->r_osd == NULL) {
1643 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1644 needmap++; /* request a newer map */
1645 } else if (err > 0) {
Sage Weil6194ea82012-07-30 16:19:28 -07001646 if (!req->r_linger) {
1647 dout("%p tid %llu requeued on osd%d\n", req,
1648 req->r_tid,
1649 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001650 req->r_flags |= CEPH_OSD_FLAG_RETRY;
Sage Weil6194ea82012-07-30 16:19:28 -07001651 }
1652 }
Sage Weil6f6c7002011-01-17 20:34:08 -08001653 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001654
1655 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1656 r_linger_item) {
1657 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1658
Sage Weil38d64532011-10-14 13:33:55 -07001659 err = __map_request(osdc, req, force_resend);
Alex Elderab60b162012-12-19 15:52:36 -06001660 dout("__map_request returned %d\n", err);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001661 if (err == 0)
1662 continue; /* no change and no osd was specified */
1663 if (err < 0)
1664 continue; /* hrm! */
1665 if (req->r_osd == NULL) {
1666 dout("tid %llu maps to no valid osd\n", req->r_tid);
1667 needmap++; /* request a newer map */
1668 continue;
1669 }
1670
1671 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1672 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001673 __register_request(osdc, req);
Alex Elderc89ce052012-12-06 07:22:04 -06001674 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001675 }
Alex Elder14d2f382013-05-15 16:28:33 -05001676 reset_changed_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001677 mutex_unlock(&osdc->request_mutex);
1678
1679 if (needmap) {
1680 dout("%d requests for down osds, need new map\n", needmap);
1681 ceph_monc_request_next_osdmap(&osdc->client->monc);
1682 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001683}
Sage Weil6f6c7002011-01-17 20:34:08 -08001684
1685
Sage Weilf24e9982009-10-06 11:31:10 -07001686/*
1687 * Process updated osd map.
1688 *
1689 * The message contains any number of incremental and full maps, normally
1690 * indicating some sort of topology change in the cluster. Kick requests
1691 * off to different OSDs as needed.
1692 */
1693void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1694{
1695 void *p, *end, *next;
1696 u32 nr_maps, maplen;
1697 u32 epoch;
1698 struct ceph_osdmap *newmap = NULL, *oldmap;
1699 int err;
1700 struct ceph_fsid fsid;
1701
1702 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1703 p = msg->front.iov_base;
1704 end = p + msg->front.iov_len;
1705
1706 /* verify fsid */
1707 ceph_decode_need(&p, end, sizeof(fsid), bad);
1708 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Sage Weil07433042009-11-18 16:50:41 -08001709 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1710 return;
Sage Weilf24e9982009-10-06 11:31:10 -07001711
1712 down_write(&osdc->map_sem);
1713
1714 /* incremental maps */
1715 ceph_decode_32_safe(&p, end, nr_maps, bad);
1716 dout(" %d inc maps\n", nr_maps);
1717 while (nr_maps > 0) {
1718 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07001719 epoch = ceph_decode_32(&p);
1720 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001721 ceph_decode_need(&p, end, maplen, bad);
1722 next = p + maplen;
1723 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1724 dout("applying incremental map %u len %d\n",
1725 epoch, maplen);
1726 newmap = osdmap_apply_incremental(&p, next,
1727 osdc->osdmap,
Alex Elder15d98822012-05-26 23:26:43 -05001728 &osdc->client->msgr);
Sage Weilf24e9982009-10-06 11:31:10 -07001729 if (IS_ERR(newmap)) {
1730 err = PTR_ERR(newmap);
1731 goto bad;
1732 }
Sage Weil30dc6382009-12-21 14:49:37 -08001733 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07001734 if (newmap != osdc->osdmap) {
1735 ceph_osdmap_destroy(osdc->osdmap);
1736 osdc->osdmap = newmap;
1737 }
Sage Weil38d64532011-10-14 13:33:55 -07001738 kick_requests(osdc, 0);
Sage Weilf24e9982009-10-06 11:31:10 -07001739 } else {
1740 dout("ignoring incremental map %u len %d\n",
1741 epoch, maplen);
1742 }
1743 p = next;
1744 nr_maps--;
1745 }
1746 if (newmap)
1747 goto done;
1748
1749 /* full maps */
1750 ceph_decode_32_safe(&p, end, nr_maps, bad);
1751 dout(" %d full maps\n", nr_maps);
1752 while (nr_maps) {
1753 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07001754 epoch = ceph_decode_32(&p);
1755 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001756 ceph_decode_need(&p, end, maplen, bad);
1757 if (nr_maps > 1) {
1758 dout("skipping non-latest full map %u len %d\n",
1759 epoch, maplen);
1760 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1761 dout("skipping full map %u len %d, "
1762 "older than our %u\n", epoch, maplen,
1763 osdc->osdmap->epoch);
1764 } else {
Sage Weil38d64532011-10-14 13:33:55 -07001765 int skipped_map = 0;
1766
Sage Weilf24e9982009-10-06 11:31:10 -07001767 dout("taking full map %u len %d\n", epoch, maplen);
1768 newmap = osdmap_decode(&p, p+maplen);
1769 if (IS_ERR(newmap)) {
1770 err = PTR_ERR(newmap);
1771 goto bad;
1772 }
Sage Weil30dc6382009-12-21 14:49:37 -08001773 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07001774 oldmap = osdc->osdmap;
1775 osdc->osdmap = newmap;
Sage Weil38d64532011-10-14 13:33:55 -07001776 if (oldmap) {
1777 if (oldmap->epoch + 1 < newmap->epoch)
1778 skipped_map = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07001779 ceph_osdmap_destroy(oldmap);
Sage Weil38d64532011-10-14 13:33:55 -07001780 }
1781 kick_requests(osdc, skipped_map);
Sage Weilf24e9982009-10-06 11:31:10 -07001782 }
1783 p += maplen;
1784 nr_maps--;
1785 }
1786
1787done:
1788 downgrade_write(&osdc->map_sem);
1789 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
Sage Weilcd634fb2011-05-12 09:29:18 -07001790
1791 /*
1792 * subscribe to subsequent osdmap updates if full to ensure
1793 * we find out when we are no longer full and stop returning
1794 * ENOSPC.
1795 */
1796 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1797 ceph_monc_request_next_osdmap(&osdc->client->monc);
1798
Alex Elderf9d25192013-02-15 11:42:29 -06001799 mutex_lock(&osdc->request_mutex);
1800 __send_queued(osdc);
1801 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001802 up_read(&osdc->map_sem);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001803 wake_up_all(&osdc->client->auth_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07001804 return;
1805
1806bad:
1807 pr_err("osdc handle_map corrupt msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08001808 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001809 up_write(&osdc->map_sem);
1810 return;
1811}
1812
Sage Weilf24e9982009-10-06 11:31:10 -07001813/*
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001814 * watch/notify callback event infrastructure
1815 *
1816 * These callbacks are used both for watch and notify operations.
1817 */
1818static void __release_event(struct kref *kref)
1819{
1820 struct ceph_osd_event *event =
1821 container_of(kref, struct ceph_osd_event, kref);
1822
1823 dout("__release_event %p\n", event);
1824 kfree(event);
1825}
1826
1827static void get_event(struct ceph_osd_event *event)
1828{
1829 kref_get(&event->kref);
1830}
1831
1832void ceph_osdc_put_event(struct ceph_osd_event *event)
1833{
1834 kref_put(&event->kref, __release_event);
1835}
1836EXPORT_SYMBOL(ceph_osdc_put_event);
1837
1838static void __insert_event(struct ceph_osd_client *osdc,
1839 struct ceph_osd_event *new)
1840{
1841 struct rb_node **p = &osdc->event_tree.rb_node;
1842 struct rb_node *parent = NULL;
1843 struct ceph_osd_event *event = NULL;
1844
1845 while (*p) {
1846 parent = *p;
1847 event = rb_entry(parent, struct ceph_osd_event, node);
1848 if (new->cookie < event->cookie)
1849 p = &(*p)->rb_left;
1850 else if (new->cookie > event->cookie)
1851 p = &(*p)->rb_right;
1852 else
1853 BUG();
1854 }
1855
1856 rb_link_node(&new->node, parent, p);
1857 rb_insert_color(&new->node, &osdc->event_tree);
1858}
1859
1860static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1861 u64 cookie)
1862{
1863 struct rb_node **p = &osdc->event_tree.rb_node;
1864 struct rb_node *parent = NULL;
1865 struct ceph_osd_event *event = NULL;
1866
1867 while (*p) {
1868 parent = *p;
1869 event = rb_entry(parent, struct ceph_osd_event, node);
1870 if (cookie < event->cookie)
1871 p = &(*p)->rb_left;
1872 else if (cookie > event->cookie)
1873 p = &(*p)->rb_right;
1874 else
1875 return event;
1876 }
1877 return NULL;
1878}
1879
1880static void __remove_event(struct ceph_osd_event *event)
1881{
1882 struct ceph_osd_client *osdc = event->osdc;
1883
1884 if (!RB_EMPTY_NODE(&event->node)) {
1885 dout("__remove_event removed %p\n", event);
1886 rb_erase(&event->node, &osdc->event_tree);
1887 ceph_osdc_put_event(event);
1888 } else {
1889 dout("__remove_event didn't remove %p\n", event);
1890 }
1891}
1892
1893int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1894 void (*event_cb)(u64, u64, u8, void *),
Alex Elder3c663bb2013-02-15 11:42:30 -06001895 void *data, struct ceph_osd_event **pevent)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001896{
1897 struct ceph_osd_event *event;
1898
1899 event = kmalloc(sizeof(*event), GFP_NOIO);
1900 if (!event)
1901 return -ENOMEM;
1902
1903 dout("create_event %p\n", event);
1904 event->cb = event_cb;
Alex Elder3c663bb2013-02-15 11:42:30 -06001905 event->one_shot = 0;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001906 event->data = data;
1907 event->osdc = osdc;
1908 INIT_LIST_HEAD(&event->osd_node);
Alex Elder3ee52342012-12-17 12:23:48 -06001909 RB_CLEAR_NODE(&event->node);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001910 kref_init(&event->kref); /* one ref for us */
1911 kref_get(&event->kref); /* one ref for the caller */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001912
1913 spin_lock(&osdc->event_lock);
1914 event->cookie = ++osdc->event_count;
1915 __insert_event(osdc, event);
1916 spin_unlock(&osdc->event_lock);
1917
1918 *pevent = event;
1919 return 0;
1920}
1921EXPORT_SYMBOL(ceph_osdc_create_event);
1922
1923void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1924{
1925 struct ceph_osd_client *osdc = event->osdc;
1926
1927 dout("cancel_event %p\n", event);
1928 spin_lock(&osdc->event_lock);
1929 __remove_event(event);
1930 spin_unlock(&osdc->event_lock);
1931 ceph_osdc_put_event(event); /* caller's */
1932}
1933EXPORT_SYMBOL(ceph_osdc_cancel_event);
1934
1935
1936static void do_event_work(struct work_struct *work)
1937{
1938 struct ceph_osd_event_work *event_work =
1939 container_of(work, struct ceph_osd_event_work, work);
1940 struct ceph_osd_event *event = event_work->event;
1941 u64 ver = event_work->ver;
1942 u64 notify_id = event_work->notify_id;
1943 u8 opcode = event_work->opcode;
1944
1945 dout("do_event_work completing %p\n", event);
1946 event->cb(ver, notify_id, opcode, event->data);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001947 dout("do_event_work completed %p\n", event);
1948 ceph_osdc_put_event(event);
1949 kfree(event_work);
1950}
1951
1952
1953/*
1954 * Process osd watch notifications
1955 */
Alex Elder3c663bb2013-02-15 11:42:30 -06001956static void handle_watch_notify(struct ceph_osd_client *osdc,
1957 struct ceph_msg *msg)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001958{
1959 void *p, *end;
1960 u8 proto_ver;
1961 u64 cookie, ver, notify_id;
1962 u8 opcode;
1963 struct ceph_osd_event *event;
1964 struct ceph_osd_event_work *event_work;
1965
1966 p = msg->front.iov_base;
1967 end = p + msg->front.iov_len;
1968
1969 ceph_decode_8_safe(&p, end, proto_ver, bad);
1970 ceph_decode_8_safe(&p, end, opcode, bad);
1971 ceph_decode_64_safe(&p, end, cookie, bad);
1972 ceph_decode_64_safe(&p, end, ver, bad);
1973 ceph_decode_64_safe(&p, end, notify_id, bad);
1974
1975 spin_lock(&osdc->event_lock);
1976 event = __find_event(osdc, cookie);
1977 if (event) {
Alex Elder3c663bb2013-02-15 11:42:30 -06001978 BUG_ON(event->one_shot);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001979 get_event(event);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001980 }
1981 spin_unlock(&osdc->event_lock);
1982 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1983 cookie, ver, event);
1984 if (event) {
1985 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001986 if (!event_work) {
1987 dout("ERROR: could not allocate event_work\n");
1988 goto done_err;
1989 }
Mariusz Kozlowski6b0ae402011-03-26 19:29:34 +01001990 INIT_WORK(&event_work->work, do_event_work);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001991 event_work->event = event;
1992 event_work->ver = ver;
1993 event_work->notify_id = notify_id;
1994 event_work->opcode = opcode;
1995 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1996 dout("WARNING: failed to queue notify event work\n");
1997 goto done_err;
1998 }
1999 }
2000
2001 return;
2002
2003done_err:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002004 ceph_osdc_put_event(event);
2005 return;
2006
2007bad:
2008 pr_err("osdc handle_watch_notify corrupt msg\n");
2009 return;
2010}
2011
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002012/*
Alex Eldere65550f2013-04-05 01:27:12 -05002013 * build new request AND message
2014 *
2015 */
2016void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2017 struct ceph_snap_context *snapc, u64 snap_id,
2018 struct timespec *mtime)
2019{
2020 struct ceph_msg *msg = req->r_request;
2021 void *p;
2022 size_t msg_size;
2023 int flags = req->r_flags;
2024 u64 data_len;
2025 unsigned int i;
2026
2027 req->r_snapid = snap_id;
2028 req->r_snapc = ceph_get_snap_context(snapc);
2029
2030 /* encode request */
2031 msg->hdr.version = cpu_to_le16(4);
2032
2033 p = msg->front.iov_base;
2034 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2035 req->r_request_osdmap_epoch = p;
2036 p += 4;
2037 req->r_request_flags = p;
2038 p += 4;
2039 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2040 ceph_encode_timespec(p, mtime);
2041 p += sizeof(struct ceph_timespec);
2042 req->r_request_reassert_version = p;
2043 p += sizeof(struct ceph_eversion); /* will get filled in */
2044
2045 /* oloc */
2046 ceph_encode_8(&p, 4);
2047 ceph_encode_8(&p, 4);
2048 ceph_encode_32(&p, 8 + 4 + 4);
2049 req->r_request_pool = p;
2050 p += 8;
2051 ceph_encode_32(&p, -1); /* preferred */
2052 ceph_encode_32(&p, 0); /* key len */
2053
2054 ceph_encode_8(&p, 1);
2055 req->r_request_pgid = p;
2056 p += 8 + 4;
2057 ceph_encode_32(&p, -1); /* preferred */
2058
2059 /* oid */
2060 ceph_encode_32(&p, req->r_oid_len);
2061 memcpy(p, req->r_oid, req->r_oid_len);
2062 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2063 p += req->r_oid_len;
2064
2065 /* ops--can imply data */
2066 ceph_encode_16(&p, (u16)req->r_num_ops);
2067 data_len = 0;
2068 for (i = 0; i < req->r_num_ops; i++) {
2069 data_len += osd_req_encode_op(req, p, i);
2070 p += sizeof(struct ceph_osd_op);
2071 }
2072
2073 /* snaps */
2074 ceph_encode_64(&p, req->r_snapid);
2075 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2076 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2077 if (req->r_snapc) {
2078 for (i = 0; i < snapc->num_snaps; i++) {
2079 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2080 }
2081 }
2082
2083 req->r_request_attempts = p;
2084 p += 4;
2085
2086 /* data */
2087 if (flags & CEPH_OSD_FLAG_WRITE) {
2088 u16 data_off;
2089
2090 /*
2091 * The header "data_off" is a hint to the receiver
2092 * allowing it to align received data into its
2093 * buffers such that there's no need to re-copy
2094 * it before writing it to disk (direct I/O).
2095 */
2096 data_off = (u16) (off & 0xffff);
2097 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2098 }
2099 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2100
2101 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2102 msg_size = p - msg->front.iov_base;
2103 msg->front.iov_len = msg_size;
2104 msg->hdr.front_len = cpu_to_le32(msg_size);
2105
2106 dout("build_request msg_size was %d\n", (int)msg_size);
2107}
2108EXPORT_SYMBOL(ceph_osdc_build_request);
2109
2110/*
Sage Weilf24e9982009-10-06 11:31:10 -07002111 * Register request, send initial attempt.
2112 */
2113int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2114 struct ceph_osd_request *req,
2115 bool nofail)
2116{
Sage Weilc1ea8822009-10-08 16:55:47 -07002117 int rc = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07002118
Sage Weilf24e9982009-10-06 11:31:10 -07002119 down_read(&osdc->map_sem);
2120 mutex_lock(&osdc->request_mutex);
Alex Elderdc4b8702013-03-25 18:16:11 -05002121 __register_request(osdc, req);
Alex Elderc10ebbf2013-05-09 14:56:32 -05002122 req->r_sent = 0;
2123 req->r_got_reply = 0;
2124 req->r_completed = 0;
Alex Elder92451b42013-03-25 18:16:11 -05002125 rc = __map_request(osdc, req, 0);
2126 if (rc < 0) {
2127 if (nofail) {
2128 dout("osdc_start_request failed map, "
2129 " will retry %lld\n", req->r_tid);
2130 rc = 0;
Sage Weil9d6fcb02011-05-12 15:48:16 -07002131 }
Alex Elder92451b42013-03-25 18:16:11 -05002132 goto out_unlock;
Sage Weilf24e9982009-10-06 11:31:10 -07002133 }
Alex Elder92451b42013-03-25 18:16:11 -05002134 if (req->r_osd == NULL) {
2135 dout("send_request %p no up osds in pg\n", req);
2136 ceph_monc_request_next_osdmap(&osdc->client->monc);
2137 } else {
Alex Elder7e2766a2013-03-25 18:16:11 -05002138 __send_queued(osdc);
Alex Elder92451b42013-03-25 18:16:11 -05002139 }
2140 rc = 0;
Dan Carpenter234af26f2011-03-29 06:25:59 +03002141out_unlock:
Sage Weilf24e9982009-10-06 11:31:10 -07002142 mutex_unlock(&osdc->request_mutex);
2143 up_read(&osdc->map_sem);
2144 return rc;
2145}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002146EXPORT_SYMBOL(ceph_osdc_start_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002147
2148/*
2149 * wait for a request to complete
2150 */
2151int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2152 struct ceph_osd_request *req)
2153{
2154 int rc;
2155
2156 rc = wait_for_completion_interruptible(&req->r_completion);
2157 if (rc < 0) {
2158 mutex_lock(&osdc->request_mutex);
2159 __cancel_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08002160 __unregister_request(osdc, req);
Sage Weilf24e9982009-10-06 11:31:10 -07002161 mutex_unlock(&osdc->request_mutex);
Sage Weil25845472011-06-03 09:37:09 -07002162 complete_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08002163 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07002164 return rc;
2165 }
2166
2167 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2168 return req->r_result;
2169}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002170EXPORT_SYMBOL(ceph_osdc_wait_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002171
2172/*
2173 * sync - wait for all in-flight requests to flush. avoid starvation.
2174 */
2175void ceph_osdc_sync(struct ceph_osd_client *osdc)
2176{
2177 struct ceph_osd_request *req;
2178 u64 last_tid, next_tid = 0;
2179
2180 mutex_lock(&osdc->request_mutex);
2181 last_tid = osdc->last_tid;
2182 while (1) {
2183 req = __lookup_request_ge(osdc, next_tid);
2184 if (!req)
2185 break;
2186 if (req->r_tid > last_tid)
2187 break;
2188
2189 next_tid = req->r_tid + 1;
2190 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2191 continue;
2192
2193 ceph_osdc_get_request(req);
2194 mutex_unlock(&osdc->request_mutex);
2195 dout("sync waiting on tid %llu (last is %llu)\n",
2196 req->r_tid, last_tid);
2197 wait_for_completion(&req->r_safe_completion);
2198 mutex_lock(&osdc->request_mutex);
2199 ceph_osdc_put_request(req);
2200 }
2201 mutex_unlock(&osdc->request_mutex);
2202 dout("sync done (thru tid %llu)\n", last_tid);
2203}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002204EXPORT_SYMBOL(ceph_osdc_sync);
Sage Weilf24e9982009-10-06 11:31:10 -07002205
2206/*
2207 * init, shutdown
2208 */
2209int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2210{
2211 int err;
2212
2213 dout("init\n");
2214 osdc->client = client;
2215 osdc->osdmap = NULL;
2216 init_rwsem(&osdc->map_sem);
2217 init_completion(&osdc->map_waiters);
2218 osdc->last_requested_map = 0;
2219 mutex_init(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07002220 osdc->last_tid = 0;
2221 osdc->osds = RB_ROOT;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002222 INIT_LIST_HEAD(&osdc->osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07002223 osdc->requests = RB_ROOT;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08002224 INIT_LIST_HEAD(&osdc->req_lru);
Sage Weil6f6c7002011-01-17 20:34:08 -08002225 INIT_LIST_HEAD(&osdc->req_unsent);
2226 INIT_LIST_HEAD(&osdc->req_notarget);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002227 INIT_LIST_HEAD(&osdc->req_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07002228 osdc->num_requests = 0;
2229 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002230 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002231 spin_lock_init(&osdc->event_lock);
2232 osdc->event_tree = RB_ROOT;
2233 osdc->event_count = 0;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002234
2235 schedule_delayed_work(&osdc->osds_timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002236 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
Sage Weilf24e9982009-10-06 11:31:10 -07002237
Sage Weil5f44f142009-11-18 14:52:18 -08002238 err = -ENOMEM;
Sage Weilf24e9982009-10-06 11:31:10 -07002239 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2240 sizeof(struct ceph_osd_request));
2241 if (!osdc->req_mempool)
Sage Weil5f44f142009-11-18 14:52:18 -08002242 goto out;
Sage Weilf24e9982009-10-06 11:31:10 -07002243
Sage Weild50b4092012-07-09 14:22:34 -07002244 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2245 OSD_OP_FRONT_LEN, 10, true,
Sage Weil4f482802010-04-24 09:56:35 -07002246 "osd_op");
Sage Weilf24e9982009-10-06 11:31:10 -07002247 if (err < 0)
Sage Weil5f44f142009-11-18 14:52:18 -08002248 goto out_mempool;
Sage Weild50b4092012-07-09 14:22:34 -07002249 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
Sage Weil4f482802010-04-24 09:56:35 -07002250 OSD_OPREPLY_FRONT_LEN, 10, true,
2251 "osd_op_reply");
Sage Weilc16e7862010-03-01 13:02:00 -08002252 if (err < 0)
2253 goto out_msgpool;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002254
2255 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2256 if (IS_ERR(osdc->notify_wq)) {
2257 err = PTR_ERR(osdc->notify_wq);
2258 osdc->notify_wq = NULL;
2259 goto out_msgpool;
2260 }
Sage Weilf24e9982009-10-06 11:31:10 -07002261 return 0;
Sage Weil5f44f142009-11-18 14:52:18 -08002262
Sage Weilc16e7862010-03-01 13:02:00 -08002263out_msgpool:
2264 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weil5f44f142009-11-18 14:52:18 -08002265out_mempool:
2266 mempool_destroy(osdc->req_mempool);
2267out:
2268 return err;
Sage Weilf24e9982009-10-06 11:31:10 -07002269}
2270
2271void ceph_osdc_stop(struct ceph_osd_client *osdc)
2272{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002273 flush_workqueue(osdc->notify_wq);
2274 destroy_workqueue(osdc->notify_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07002275 cancel_delayed_work_sync(&osdc->timeout_work);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002276 cancel_delayed_work_sync(&osdc->osds_timeout_work);
Sage Weilf24e9982009-10-06 11:31:10 -07002277 if (osdc->osdmap) {
2278 ceph_osdmap_destroy(osdc->osdmap);
2279 osdc->osdmap = NULL;
2280 }
Sage Weilaca420b2011-08-31 14:45:53 -07002281 remove_all_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07002282 mempool_destroy(osdc->req_mempool);
2283 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weilc16e7862010-03-01 13:02:00 -08002284 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
Sage Weilf24e9982009-10-06 11:31:10 -07002285}
2286
2287/*
2288 * Read some contiguous pages. If we cross a stripe boundary, shorten
2289 * *plen. Return number of bytes read, or error.
2290 */
2291int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2292 struct ceph_vino vino, struct ceph_file_layout *layout,
2293 u64 off, u64 *plen,
2294 u32 truncate_seq, u64 truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -08002295 struct page **pages, int num_pages, int page_align)
Sage Weilf24e9982009-10-06 11:31:10 -07002296{
2297 struct ceph_osd_request *req;
2298 int rc = 0;
2299
2300 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2301 vino.snap, off, *plen);
Alex Elder79528732013-04-03 21:32:51 -05002302 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002303 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
Alex Elderacead002013-03-14 14:09:05 -05002304 NULL, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002305 false);
Sage Weil68162822012-09-24 21:01:02 -07002306 if (IS_ERR(req))
2307 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002308
2309 /* it may be a short read due to an object boundary */
Alex Elder0fff87e2013-02-14 12:16:43 -06002310
Alex Elder406e2c92013-04-15 14:50:36 -05002311 osd_req_op_extent_osd_data_pages(req, 0,
Alex Eldera4ce40a2013-04-05 01:27:12 -05002312 pages, *plen, page_align, false, false);
Sage Weilf24e9982009-10-06 11:31:10 -07002313
Alex Eldere0c59482013-03-07 15:38:25 -06002314 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
Alex Elder43bfe5d2013-04-03 01:28:57 -05002315 off, *plen, *plen, page_align);
Sage Weilf24e9982009-10-06 11:31:10 -07002316
Alex Elder79528732013-04-03 21:32:51 -05002317 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
Alex Elder02ee07d2013-03-14 14:09:06 -05002318
Sage Weilf24e9982009-10-06 11:31:10 -07002319 rc = ceph_osdc_start_request(osdc, req, false);
2320 if (!rc)
2321 rc = ceph_osdc_wait_request(osdc, req);
2322
2323 ceph_osdc_put_request(req);
2324 dout("readpages result %d\n", rc);
2325 return rc;
2326}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002327EXPORT_SYMBOL(ceph_osdc_readpages);
Sage Weilf24e9982009-10-06 11:31:10 -07002328
2329/*
2330 * do a synchronous write on N pages
2331 */
2332int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2333 struct ceph_file_layout *layout,
2334 struct ceph_snap_context *snapc,
2335 u64 off, u64 len,
2336 u32 truncate_seq, u64 truncate_size,
2337 struct timespec *mtime,
Alex Elder24808822013-02-15 11:42:29 -06002338 struct page **pages, int num_pages)
Sage Weilf24e9982009-10-06 11:31:10 -07002339{
2340 struct ceph_osd_request *req;
2341 int rc = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -08002342 int page_align = off & ~PAGE_MASK;
Sage Weilf24e9982009-10-06 11:31:10 -07002343
Alex Elderacead002013-03-14 14:09:05 -05002344 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
Alex Elder79528732013-04-03 21:32:51 -05002345 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002346 CEPH_OSD_OP_WRITE,
Alex Elder24808822013-02-15 11:42:29 -06002347 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
Alex Elderacead002013-03-14 14:09:05 -05002348 snapc, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002349 true);
Sage Weil68162822012-09-24 21:01:02 -07002350 if (IS_ERR(req))
2351 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002352
2353 /* it may be a short write due to an object boundary */
Alex Elder406e2c92013-04-15 14:50:36 -05002354 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
Alex Elder43bfe5d2013-04-03 01:28:57 -05002355 false, false);
2356 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
Sage Weilf24e9982009-10-06 11:31:10 -07002357
Alex Elder79528732013-04-03 21:32:51 -05002358 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
Alex Elder02ee07d2013-03-14 14:09:06 -05002359
Alex Elder87f979d2013-02-15 11:42:29 -06002360 rc = ceph_osdc_start_request(osdc, req, true);
Sage Weilf24e9982009-10-06 11:31:10 -07002361 if (!rc)
2362 rc = ceph_osdc_wait_request(osdc, req);
2363
2364 ceph_osdc_put_request(req);
2365 if (rc == 0)
2366 rc = len;
2367 dout("writepages result %d\n", rc);
2368 return rc;
2369}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002370EXPORT_SYMBOL(ceph_osdc_writepages);
Sage Weilf24e9982009-10-06 11:31:10 -07002371
Alex Elder5522ae02013-05-01 12:43:04 -05002372int ceph_osdc_setup(void)
2373{
2374 BUG_ON(ceph_osd_request_cache);
2375 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2376 sizeof (struct ceph_osd_request),
2377 __alignof__(struct ceph_osd_request),
2378 0, NULL);
2379
2380 return ceph_osd_request_cache ? 0 : -ENOMEM;
2381}
2382EXPORT_SYMBOL(ceph_osdc_setup);
2383
2384void ceph_osdc_cleanup(void)
2385{
2386 BUG_ON(!ceph_osd_request_cache);
2387 kmem_cache_destroy(ceph_osd_request_cache);
2388 ceph_osd_request_cache = NULL;
2389}
2390EXPORT_SYMBOL(ceph_osdc_cleanup);
2391
Sage Weilf24e9982009-10-06 11:31:10 -07002392/*
2393 * handle incoming message
2394 */
2395static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2396{
2397 struct ceph_osd *osd = con->private;
Julia Lawall32c895e2009-11-21 16:53:16 +01002398 struct ceph_osd_client *osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002399 int type = le16_to_cpu(msg->hdr.type);
2400
2401 if (!osd)
Sage Weil4a32f932010-06-13 10:27:53 -07002402 goto out;
Julia Lawall32c895e2009-11-21 16:53:16 +01002403 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002404
2405 switch (type) {
2406 case CEPH_MSG_OSD_MAP:
2407 ceph_osdc_handle_map(osdc, msg);
2408 break;
2409 case CEPH_MSG_OSD_OPREPLY:
Sage Weil350b1c32009-12-22 10:45:45 -08002410 handle_reply(osdc, msg, con);
Sage Weilf24e9982009-10-06 11:31:10 -07002411 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002412 case CEPH_MSG_WATCH_NOTIFY:
2413 handle_watch_notify(osdc, msg);
2414 break;
Sage Weilf24e9982009-10-06 11:31:10 -07002415
2416 default:
2417 pr_err("received unknown message type %d %s\n", type,
2418 ceph_msg_type_name(type));
2419 }
Sage Weil4a32f932010-06-13 10:27:53 -07002420out:
Sage Weilf24e9982009-10-06 11:31:10 -07002421 ceph_msg_put(msg);
2422}
2423
Sage Weil5b3a4db2010-02-19 21:43:23 -08002424/*
Sage Weil21b667f2010-03-04 10:22:59 -08002425 * lookup and return message for incoming reply. set up reply message
2426 * pages.
Sage Weil5b3a4db2010-02-19 21:43:23 -08002427 */
2428static struct ceph_msg *get_reply(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002429 struct ceph_msg_header *hdr,
2430 int *skip)
Sage Weilf24e9982009-10-06 11:31:10 -07002431{
2432 struct ceph_osd *osd = con->private;
2433 struct ceph_osd_client *osdc = osd->o_osdc;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002434 struct ceph_msg *m;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002435 struct ceph_osd_request *req;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002436 int front = le32_to_cpu(hdr->front_len);
2437 int data_len = le32_to_cpu(hdr->data_len);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002438 u64 tid;
Sage Weilf24e9982009-10-06 11:31:10 -07002439
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002440 tid = le64_to_cpu(hdr->tid);
2441 mutex_lock(&osdc->request_mutex);
2442 req = __lookup_request(osdc, tid);
2443 if (!req) {
2444 *skip = 1;
2445 m = NULL;
Sage Weil756a16a2012-07-30 16:26:13 -07002446 dout("get_reply unknown tid %llu from osd%d\n", tid,
2447 osd->o_osd);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002448 goto out;
2449 }
Sage Weilc16e7862010-03-01 13:02:00 -08002450
Alex Elderace6d3a2013-04-01 16:12:14 -05002451 if (req->r_reply->con)
Alex Elder8921d112012-06-01 14:56:43 -05002452 dout("%s revoking msg %p from old con %p\n", __func__,
Alex Elderace6d3a2013-04-01 16:12:14 -05002453 req->r_reply, req->r_reply->con);
2454 ceph_msg_revoke_incoming(req->r_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002455
Sage Weilc16e7862010-03-01 13:02:00 -08002456 if (front > req->r_reply->front.iov_len) {
Alex Elder49743412013-05-29 11:19:00 -05002457 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2458 front, (int)req->r_reply->front.iov_len,
2459 (unsigned int)con->peer_name.type,
2460 le64_to_cpu(con->peer_name.num));
Sage Weilb61c2762011-08-09 15:03:46 -07002461 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07002462 if (!m)
Sage Weilc16e7862010-03-01 13:02:00 -08002463 goto out;
2464 ceph_msg_put(req->r_reply);
2465 req->r_reply = m;
2466 }
2467 m = ceph_msg_get(req->r_reply);
2468
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002469 if (data_len > 0) {
Alex Eldera4ce40a2013-04-05 01:27:12 -05002470 struct ceph_osd_data *osd_data;
Alex Elder0fff87e2013-02-14 12:16:43 -06002471
Alex Eldera4ce40a2013-04-05 01:27:12 -05002472 /*
2473 * XXX This is assuming there is only one op containing
2474 * XXX page data. Probably OK for reads, but this
2475 * XXX ought to be done more generally.
2476 */
Alex Elder406e2c92013-04-15 14:50:36 -05002477 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder0fff87e2013-02-14 12:16:43 -06002478 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
Alex Elder0fff87e2013-02-14 12:16:43 -06002479 if (osd_data->pages &&
Alex Eldere0c59482013-03-07 15:38:25 -06002480 unlikely(osd_data->length < data_len)) {
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002481
Alex Eldere0c59482013-03-07 15:38:25 -06002482 pr_warning("tid %lld reply has %d bytes "
2483 "we had only %llu bytes ready\n",
2484 tid, data_len, osd_data->length);
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002485 *skip = 1;
2486 ceph_msg_put(m);
2487 m = NULL;
2488 goto out;
2489 }
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002490 }
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002491 }
Sage Weil5b3a4db2010-02-19 21:43:23 -08002492 *skip = 0;
Sage Weilc16e7862010-03-01 13:02:00 -08002493 dout("get_reply tid %lld %p\n", tid, m);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002494
2495out:
2496 mutex_unlock(&osdc->request_mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002497 return m;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002498
2499}
2500
2501static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2502 struct ceph_msg_header *hdr,
2503 int *skip)
2504{
2505 struct ceph_osd *osd = con->private;
2506 int type = le16_to_cpu(hdr->type);
2507 int front = le32_to_cpu(hdr->front_len);
2508
Alex Elder1c20f2d2012-06-04 14:43:32 -05002509 *skip = 0;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002510 switch (type) {
2511 case CEPH_MSG_OSD_MAP:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002512 case CEPH_MSG_WATCH_NOTIFY:
Sage Weilb61c2762011-08-09 15:03:46 -07002513 return ceph_msg_new(type, front, GFP_NOFS, false);
Sage Weil5b3a4db2010-02-19 21:43:23 -08002514 case CEPH_MSG_OSD_OPREPLY:
2515 return get_reply(con, hdr, skip);
2516 default:
2517 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2518 osd->o_osd);
2519 *skip = 1;
2520 return NULL;
2521 }
Sage Weilf24e9982009-10-06 11:31:10 -07002522}
2523
2524/*
2525 * Wrappers to refcount containing ceph_osd struct
2526 */
2527static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2528{
2529 struct ceph_osd *osd = con->private;
2530 if (get_osd(osd))
2531 return con;
2532 return NULL;
2533}
2534
2535static void put_osd_con(struct ceph_connection *con)
2536{
2537 struct ceph_osd *osd = con->private;
2538 put_osd(osd);
2539}
2540
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002541/*
2542 * authentication
2543 */
Alex Eldera3530df2012-05-16 15:16:39 -05002544/*
2545 * Note: returned pointer is the address of a structure that's
2546 * managed separately. Caller must *not* attempt to free it.
2547 */
2548static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
Alex Elder8f43fb52012-05-16 15:16:39 -05002549 int *proto, int force_new)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002550{
2551 struct ceph_osd *o = con->private;
2552 struct ceph_osd_client *osdc = o->o_osdc;
2553 struct ceph_auth_client *ac = osdc->client->monc.auth;
Alex Elder74f18692012-05-16 15:16:39 -05002554 struct ceph_auth_handshake *auth = &o->o_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002555
Alex Elder74f18692012-05-16 15:16:39 -05002556 if (force_new && auth->authorizer) {
Sage Weil27859f92013-03-25 10:26:14 -07002557 ceph_auth_destroy_authorizer(ac, auth->authorizer);
Alex Elder74f18692012-05-16 15:16:39 -05002558 auth->authorizer = NULL;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002559 }
Sage Weil27859f92013-03-25 10:26:14 -07002560 if (!auth->authorizer) {
2561 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2562 auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002563 if (ret)
Alex Eldera3530df2012-05-16 15:16:39 -05002564 return ERR_PTR(ret);
Sage Weil27859f92013-03-25 10:26:14 -07002565 } else {
2566 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
Sage Weil0bed9b52013-03-25 10:26:01 -07002567 auth);
2568 if (ret)
2569 return ERR_PTR(ret);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002570 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002571 *proto = ac->protocol;
Alex Elder74f18692012-05-16 15:16:39 -05002572
Alex Eldera3530df2012-05-16 15:16:39 -05002573 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002574}
2575
2576
2577static int verify_authorizer_reply(struct ceph_connection *con, int len)
2578{
2579 struct ceph_osd *o = con->private;
2580 struct ceph_osd_client *osdc = o->o_osdc;
2581 struct ceph_auth_client *ac = osdc->client->monc.auth;
2582
Sage Weil27859f92013-03-25 10:26:14 -07002583 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002584}
2585
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002586static int invalidate_authorizer(struct ceph_connection *con)
2587{
2588 struct ceph_osd *o = con->private;
2589 struct ceph_osd_client *osdc = o->o_osdc;
2590 struct ceph_auth_client *ac = osdc->client->monc.auth;
2591
Sage Weil27859f92013-03-25 10:26:14 -07002592 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002593 return ceph_monc_validate_auth(&osdc->client->monc);
2594}
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002595
Tobias Klauser9e327892010-05-20 10:40:19 +02002596static const struct ceph_connection_operations osd_con_ops = {
Sage Weilf24e9982009-10-06 11:31:10 -07002597 .get = get_osd_con,
2598 .put = put_osd_con,
2599 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002600 .get_authorizer = get_authorizer,
2601 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002602 .invalidate_authorizer = invalidate_authorizer,
Sage Weilf24e9982009-10-06 11:31:10 -07002603 .alloc_msg = alloc_msg,
Sage Weil81b024e2009-10-09 10:29:18 -07002604 .fault = osd_reset,
Sage Weilf24e9982009-10-06 11:31:10 -07002605};