blob: dd47889adc4aec94941d6f17105878ebe235db8f [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;
Yan, Zhengccca4e32013-06-02 18:40:23 +0800736 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
737 if (truncate_size <= object_base) {
738 truncate_size = 0;
739 } else {
740 truncate_size -= object_base;
741 if (truncate_size > object_size)
742 truncate_size = object_size;
743 }
Alex Eldera19dadf2013-03-13 20:50:01 -0500744 }
Alex Elderd18d1e22013-03-13 20:50:01 -0500745
Alex Elderc99d2d42013-04-05 01:27:11 -0500746 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
Alex Elderb0270322013-03-13 20:50:01 -0500747 truncate_size, truncate_seq);
Alex Elder8c042b02013-04-03 01:28:58 -0500748
Alex Elderacead002013-03-14 14:09:05 -0500749 /*
750 * A second op in the ops array means the caller wants to
751 * also issue a include a 'startsync' command so that the
752 * osd will flush data quickly.
753 */
754 if (num_ops > 1)
Alex Elderc99d2d42013-04-05 01:27:11 -0500755 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
Alex Elderd18d1e22013-03-13 20:50:01 -0500756
Sage Weilf24e9982009-10-06 11:31:10 -0700757 req->r_file_layout = *layout; /* keep a copy */
758
Alex Elder75d1c942013-03-13 20:50:00 -0500759 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
760 vino.ino, objnum);
Alex Elderdbe0fc42013-02-15 22:10:17 -0600761 req->r_oid_len = strlen(req->r_oid);
762
Sage Weilf24e9982009-10-06 11:31:10 -0700763 return req;
764}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700765EXPORT_SYMBOL(ceph_osdc_new_request);
Sage Weilf24e9982009-10-06 11:31:10 -0700766
767/*
768 * We keep osd requests in an rbtree, sorted by ->r_tid.
769 */
770static void __insert_request(struct ceph_osd_client *osdc,
771 struct ceph_osd_request *new)
772{
773 struct rb_node **p = &osdc->requests.rb_node;
774 struct rb_node *parent = NULL;
775 struct ceph_osd_request *req = NULL;
776
777 while (*p) {
778 parent = *p;
779 req = rb_entry(parent, struct ceph_osd_request, r_node);
780 if (new->r_tid < req->r_tid)
781 p = &(*p)->rb_left;
782 else if (new->r_tid > req->r_tid)
783 p = &(*p)->rb_right;
784 else
785 BUG();
786 }
787
788 rb_link_node(&new->r_node, parent, p);
789 rb_insert_color(&new->r_node, &osdc->requests);
790}
791
792static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
793 u64 tid)
794{
795 struct ceph_osd_request *req;
796 struct rb_node *n = osdc->requests.rb_node;
797
798 while (n) {
799 req = rb_entry(n, struct ceph_osd_request, r_node);
800 if (tid < req->r_tid)
801 n = n->rb_left;
802 else if (tid > req->r_tid)
803 n = n->rb_right;
804 else
805 return req;
806 }
807 return NULL;
808}
809
810static struct ceph_osd_request *
811__lookup_request_ge(struct ceph_osd_client *osdc,
812 u64 tid)
813{
814 struct ceph_osd_request *req;
815 struct rb_node *n = osdc->requests.rb_node;
816
817 while (n) {
818 req = rb_entry(n, struct ceph_osd_request, r_node);
819 if (tid < req->r_tid) {
820 if (!n->rb_left)
821 return req;
822 n = n->rb_left;
823 } else if (tid > req->r_tid) {
824 n = n->rb_right;
825 } else {
826 return req;
827 }
828 }
829 return NULL;
830}
831
Sage Weil6f6c7002011-01-17 20:34:08 -0800832/*
833 * Resubmit requests pending on the given osd.
834 */
835static void __kick_osd_requests(struct ceph_osd_client *osdc,
836 struct ceph_osd *osd)
837{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700838 struct ceph_osd_request *req, *nreq;
Alex Eldere02493c2013-03-25 18:16:11 -0500839 LIST_HEAD(resend);
Sage Weil6f6c7002011-01-17 20:34:08 -0800840 int err;
841
842 dout("__kick_osd_requests osd%d\n", osd->o_osd);
843 err = __reset_osd(osdc, osd);
Alex Elder685a7552012-12-07 09:57:58 -0600844 if (err)
Sage Weil6f6c7002011-01-17 20:34:08 -0800845 return;
Alex Eldere02493c2013-03-25 18:16:11 -0500846 /*
847 * Build up a list of requests to resend by traversing the
848 * osd's list of requests. Requests for a given object are
849 * sent in tid order, and that is also the order they're
850 * kept on this list. Therefore all requests that are in
851 * flight will be found first, followed by all requests that
852 * have not yet been sent. And to resend requests while
853 * preserving this order we will want to put any sent
854 * requests back on the front of the osd client's unsent
855 * list.
856 *
857 * So we build a separate ordered list of already-sent
858 * requests for the affected osd and splice it onto the
859 * front of the osd client's unsent list. Once we've seen a
860 * request that has not yet been sent we're done. Those
861 * requests are already sitting right where they belong.
862 */
Sage Weil6f6c7002011-01-17 20:34:08 -0800863 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
Alex Eldere02493c2013-03-25 18:16:11 -0500864 if (!req->r_sent)
865 break;
866 list_move_tail(&req->r_req_lru_item, &resend);
867 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
Sage Weil6f6c7002011-01-17 20:34:08 -0800868 osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700869 if (!req->r_linger)
870 req->r_flags |= CEPH_OSD_FLAG_RETRY;
871 }
Alex Eldere02493c2013-03-25 18:16:11 -0500872 list_splice(&resend, &osdc->req_unsent);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700873
Alex Eldere02493c2013-03-25 18:16:11 -0500874 /*
875 * Linger requests are re-registered before sending, which
876 * sets up a new tid for each. We add them to the unsent
877 * list at the end to keep things in tid order.
878 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700879 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
880 r_linger_osd) {
Sage Weil77f38e02011-04-06 09:09:16 -0700881 /*
882 * reregister request prior to unregistering linger so
883 * that r_osd is preserved.
884 */
885 BUG_ON(!list_empty(&req->r_req_lru_item));
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700886 __register_request(osdc, req);
Alex Eldere02493c2013-03-25 18:16:11 -0500887 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
Alex Elderad885922013-03-25 18:16:11 -0500888 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
Sage Weil77f38e02011-04-06 09:09:16 -0700889 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700890 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
891 osd->o_osd);
Sage Weil6f6c7002011-01-17 20:34:08 -0800892 }
893}
894
Sage Weilf24e9982009-10-06 11:31:10 -0700895/*
Sage Weil81b024e2009-10-09 10:29:18 -0700896 * If the osd connection drops, we need to resubmit all requests.
Sage Weilf24e9982009-10-06 11:31:10 -0700897 */
898static void osd_reset(struct ceph_connection *con)
899{
900 struct ceph_osd *osd = con->private;
901 struct ceph_osd_client *osdc;
902
903 if (!osd)
904 return;
905 dout("osd_reset osd%d\n", osd->o_osd);
906 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -0700907 down_read(&osdc->map_sem);
Sage Weil83aff952012-11-28 12:28:24 -0800908 mutex_lock(&osdc->request_mutex);
909 __kick_osd_requests(osdc, osd);
Alex Elderf9d25192013-02-15 11:42:29 -0600910 __send_queued(osdc);
Sage Weil83aff952012-11-28 12:28:24 -0800911 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -0700912 up_read(&osdc->map_sem);
913}
914
915/*
916 * Track open sessions with osds.
917 */
Alex Eldere10006f2012-05-26 23:26:43 -0500918static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
Sage Weilf24e9982009-10-06 11:31:10 -0700919{
920 struct ceph_osd *osd;
921
922 osd = kzalloc(sizeof(*osd), GFP_NOFS);
923 if (!osd)
924 return NULL;
925
926 atomic_set(&osd->o_ref, 1);
927 osd->o_osdc = osdc;
Alex Eldere10006f2012-05-26 23:26:43 -0500928 osd->o_osd = onum;
Alex Elderf4077312012-12-06 07:22:04 -0600929 RB_CLEAR_NODE(&osd->o_node);
Sage Weilf24e9982009-10-06 11:31:10 -0700930 INIT_LIST_HEAD(&osd->o_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700931 INIT_LIST_HEAD(&osd->o_linger_requests);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800932 INIT_LIST_HEAD(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700933 osd->o_incarnation = 1;
934
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700935 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800936
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800937 INIT_LIST_HEAD(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -0700938 return osd;
939}
940
941static struct ceph_osd *get_osd(struct ceph_osd *osd)
942{
943 if (atomic_inc_not_zero(&osd->o_ref)) {
944 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
945 atomic_read(&osd->o_ref));
946 return osd;
947 } else {
948 dout("get_osd %p FAIL\n", osd);
949 return NULL;
950 }
951}
952
953static void put_osd(struct ceph_osd *osd)
954{
955 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
956 atomic_read(&osd->o_ref) - 1);
Alex Eldera2556512012-05-16 15:16:39 -0500957 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
Sage Weil79494d12010-05-27 14:15:49 -0700958 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
959
Sage Weil27859f92013-03-25 10:26:14 -0700960 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
Sage Weilf24e9982009-10-06 11:31:10 -0700961 kfree(osd);
Sage Weil79494d12010-05-27 14:15:49 -0700962 }
Sage Weilf24e9982009-10-06 11:31:10 -0700963}
964
965/*
966 * remove an osd from our map
967 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800968static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -0700969{
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800970 dout("__remove_osd %p\n", osd);
Sage Weilf24e9982009-10-06 11:31:10 -0700971 BUG_ON(!list_empty(&osd->o_requests));
972 rb_erase(&osd->o_node, &osdc->osds);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800973 list_del_init(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700974 ceph_con_close(&osd->o_con);
975 put_osd(osd);
976}
977
Sage Weilaca420b2011-08-31 14:45:53 -0700978static void remove_all_osds(struct ceph_osd_client *osdc)
979{
Jiaju Zhang048a9d22012-07-20 08:18:36 -0500980 dout("%s %p\n", __func__, osdc);
Sage Weilaca420b2011-08-31 14:45:53 -0700981 mutex_lock(&osdc->request_mutex);
982 while (!RB_EMPTY_ROOT(&osdc->osds)) {
983 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
984 struct ceph_osd, o_node);
985 __remove_osd(osdc, osd);
986 }
987 mutex_unlock(&osdc->request_mutex);
988}
989
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800990static void __move_osd_to_lru(struct ceph_osd_client *osdc,
991 struct ceph_osd *osd)
992{
993 dout("__move_osd_to_lru %p\n", osd);
994 BUG_ON(!list_empty(&osd->o_osd_lru));
995 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700996 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800997}
998
999static void __remove_osd_from_lru(struct ceph_osd *osd)
1000{
1001 dout("__remove_osd_from_lru %p\n", osd);
1002 if (!list_empty(&osd->o_osd_lru))
1003 list_del_init(&osd->o_osd_lru);
1004}
1005
Sage Weilaca420b2011-08-31 14:45:53 -07001006static void remove_old_osds(struct ceph_osd_client *osdc)
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001007{
1008 struct ceph_osd *osd, *nosd;
1009
1010 dout("__remove_old_osds %p\n", osdc);
1011 mutex_lock(&osdc->request_mutex);
1012 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
Sage Weilaca420b2011-08-31 14:45:53 -07001013 if (time_before(jiffies, osd->lru_ttl))
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001014 break;
1015 __remove_osd(osdc, osd);
1016 }
1017 mutex_unlock(&osdc->request_mutex);
1018}
1019
Sage Weilf24e9982009-10-06 11:31:10 -07001020/*
1021 * reset osd connect
1022 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001023static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -07001024{
Alex Elderc3acb182012-12-07 09:57:58 -06001025 struct ceph_entity_addr *peer_addr;
Sage Weilf24e9982009-10-06 11:31:10 -07001026
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001027 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001028 if (list_empty(&osd->o_requests) &&
1029 list_empty(&osd->o_linger_requests)) {
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001030 __remove_osd(osdc, osd);
Alex Elderc3acb182012-12-07 09:57:58 -06001031
1032 return -ENODEV;
1033 }
1034
1035 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1036 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1037 !ceph_con_opened(&osd->o_con)) {
1038 struct ceph_osd_request *req;
1039
Sage Weil87b315a2010-03-22 14:51:18 -07001040 dout(" osd addr hasn't changed and connection never opened,"
1041 " letting msgr retry");
1042 /* touch each r_stamp for handle_timeout()'s benfit */
1043 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1044 req->r_stamp = jiffies;
Alex Elderc3acb182012-12-07 09:57:58 -06001045
1046 return -EAGAIN;
Sage Weilf24e9982009-10-06 11:31:10 -07001047 }
Alex Elderc3acb182012-12-07 09:57:58 -06001048
1049 ceph_con_close(&osd->o_con);
1050 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1051 osd->o_incarnation++;
1052
1053 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001054}
1055
1056static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1057{
1058 struct rb_node **p = &osdc->osds.rb_node;
1059 struct rb_node *parent = NULL;
1060 struct ceph_osd *osd = NULL;
1061
Sage Weilaca420b2011-08-31 14:45:53 -07001062 dout("__insert_osd %p osd%d\n", new, new->o_osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001063 while (*p) {
1064 parent = *p;
1065 osd = rb_entry(parent, struct ceph_osd, o_node);
1066 if (new->o_osd < osd->o_osd)
1067 p = &(*p)->rb_left;
1068 else if (new->o_osd > osd->o_osd)
1069 p = &(*p)->rb_right;
1070 else
1071 BUG();
1072 }
1073
1074 rb_link_node(&new->o_node, parent, p);
1075 rb_insert_color(&new->o_node, &osdc->osds);
1076}
1077
1078static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1079{
1080 struct ceph_osd *osd;
1081 struct rb_node *n = osdc->osds.rb_node;
1082
1083 while (n) {
1084 osd = rb_entry(n, struct ceph_osd, o_node);
1085 if (o < osd->o_osd)
1086 n = n->rb_left;
1087 else if (o > osd->o_osd)
1088 n = n->rb_right;
1089 else
1090 return osd;
1091 }
1092 return NULL;
1093}
1094
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001095static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1096{
1097 schedule_delayed_work(&osdc->timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001098 osdc->client->options->osd_keepalive_timeout * HZ);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001099}
1100
1101static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1102{
1103 cancel_delayed_work(&osdc->timeout_work);
1104}
Sage Weilf24e9982009-10-06 11:31:10 -07001105
1106/*
1107 * Register request, assign tid. If this is the first request, set up
1108 * the timeout event.
1109 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001110static void __register_request(struct ceph_osd_client *osdc,
1111 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001112{
Sage Weilf24e9982009-10-06 11:31:10 -07001113 req->r_tid = ++osdc->last_tid;
Sage Weil6df058c2009-12-22 11:24:33 -08001114 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
Sage Weil77f38e02011-04-06 09:09:16 -07001115 dout("__register_request %p tid %lld\n", req, req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07001116 __insert_request(osdc, req);
1117 ceph_osdc_get_request(req);
1118 osdc->num_requests++;
Sage Weilf24e9982009-10-06 11:31:10 -07001119 if (osdc->num_requests == 1) {
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001120 dout(" first request, scheduling timeout\n");
1121 __schedule_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001122 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001123}
1124
Sage Weilf24e9982009-10-06 11:31:10 -07001125/*
1126 * called under osdc->request_mutex
1127 */
1128static void __unregister_request(struct ceph_osd_client *osdc,
1129 struct ceph_osd_request *req)
1130{
Sage Weil35f9f8a2012-05-16 15:16:38 -05001131 if (RB_EMPTY_NODE(&req->r_node)) {
1132 dout("__unregister_request %p tid %lld not registered\n",
1133 req, req->r_tid);
1134 return;
1135 }
1136
Sage Weilf24e9982009-10-06 11:31:10 -07001137 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1138 rb_erase(&req->r_node, &osdc->requests);
1139 osdc->num_requests--;
1140
Sage Weil0ba64782009-10-08 16:57:16 -07001141 if (req->r_osd) {
1142 /* make sure the original request isn't in flight. */
Alex Elder6740a842012-06-01 14:56:43 -05001143 ceph_msg_revoke(req->r_request);
Sage Weil0ba64782009-10-08 16:57:16 -07001144
1145 list_del_init(&req->r_osd_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001146 if (list_empty(&req->r_osd->o_requests) &&
1147 list_empty(&req->r_osd->o_linger_requests)) {
1148 dout("moving osd to %p lru\n", req->r_osd);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001149 __move_osd_to_lru(osdc, req->r_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001150 }
Sage Weilfbdb9192011-03-29 12:11:06 -07001151 if (list_empty(&req->r_linger_item))
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001152 req->r_osd = NULL;
Sage Weil0ba64782009-10-08 16:57:16 -07001153 }
Sage Weilf24e9982009-10-06 11:31:10 -07001154
Alex Elder7d5f2482012-11-29 08:37:03 -06001155 list_del_init(&req->r_req_lru_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001156 ceph_osdc_put_request(req);
1157
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001158 if (osdc->num_requests == 0) {
1159 dout(" no requests, canceling timeout\n");
1160 __cancel_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001161 }
1162}
1163
1164/*
1165 * Cancel a previously queued request message
1166 */
1167static void __cancel_request(struct ceph_osd_request *req)
1168{
Sage Weil6bc18872010-09-27 10:18:52 -07001169 if (req->r_sent && req->r_osd) {
Alex Elder6740a842012-06-01 14:56:43 -05001170 ceph_msg_revoke(req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001171 req->r_sent = 0;
1172 }
1173}
1174
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001175static void __register_linger_request(struct ceph_osd_client *osdc,
1176 struct ceph_osd_request *req)
1177{
1178 dout("__register_linger_request %p\n", req);
Alex Elder96e4dac2013-05-22 20:54:25 -05001179 ceph_osdc_get_request(req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001180 list_add_tail(&req->r_linger_item, &osdc->req_linger);
Sage Weil6194ea82012-07-30 16:19:28 -07001181 if (req->r_osd)
1182 list_add_tail(&req->r_linger_osd,
1183 &req->r_osd->o_linger_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001184}
1185
1186static void __unregister_linger_request(struct ceph_osd_client *osdc,
1187 struct ceph_osd_request *req)
1188{
1189 dout("__unregister_linger_request %p\n", req);
Alex Elder61c74032012-12-06 09:37:23 -06001190 list_del_init(&req->r_linger_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001191 if (req->r_osd) {
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001192 list_del_init(&req->r_linger_osd);
1193
1194 if (list_empty(&req->r_osd->o_requests) &&
1195 list_empty(&req->r_osd->o_linger_requests)) {
1196 dout("moving osd to %p lru\n", req->r_osd);
1197 __move_osd_to_lru(osdc, req->r_osd);
1198 }
Sage Weilfbdb9192011-03-29 12:11:06 -07001199 if (list_empty(&req->r_osd_item))
1200 req->r_osd = NULL;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001201 }
Alex Elder96e4dac2013-05-22 20:54:25 -05001202 ceph_osdc_put_request(req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001203}
1204
1205void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1206 struct ceph_osd_request *req)
1207{
1208 mutex_lock(&osdc->request_mutex);
1209 if (req->r_linger) {
Alex Elderc10ebbf2013-05-09 14:56:32 -05001210 req->r_linger = 0;
Alex Elder96e4dac2013-05-22 20:54:25 -05001211 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001212 }
1213 mutex_unlock(&osdc->request_mutex);
1214}
1215EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1216
1217void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1218 struct ceph_osd_request *req)
1219{
1220 if (!req->r_linger) {
1221 dout("set_request_linger %p\n", req);
1222 req->r_linger = 1;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001223 }
1224}
1225EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1226
Sage Weilf24e9982009-10-06 11:31:10 -07001227/*
1228 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1229 * (as needed), and set the request r_osd appropriately. If there is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001230 * no up osd, set r_osd to NULL. Move the request to the appropriate list
Sage Weil6f6c7002011-01-17 20:34:08 -08001231 * (unsent, homeless) or leave on in-flight lru.
Sage Weilf24e9982009-10-06 11:31:10 -07001232 *
1233 * Return 0 if unchanged, 1 if changed, or negative on error.
1234 *
1235 * Caller should hold map_sem for read and request_mutex.
1236 */
Sage Weil6f6c7002011-01-17 20:34:08 -08001237static int __map_request(struct ceph_osd_client *osdc,
Sage Weil38d64532011-10-14 13:33:55 -07001238 struct ceph_osd_request *req, int force_resend)
Sage Weilf24e9982009-10-06 11:31:10 -07001239{
Sage Weil5b191d92013-02-23 10:38:16 -08001240 struct ceph_pg pgid;
Sage Weild85b7052010-05-10 10:24:48 -07001241 int acting[CEPH_PG_MAX_SIZE];
1242 int o = -1, num = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001243 int err;
Sage Weilf24e9982009-10-06 11:31:10 -07001244
Sage Weil6f6c7002011-01-17 20:34:08 -08001245 dout("map_request %p tid %lld\n", req, req->r_tid);
Alex Elder41766f82013-03-01 18:00:15 -06001246 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1247 ceph_file_layout_pg_pool(req->r_file_layout));
Sage Weil6f6c7002011-01-17 20:34:08 -08001248 if (err) {
1249 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilf24e9982009-10-06 11:31:10 -07001250 return err;
Sage Weil6f6c7002011-01-17 20:34:08 -08001251 }
Sage Weil7740a422010-01-08 15:58:25 -08001252 req->r_pgid = pgid;
1253
Sage Weild85b7052010-05-10 10:24:48 -07001254 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1255 if (err > 0) {
1256 o = acting[0];
1257 num = err;
1258 }
Sage Weilf24e9982009-10-06 11:31:10 -07001259
Sage Weil38d64532011-10-14 13:33:55 -07001260 if ((!force_resend &&
1261 req->r_osd && req->r_osd->o_osd == o &&
Sage Weild85b7052010-05-10 10:24:48 -07001262 req->r_sent >= req->r_osd->o_incarnation &&
1263 req->r_num_pg_osds == num &&
1264 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001265 (req->r_osd == NULL && o == -1))
1266 return 0; /* no change */
1267
Sage Weil5b191d92013-02-23 10:38:16 -08001268 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1269 req->r_tid, pgid.pool, pgid.seed, o,
Sage Weilf24e9982009-10-06 11:31:10 -07001270 req->r_osd ? req->r_osd->o_osd : -1);
1271
Sage Weild85b7052010-05-10 10:24:48 -07001272 /* record full pg acting set */
1273 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1274 req->r_num_pg_osds = num;
1275
Sage Weilf24e9982009-10-06 11:31:10 -07001276 if (req->r_osd) {
1277 __cancel_request(req);
1278 list_del_init(&req->r_osd_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001279 req->r_osd = NULL;
1280 }
1281
1282 req->r_osd = __lookup_osd(osdc, o);
1283 if (!req->r_osd && o >= 0) {
Sage Weilc99eb1c2010-02-26 09:37:33 -08001284 err = -ENOMEM;
Alex Eldere10006f2012-05-26 23:26:43 -05001285 req->r_osd = create_osd(osdc, o);
Sage Weil6f6c7002011-01-17 20:34:08 -08001286 if (!req->r_osd) {
1287 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilc99eb1c2010-02-26 09:37:33 -08001288 goto out;
Sage Weil6f6c7002011-01-17 20:34:08 -08001289 }
Sage Weilf24e9982009-10-06 11:31:10 -07001290
Sage Weil6f6c7002011-01-17 20:34:08 -08001291 dout("map_request osd %p is osd%d\n", req->r_osd, o);
Sage Weilf24e9982009-10-06 11:31:10 -07001292 __insert_osd(osdc, req->r_osd);
1293
Sage Weilb7a9e5d2012-06-27 12:24:08 -07001294 ceph_con_open(&req->r_osd->o_con,
1295 CEPH_ENTITY_TYPE_OSD, o,
1296 &osdc->osdmap->osd_addr[o]);
Sage Weilf24e9982009-10-06 11:31:10 -07001297 }
1298
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001299 if (req->r_osd) {
1300 __remove_osd_from_lru(req->r_osd);
Alex Elderad885922013-03-25 18:16:11 -05001301 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1302 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
Sage Weil6f6c7002011-01-17 20:34:08 -08001303 } else {
Alex Elderad885922013-03-25 18:16:11 -05001304 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001305 }
Sage Weild85b7052010-05-10 10:24:48 -07001306 err = 1; /* osd or pg changed */
Sage Weilf24e9982009-10-06 11:31:10 -07001307
1308out:
Sage Weilf24e9982009-10-06 11:31:10 -07001309 return err;
1310}
1311
1312/*
1313 * caller should hold map_sem (for read) and request_mutex
1314 */
Sage Weil56e925b2012-01-03 12:34:34 -08001315static void __send_request(struct ceph_osd_client *osdc,
1316 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001317{
Sage Weil1b83bef2013-02-25 16:11:12 -08001318 void *p;
Sage Weilf24e9982009-10-06 11:31:10 -07001319
Sage Weil1b83bef2013-02-25 16:11:12 -08001320 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1321 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1322 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
Sage Weilf24e9982009-10-06 11:31:10 -07001323
Sage Weil1b83bef2013-02-25 16:11:12 -08001324 /* fill in message content that changes each time we send it */
1325 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1326 put_unaligned_le32(req->r_flags, req->r_request_flags);
1327 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1328 p = req->r_request_pgid;
1329 ceph_encode_64(&p, req->r_pgid.pool);
1330 ceph_encode_32(&p, req->r_pgid.seed);
1331 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1332 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1333 sizeof(req->r_reassert_version));
Sage Weil2169aea2013-02-25 16:13:08 -08001334
Sage Weil3dd72fc2010-03-22 14:42:30 -07001335 req->r_stamp = jiffies;
Henry C Chang07a27e22010-08-22 21:34:27 -07001336 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001337
1338 ceph_msg_get(req->r_request); /* send consumes a ref */
Alex Elder26be8802013-04-15 11:20:42 -05001339
Sage Weilf24e9982009-10-06 11:31:10 -07001340 req->r_sent = req->r_osd->o_incarnation;
Alex Elder26be8802013-04-15 11:20:42 -05001341
1342 ceph_con_send(&req->r_osd->o_con, req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001343}
1344
1345/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001346 * Send any requests in the queue (req_unsent).
1347 */
Alex Elderf9d25192013-02-15 11:42:29 -06001348static void __send_queued(struct ceph_osd_client *osdc)
Sage Weil6f6c7002011-01-17 20:34:08 -08001349{
1350 struct ceph_osd_request *req, *tmp;
1351
Alex Elderf9d25192013-02-15 11:42:29 -06001352 dout("__send_queued\n");
1353 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
Sage Weil6f6c7002011-01-17 20:34:08 -08001354 __send_request(osdc, req);
Sage Weil6f6c7002011-01-17 20:34:08 -08001355}
1356
1357/*
Sage Weilf24e9982009-10-06 11:31:10 -07001358 * Timeout callback, called every N seconds when 1 or more osd
1359 * requests has been active for more than N seconds. When this
1360 * happens, we ping all OSDs with requests who have timed out to
1361 * ensure any communications channel reset is detected. Reset the
1362 * request timeouts another N seconds in the future as we go.
1363 * Reschedule the timeout event another N seconds in future (unless
1364 * there are no open requests).
1365 */
1366static void handle_timeout(struct work_struct *work)
1367{
1368 struct ceph_osd_client *osdc =
1369 container_of(work, struct ceph_osd_client, timeout_work.work);
Sage Weil83aff952012-11-28 12:28:24 -08001370 struct ceph_osd_request *req;
Sage Weilf24e9982009-10-06 11:31:10 -07001371 struct ceph_osd *osd;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001372 unsigned long keepalive =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001373 osdc->client->options->osd_keepalive_timeout * HZ;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001374 struct list_head slow_osds;
Sage Weilf24e9982009-10-06 11:31:10 -07001375 dout("timeout\n");
1376 down_read(&osdc->map_sem);
1377
1378 ceph_monc_request_next_osdmap(&osdc->client->monc);
1379
1380 mutex_lock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001381
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001382 /*
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001383 * ping osds that are a bit slow. this ensures that if there
1384 * is a break in the TCP connection we will notice, and reopen
1385 * a connection with that osd (from the fault callback).
1386 */
1387 INIT_LIST_HEAD(&slow_osds);
1388 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
Sage Weil3dd72fc2010-03-22 14:42:30 -07001389 if (time_before(jiffies, req->r_stamp + keepalive))
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001390 break;
1391
1392 osd = req->r_osd;
1393 BUG_ON(!osd);
1394 dout(" tid %llu is slow, will send keepalive on osd%d\n",
Sage Weilf24e9982009-10-06 11:31:10 -07001395 req->r_tid, osd->o_osd);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001396 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1397 }
1398 while (!list_empty(&slow_osds)) {
1399 osd = list_entry(slow_osds.next, struct ceph_osd,
1400 o_keepalive_item);
1401 list_del_init(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001402 ceph_con_keepalive(&osd->o_con);
1403 }
1404
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001405 __schedule_osd_timeout(osdc);
Alex Elderf9d25192013-02-15 11:42:29 -06001406 __send_queued(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001407 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001408 up_read(&osdc->map_sem);
1409}
1410
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001411static void handle_osds_timeout(struct work_struct *work)
1412{
1413 struct ceph_osd_client *osdc =
1414 container_of(work, struct ceph_osd_client,
1415 osds_timeout_work.work);
1416 unsigned long delay =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001417 osdc->client->options->osd_idle_ttl * HZ >> 2;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001418
1419 dout("osds timeout\n");
1420 down_read(&osdc->map_sem);
Sage Weilaca420b2011-08-31 14:45:53 -07001421 remove_old_osds(osdc);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001422 up_read(&osdc->map_sem);
1423
1424 schedule_delayed_work(&osdc->osds_timeout_work,
1425 round_jiffies_relative(delay));
1426}
1427
Sage Weil25845472011-06-03 09:37:09 -07001428static void complete_request(struct ceph_osd_request *req)
1429{
Sage Weil25845472011-06-03 09:37:09 -07001430 complete_all(&req->r_safe_completion); /* fsync waiter */
1431}
1432
Sage Weilf24e9982009-10-06 11:31:10 -07001433/*
1434 * handle osd op reply. either call the callback if it is specified,
1435 * or do the completion to wake up the waiting thread.
1436 */
Sage Weil350b1c32009-12-22 10:45:45 -08001437static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1438 struct ceph_connection *con)
Sage Weilf24e9982009-10-06 11:31:10 -07001439{
Sage Weil1b83bef2013-02-25 16:11:12 -08001440 void *p, *end;
Sage Weilf24e9982009-10-06 11:31:10 -07001441 struct ceph_osd_request *req;
1442 u64 tid;
Sage Weil1b83bef2013-02-25 16:11:12 -08001443 int object_len;
Alex Elder79528732013-04-03 21:32:51 -05001444 unsigned int numops;
1445 int payload_len, flags;
Sage Weil0ceed5d2010-05-11 09:53:18 -07001446 s32 result;
Sage Weil1b83bef2013-02-25 16:11:12 -08001447 s32 retry_attempt;
1448 struct ceph_pg pg;
1449 int err;
1450 u32 reassert_epoch;
1451 u64 reassert_version;
1452 u32 osdmap_epoch;
Alex Elder0d5af162013-02-27 10:26:25 -06001453 int already_completed;
Alex Elder9fc6e062013-04-03 01:28:57 -05001454 u32 bytes;
Alex Elder79528732013-04-03 21:32:51 -05001455 unsigned int i;
Sage Weilf24e9982009-10-06 11:31:10 -07001456
Sage Weil6df058c2009-12-22 11:24:33 -08001457 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil1b83bef2013-02-25 16:11:12 -08001458 dout("handle_reply %p tid %llu\n", msg, tid);
1459
1460 p = msg->front.iov_base;
1461 end = p + msg->front.iov_len;
1462
1463 ceph_decode_need(&p, end, 4, bad);
1464 object_len = ceph_decode_32(&p);
1465 ceph_decode_need(&p, end, object_len, bad);
1466 p += object_len;
1467
Alex Elderef4859d2013-04-01 18:58:26 -05001468 err = ceph_decode_pgid(&p, end, &pg);
Sage Weil1b83bef2013-02-25 16:11:12 -08001469 if (err)
Sage Weilf24e9982009-10-06 11:31:10 -07001470 goto bad;
Sage Weil1b83bef2013-02-25 16:11:12 -08001471
1472 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1473 flags = ceph_decode_64(&p);
1474 result = ceph_decode_32(&p);
1475 reassert_epoch = ceph_decode_32(&p);
1476 reassert_version = ceph_decode_64(&p);
1477 osdmap_epoch = ceph_decode_32(&p);
1478
Sage Weilf24e9982009-10-06 11:31:10 -07001479 /* lookup */
1480 mutex_lock(&osdc->request_mutex);
1481 req = __lookup_request(osdc, tid);
1482 if (req == NULL) {
1483 dout("handle_reply tid %llu dne\n", tid);
Alex Elder8058fd42013-04-01 18:58:26 -05001484 goto bad_mutex;
Sage Weilf24e9982009-10-06 11:31:10 -07001485 }
1486 ceph_osdc_get_request(req);
Sage Weil1b83bef2013-02-25 16:11:12 -08001487
1488 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1489 req, result);
1490
1491 ceph_decode_need(&p, end, 4, bad);
1492 numops = ceph_decode_32(&p);
1493 if (numops > CEPH_OSD_MAX_OP)
1494 goto bad_put;
1495 if (numops != req->r_num_ops)
1496 goto bad_put;
1497 payload_len = 0;
1498 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1499 for (i = 0; i < numops; i++) {
1500 struct ceph_osd_op *op = p;
1501 int len;
1502
1503 len = le32_to_cpu(op->payload_len);
1504 req->r_reply_op_len[i] = len;
1505 dout(" op %d has %d bytes\n", i, len);
1506 payload_len += len;
1507 p += sizeof(*op);
1508 }
Alex Elder9fc6e062013-04-03 01:28:57 -05001509 bytes = le32_to_cpu(msg->hdr.data_len);
1510 if (payload_len != bytes) {
Sage Weil1b83bef2013-02-25 16:11:12 -08001511 pr_warning("sum of op payload lens %d != data_len %d",
Alex Elder9fc6e062013-04-03 01:28:57 -05001512 payload_len, bytes);
Sage Weil1b83bef2013-02-25 16:11:12 -08001513 goto bad_put;
1514 }
1515
1516 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1517 retry_attempt = ceph_decode_32(&p);
1518 for (i = 0; i < numops; i++)
1519 req->r_reply_op_result[i] = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001520
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001521 already_completed = req->r_got_reply;
1522
Sage Weilf24e9982009-10-06 11:31:10 -07001523 if (!req->r_got_reply) {
Sage Weilf24e9982009-10-06 11:31:10 -07001524
Sage Weil1b83bef2013-02-25 16:11:12 -08001525 req->r_result = result;
Sage Weilf24e9982009-10-06 11:31:10 -07001526 dout("handle_reply result %d bytes %d\n", req->r_result,
1527 bytes);
1528 if (req->r_result == 0)
1529 req->r_result = bytes;
1530
1531 /* in case this is a write and we need to replay, */
Sage Weil1b83bef2013-02-25 16:11:12 -08001532 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1533 req->r_reassert_version.version = cpu_to_le64(reassert_version);
Sage Weilf24e9982009-10-06 11:31:10 -07001534
1535 req->r_got_reply = 1;
1536 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1537 dout("handle_reply tid %llu dup ack\n", tid);
Sage Weil34b43a52009-12-01 12:23:54 -08001538 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001539 goto done;
1540 }
1541
1542 dout("handle_reply tid %llu flags %d\n", tid, flags);
1543
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001544 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1545 __register_linger_request(osdc, req);
1546
Sage Weilf24e9982009-10-06 11:31:10 -07001547 /* either this is a read, or we got the safe response */
Sage Weil0ceed5d2010-05-11 09:53:18 -07001548 if (result < 0 ||
1549 (flags & CEPH_OSD_FLAG_ONDISK) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001550 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1551 __unregister_request(osdc, req);
1552
1553 mutex_unlock(&osdc->request_mutex);
1554
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001555 if (!already_completed) {
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001556 if (req->r_unsafe_callback &&
1557 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1558 req->r_unsafe_callback(req, true);
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001559 if (req->r_callback)
1560 req->r_callback(req, msg);
1561 else
1562 complete_all(&req->r_completion);
1563 }
Sage Weilf24e9982009-10-06 11:31:10 -07001564
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001565 if (flags & CEPH_OSD_FLAG_ONDISK) {
1566 if (req->r_unsafe_callback && already_completed)
1567 req->r_unsafe_callback(req, false);
Sage Weil25845472011-06-03 09:37:09 -07001568 complete_request(req);
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001569 }
Sage Weilf24e9982009-10-06 11:31:10 -07001570
1571done:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001572 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07001573 ceph_osdc_put_request(req);
1574 return;
1575
Sage Weil1b83bef2013-02-25 16:11:12 -08001576bad_put:
1577 ceph_osdc_put_request(req);
Alex Elder8058fd42013-04-01 18:58:26 -05001578bad_mutex:
1579 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001580bad:
Sage Weil1b83bef2013-02-25 16:11:12 -08001581 pr_err("corrupt osd_op_reply got %d %d\n",
1582 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
Sage Weil9ec7cab2009-12-14 15:13:47 -08001583 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001584}
1585
Sage Weil6f6c7002011-01-17 20:34:08 -08001586static void reset_changed_osds(struct ceph_osd_client *osdc)
Sage Weilf24e9982009-10-06 11:31:10 -07001587{
Sage Weilf24e9982009-10-06 11:31:10 -07001588 struct rb_node *p, *n;
Sage Weilf24e9982009-10-06 11:31:10 -07001589
Sage Weil6f6c7002011-01-17 20:34:08 -08001590 for (p = rb_first(&osdc->osds); p; p = n) {
1591 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
Sage Weilf24e9982009-10-06 11:31:10 -07001592
Sage Weil6f6c7002011-01-17 20:34:08 -08001593 n = rb_next(p);
1594 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1595 memcmp(&osd->o_con.peer_addr,
1596 ceph_osd_addr(osdc->osdmap,
1597 osd->o_osd),
1598 sizeof(struct ceph_entity_addr)) != 0)
1599 __reset_osd(osdc, osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001600 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001601}
1602
1603/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001604 * Requeue requests whose mapping to an OSD has changed. If requests map to
1605 * no osd, request a new map.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001606 *
Alex Eldere6d50f62012-12-26 14:31:40 -06001607 * Caller should hold map_sem for read.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001608 */
Sage Weil38d64532011-10-14 13:33:55 -07001609static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001610{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001611 struct ceph_osd_request *req, *nreq;
Sage Weil6f6c7002011-01-17 20:34:08 -08001612 struct rb_node *p;
1613 int needmap = 0;
1614 int err;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001615
Sage Weil38d64532011-10-14 13:33:55 -07001616 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001617 mutex_lock(&osdc->request_mutex);
Sage Weil6194ea82012-07-30 16:19:28 -07001618 for (p = rb_first(&osdc->requests); p; ) {
Sage Weil6f6c7002011-01-17 20:34:08 -08001619 req = rb_entry(p, struct ceph_osd_request, r_node);
Sage Weil6194ea82012-07-30 16:19:28 -07001620 p = rb_next(p);
Alex Elderab60b162012-12-19 15:52:36 -06001621
1622 /*
1623 * For linger requests that have not yet been
1624 * registered, move them to the linger list; they'll
1625 * be sent to the osd in the loop below. Unregister
1626 * the request before re-registering it as a linger
1627 * request to ensure the __map_request() below
1628 * will decide it needs to be sent.
1629 */
1630 if (req->r_linger && list_empty(&req->r_linger_item)) {
1631 dout("%p tid %llu restart on osd%d\n",
1632 req, req->r_tid,
1633 req->r_osd ? req->r_osd->o_osd : -1);
Alex Elder96e4dac2013-05-22 20:54:25 -05001634 ceph_osdc_get_request(req);
Alex Elderab60b162012-12-19 15:52:36 -06001635 __unregister_request(osdc, req);
1636 __register_linger_request(osdc, req);
Alex Elder96e4dac2013-05-22 20:54:25 -05001637 ceph_osdc_put_request(req);
Alex Elderab60b162012-12-19 15:52:36 -06001638 continue;
1639 }
1640
Sage Weil38d64532011-10-14 13:33:55 -07001641 err = __map_request(osdc, req, force_resend);
Sage Weil6f6c7002011-01-17 20:34:08 -08001642 if (err < 0)
1643 continue; /* error */
1644 if (req->r_osd == NULL) {
1645 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1646 needmap++; /* request a newer map */
1647 } else if (err > 0) {
Sage Weil6194ea82012-07-30 16:19:28 -07001648 if (!req->r_linger) {
1649 dout("%p tid %llu requeued on osd%d\n", req,
1650 req->r_tid,
1651 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001652 req->r_flags |= CEPH_OSD_FLAG_RETRY;
Sage Weil6194ea82012-07-30 16:19:28 -07001653 }
1654 }
Sage Weil6f6c7002011-01-17 20:34:08 -08001655 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001656
1657 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1658 r_linger_item) {
1659 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1660
Sage Weil38d64532011-10-14 13:33:55 -07001661 err = __map_request(osdc, req, force_resend);
Alex Elderab60b162012-12-19 15:52:36 -06001662 dout("__map_request returned %d\n", err);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001663 if (err == 0)
1664 continue; /* no change and no osd was specified */
1665 if (err < 0)
1666 continue; /* hrm! */
1667 if (req->r_osd == NULL) {
1668 dout("tid %llu maps to no valid osd\n", req->r_tid);
1669 needmap++; /* request a newer map */
1670 continue;
1671 }
1672
1673 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1674 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001675 __register_request(osdc, req);
Alex Elderc89ce052012-12-06 07:22:04 -06001676 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001677 }
Alex Elder14d2f382013-05-15 16:28:33 -05001678 reset_changed_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001679 mutex_unlock(&osdc->request_mutex);
1680
1681 if (needmap) {
1682 dout("%d requests for down osds, need new map\n", needmap);
1683 ceph_monc_request_next_osdmap(&osdc->client->monc);
1684 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001685}
Sage Weil6f6c7002011-01-17 20:34:08 -08001686
1687
Sage Weilf24e9982009-10-06 11:31:10 -07001688/*
1689 * Process updated osd map.
1690 *
1691 * The message contains any number of incremental and full maps, normally
1692 * indicating some sort of topology change in the cluster. Kick requests
1693 * off to different OSDs as needed.
1694 */
1695void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1696{
1697 void *p, *end, *next;
1698 u32 nr_maps, maplen;
1699 u32 epoch;
1700 struct ceph_osdmap *newmap = NULL, *oldmap;
1701 int err;
1702 struct ceph_fsid fsid;
1703
1704 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1705 p = msg->front.iov_base;
1706 end = p + msg->front.iov_len;
1707
1708 /* verify fsid */
1709 ceph_decode_need(&p, end, sizeof(fsid), bad);
1710 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Sage Weil07433042009-11-18 16:50:41 -08001711 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1712 return;
Sage Weilf24e9982009-10-06 11:31:10 -07001713
1714 down_write(&osdc->map_sem);
1715
1716 /* incremental maps */
1717 ceph_decode_32_safe(&p, end, nr_maps, bad);
1718 dout(" %d inc maps\n", nr_maps);
1719 while (nr_maps > 0) {
1720 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07001721 epoch = ceph_decode_32(&p);
1722 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001723 ceph_decode_need(&p, end, maplen, bad);
1724 next = p + maplen;
1725 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1726 dout("applying incremental map %u len %d\n",
1727 epoch, maplen);
1728 newmap = osdmap_apply_incremental(&p, next,
1729 osdc->osdmap,
Alex Elder15d98822012-05-26 23:26:43 -05001730 &osdc->client->msgr);
Sage Weilf24e9982009-10-06 11:31:10 -07001731 if (IS_ERR(newmap)) {
1732 err = PTR_ERR(newmap);
1733 goto bad;
1734 }
Sage Weil30dc6382009-12-21 14:49:37 -08001735 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07001736 if (newmap != osdc->osdmap) {
1737 ceph_osdmap_destroy(osdc->osdmap);
1738 osdc->osdmap = newmap;
1739 }
Sage Weil38d64532011-10-14 13:33:55 -07001740 kick_requests(osdc, 0);
Sage Weilf24e9982009-10-06 11:31:10 -07001741 } else {
1742 dout("ignoring incremental map %u len %d\n",
1743 epoch, maplen);
1744 }
1745 p = next;
1746 nr_maps--;
1747 }
1748 if (newmap)
1749 goto done;
1750
1751 /* full maps */
1752 ceph_decode_32_safe(&p, end, nr_maps, bad);
1753 dout(" %d full maps\n", nr_maps);
1754 while (nr_maps) {
1755 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07001756 epoch = ceph_decode_32(&p);
1757 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001758 ceph_decode_need(&p, end, maplen, bad);
1759 if (nr_maps > 1) {
1760 dout("skipping non-latest full map %u len %d\n",
1761 epoch, maplen);
1762 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1763 dout("skipping full map %u len %d, "
1764 "older than our %u\n", epoch, maplen,
1765 osdc->osdmap->epoch);
1766 } else {
Sage Weil38d64532011-10-14 13:33:55 -07001767 int skipped_map = 0;
1768
Sage Weilf24e9982009-10-06 11:31:10 -07001769 dout("taking full map %u len %d\n", epoch, maplen);
1770 newmap = osdmap_decode(&p, p+maplen);
1771 if (IS_ERR(newmap)) {
1772 err = PTR_ERR(newmap);
1773 goto bad;
1774 }
Sage Weil30dc6382009-12-21 14:49:37 -08001775 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07001776 oldmap = osdc->osdmap;
1777 osdc->osdmap = newmap;
Sage Weil38d64532011-10-14 13:33:55 -07001778 if (oldmap) {
1779 if (oldmap->epoch + 1 < newmap->epoch)
1780 skipped_map = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07001781 ceph_osdmap_destroy(oldmap);
Sage Weil38d64532011-10-14 13:33:55 -07001782 }
1783 kick_requests(osdc, skipped_map);
Sage Weilf24e9982009-10-06 11:31:10 -07001784 }
1785 p += maplen;
1786 nr_maps--;
1787 }
1788
1789done:
1790 downgrade_write(&osdc->map_sem);
1791 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
Sage Weilcd634fb2011-05-12 09:29:18 -07001792
1793 /*
1794 * subscribe to subsequent osdmap updates if full to ensure
1795 * we find out when we are no longer full and stop returning
1796 * ENOSPC.
1797 */
1798 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1799 ceph_monc_request_next_osdmap(&osdc->client->monc);
1800
Alex Elderf9d25192013-02-15 11:42:29 -06001801 mutex_lock(&osdc->request_mutex);
1802 __send_queued(osdc);
1803 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001804 up_read(&osdc->map_sem);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001805 wake_up_all(&osdc->client->auth_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07001806 return;
1807
1808bad:
1809 pr_err("osdc handle_map corrupt msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08001810 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001811 up_write(&osdc->map_sem);
1812 return;
1813}
1814
Sage Weilf24e9982009-10-06 11:31:10 -07001815/*
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001816 * watch/notify callback event infrastructure
1817 *
1818 * These callbacks are used both for watch and notify operations.
1819 */
1820static void __release_event(struct kref *kref)
1821{
1822 struct ceph_osd_event *event =
1823 container_of(kref, struct ceph_osd_event, kref);
1824
1825 dout("__release_event %p\n", event);
1826 kfree(event);
1827}
1828
1829static void get_event(struct ceph_osd_event *event)
1830{
1831 kref_get(&event->kref);
1832}
1833
1834void ceph_osdc_put_event(struct ceph_osd_event *event)
1835{
1836 kref_put(&event->kref, __release_event);
1837}
1838EXPORT_SYMBOL(ceph_osdc_put_event);
1839
1840static void __insert_event(struct ceph_osd_client *osdc,
1841 struct ceph_osd_event *new)
1842{
1843 struct rb_node **p = &osdc->event_tree.rb_node;
1844 struct rb_node *parent = NULL;
1845 struct ceph_osd_event *event = NULL;
1846
1847 while (*p) {
1848 parent = *p;
1849 event = rb_entry(parent, struct ceph_osd_event, node);
1850 if (new->cookie < event->cookie)
1851 p = &(*p)->rb_left;
1852 else if (new->cookie > event->cookie)
1853 p = &(*p)->rb_right;
1854 else
1855 BUG();
1856 }
1857
1858 rb_link_node(&new->node, parent, p);
1859 rb_insert_color(&new->node, &osdc->event_tree);
1860}
1861
1862static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1863 u64 cookie)
1864{
1865 struct rb_node **p = &osdc->event_tree.rb_node;
1866 struct rb_node *parent = NULL;
1867 struct ceph_osd_event *event = NULL;
1868
1869 while (*p) {
1870 parent = *p;
1871 event = rb_entry(parent, struct ceph_osd_event, node);
1872 if (cookie < event->cookie)
1873 p = &(*p)->rb_left;
1874 else if (cookie > event->cookie)
1875 p = &(*p)->rb_right;
1876 else
1877 return event;
1878 }
1879 return NULL;
1880}
1881
1882static void __remove_event(struct ceph_osd_event *event)
1883{
1884 struct ceph_osd_client *osdc = event->osdc;
1885
1886 if (!RB_EMPTY_NODE(&event->node)) {
1887 dout("__remove_event removed %p\n", event);
1888 rb_erase(&event->node, &osdc->event_tree);
1889 ceph_osdc_put_event(event);
1890 } else {
1891 dout("__remove_event didn't remove %p\n", event);
1892 }
1893}
1894
1895int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1896 void (*event_cb)(u64, u64, u8, void *),
Alex Elder3c663bb2013-02-15 11:42:30 -06001897 void *data, struct ceph_osd_event **pevent)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001898{
1899 struct ceph_osd_event *event;
1900
1901 event = kmalloc(sizeof(*event), GFP_NOIO);
1902 if (!event)
1903 return -ENOMEM;
1904
1905 dout("create_event %p\n", event);
1906 event->cb = event_cb;
Alex Elder3c663bb2013-02-15 11:42:30 -06001907 event->one_shot = 0;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001908 event->data = data;
1909 event->osdc = osdc;
1910 INIT_LIST_HEAD(&event->osd_node);
Alex Elder3ee52342012-12-17 12:23:48 -06001911 RB_CLEAR_NODE(&event->node);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001912 kref_init(&event->kref); /* one ref for us */
1913 kref_get(&event->kref); /* one ref for the caller */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001914
1915 spin_lock(&osdc->event_lock);
1916 event->cookie = ++osdc->event_count;
1917 __insert_event(osdc, event);
1918 spin_unlock(&osdc->event_lock);
1919
1920 *pevent = event;
1921 return 0;
1922}
1923EXPORT_SYMBOL(ceph_osdc_create_event);
1924
1925void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1926{
1927 struct ceph_osd_client *osdc = event->osdc;
1928
1929 dout("cancel_event %p\n", event);
1930 spin_lock(&osdc->event_lock);
1931 __remove_event(event);
1932 spin_unlock(&osdc->event_lock);
1933 ceph_osdc_put_event(event); /* caller's */
1934}
1935EXPORT_SYMBOL(ceph_osdc_cancel_event);
1936
1937
1938static void do_event_work(struct work_struct *work)
1939{
1940 struct ceph_osd_event_work *event_work =
1941 container_of(work, struct ceph_osd_event_work, work);
1942 struct ceph_osd_event *event = event_work->event;
1943 u64 ver = event_work->ver;
1944 u64 notify_id = event_work->notify_id;
1945 u8 opcode = event_work->opcode;
1946
1947 dout("do_event_work completing %p\n", event);
1948 event->cb(ver, notify_id, opcode, event->data);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001949 dout("do_event_work completed %p\n", event);
1950 ceph_osdc_put_event(event);
1951 kfree(event_work);
1952}
1953
1954
1955/*
1956 * Process osd watch notifications
1957 */
Alex Elder3c663bb2013-02-15 11:42:30 -06001958static void handle_watch_notify(struct ceph_osd_client *osdc,
1959 struct ceph_msg *msg)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001960{
1961 void *p, *end;
1962 u8 proto_ver;
1963 u64 cookie, ver, notify_id;
1964 u8 opcode;
1965 struct ceph_osd_event *event;
1966 struct ceph_osd_event_work *event_work;
1967
1968 p = msg->front.iov_base;
1969 end = p + msg->front.iov_len;
1970
1971 ceph_decode_8_safe(&p, end, proto_ver, bad);
1972 ceph_decode_8_safe(&p, end, opcode, bad);
1973 ceph_decode_64_safe(&p, end, cookie, bad);
1974 ceph_decode_64_safe(&p, end, ver, bad);
1975 ceph_decode_64_safe(&p, end, notify_id, bad);
1976
1977 spin_lock(&osdc->event_lock);
1978 event = __find_event(osdc, cookie);
1979 if (event) {
Alex Elder3c663bb2013-02-15 11:42:30 -06001980 BUG_ON(event->one_shot);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001981 get_event(event);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001982 }
1983 spin_unlock(&osdc->event_lock);
1984 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1985 cookie, ver, event);
1986 if (event) {
1987 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001988 if (!event_work) {
1989 dout("ERROR: could not allocate event_work\n");
1990 goto done_err;
1991 }
Mariusz Kozlowski6b0ae402011-03-26 19:29:34 +01001992 INIT_WORK(&event_work->work, do_event_work);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001993 event_work->event = event;
1994 event_work->ver = ver;
1995 event_work->notify_id = notify_id;
1996 event_work->opcode = opcode;
1997 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1998 dout("WARNING: failed to queue notify event work\n");
1999 goto done_err;
2000 }
2001 }
2002
2003 return;
2004
2005done_err:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002006 ceph_osdc_put_event(event);
2007 return;
2008
2009bad:
2010 pr_err("osdc handle_watch_notify corrupt msg\n");
2011 return;
2012}
2013
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002014/*
Alex Eldere65550f2013-04-05 01:27:12 -05002015 * build new request AND message
2016 *
2017 */
2018void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2019 struct ceph_snap_context *snapc, u64 snap_id,
2020 struct timespec *mtime)
2021{
2022 struct ceph_msg *msg = req->r_request;
2023 void *p;
2024 size_t msg_size;
2025 int flags = req->r_flags;
2026 u64 data_len;
2027 unsigned int i;
2028
2029 req->r_snapid = snap_id;
2030 req->r_snapc = ceph_get_snap_context(snapc);
2031
2032 /* encode request */
2033 msg->hdr.version = cpu_to_le16(4);
2034
2035 p = msg->front.iov_base;
2036 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2037 req->r_request_osdmap_epoch = p;
2038 p += 4;
2039 req->r_request_flags = p;
2040 p += 4;
2041 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2042 ceph_encode_timespec(p, mtime);
2043 p += sizeof(struct ceph_timespec);
2044 req->r_request_reassert_version = p;
2045 p += sizeof(struct ceph_eversion); /* will get filled in */
2046
2047 /* oloc */
2048 ceph_encode_8(&p, 4);
2049 ceph_encode_8(&p, 4);
2050 ceph_encode_32(&p, 8 + 4 + 4);
2051 req->r_request_pool = p;
2052 p += 8;
2053 ceph_encode_32(&p, -1); /* preferred */
2054 ceph_encode_32(&p, 0); /* key len */
2055
2056 ceph_encode_8(&p, 1);
2057 req->r_request_pgid = p;
2058 p += 8 + 4;
2059 ceph_encode_32(&p, -1); /* preferred */
2060
2061 /* oid */
2062 ceph_encode_32(&p, req->r_oid_len);
2063 memcpy(p, req->r_oid, req->r_oid_len);
2064 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2065 p += req->r_oid_len;
2066
2067 /* ops--can imply data */
2068 ceph_encode_16(&p, (u16)req->r_num_ops);
2069 data_len = 0;
2070 for (i = 0; i < req->r_num_ops; i++) {
2071 data_len += osd_req_encode_op(req, p, i);
2072 p += sizeof(struct ceph_osd_op);
2073 }
2074
2075 /* snaps */
2076 ceph_encode_64(&p, req->r_snapid);
2077 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2078 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2079 if (req->r_snapc) {
2080 for (i = 0; i < snapc->num_snaps; i++) {
2081 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2082 }
2083 }
2084
2085 req->r_request_attempts = p;
2086 p += 4;
2087
2088 /* data */
2089 if (flags & CEPH_OSD_FLAG_WRITE) {
2090 u16 data_off;
2091
2092 /*
2093 * The header "data_off" is a hint to the receiver
2094 * allowing it to align received data into its
2095 * buffers such that there's no need to re-copy
2096 * it before writing it to disk (direct I/O).
2097 */
2098 data_off = (u16) (off & 0xffff);
2099 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2100 }
2101 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2102
2103 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2104 msg_size = p - msg->front.iov_base;
2105 msg->front.iov_len = msg_size;
2106 msg->hdr.front_len = cpu_to_le32(msg_size);
2107
2108 dout("build_request msg_size was %d\n", (int)msg_size);
2109}
2110EXPORT_SYMBOL(ceph_osdc_build_request);
2111
2112/*
Sage Weilf24e9982009-10-06 11:31:10 -07002113 * Register request, send initial attempt.
2114 */
2115int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2116 struct ceph_osd_request *req,
2117 bool nofail)
2118{
Sage Weilc1ea8822009-10-08 16:55:47 -07002119 int rc = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07002120
Sage Weilf24e9982009-10-06 11:31:10 -07002121 down_read(&osdc->map_sem);
2122 mutex_lock(&osdc->request_mutex);
Alex Elderdc4b8702013-03-25 18:16:11 -05002123 __register_request(osdc, req);
Alex Elderc10ebbf2013-05-09 14:56:32 -05002124 req->r_sent = 0;
2125 req->r_got_reply = 0;
Alex Elder92451b42013-03-25 18:16:11 -05002126 rc = __map_request(osdc, req, 0);
2127 if (rc < 0) {
2128 if (nofail) {
2129 dout("osdc_start_request failed map, "
2130 " will retry %lld\n", req->r_tid);
2131 rc = 0;
Sage Weil9d6fcb02011-05-12 15:48:16 -07002132 }
Alex Elder92451b42013-03-25 18:16:11 -05002133 goto out_unlock;
Sage Weilf24e9982009-10-06 11:31:10 -07002134 }
Alex Elder92451b42013-03-25 18:16:11 -05002135 if (req->r_osd == NULL) {
2136 dout("send_request %p no up osds in pg\n", req);
2137 ceph_monc_request_next_osdmap(&osdc->client->monc);
2138 } else {
Alex Elder7e2766a2013-03-25 18:16:11 -05002139 __send_queued(osdc);
Alex Elder92451b42013-03-25 18:16:11 -05002140 }
2141 rc = 0;
Dan Carpenter234af26f2011-03-29 06:25:59 +03002142out_unlock:
Sage Weilf24e9982009-10-06 11:31:10 -07002143 mutex_unlock(&osdc->request_mutex);
2144 up_read(&osdc->map_sem);
2145 return rc;
2146}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002147EXPORT_SYMBOL(ceph_osdc_start_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002148
2149/*
2150 * wait for a request to complete
2151 */
2152int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2153 struct ceph_osd_request *req)
2154{
2155 int rc;
2156
2157 rc = wait_for_completion_interruptible(&req->r_completion);
2158 if (rc < 0) {
2159 mutex_lock(&osdc->request_mutex);
2160 __cancel_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08002161 __unregister_request(osdc, req);
Sage Weilf24e9982009-10-06 11:31:10 -07002162 mutex_unlock(&osdc->request_mutex);
Sage Weil25845472011-06-03 09:37:09 -07002163 complete_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08002164 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07002165 return rc;
2166 }
2167
2168 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2169 return req->r_result;
2170}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002171EXPORT_SYMBOL(ceph_osdc_wait_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002172
2173/*
2174 * sync - wait for all in-flight requests to flush. avoid starvation.
2175 */
2176void ceph_osdc_sync(struct ceph_osd_client *osdc)
2177{
2178 struct ceph_osd_request *req;
2179 u64 last_tid, next_tid = 0;
2180
2181 mutex_lock(&osdc->request_mutex);
2182 last_tid = osdc->last_tid;
2183 while (1) {
2184 req = __lookup_request_ge(osdc, next_tid);
2185 if (!req)
2186 break;
2187 if (req->r_tid > last_tid)
2188 break;
2189
2190 next_tid = req->r_tid + 1;
2191 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2192 continue;
2193
2194 ceph_osdc_get_request(req);
2195 mutex_unlock(&osdc->request_mutex);
2196 dout("sync waiting on tid %llu (last is %llu)\n",
2197 req->r_tid, last_tid);
2198 wait_for_completion(&req->r_safe_completion);
2199 mutex_lock(&osdc->request_mutex);
2200 ceph_osdc_put_request(req);
2201 }
2202 mutex_unlock(&osdc->request_mutex);
2203 dout("sync done (thru tid %llu)\n", last_tid);
2204}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002205EXPORT_SYMBOL(ceph_osdc_sync);
Sage Weilf24e9982009-10-06 11:31:10 -07002206
2207/*
2208 * init, shutdown
2209 */
2210int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2211{
2212 int err;
2213
2214 dout("init\n");
2215 osdc->client = client;
2216 osdc->osdmap = NULL;
2217 init_rwsem(&osdc->map_sem);
2218 init_completion(&osdc->map_waiters);
2219 osdc->last_requested_map = 0;
2220 mutex_init(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07002221 osdc->last_tid = 0;
2222 osdc->osds = RB_ROOT;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002223 INIT_LIST_HEAD(&osdc->osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07002224 osdc->requests = RB_ROOT;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08002225 INIT_LIST_HEAD(&osdc->req_lru);
Sage Weil6f6c7002011-01-17 20:34:08 -08002226 INIT_LIST_HEAD(&osdc->req_unsent);
2227 INIT_LIST_HEAD(&osdc->req_notarget);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002228 INIT_LIST_HEAD(&osdc->req_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07002229 osdc->num_requests = 0;
2230 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002231 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002232 spin_lock_init(&osdc->event_lock);
2233 osdc->event_tree = RB_ROOT;
2234 osdc->event_count = 0;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002235
2236 schedule_delayed_work(&osdc->osds_timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002237 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
Sage Weilf24e9982009-10-06 11:31:10 -07002238
Sage Weil5f44f142009-11-18 14:52:18 -08002239 err = -ENOMEM;
Sage Weilf24e9982009-10-06 11:31:10 -07002240 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2241 sizeof(struct ceph_osd_request));
2242 if (!osdc->req_mempool)
Sage Weil5f44f142009-11-18 14:52:18 -08002243 goto out;
Sage Weilf24e9982009-10-06 11:31:10 -07002244
Sage Weild50b4092012-07-09 14:22:34 -07002245 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2246 OSD_OP_FRONT_LEN, 10, true,
Sage Weil4f482802010-04-24 09:56:35 -07002247 "osd_op");
Sage Weilf24e9982009-10-06 11:31:10 -07002248 if (err < 0)
Sage Weil5f44f142009-11-18 14:52:18 -08002249 goto out_mempool;
Sage Weild50b4092012-07-09 14:22:34 -07002250 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
Sage Weil4f482802010-04-24 09:56:35 -07002251 OSD_OPREPLY_FRONT_LEN, 10, true,
2252 "osd_op_reply");
Sage Weilc16e7862010-03-01 13:02:00 -08002253 if (err < 0)
2254 goto out_msgpool;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002255
2256 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2257 if (IS_ERR(osdc->notify_wq)) {
2258 err = PTR_ERR(osdc->notify_wq);
2259 osdc->notify_wq = NULL;
2260 goto out_msgpool;
2261 }
Sage Weilf24e9982009-10-06 11:31:10 -07002262 return 0;
Sage Weil5f44f142009-11-18 14:52:18 -08002263
Sage Weilc16e7862010-03-01 13:02:00 -08002264out_msgpool:
2265 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weil5f44f142009-11-18 14:52:18 -08002266out_mempool:
2267 mempool_destroy(osdc->req_mempool);
2268out:
2269 return err;
Sage Weilf24e9982009-10-06 11:31:10 -07002270}
2271
2272void ceph_osdc_stop(struct ceph_osd_client *osdc)
2273{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002274 flush_workqueue(osdc->notify_wq);
2275 destroy_workqueue(osdc->notify_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07002276 cancel_delayed_work_sync(&osdc->timeout_work);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002277 cancel_delayed_work_sync(&osdc->osds_timeout_work);
Sage Weilf24e9982009-10-06 11:31:10 -07002278 if (osdc->osdmap) {
2279 ceph_osdmap_destroy(osdc->osdmap);
2280 osdc->osdmap = NULL;
2281 }
Sage Weilaca420b2011-08-31 14:45:53 -07002282 remove_all_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07002283 mempool_destroy(osdc->req_mempool);
2284 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weilc16e7862010-03-01 13:02:00 -08002285 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
Sage Weilf24e9982009-10-06 11:31:10 -07002286}
2287
2288/*
2289 * Read some contiguous pages. If we cross a stripe boundary, shorten
2290 * *plen. Return number of bytes read, or error.
2291 */
2292int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2293 struct ceph_vino vino, struct ceph_file_layout *layout,
2294 u64 off, u64 *plen,
2295 u32 truncate_seq, u64 truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -08002296 struct page **pages, int num_pages, int page_align)
Sage Weilf24e9982009-10-06 11:31:10 -07002297{
2298 struct ceph_osd_request *req;
2299 int rc = 0;
2300
2301 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2302 vino.snap, off, *plen);
Alex Elder79528732013-04-03 21:32:51 -05002303 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002304 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
Alex Elderacead002013-03-14 14:09:05 -05002305 NULL, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002306 false);
Sage Weil68162822012-09-24 21:01:02 -07002307 if (IS_ERR(req))
2308 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002309
2310 /* it may be a short read due to an object boundary */
Alex Elder0fff87e2013-02-14 12:16:43 -06002311
Alex Elder406e2c92013-04-15 14:50:36 -05002312 osd_req_op_extent_osd_data_pages(req, 0,
Alex Eldera4ce40a2013-04-05 01:27:12 -05002313 pages, *plen, page_align, false, false);
Sage Weilf24e9982009-10-06 11:31:10 -07002314
Alex Eldere0c59482013-03-07 15:38:25 -06002315 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
Alex Elder43bfe5d2013-04-03 01:28:57 -05002316 off, *plen, *plen, page_align);
Sage Weilf24e9982009-10-06 11:31:10 -07002317
Alex Elder79528732013-04-03 21:32:51 -05002318 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
Alex Elder02ee07d2013-03-14 14:09:06 -05002319
Sage Weilf24e9982009-10-06 11:31:10 -07002320 rc = ceph_osdc_start_request(osdc, req, false);
2321 if (!rc)
2322 rc = ceph_osdc_wait_request(osdc, req);
2323
2324 ceph_osdc_put_request(req);
2325 dout("readpages result %d\n", rc);
2326 return rc;
2327}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002328EXPORT_SYMBOL(ceph_osdc_readpages);
Sage Weilf24e9982009-10-06 11:31:10 -07002329
2330/*
2331 * do a synchronous write on N pages
2332 */
2333int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2334 struct ceph_file_layout *layout,
2335 struct ceph_snap_context *snapc,
2336 u64 off, u64 len,
2337 u32 truncate_seq, u64 truncate_size,
2338 struct timespec *mtime,
Alex Elder24808822013-02-15 11:42:29 -06002339 struct page **pages, int num_pages)
Sage Weilf24e9982009-10-06 11:31:10 -07002340{
2341 struct ceph_osd_request *req;
2342 int rc = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -08002343 int page_align = off & ~PAGE_MASK;
Sage Weilf24e9982009-10-06 11:31:10 -07002344
Alex Elderacead002013-03-14 14:09:05 -05002345 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
Alex Elder79528732013-04-03 21:32:51 -05002346 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002347 CEPH_OSD_OP_WRITE,
Alex Elder24808822013-02-15 11:42:29 -06002348 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
Alex Elderacead002013-03-14 14:09:05 -05002349 snapc, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002350 true);
Sage Weil68162822012-09-24 21:01:02 -07002351 if (IS_ERR(req))
2352 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002353
2354 /* it may be a short write due to an object boundary */
Alex Elder406e2c92013-04-15 14:50:36 -05002355 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
Alex Elder43bfe5d2013-04-03 01:28:57 -05002356 false, false);
2357 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
Sage Weilf24e9982009-10-06 11:31:10 -07002358
Alex Elder79528732013-04-03 21:32:51 -05002359 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
Alex Elder02ee07d2013-03-14 14:09:06 -05002360
Alex Elder87f979d2013-02-15 11:42:29 -06002361 rc = ceph_osdc_start_request(osdc, req, true);
Sage Weilf24e9982009-10-06 11:31:10 -07002362 if (!rc)
2363 rc = ceph_osdc_wait_request(osdc, req);
2364
2365 ceph_osdc_put_request(req);
2366 if (rc == 0)
2367 rc = len;
2368 dout("writepages result %d\n", rc);
2369 return rc;
2370}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002371EXPORT_SYMBOL(ceph_osdc_writepages);
Sage Weilf24e9982009-10-06 11:31:10 -07002372
Alex Elder5522ae02013-05-01 12:43:04 -05002373int ceph_osdc_setup(void)
2374{
2375 BUG_ON(ceph_osd_request_cache);
2376 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2377 sizeof (struct ceph_osd_request),
2378 __alignof__(struct ceph_osd_request),
2379 0, NULL);
2380
2381 return ceph_osd_request_cache ? 0 : -ENOMEM;
2382}
2383EXPORT_SYMBOL(ceph_osdc_setup);
2384
2385void ceph_osdc_cleanup(void)
2386{
2387 BUG_ON(!ceph_osd_request_cache);
2388 kmem_cache_destroy(ceph_osd_request_cache);
2389 ceph_osd_request_cache = NULL;
2390}
2391EXPORT_SYMBOL(ceph_osdc_cleanup);
2392
Sage Weilf24e9982009-10-06 11:31:10 -07002393/*
2394 * handle incoming message
2395 */
2396static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2397{
2398 struct ceph_osd *osd = con->private;
Julia Lawall32c895e2009-11-21 16:53:16 +01002399 struct ceph_osd_client *osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002400 int type = le16_to_cpu(msg->hdr.type);
2401
2402 if (!osd)
Sage Weil4a32f932010-06-13 10:27:53 -07002403 goto out;
Julia Lawall32c895e2009-11-21 16:53:16 +01002404 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002405
2406 switch (type) {
2407 case CEPH_MSG_OSD_MAP:
2408 ceph_osdc_handle_map(osdc, msg);
2409 break;
2410 case CEPH_MSG_OSD_OPREPLY:
Sage Weil350b1c32009-12-22 10:45:45 -08002411 handle_reply(osdc, msg, con);
Sage Weilf24e9982009-10-06 11:31:10 -07002412 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002413 case CEPH_MSG_WATCH_NOTIFY:
2414 handle_watch_notify(osdc, msg);
2415 break;
Sage Weilf24e9982009-10-06 11:31:10 -07002416
2417 default:
2418 pr_err("received unknown message type %d %s\n", type,
2419 ceph_msg_type_name(type));
2420 }
Sage Weil4a32f932010-06-13 10:27:53 -07002421out:
Sage Weilf24e9982009-10-06 11:31:10 -07002422 ceph_msg_put(msg);
2423}
2424
Sage Weil5b3a4db2010-02-19 21:43:23 -08002425/*
Sage Weil21b667f2010-03-04 10:22:59 -08002426 * lookup and return message for incoming reply. set up reply message
2427 * pages.
Sage Weil5b3a4db2010-02-19 21:43:23 -08002428 */
2429static struct ceph_msg *get_reply(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002430 struct ceph_msg_header *hdr,
2431 int *skip)
Sage Weilf24e9982009-10-06 11:31:10 -07002432{
2433 struct ceph_osd *osd = con->private;
2434 struct ceph_osd_client *osdc = osd->o_osdc;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002435 struct ceph_msg *m;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002436 struct ceph_osd_request *req;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002437 int front = le32_to_cpu(hdr->front_len);
2438 int data_len = le32_to_cpu(hdr->data_len);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002439 u64 tid;
Sage Weilf24e9982009-10-06 11:31:10 -07002440
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002441 tid = le64_to_cpu(hdr->tid);
2442 mutex_lock(&osdc->request_mutex);
2443 req = __lookup_request(osdc, tid);
2444 if (!req) {
2445 *skip = 1;
2446 m = NULL;
Sage Weil756a16a2012-07-30 16:26:13 -07002447 dout("get_reply unknown tid %llu from osd%d\n", tid,
2448 osd->o_osd);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002449 goto out;
2450 }
Sage Weilc16e7862010-03-01 13:02:00 -08002451
Alex Elderace6d3a2013-04-01 16:12:14 -05002452 if (req->r_reply->con)
Alex Elder8921d112012-06-01 14:56:43 -05002453 dout("%s revoking msg %p from old con %p\n", __func__,
Alex Elderace6d3a2013-04-01 16:12:14 -05002454 req->r_reply, req->r_reply->con);
2455 ceph_msg_revoke_incoming(req->r_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002456
Sage Weilc16e7862010-03-01 13:02:00 -08002457 if (front > req->r_reply->front.iov_len) {
Alex Elder49743412013-05-29 11:19:00 -05002458 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2459 front, (int)req->r_reply->front.iov_len,
2460 (unsigned int)con->peer_name.type,
2461 le64_to_cpu(con->peer_name.num));
Sage Weilb61c2762011-08-09 15:03:46 -07002462 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07002463 if (!m)
Sage Weilc16e7862010-03-01 13:02:00 -08002464 goto out;
2465 ceph_msg_put(req->r_reply);
2466 req->r_reply = m;
2467 }
2468 m = ceph_msg_get(req->r_reply);
2469
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002470 if (data_len > 0) {
Alex Eldera4ce40a2013-04-05 01:27:12 -05002471 struct ceph_osd_data *osd_data;
Alex Elder0fff87e2013-02-14 12:16:43 -06002472
Alex Eldera4ce40a2013-04-05 01:27:12 -05002473 /*
2474 * XXX This is assuming there is only one op containing
2475 * XXX page data. Probably OK for reads, but this
2476 * XXX ought to be done more generally.
2477 */
Alex Elder406e2c92013-04-15 14:50:36 -05002478 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder0fff87e2013-02-14 12:16:43 -06002479 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
Alex Elder0fff87e2013-02-14 12:16:43 -06002480 if (osd_data->pages &&
Alex Eldere0c59482013-03-07 15:38:25 -06002481 unlikely(osd_data->length < data_len)) {
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002482
Alex Eldere0c59482013-03-07 15:38:25 -06002483 pr_warning("tid %lld reply has %d bytes "
2484 "we had only %llu bytes ready\n",
2485 tid, data_len, osd_data->length);
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002486 *skip = 1;
2487 ceph_msg_put(m);
2488 m = NULL;
2489 goto out;
2490 }
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002491 }
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002492 }
Sage Weil5b3a4db2010-02-19 21:43:23 -08002493 *skip = 0;
Sage Weilc16e7862010-03-01 13:02:00 -08002494 dout("get_reply tid %lld %p\n", tid, m);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002495
2496out:
2497 mutex_unlock(&osdc->request_mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002498 return m;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002499
2500}
2501
2502static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2503 struct ceph_msg_header *hdr,
2504 int *skip)
2505{
2506 struct ceph_osd *osd = con->private;
2507 int type = le16_to_cpu(hdr->type);
2508 int front = le32_to_cpu(hdr->front_len);
2509
Alex Elder1c20f2d2012-06-04 14:43:32 -05002510 *skip = 0;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002511 switch (type) {
2512 case CEPH_MSG_OSD_MAP:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002513 case CEPH_MSG_WATCH_NOTIFY:
Sage Weilb61c2762011-08-09 15:03:46 -07002514 return ceph_msg_new(type, front, GFP_NOFS, false);
Sage Weil5b3a4db2010-02-19 21:43:23 -08002515 case CEPH_MSG_OSD_OPREPLY:
2516 return get_reply(con, hdr, skip);
2517 default:
2518 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2519 osd->o_osd);
2520 *skip = 1;
2521 return NULL;
2522 }
Sage Weilf24e9982009-10-06 11:31:10 -07002523}
2524
2525/*
2526 * Wrappers to refcount containing ceph_osd struct
2527 */
2528static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2529{
2530 struct ceph_osd *osd = con->private;
2531 if (get_osd(osd))
2532 return con;
2533 return NULL;
2534}
2535
2536static void put_osd_con(struct ceph_connection *con)
2537{
2538 struct ceph_osd *osd = con->private;
2539 put_osd(osd);
2540}
2541
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002542/*
2543 * authentication
2544 */
Alex Eldera3530df2012-05-16 15:16:39 -05002545/*
2546 * Note: returned pointer is the address of a structure that's
2547 * managed separately. Caller must *not* attempt to free it.
2548 */
2549static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
Alex Elder8f43fb52012-05-16 15:16:39 -05002550 int *proto, int force_new)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002551{
2552 struct ceph_osd *o = con->private;
2553 struct ceph_osd_client *osdc = o->o_osdc;
2554 struct ceph_auth_client *ac = osdc->client->monc.auth;
Alex Elder74f18692012-05-16 15:16:39 -05002555 struct ceph_auth_handshake *auth = &o->o_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002556
Alex Elder74f18692012-05-16 15:16:39 -05002557 if (force_new && auth->authorizer) {
Sage Weil27859f92013-03-25 10:26:14 -07002558 ceph_auth_destroy_authorizer(ac, auth->authorizer);
Alex Elder74f18692012-05-16 15:16:39 -05002559 auth->authorizer = NULL;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002560 }
Sage Weil27859f92013-03-25 10:26:14 -07002561 if (!auth->authorizer) {
2562 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2563 auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002564 if (ret)
Alex Eldera3530df2012-05-16 15:16:39 -05002565 return ERR_PTR(ret);
Sage Weil27859f92013-03-25 10:26:14 -07002566 } else {
2567 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
Sage Weil0bed9b52013-03-25 10:26:01 -07002568 auth);
2569 if (ret)
2570 return ERR_PTR(ret);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002571 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002572 *proto = ac->protocol;
Alex Elder74f18692012-05-16 15:16:39 -05002573
Alex Eldera3530df2012-05-16 15:16:39 -05002574 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002575}
2576
2577
2578static int verify_authorizer_reply(struct ceph_connection *con, int len)
2579{
2580 struct ceph_osd *o = con->private;
2581 struct ceph_osd_client *osdc = o->o_osdc;
2582 struct ceph_auth_client *ac = osdc->client->monc.auth;
2583
Sage Weil27859f92013-03-25 10:26:14 -07002584 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002585}
2586
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002587static int invalidate_authorizer(struct ceph_connection *con)
2588{
2589 struct ceph_osd *o = con->private;
2590 struct ceph_osd_client *osdc = o->o_osdc;
2591 struct ceph_auth_client *ac = osdc->client->monc.auth;
2592
Sage Weil27859f92013-03-25 10:26:14 -07002593 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002594 return ceph_monc_validate_auth(&osdc->client->monc);
2595}
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002596
Tobias Klauser9e327892010-05-20 10:40:19 +02002597static const struct ceph_connection_operations osd_con_ops = {
Sage Weilf24e9982009-10-06 11:31:10 -07002598 .get = get_osd_con,
2599 .put = put_osd_con,
2600 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002601 .get_authorizer = get_authorizer,
2602 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002603 .invalidate_authorizer = invalidate_authorizer,
Sage Weilf24e9982009-10-06 11:31:10 -07002604 .alloc_msg = alloc_msg,
Sage Weil81b024e2009-10-09 10:29:18 -07002605 .fault = osd_reset,
Sage Weilf24e9982009-10-06 11:31:10 -07002606};