blob: a3395fdfbd4f6345cdf3bd3a12ebdc1d7f26ecbb [file] [log] [blame]
Alex Eldera4ce40a2013-04-05 01:27:12 -05001
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002#include <linux/ceph/ceph_debug.h>
Sage Weilf24e9982009-10-06 11:31:10 -07003
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07004#include <linux/module.h>
Sage Weilf24e9982009-10-06 11:31:10 -07005#include <linux/err.h>
6#include <linux/highmem.h>
7#include <linux/mm.h>
8#include <linux/pagemap.h>
9#include <linux/slab.h>
10#include <linux/uaccess.h>
Yehuda Sadeh68b44762010-04-06 15:01:27 -070011#ifdef CONFIG_BLOCK
12#include <linux/bio.h>
13#endif
Sage Weilf24e9982009-10-06 11:31:10 -070014
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070015#include <linux/ceph/libceph.h>
16#include <linux/ceph/osd_client.h>
17#include <linux/ceph/messenger.h>
18#include <linux/ceph/decode.h>
19#include <linux/ceph/auth.h>
20#include <linux/ceph/pagelist.h>
Sage Weilf24e9982009-10-06 11:31:10 -070021
Sage Weilc16e7862010-03-01 13:02:00 -080022#define OSD_OP_FRONT_LEN 4096
23#define OSD_OPREPLY_FRONT_LEN 512
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -080024
Alex Elder5522ae02013-05-01 12:43:04 -050025static struct kmem_cache *ceph_osd_request_cache;
26
Tobias Klauser9e327892010-05-20 10:40:19 +020027static const struct ceph_connection_operations osd_con_ops;
Sage Weilf24e9982009-10-06 11:31:10 -070028
Alex Elderf9d25192013-02-15 11:42:29 -060029static void __send_queued(struct ceph_osd_client *osdc);
Sage Weil6f6c7002011-01-17 20:34:08 -080030static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -070031static void __register_request(struct ceph_osd_client *osdc,
32 struct ceph_osd_request *req);
33static void __unregister_linger_request(struct ceph_osd_client *osdc,
34 struct ceph_osd_request *req);
Sage Weil56e925b2012-01-03 12:34:34 -080035static void __send_request(struct ceph_osd_client *osdc,
36 struct ceph_osd_request *req);
Sage Weilf24e9982009-10-06 11:31:10 -070037
38/*
39 * Implement client access to distributed object storage cluster.
40 *
41 * All data objects are stored within a cluster/cloud of OSDs, or
42 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
43 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
44 * remote daemons serving up and coordinating consistent and safe
45 * access to storage.
46 *
47 * Cluster membership and the mapping of data objects onto storage devices
48 * are described by the osd map.
49 *
50 * We keep track of pending OSD requests (read, write), resubmit
51 * requests to different OSDs when the cluster topology/data layout
52 * change, or retry the affected requests when the communications
53 * channel with an OSD is reset.
54 */
55
56/*
57 * calculate the mapping of a file extent onto an object, and fill out the
58 * request accordingly. shorten extent as necessary if it crosses an
59 * object boundary.
60 *
61 * fill osd op in request message.
62 */
Alex Elderdbe0fc42013-02-15 22:10:17 -060063static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
Alex Eldera19dadf2013-03-13 20:50:01 -050064 u64 *objnum, u64 *objoff, u64 *objlen)
Sage Weilf24e9982009-10-06 11:31:10 -070065{
Alex Elder60e56f12013-02-15 11:42:29 -060066 u64 orig_len = *plen;
Sage Weild63b77f2012-09-24 20:59:48 -070067 int r;
Sage Weilf24e9982009-10-06 11:31:10 -070068
Alex Elder60e56f12013-02-15 11:42:29 -060069 /* object extent? */
Alex Elder75d1c942013-03-13 20:50:00 -050070 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
71 objoff, objlen);
Sage Weild63b77f2012-09-24 20:59:48 -070072 if (r < 0)
73 return r;
Alex Elder75d1c942013-03-13 20:50:00 -050074 if (*objlen < orig_len) {
75 *plen = *objlen;
Alex Elder60e56f12013-02-15 11:42:29 -060076 dout(" skipping last %llu, final file extent %llu~%llu\n",
77 orig_len - *plen, off, *plen);
78 }
79
Alex Elder75d1c942013-03-13 20:50:00 -050080 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
Sage Weilf24e9982009-10-06 11:31:10 -070081
Alex Elder3ff5f382013-02-15 22:10:17 -060082 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -070083}
84
Alex Elderc54d47b2013-04-03 01:28:57 -050085static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
86{
87 memset(osd_data, 0, sizeof (*osd_data));
88 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
89}
90
Alex Eldera4ce40a2013-04-05 01:27:12 -050091static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -050092 struct page **pages, u64 length, u32 alignment,
93 bool pages_from_pool, bool own_pages)
94{
95 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
96 osd_data->pages = pages;
97 osd_data->length = length;
98 osd_data->alignment = alignment;
99 osd_data->pages_from_pool = pages_from_pool;
100 osd_data->own_pages = own_pages;
101}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500102
Alex Eldera4ce40a2013-04-05 01:27:12 -0500103static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -0500104 struct ceph_pagelist *pagelist)
105{
106 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
107 osd_data->pagelist = pagelist;
108}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500109
110#ifdef CONFIG_BLOCK
Alex Eldera4ce40a2013-04-05 01:27:12 -0500111static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -0500112 struct bio *bio, size_t bio_length)
113{
114 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
115 osd_data->bio = bio;
116 osd_data->bio_length = bio_length;
117}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500118#endif /* CONFIG_BLOCK */
119
Alex Elder863c7eb2013-04-15 14:50:36 -0500120#define osd_req_op_data(oreq, whch, typ, fld) \
121 ({ \
122 BUG_ON(whch >= (oreq)->r_num_ops); \
123 &(oreq)->r_ops[whch].typ.fld; \
124 })
125
Alex Elder49719772013-02-11 12:33:24 -0600126static struct ceph_osd_data *
127osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
128{
129 BUG_ON(which >= osd_req->r_num_ops);
130
131 return &osd_req->r_ops[which].raw_data_in;
132}
133
Alex Eldera4ce40a2013-04-05 01:27:12 -0500134struct ceph_osd_data *
135osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500136 unsigned int which)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500137{
Alex Elder863c7eb2013-04-15 14:50:36 -0500138 return osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500139}
140EXPORT_SYMBOL(osd_req_op_extent_osd_data);
141
142struct ceph_osd_data *
Alex Eldera4ce40a2013-04-05 01:27:12 -0500143osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
144 unsigned int which)
145{
Alex Elder863c7eb2013-04-15 14:50:36 -0500146 return osd_req_op_data(osd_req, which, cls, response_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500147}
148EXPORT_SYMBOL(osd_req_op_cls_response_data); /* ??? */
149
Alex Elder49719772013-02-11 12:33:24 -0600150void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
151 unsigned int which, struct page **pages,
152 u64 length, u32 alignment,
153 bool pages_from_pool, bool own_pages)
154{
155 struct ceph_osd_data *osd_data;
156
157 osd_data = osd_req_op_raw_data_in(osd_req, which);
158 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
159 pages_from_pool, own_pages);
160}
161EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
162
Alex Eldera4ce40a2013-04-05 01:27:12 -0500163void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500164 unsigned int which, struct page **pages,
165 u64 length, u32 alignment,
Alex Eldera4ce40a2013-04-05 01:27:12 -0500166 bool pages_from_pool, bool own_pages)
167{
168 struct ceph_osd_data *osd_data;
169
Alex Elder863c7eb2013-04-15 14:50:36 -0500170 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500171 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
172 pages_from_pool, own_pages);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500173}
174EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
175
176void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500177 unsigned int which, struct ceph_pagelist *pagelist)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500178{
179 struct ceph_osd_data *osd_data;
180
Alex Elder863c7eb2013-04-15 14:50:36 -0500181 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500182 ceph_osd_data_pagelist_init(osd_data, pagelist);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500183}
184EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
185
186#ifdef CONFIG_BLOCK
187void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500188 unsigned int which, struct bio *bio, size_t bio_length)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500189{
190 struct ceph_osd_data *osd_data;
Alex Elder863c7eb2013-04-15 14:50:36 -0500191
192 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500193 ceph_osd_data_bio_init(osd_data, bio, bio_length);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500194}
195EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
196#endif /* CONFIG_BLOCK */
197
198static void osd_req_op_cls_request_info_pagelist(
199 struct ceph_osd_request *osd_req,
200 unsigned int which, struct ceph_pagelist *pagelist)
201{
202 struct ceph_osd_data *osd_data;
203
Alex Elder863c7eb2013-04-15 14:50:36 -0500204 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500205 ceph_osd_data_pagelist_init(osd_data, pagelist);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500206}
207
Alex Elder04017e22013-04-05 14:46:02 -0500208void osd_req_op_cls_request_data_pagelist(
209 struct ceph_osd_request *osd_req,
210 unsigned int which, struct ceph_pagelist *pagelist)
211{
212 struct ceph_osd_data *osd_data;
213
Alex Elder863c7eb2013-04-15 14:50:36 -0500214 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
Alex Elder04017e22013-04-05 14:46:02 -0500215 ceph_osd_data_pagelist_init(osd_data, pagelist);
216}
217EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
218
Alex Elder6c57b552013-04-19 15:34:49 -0500219void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
220 unsigned int which, struct page **pages, u64 length,
221 u32 alignment, bool pages_from_pool, bool own_pages)
222{
223 struct ceph_osd_data *osd_data;
224
225 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
226 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
227 pages_from_pool, own_pages);
228}
229EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
230
Alex Eldera4ce40a2013-04-05 01:27:12 -0500231void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
232 unsigned int which, struct page **pages, u64 length,
233 u32 alignment, bool pages_from_pool, bool own_pages)
234{
235 struct ceph_osd_data *osd_data;
236
Alex Elder863c7eb2013-04-15 14:50:36 -0500237 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500238 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
239 pages_from_pool, own_pages);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500240}
241EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
242
Alex Elder23c08a9cb2013-04-03 01:28:58 -0500243static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
244{
245 switch (osd_data->type) {
246 case CEPH_OSD_DATA_TYPE_NONE:
247 return 0;
248 case CEPH_OSD_DATA_TYPE_PAGES:
249 return osd_data->length;
250 case CEPH_OSD_DATA_TYPE_PAGELIST:
251 return (u64)osd_data->pagelist->length;
252#ifdef CONFIG_BLOCK
253 case CEPH_OSD_DATA_TYPE_BIO:
254 return (u64)osd_data->bio_length;
255#endif /* CONFIG_BLOCK */
256 default:
257 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
258 return 0;
259 }
260}
261
Alex Elderc54d47b2013-04-03 01:28:57 -0500262static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
263{
Alex Elder54764922013-04-05 01:27:12 -0500264 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
Alex Elderc54d47b2013-04-03 01:28:57 -0500265 int num_pages;
266
267 num_pages = calc_pages_for((u64)osd_data->alignment,
268 (u64)osd_data->length);
269 ceph_release_page_vector(osd_data->pages, num_pages);
270 }
Alex Elder54764922013-04-05 01:27:12 -0500271 ceph_osd_data_init(osd_data);
272}
273
274static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
275 unsigned int which)
276{
277 struct ceph_osd_req_op *op;
278
279 BUG_ON(which >= osd_req->r_num_ops);
280 op = &osd_req->r_ops[which];
281
282 switch (op->op) {
283 case CEPH_OSD_OP_READ:
284 case CEPH_OSD_OP_WRITE:
285 ceph_osd_data_release(&op->extent.osd_data);
286 break;
287 case CEPH_OSD_OP_CALL:
288 ceph_osd_data_release(&op->cls.request_info);
Alex Elder04017e22013-04-05 14:46:02 -0500289 ceph_osd_data_release(&op->cls.request_data);
Alex Elder54764922013-04-05 01:27:12 -0500290 ceph_osd_data_release(&op->cls.response_data);
291 break;
292 default:
293 break;
294 }
Alex Elderc54d47b2013-04-03 01:28:57 -0500295}
296
Sage Weilf24e9982009-10-06 11:31:10 -0700297/*
298 * requests
299 */
Sage Weil415e49a2009-12-07 13:37:03 -0800300void ceph_osdc_release_request(struct kref *kref)
Sage Weilf24e9982009-10-06 11:31:10 -0700301{
Alex Elderc54d47b2013-04-03 01:28:57 -0500302 struct ceph_osd_request *req;
Alex Elder54764922013-04-05 01:27:12 -0500303 unsigned int which;
Sage Weil415e49a2009-12-07 13:37:03 -0800304
Alex Elderc54d47b2013-04-03 01:28:57 -0500305 req = container_of(kref, struct ceph_osd_request, r_kref);
Sage Weil415e49a2009-12-07 13:37:03 -0800306 if (req->r_request)
307 ceph_msg_put(req->r_request);
Alex Elderace6d3a2013-04-01 16:12:14 -0500308 if (req->r_reply) {
Alex Elder8921d112012-06-01 14:56:43 -0500309 ceph_msg_revoke_incoming(req->r_reply);
Alex Elderab8cb342012-06-04 14:43:32 -0500310 ceph_msg_put(req->r_reply);
Alex Elderace6d3a2013-04-01 16:12:14 -0500311 }
Alex Elder0fff87e2013-02-14 12:16:43 -0600312
Alex Elder54764922013-04-05 01:27:12 -0500313 for (which = 0; which < req->r_num_ops; which++)
314 osd_req_op_data_release(req, which);
Alex Elder0fff87e2013-02-14 12:16:43 -0600315
Sage Weil415e49a2009-12-07 13:37:03 -0800316 ceph_put_snap_context(req->r_snapc);
317 if (req->r_mempool)
318 mempool_free(req, req->r_osdc->req_mempool);
319 else
Alex Elder5522ae02013-05-01 12:43:04 -0500320 kmem_cache_free(ceph_osd_request_cache, req);
321
Sage Weilf24e9982009-10-06 11:31:10 -0700322}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700323EXPORT_SYMBOL(ceph_osdc_release_request);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700324
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700325struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700326 struct ceph_snap_context *snapc,
Sage Weil1b83bef2013-02-25 16:11:12 -0800327 unsigned int num_ops,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700328 bool use_mempool,
Alex Elder54a54002012-11-13 21:11:15 -0600329 gfp_t gfp_flags)
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700330{
331 struct ceph_osd_request *req;
332 struct ceph_msg *msg;
Sage Weil1b83bef2013-02-25 16:11:12 -0800333 size_t msg_size;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700334
Alex Elder79528732013-04-03 21:32:51 -0500335 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
336 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
337
Sage Weil1b83bef2013-02-25 16:11:12 -0800338 msg_size = 4 + 4 + 8 + 8 + 4+8;
339 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
340 msg_size += 1 + 8 + 4 + 4; /* pg_t */
341 msg_size += 4 + MAX_OBJ_NAME_SIZE;
342 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
343 msg_size += 8; /* snapid */
344 msg_size += 8; /* snap_seq */
345 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
346 msg_size += 4;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700347
348 if (use_mempool) {
349 req = mempool_alloc(osdc->req_mempool, gfp_flags);
350 memset(req, 0, sizeof(*req));
351 } else {
Alex Elder5522ae02013-05-01 12:43:04 -0500352 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700353 }
354 if (req == NULL)
355 return NULL;
356
357 req->r_osdc = osdc;
358 req->r_mempool = use_mempool;
Alex Elder79528732013-04-03 21:32:51 -0500359 req->r_num_ops = num_ops;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700360
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700361 kref_init(&req->r_kref);
362 init_completion(&req->r_completion);
363 init_completion(&req->r_safe_completion);
Alex Eldera978fa22012-12-17 12:23:48 -0600364 RB_CLEAR_NODE(&req->r_node);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700365 INIT_LIST_HEAD(&req->r_unsafe_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700366 INIT_LIST_HEAD(&req->r_linger_item);
367 INIT_LIST_HEAD(&req->r_linger_osd);
Sage Weil935b6392011-09-16 11:13:17 -0700368 INIT_LIST_HEAD(&req->r_req_lru_item);
Sage Weilcd430452012-07-09 14:31:41 -0700369 INIT_LIST_HEAD(&req->r_osd_item);
370
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700371 /* create reply message */
372 if (use_mempool)
373 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
374 else
375 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
Sage Weilb61c2762011-08-09 15:03:46 -0700376 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700377 if (!msg) {
378 ceph_osdc_put_request(req);
379 return NULL;
380 }
381 req->r_reply = msg;
382
383 /* create request message; allow space for oid */
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700384 if (use_mempool)
385 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
386 else
Sage Weilb61c2762011-08-09 15:03:46 -0700387 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700388 if (!msg) {
389 ceph_osdc_put_request(req);
390 return NULL;
391 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700392
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700393 memset(msg->front.iov_base, 0, msg->front.iov_len);
394
395 req->r_request = msg;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700396
397 return req;
398}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700399EXPORT_SYMBOL(ceph_osdc_alloc_request);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700400
Alex Eldera8dd0a32013-03-13 20:50:00 -0500401static bool osd_req_opcode_valid(u16 opcode)
402{
403 switch (opcode) {
404 case CEPH_OSD_OP_READ:
405 case CEPH_OSD_OP_STAT:
406 case CEPH_OSD_OP_MAPEXT:
407 case CEPH_OSD_OP_MASKTRUNC:
408 case CEPH_OSD_OP_SPARSE_READ:
409 case CEPH_OSD_OP_NOTIFY:
410 case CEPH_OSD_OP_NOTIFY_ACK:
411 case CEPH_OSD_OP_ASSERT_VER:
412 case CEPH_OSD_OP_WRITE:
413 case CEPH_OSD_OP_WRITEFULL:
414 case CEPH_OSD_OP_TRUNCATE:
415 case CEPH_OSD_OP_ZERO:
416 case CEPH_OSD_OP_DELETE:
417 case CEPH_OSD_OP_APPEND:
418 case CEPH_OSD_OP_STARTSYNC:
419 case CEPH_OSD_OP_SETTRUNC:
420 case CEPH_OSD_OP_TRIMTRUNC:
421 case CEPH_OSD_OP_TMAPUP:
422 case CEPH_OSD_OP_TMAPPUT:
423 case CEPH_OSD_OP_TMAPGET:
424 case CEPH_OSD_OP_CREATE:
425 case CEPH_OSD_OP_ROLLBACK:
426 case CEPH_OSD_OP_WATCH:
427 case CEPH_OSD_OP_OMAPGETKEYS:
428 case CEPH_OSD_OP_OMAPGETVALS:
429 case CEPH_OSD_OP_OMAPGETHEADER:
430 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
431 case CEPH_OSD_OP_OMAPSETVALS:
432 case CEPH_OSD_OP_OMAPSETHEADER:
433 case CEPH_OSD_OP_OMAPCLEAR:
434 case CEPH_OSD_OP_OMAPRMKEYS:
435 case CEPH_OSD_OP_OMAP_CMP:
436 case CEPH_OSD_OP_CLONERANGE:
437 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
438 case CEPH_OSD_OP_SRC_CMPXATTR:
439 case CEPH_OSD_OP_GETXATTR:
440 case CEPH_OSD_OP_GETXATTRS:
441 case CEPH_OSD_OP_CMPXATTR:
442 case CEPH_OSD_OP_SETXATTR:
443 case CEPH_OSD_OP_SETXATTRS:
444 case CEPH_OSD_OP_RESETXATTRS:
445 case CEPH_OSD_OP_RMXATTR:
446 case CEPH_OSD_OP_PULL:
447 case CEPH_OSD_OP_PUSH:
448 case CEPH_OSD_OP_BALANCEREADS:
449 case CEPH_OSD_OP_UNBALANCEREADS:
450 case CEPH_OSD_OP_SCRUB:
451 case CEPH_OSD_OP_SCRUB_RESERVE:
452 case CEPH_OSD_OP_SCRUB_UNRESERVE:
453 case CEPH_OSD_OP_SCRUB_STOP:
454 case CEPH_OSD_OP_SCRUB_MAP:
455 case CEPH_OSD_OP_WRLOCK:
456 case CEPH_OSD_OP_WRUNLOCK:
457 case CEPH_OSD_OP_RDLOCK:
458 case CEPH_OSD_OP_RDUNLOCK:
459 case CEPH_OSD_OP_UPLOCK:
460 case CEPH_OSD_OP_DNLOCK:
461 case CEPH_OSD_OP_CALL:
462 case CEPH_OSD_OP_PGLS:
463 case CEPH_OSD_OP_PGLS_FILTER:
464 return true;
465 default:
466 return false;
467 }
468}
469
Alex Elder33803f32013-03-13 20:50:00 -0500470/*
471 * This is an osd op init function for opcodes that have no data or
472 * other information associated with them. It also serves as a
473 * common init routine for all the other init functions, below.
474 */
Alex Elderc99d2d42013-04-05 01:27:11 -0500475static struct ceph_osd_req_op *
Alex Elder49719772013-02-11 12:33:24 -0600476_osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
Alex Elderc99d2d42013-04-05 01:27:11 -0500477 u16 opcode)
Alex Elder33803f32013-03-13 20:50:00 -0500478{
Alex Elderc99d2d42013-04-05 01:27:11 -0500479 struct ceph_osd_req_op *op;
480
481 BUG_ON(which >= osd_req->r_num_ops);
Alex Elder33803f32013-03-13 20:50:00 -0500482 BUG_ON(!osd_req_opcode_valid(opcode));
483
Alex Elderc99d2d42013-04-05 01:27:11 -0500484 op = &osd_req->r_ops[which];
Alex Elder33803f32013-03-13 20:50:00 -0500485 memset(op, 0, sizeof (*op));
Alex Elder33803f32013-03-13 20:50:00 -0500486 op->op = opcode;
Alex Elderc99d2d42013-04-05 01:27:11 -0500487
488 return op;
Alex Elder33803f32013-03-13 20:50:00 -0500489}
490
Alex Elder49719772013-02-11 12:33:24 -0600491void osd_req_op_init(struct ceph_osd_request *osd_req,
492 unsigned int which, u16 opcode)
493{
494 (void)_osd_req_op_init(osd_req, which, opcode);
495}
496EXPORT_SYMBOL(osd_req_op_init);
497
Alex Elderc99d2d42013-04-05 01:27:11 -0500498void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
499 unsigned int which, u16 opcode,
Alex Elder33803f32013-03-13 20:50:00 -0500500 u64 offset, u64 length,
501 u64 truncate_size, u32 truncate_seq)
502{
Alex Elder49719772013-02-11 12:33:24 -0600503 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder33803f32013-03-13 20:50:00 -0500504 size_t payload_len = 0;
505
506 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
507
Alex Elder33803f32013-03-13 20:50:00 -0500508 op->extent.offset = offset;
509 op->extent.length = length;
510 op->extent.truncate_size = truncate_size;
511 op->extent.truncate_seq = truncate_seq;
512 if (opcode == CEPH_OSD_OP_WRITE)
513 payload_len += length;
514
515 op->payload_len = payload_len;
516}
517EXPORT_SYMBOL(osd_req_op_extent_init);
518
Alex Elderc99d2d42013-04-05 01:27:11 -0500519void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
520 unsigned int which, u64 length)
Alex Eldere5975c72013-03-14 14:09:05 -0500521{
Alex Elderc99d2d42013-04-05 01:27:11 -0500522 struct ceph_osd_req_op *op;
523 u64 previous;
524
525 BUG_ON(which >= osd_req->r_num_ops);
526 op = &osd_req->r_ops[which];
527 previous = op->extent.length;
Alex Eldere5975c72013-03-14 14:09:05 -0500528
529 if (length == previous)
530 return; /* Nothing to do */
531 BUG_ON(length > previous);
532
533 op->extent.length = length;
534 op->payload_len -= previous - length;
535}
536EXPORT_SYMBOL(osd_req_op_extent_update);
537
Alex Elderc99d2d42013-04-05 01:27:11 -0500538void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
Alex Elder04017e22013-04-05 14:46:02 -0500539 u16 opcode, const char *class, const char *method)
Alex Elder33803f32013-03-13 20:50:00 -0500540{
Alex Elder49719772013-02-11 12:33:24 -0600541 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder5f562df2013-04-05 01:27:12 -0500542 struct ceph_pagelist *pagelist;
Alex Elder33803f32013-03-13 20:50:00 -0500543 size_t payload_len = 0;
544 size_t size;
545
546 BUG_ON(opcode != CEPH_OSD_OP_CALL);
547
Alex Elder5f562df2013-04-05 01:27:12 -0500548 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
549 BUG_ON(!pagelist);
550 ceph_pagelist_init(pagelist);
551
Alex Elder33803f32013-03-13 20:50:00 -0500552 op->cls.class_name = class;
553 size = strlen(class);
554 BUG_ON(size > (size_t) U8_MAX);
555 op->cls.class_len = size;
Alex Elder5f562df2013-04-05 01:27:12 -0500556 ceph_pagelist_append(pagelist, class, size);
Alex Elder33803f32013-03-13 20:50:00 -0500557 payload_len += size;
558
559 op->cls.method_name = method;
560 size = strlen(method);
561 BUG_ON(size > (size_t) U8_MAX);
562 op->cls.method_len = size;
Alex Elder5f562df2013-04-05 01:27:12 -0500563 ceph_pagelist_append(pagelist, method, size);
Alex Elder33803f32013-03-13 20:50:00 -0500564 payload_len += size;
565
Alex Eldera4ce40a2013-04-05 01:27:12 -0500566 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
Alex Elder5f562df2013-04-05 01:27:12 -0500567
Alex Elder33803f32013-03-13 20:50:00 -0500568 op->cls.argc = 0; /* currently unused */
569
570 op->payload_len = payload_len;
571}
572EXPORT_SYMBOL(osd_req_op_cls_init);
Alex Elder8c042b02013-04-03 01:28:58 -0500573
Alex Elderc99d2d42013-04-05 01:27:11 -0500574void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
575 unsigned int which, u16 opcode,
Alex Elder33803f32013-03-13 20:50:00 -0500576 u64 cookie, u64 version, int flag)
577{
Alex Elder49719772013-02-11 12:33:24 -0600578 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder33803f32013-03-13 20:50:00 -0500579
Alex Elderc99d2d42013-04-05 01:27:11 -0500580 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
Alex Elder33803f32013-03-13 20:50:00 -0500581
582 op->watch.cookie = cookie;
Alex Elder9ef1ee52013-04-21 16:51:50 -0500583 op->watch.ver = version;
Alex Elder33803f32013-03-13 20:50:00 -0500584 if (opcode == CEPH_OSD_OP_WATCH && flag)
Alex Elderc99d2d42013-04-05 01:27:11 -0500585 op->watch.flag = (u8)1;
Alex Elder33803f32013-03-13 20:50:00 -0500586}
587EXPORT_SYMBOL(osd_req_op_watch_init);
588
Alex Elder90af3602013-04-05 14:46:01 -0500589static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
Alex Elderec9123c2013-04-05 01:27:12 -0500590 struct ceph_osd_data *osd_data)
591{
592 u64 length = ceph_osd_data_length(osd_data);
593
594 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
595 BUG_ON(length > (u64) SIZE_MAX);
596 if (length)
Alex Elder90af3602013-04-05 14:46:01 -0500597 ceph_msg_data_add_pages(msg, osd_data->pages,
Alex Elderec9123c2013-04-05 01:27:12 -0500598 length, osd_data->alignment);
599 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
600 BUG_ON(!length);
Alex Elder90af3602013-04-05 14:46:01 -0500601 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
Alex Elderec9123c2013-04-05 01:27:12 -0500602#ifdef CONFIG_BLOCK
603 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
Alex Elder90af3602013-04-05 14:46:01 -0500604 ceph_msg_data_add_bio(msg, osd_data->bio, length);
Alex Elderec9123c2013-04-05 01:27:12 -0500605#endif
606 } else {
607 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
608 }
609}
610
Alex Elder175face2013-03-08 13:35:36 -0600611static u64 osd_req_encode_op(struct ceph_osd_request *req,
Alex Elder79528732013-04-03 21:32:51 -0500612 struct ceph_osd_op *dst, unsigned int which)
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700613{
Alex Elder79528732013-04-03 21:32:51 -0500614 struct ceph_osd_req_op *src;
Alex Elder04017e22013-04-05 14:46:02 -0500615 struct ceph_osd_data *osd_data;
Alex Elder54d50642013-04-03 01:28:58 -0500616 u64 request_data_len = 0;
Alex Elder04017e22013-04-05 14:46:02 -0500617 u64 data_length;
Alex Elder175face2013-03-08 13:35:36 -0600618
Alex Elder79528732013-04-03 21:32:51 -0500619 BUG_ON(which >= req->r_num_ops);
620 src = &req->r_ops[which];
Alex Eldera8dd0a32013-03-13 20:50:00 -0500621 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
622 pr_err("unrecognized osd opcode %d\n", src->op);
623
624 return 0;
625 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700626
Alex Elder065a68f2012-04-20 15:49:43 -0500627 switch (src->op) {
Alex Elderfbfab532013-02-08 09:55:48 -0600628 case CEPH_OSD_OP_STAT:
Alex Elder49719772013-02-11 12:33:24 -0600629 osd_data = &src->raw_data_in;
630 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Alex Elderfbfab532013-02-08 09:55:48 -0600631 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700632 case CEPH_OSD_OP_READ:
633 case CEPH_OSD_OP_WRITE:
Alex Elder175face2013-03-08 13:35:36 -0600634 if (src->op == CEPH_OSD_OP_WRITE)
Alex Elder54d50642013-04-03 01:28:58 -0500635 request_data_len = src->extent.length;
Alex Elder175face2013-03-08 13:35:36 -0600636 dst->extent.offset = cpu_to_le64(src->extent.offset);
637 dst->extent.length = cpu_to_le64(src->extent.length);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700638 dst->extent.truncate_size =
639 cpu_to_le64(src->extent.truncate_size);
640 dst->extent.truncate_seq =
641 cpu_to_le32(src->extent.truncate_seq);
Alex Elder04017e22013-04-05 14:46:02 -0500642 osd_data = &src->extent.osd_data;
Alex Elder54764922013-04-05 01:27:12 -0500643 if (src->op == CEPH_OSD_OP_WRITE)
Alex Elder04017e22013-04-05 14:46:02 -0500644 ceph_osdc_msg_data_add(req->r_request, osd_data);
Alex Elder54764922013-04-05 01:27:12 -0500645 else
Alex Elder04017e22013-04-05 14:46:02 -0500646 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700647 break;
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700648 case CEPH_OSD_OP_CALL:
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700649 dst->cls.class_len = src->cls.class_len;
650 dst->cls.method_len = src->cls.method_len;
Alex Elder04017e22013-04-05 14:46:02 -0500651 osd_data = &src->cls.request_info;
652 ceph_osdc_msg_data_add(req->r_request, osd_data);
653 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
654 request_data_len = osd_data->pagelist->length;
655
656 osd_data = &src->cls.request_data;
657 data_length = ceph_osd_data_length(osd_data);
658 if (data_length) {
659 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
660 dst->cls.indata_len = cpu_to_le32(data_length);
661 ceph_osdc_msg_data_add(req->r_request, osd_data);
662 src->payload_len += data_length;
663 request_data_len += data_length;
664 }
665 osd_data = &src->cls.response_data;
666 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700667 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700668 case CEPH_OSD_OP_STARTSYNC:
669 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700670 case CEPH_OSD_OP_NOTIFY_ACK:
671 case CEPH_OSD_OP_WATCH:
672 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
673 dst->watch.ver = cpu_to_le64(src->watch.ver);
674 dst->watch.flag = src->watch.flag;
675 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700676 default:
Alex Elder4c464592013-02-15 11:42:30 -0600677 pr_err("unsupported osd opcode %s\n",
Alex Elder8f63ca22013-03-04 11:08:29 -0600678 ceph_osd_op_name(src->op));
Alex Elder4c464592013-02-15 11:42:30 -0600679 WARN_ON(1);
Alex Eldera8dd0a32013-03-13 20:50:00 -0500680
681 return 0;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700682 }
Alex Eldera8dd0a32013-03-13 20:50:00 -0500683 dst->op = cpu_to_le16(src->op);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700684 dst->payload_len = cpu_to_le32(src->payload_len);
Alex Elder175face2013-03-08 13:35:36 -0600685
Alex Elder54d50642013-04-03 01:28:58 -0500686 return request_data_len;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700687}
688
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700689/*
Sage Weilf24e9982009-10-06 11:31:10 -0700690 * build new request AND message, calculate layout, and adjust file
691 * extent as needed.
692 *
693 * if the file was recently truncated, we include information about its
694 * old and new size so that the object can be updated appropriately. (we
695 * avoid synchronously deleting truncated objects because it's slow.)
696 *
697 * if @do_sync, include a 'startsync' command so that the osd will flush
698 * data quickly.
699 */
700struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
701 struct ceph_file_layout *layout,
702 struct ceph_vino vino,
Alex Elderacead002013-03-14 14:09:05 -0500703 u64 off, u64 *plen, int num_ops,
Sage Weilf24e9982009-10-06 11:31:10 -0700704 int opcode, int flags,
705 struct ceph_snap_context *snapc,
Sage Weilf24e9982009-10-06 11:31:10 -0700706 u32 truncate_seq,
707 u64 truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -0600708 bool use_mempool)
Sage Weilf24e9982009-10-06 11:31:10 -0700709{
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700710 struct ceph_osd_request *req;
Alex Elder75d1c942013-03-13 20:50:00 -0500711 u64 objnum = 0;
712 u64 objoff = 0;
713 u64 objlen = 0;
Alex Elderd18d1e22013-03-13 20:50:01 -0500714 u32 object_size;
715 u64 object_base;
Sage Weil68162822012-09-24 21:01:02 -0700716 int r;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700717
Alex Elderd18d1e22013-03-13 20:50:01 -0500718 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700719
Alex Elderacead002013-03-14 14:09:05 -0500720 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
Alex Elderae7ca4a32012-11-13 21:11:15 -0600721 GFP_NOFS);
Sage Weil4ad12622011-05-03 09:23:36 -0700722 if (!req)
Sage Weil68162822012-09-24 21:01:02 -0700723 return ERR_PTR(-ENOMEM);
Alex Elder79528732013-04-03 21:32:51 -0500724
Alex Elderd178a9e2012-11-13 21:11:15 -0600725 req->r_flags = flags;
Sage Weilf24e9982009-10-06 11:31:10 -0700726
727 /* calculate max write size */
Alex Eldera19dadf2013-03-13 20:50:01 -0500728 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
Alex Elder3ff5f382013-02-15 22:10:17 -0600729 if (r < 0) {
730 ceph_osdc_put_request(req);
Sage Weil68162822012-09-24 21:01:02 -0700731 return ERR_PTR(r);
Alex Elder3ff5f382013-02-15 22:10:17 -0600732 }
Alex Eldera19dadf2013-03-13 20:50:01 -0500733
Alex Elderd18d1e22013-03-13 20:50:01 -0500734 object_size = le32_to_cpu(layout->fl_object_size);
735 object_base = off - objoff;
736 if (truncate_size <= object_base) {
737 truncate_size = 0;
738 } else {
739 truncate_size -= object_base;
740 if (truncate_size > object_size)
741 truncate_size = object_size;
Alex Eldera19dadf2013-03-13 20:50:01 -0500742 }
Alex Elderd18d1e22013-03-13 20:50:01 -0500743
Alex Elderc99d2d42013-04-05 01:27:11 -0500744 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
Alex Elderb0270322013-03-13 20:50:01 -0500745 truncate_size, truncate_seq);
Alex Elder8c042b02013-04-03 01:28:58 -0500746
Alex Elderacead002013-03-14 14:09:05 -0500747 /*
748 * A second op in the ops array means the caller wants to
749 * also issue a include a 'startsync' command so that the
750 * osd will flush data quickly.
751 */
752 if (num_ops > 1)
Alex Elderc99d2d42013-04-05 01:27:11 -0500753 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
Alex Elderd18d1e22013-03-13 20:50:01 -0500754
Sage Weilf24e9982009-10-06 11:31:10 -0700755 req->r_file_layout = *layout; /* keep a copy */
756
Alex Elder75d1c942013-03-13 20:50:00 -0500757 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
758 vino.ino, objnum);
Alex Elderdbe0fc42013-02-15 22:10:17 -0600759 req->r_oid_len = strlen(req->r_oid);
760
Sage Weilf24e9982009-10-06 11:31:10 -0700761 return req;
762}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700763EXPORT_SYMBOL(ceph_osdc_new_request);
Sage Weilf24e9982009-10-06 11:31:10 -0700764
765/*
766 * We keep osd requests in an rbtree, sorted by ->r_tid.
767 */
768static void __insert_request(struct ceph_osd_client *osdc,
769 struct ceph_osd_request *new)
770{
771 struct rb_node **p = &osdc->requests.rb_node;
772 struct rb_node *parent = NULL;
773 struct ceph_osd_request *req = NULL;
774
775 while (*p) {
776 parent = *p;
777 req = rb_entry(parent, struct ceph_osd_request, r_node);
778 if (new->r_tid < req->r_tid)
779 p = &(*p)->rb_left;
780 else if (new->r_tid > req->r_tid)
781 p = &(*p)->rb_right;
782 else
783 BUG();
784 }
785
786 rb_link_node(&new->r_node, parent, p);
787 rb_insert_color(&new->r_node, &osdc->requests);
788}
789
790static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
791 u64 tid)
792{
793 struct ceph_osd_request *req;
794 struct rb_node *n = osdc->requests.rb_node;
795
796 while (n) {
797 req = rb_entry(n, struct ceph_osd_request, r_node);
798 if (tid < req->r_tid)
799 n = n->rb_left;
800 else if (tid > req->r_tid)
801 n = n->rb_right;
802 else
803 return req;
804 }
805 return NULL;
806}
807
808static struct ceph_osd_request *
809__lookup_request_ge(struct ceph_osd_client *osdc,
810 u64 tid)
811{
812 struct ceph_osd_request *req;
813 struct rb_node *n = osdc->requests.rb_node;
814
815 while (n) {
816 req = rb_entry(n, struct ceph_osd_request, r_node);
817 if (tid < req->r_tid) {
818 if (!n->rb_left)
819 return req;
820 n = n->rb_left;
821 } else if (tid > req->r_tid) {
822 n = n->rb_right;
823 } else {
824 return req;
825 }
826 }
827 return NULL;
828}
829
Sage Weil6f6c7002011-01-17 20:34:08 -0800830/*
831 * Resubmit requests pending on the given osd.
832 */
833static void __kick_osd_requests(struct ceph_osd_client *osdc,
834 struct ceph_osd *osd)
835{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700836 struct ceph_osd_request *req, *nreq;
Alex Eldere02493c2013-03-25 18:16:11 -0500837 LIST_HEAD(resend);
Sage Weil6f6c7002011-01-17 20:34:08 -0800838 int err;
839
840 dout("__kick_osd_requests osd%d\n", osd->o_osd);
841 err = __reset_osd(osdc, osd);
Alex Elder685a7552012-12-07 09:57:58 -0600842 if (err)
Sage Weil6f6c7002011-01-17 20:34:08 -0800843 return;
Alex Eldere02493c2013-03-25 18:16:11 -0500844 /*
845 * Build up a list of requests to resend by traversing the
846 * osd's list of requests. Requests for a given object are
847 * sent in tid order, and that is also the order they're
848 * kept on this list. Therefore all requests that are in
849 * flight will be found first, followed by all requests that
850 * have not yet been sent. And to resend requests while
851 * preserving this order we will want to put any sent
852 * requests back on the front of the osd client's unsent
853 * list.
854 *
855 * So we build a separate ordered list of already-sent
856 * requests for the affected osd and splice it onto the
857 * front of the osd client's unsent list. Once we've seen a
858 * request that has not yet been sent we're done. Those
859 * requests are already sitting right where they belong.
860 */
Sage Weil6f6c7002011-01-17 20:34:08 -0800861 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
Alex Eldere02493c2013-03-25 18:16:11 -0500862 if (!req->r_sent)
863 break;
864 list_move_tail(&req->r_req_lru_item, &resend);
865 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
Sage Weil6f6c7002011-01-17 20:34:08 -0800866 osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700867 if (!req->r_linger)
868 req->r_flags |= CEPH_OSD_FLAG_RETRY;
869 }
Alex Eldere02493c2013-03-25 18:16:11 -0500870 list_splice(&resend, &osdc->req_unsent);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700871
Alex Eldere02493c2013-03-25 18:16:11 -0500872 /*
873 * Linger requests are re-registered before sending, which
874 * sets up a new tid for each. We add them to the unsent
875 * list at the end to keep things in tid order.
876 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700877 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
878 r_linger_osd) {
Sage Weil77f38e02011-04-06 09:09:16 -0700879 /*
880 * reregister request prior to unregistering linger so
881 * that r_osd is preserved.
882 */
883 BUG_ON(!list_empty(&req->r_req_lru_item));
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700884 __register_request(osdc, req);
Alex Eldere02493c2013-03-25 18:16:11 -0500885 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
Alex Elderad885922013-03-25 18:16:11 -0500886 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
Sage Weil77f38e02011-04-06 09:09:16 -0700887 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700888 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
889 osd->o_osd);
Sage Weil6f6c7002011-01-17 20:34:08 -0800890 }
891}
892
Sage Weilf24e9982009-10-06 11:31:10 -0700893/*
Sage Weil81b024e2009-10-09 10:29:18 -0700894 * If the osd connection drops, we need to resubmit all requests.
Sage Weilf24e9982009-10-06 11:31:10 -0700895 */
896static void osd_reset(struct ceph_connection *con)
897{
898 struct ceph_osd *osd = con->private;
899 struct ceph_osd_client *osdc;
900
901 if (!osd)
902 return;
903 dout("osd_reset osd%d\n", osd->o_osd);
904 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -0700905 down_read(&osdc->map_sem);
Sage Weil83aff952012-11-28 12:28:24 -0800906 mutex_lock(&osdc->request_mutex);
907 __kick_osd_requests(osdc, osd);
Alex Elderf9d25192013-02-15 11:42:29 -0600908 __send_queued(osdc);
Sage Weil83aff952012-11-28 12:28:24 -0800909 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -0700910 up_read(&osdc->map_sem);
911}
912
913/*
914 * Track open sessions with osds.
915 */
Alex Eldere10006f2012-05-26 23:26:43 -0500916static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
Sage Weilf24e9982009-10-06 11:31:10 -0700917{
918 struct ceph_osd *osd;
919
920 osd = kzalloc(sizeof(*osd), GFP_NOFS);
921 if (!osd)
922 return NULL;
923
924 atomic_set(&osd->o_ref, 1);
925 osd->o_osdc = osdc;
Alex Eldere10006f2012-05-26 23:26:43 -0500926 osd->o_osd = onum;
Alex Elderf4077312012-12-06 07:22:04 -0600927 RB_CLEAR_NODE(&osd->o_node);
Sage Weilf24e9982009-10-06 11:31:10 -0700928 INIT_LIST_HEAD(&osd->o_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700929 INIT_LIST_HEAD(&osd->o_linger_requests);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800930 INIT_LIST_HEAD(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700931 osd->o_incarnation = 1;
932
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700933 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800934
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800935 INIT_LIST_HEAD(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -0700936 return osd;
937}
938
939static struct ceph_osd *get_osd(struct ceph_osd *osd)
940{
941 if (atomic_inc_not_zero(&osd->o_ref)) {
942 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
943 atomic_read(&osd->o_ref));
944 return osd;
945 } else {
946 dout("get_osd %p FAIL\n", osd);
947 return NULL;
948 }
949}
950
951static void put_osd(struct ceph_osd *osd)
952{
953 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
954 atomic_read(&osd->o_ref) - 1);
Alex Eldera2556512012-05-16 15:16:39 -0500955 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
Sage Weil79494d12010-05-27 14:15:49 -0700956 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
957
Sage Weil27859f92013-03-25 10:26:14 -0700958 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
Sage Weilf24e9982009-10-06 11:31:10 -0700959 kfree(osd);
Sage Weil79494d12010-05-27 14:15:49 -0700960 }
Sage Weilf24e9982009-10-06 11:31:10 -0700961}
962
963/*
964 * remove an osd from our map
965 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800966static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -0700967{
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800968 dout("__remove_osd %p\n", osd);
Sage Weilf24e9982009-10-06 11:31:10 -0700969 BUG_ON(!list_empty(&osd->o_requests));
970 rb_erase(&osd->o_node, &osdc->osds);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800971 list_del_init(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700972 ceph_con_close(&osd->o_con);
973 put_osd(osd);
974}
975
Sage Weilaca420b2011-08-31 14:45:53 -0700976static void remove_all_osds(struct ceph_osd_client *osdc)
977{
Jiaju Zhang048a9d22012-07-20 08:18:36 -0500978 dout("%s %p\n", __func__, osdc);
Sage Weilaca420b2011-08-31 14:45:53 -0700979 mutex_lock(&osdc->request_mutex);
980 while (!RB_EMPTY_ROOT(&osdc->osds)) {
981 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
982 struct ceph_osd, o_node);
983 __remove_osd(osdc, osd);
984 }
985 mutex_unlock(&osdc->request_mutex);
986}
987
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800988static void __move_osd_to_lru(struct ceph_osd_client *osdc,
989 struct ceph_osd *osd)
990{
991 dout("__move_osd_to_lru %p\n", osd);
992 BUG_ON(!list_empty(&osd->o_osd_lru));
993 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700994 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800995}
996
997static void __remove_osd_from_lru(struct ceph_osd *osd)
998{
999 dout("__remove_osd_from_lru %p\n", osd);
1000 if (!list_empty(&osd->o_osd_lru))
1001 list_del_init(&osd->o_osd_lru);
1002}
1003
Sage Weilaca420b2011-08-31 14:45:53 -07001004static void remove_old_osds(struct ceph_osd_client *osdc)
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001005{
1006 struct ceph_osd *osd, *nosd;
1007
1008 dout("__remove_old_osds %p\n", osdc);
1009 mutex_lock(&osdc->request_mutex);
1010 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
Sage Weilaca420b2011-08-31 14:45:53 -07001011 if (time_before(jiffies, osd->lru_ttl))
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001012 break;
1013 __remove_osd(osdc, osd);
1014 }
1015 mutex_unlock(&osdc->request_mutex);
1016}
1017
Sage Weilf24e9982009-10-06 11:31:10 -07001018/*
1019 * reset osd connect
1020 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001021static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -07001022{
Alex Elderc3acb182012-12-07 09:57:58 -06001023 struct ceph_entity_addr *peer_addr;
Sage Weilf24e9982009-10-06 11:31:10 -07001024
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001025 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001026 if (list_empty(&osd->o_requests) &&
1027 list_empty(&osd->o_linger_requests)) {
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001028 __remove_osd(osdc, osd);
Alex Elderc3acb182012-12-07 09:57:58 -06001029
1030 return -ENODEV;
1031 }
1032
1033 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1034 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1035 !ceph_con_opened(&osd->o_con)) {
1036 struct ceph_osd_request *req;
1037
Sage Weil87b315a2010-03-22 14:51:18 -07001038 dout(" osd addr hasn't changed and connection never opened,"
1039 " letting msgr retry");
1040 /* touch each r_stamp for handle_timeout()'s benfit */
1041 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1042 req->r_stamp = jiffies;
Alex Elderc3acb182012-12-07 09:57:58 -06001043
1044 return -EAGAIN;
Sage Weilf24e9982009-10-06 11:31:10 -07001045 }
Alex Elderc3acb182012-12-07 09:57:58 -06001046
1047 ceph_con_close(&osd->o_con);
1048 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1049 osd->o_incarnation++;
1050
1051 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001052}
1053
1054static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1055{
1056 struct rb_node **p = &osdc->osds.rb_node;
1057 struct rb_node *parent = NULL;
1058 struct ceph_osd *osd = NULL;
1059
Sage Weilaca420b2011-08-31 14:45:53 -07001060 dout("__insert_osd %p osd%d\n", new, new->o_osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001061 while (*p) {
1062 parent = *p;
1063 osd = rb_entry(parent, struct ceph_osd, o_node);
1064 if (new->o_osd < osd->o_osd)
1065 p = &(*p)->rb_left;
1066 else if (new->o_osd > osd->o_osd)
1067 p = &(*p)->rb_right;
1068 else
1069 BUG();
1070 }
1071
1072 rb_link_node(&new->o_node, parent, p);
1073 rb_insert_color(&new->o_node, &osdc->osds);
1074}
1075
1076static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1077{
1078 struct ceph_osd *osd;
1079 struct rb_node *n = osdc->osds.rb_node;
1080
1081 while (n) {
1082 osd = rb_entry(n, struct ceph_osd, o_node);
1083 if (o < osd->o_osd)
1084 n = n->rb_left;
1085 else if (o > osd->o_osd)
1086 n = n->rb_right;
1087 else
1088 return osd;
1089 }
1090 return NULL;
1091}
1092
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001093static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1094{
1095 schedule_delayed_work(&osdc->timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001096 osdc->client->options->osd_keepalive_timeout * HZ);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001097}
1098
1099static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1100{
1101 cancel_delayed_work(&osdc->timeout_work);
1102}
Sage Weilf24e9982009-10-06 11:31:10 -07001103
1104/*
1105 * Register request, assign tid. If this is the first request, set up
1106 * the timeout event.
1107 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001108static void __register_request(struct ceph_osd_client *osdc,
1109 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001110{
Sage Weilf24e9982009-10-06 11:31:10 -07001111 req->r_tid = ++osdc->last_tid;
Sage Weil6df058c2009-12-22 11:24:33 -08001112 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
Sage Weil77f38e02011-04-06 09:09:16 -07001113 dout("__register_request %p tid %lld\n", req, req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07001114 __insert_request(osdc, req);
1115 ceph_osdc_get_request(req);
1116 osdc->num_requests++;
Sage Weilf24e9982009-10-06 11:31:10 -07001117 if (osdc->num_requests == 1) {
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001118 dout(" first request, scheduling timeout\n");
1119 __schedule_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001120 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001121}
1122
Sage Weilf24e9982009-10-06 11:31:10 -07001123/*
1124 * called under osdc->request_mutex
1125 */
1126static void __unregister_request(struct ceph_osd_client *osdc,
1127 struct ceph_osd_request *req)
1128{
Sage Weil35f9f8a2012-05-16 15:16:38 -05001129 if (RB_EMPTY_NODE(&req->r_node)) {
1130 dout("__unregister_request %p tid %lld not registered\n",
1131 req, req->r_tid);
1132 return;
1133 }
1134
Sage Weilf24e9982009-10-06 11:31:10 -07001135 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1136 rb_erase(&req->r_node, &osdc->requests);
1137 osdc->num_requests--;
1138
Sage Weil0ba64782009-10-08 16:57:16 -07001139 if (req->r_osd) {
1140 /* make sure the original request isn't in flight. */
Alex Elder6740a842012-06-01 14:56:43 -05001141 ceph_msg_revoke(req->r_request);
Sage Weil0ba64782009-10-08 16:57:16 -07001142
1143 list_del_init(&req->r_osd_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001144 if (list_empty(&req->r_osd->o_requests) &&
1145 list_empty(&req->r_osd->o_linger_requests)) {
1146 dout("moving osd to %p lru\n", req->r_osd);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001147 __move_osd_to_lru(osdc, req->r_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001148 }
Sage Weilfbdb9192011-03-29 12:11:06 -07001149 if (list_empty(&req->r_linger_item))
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001150 req->r_osd = NULL;
Sage Weil0ba64782009-10-08 16:57:16 -07001151 }
Sage Weilf24e9982009-10-06 11:31:10 -07001152
Alex Elder7d5f2482012-11-29 08:37:03 -06001153 list_del_init(&req->r_req_lru_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001154 ceph_osdc_put_request(req);
1155
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001156 if (osdc->num_requests == 0) {
1157 dout(" no requests, canceling timeout\n");
1158 __cancel_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001159 }
1160}
1161
1162/*
1163 * Cancel a previously queued request message
1164 */
1165static void __cancel_request(struct ceph_osd_request *req)
1166{
Sage Weil6bc18872010-09-27 10:18:52 -07001167 if (req->r_sent && req->r_osd) {
Alex Elder6740a842012-06-01 14:56:43 -05001168 ceph_msg_revoke(req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001169 req->r_sent = 0;
1170 }
1171}
1172
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001173static void __register_linger_request(struct ceph_osd_client *osdc,
1174 struct ceph_osd_request *req)
1175{
1176 dout("__register_linger_request %p\n", req);
1177 list_add_tail(&req->r_linger_item, &osdc->req_linger);
Sage Weil6194ea82012-07-30 16:19:28 -07001178 if (req->r_osd)
1179 list_add_tail(&req->r_linger_osd,
1180 &req->r_osd->o_linger_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001181}
1182
1183static void __unregister_linger_request(struct ceph_osd_client *osdc,
1184 struct ceph_osd_request *req)
1185{
1186 dout("__unregister_linger_request %p\n", req);
Alex Elder61c74032012-12-06 09:37:23 -06001187 list_del_init(&req->r_linger_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001188 if (req->r_osd) {
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001189 list_del_init(&req->r_linger_osd);
1190
1191 if (list_empty(&req->r_osd->o_requests) &&
1192 list_empty(&req->r_osd->o_linger_requests)) {
1193 dout("moving osd to %p lru\n", req->r_osd);
1194 __move_osd_to_lru(osdc, req->r_osd);
1195 }
Sage Weilfbdb9192011-03-29 12:11:06 -07001196 if (list_empty(&req->r_osd_item))
1197 req->r_osd = NULL;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001198 }
1199}
1200
1201void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1202 struct ceph_osd_request *req)
1203{
1204 mutex_lock(&osdc->request_mutex);
1205 if (req->r_linger) {
1206 __unregister_linger_request(osdc, req);
1207 ceph_osdc_put_request(req);
1208 }
1209 mutex_unlock(&osdc->request_mutex);
1210}
1211EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1212
1213void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1214 struct ceph_osd_request *req)
1215{
1216 if (!req->r_linger) {
1217 dout("set_request_linger %p\n", req);
1218 req->r_linger = 1;
1219 /*
1220 * caller is now responsible for calling
1221 * unregister_linger_request
1222 */
1223 ceph_osdc_get_request(req);
1224 }
1225}
1226EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1227
Sage Weilf24e9982009-10-06 11:31:10 -07001228/*
1229 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1230 * (as needed), and set the request r_osd appropriately. If there is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001231 * no up osd, set r_osd to NULL. Move the request to the appropriate list
Sage Weil6f6c7002011-01-17 20:34:08 -08001232 * (unsent, homeless) or leave on in-flight lru.
Sage Weilf24e9982009-10-06 11:31:10 -07001233 *
1234 * Return 0 if unchanged, 1 if changed, or negative on error.
1235 *
1236 * Caller should hold map_sem for read and request_mutex.
1237 */
Sage Weil6f6c7002011-01-17 20:34:08 -08001238static int __map_request(struct ceph_osd_client *osdc,
Sage Weil38d64532011-10-14 13:33:55 -07001239 struct ceph_osd_request *req, int force_resend)
Sage Weilf24e9982009-10-06 11:31:10 -07001240{
Sage Weil5b191d92013-02-23 10:38:16 -08001241 struct ceph_pg pgid;
Sage Weild85b7052010-05-10 10:24:48 -07001242 int acting[CEPH_PG_MAX_SIZE];
1243 int o = -1, num = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001244 int err;
Sage Weilf24e9982009-10-06 11:31:10 -07001245
Sage Weil6f6c7002011-01-17 20:34:08 -08001246 dout("map_request %p tid %lld\n", req, req->r_tid);
Alex Elder41766f82013-03-01 18:00:15 -06001247 err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1248 ceph_file_layout_pg_pool(req->r_file_layout));
Sage Weil6f6c7002011-01-17 20:34:08 -08001249 if (err) {
1250 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilf24e9982009-10-06 11:31:10 -07001251 return err;
Sage Weil6f6c7002011-01-17 20:34:08 -08001252 }
Sage Weil7740a422010-01-08 15:58:25 -08001253 req->r_pgid = pgid;
1254
Sage Weild85b7052010-05-10 10:24:48 -07001255 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1256 if (err > 0) {
1257 o = acting[0];
1258 num = err;
1259 }
Sage Weilf24e9982009-10-06 11:31:10 -07001260
Sage Weil38d64532011-10-14 13:33:55 -07001261 if ((!force_resend &&
1262 req->r_osd && req->r_osd->o_osd == o &&
Sage Weild85b7052010-05-10 10:24:48 -07001263 req->r_sent >= req->r_osd->o_incarnation &&
1264 req->r_num_pg_osds == num &&
1265 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001266 (req->r_osd == NULL && o == -1))
1267 return 0; /* no change */
1268
Sage Weil5b191d92013-02-23 10:38:16 -08001269 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1270 req->r_tid, pgid.pool, pgid.seed, o,
Sage Weilf24e9982009-10-06 11:31:10 -07001271 req->r_osd ? req->r_osd->o_osd : -1);
1272
Sage Weild85b7052010-05-10 10:24:48 -07001273 /* record full pg acting set */
1274 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1275 req->r_num_pg_osds = num;
1276
Sage Weilf24e9982009-10-06 11:31:10 -07001277 if (req->r_osd) {
1278 __cancel_request(req);
1279 list_del_init(&req->r_osd_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001280 req->r_osd = NULL;
1281 }
1282
1283 req->r_osd = __lookup_osd(osdc, o);
1284 if (!req->r_osd && o >= 0) {
Sage Weilc99eb1c2010-02-26 09:37:33 -08001285 err = -ENOMEM;
Alex Eldere10006f2012-05-26 23:26:43 -05001286 req->r_osd = create_osd(osdc, o);
Sage Weil6f6c7002011-01-17 20:34:08 -08001287 if (!req->r_osd) {
1288 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilc99eb1c2010-02-26 09:37:33 -08001289 goto out;
Sage Weil6f6c7002011-01-17 20:34:08 -08001290 }
Sage Weilf24e9982009-10-06 11:31:10 -07001291
Sage Weil6f6c7002011-01-17 20:34:08 -08001292 dout("map_request osd %p is osd%d\n", req->r_osd, o);
Sage Weilf24e9982009-10-06 11:31:10 -07001293 __insert_osd(osdc, req->r_osd);
1294
Sage Weilb7a9e5d2012-06-27 12:24:08 -07001295 ceph_con_open(&req->r_osd->o_con,
1296 CEPH_ENTITY_TYPE_OSD, o,
1297 &osdc->osdmap->osd_addr[o]);
Sage Weilf24e9982009-10-06 11:31:10 -07001298 }
1299
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001300 if (req->r_osd) {
1301 __remove_osd_from_lru(req->r_osd);
Alex Elderad885922013-03-25 18:16:11 -05001302 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1303 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
Sage Weil6f6c7002011-01-17 20:34:08 -08001304 } else {
Alex Elderad885922013-03-25 18:16:11 -05001305 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001306 }
Sage Weild85b7052010-05-10 10:24:48 -07001307 err = 1; /* osd or pg changed */
Sage Weilf24e9982009-10-06 11:31:10 -07001308
1309out:
Sage Weilf24e9982009-10-06 11:31:10 -07001310 return err;
1311}
1312
1313/*
1314 * caller should hold map_sem (for read) and request_mutex
1315 */
Sage Weil56e925b2012-01-03 12:34:34 -08001316static void __send_request(struct ceph_osd_client *osdc,
1317 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001318{
Sage Weil1b83bef2013-02-25 16:11:12 -08001319 void *p;
Sage Weilf24e9982009-10-06 11:31:10 -07001320
Sage Weil1b83bef2013-02-25 16:11:12 -08001321 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1322 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1323 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
Sage Weilf24e9982009-10-06 11:31:10 -07001324
Sage Weil1b83bef2013-02-25 16:11:12 -08001325 /* fill in message content that changes each time we send it */
1326 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1327 put_unaligned_le32(req->r_flags, req->r_request_flags);
1328 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1329 p = req->r_request_pgid;
1330 ceph_encode_64(&p, req->r_pgid.pool);
1331 ceph_encode_32(&p, req->r_pgid.seed);
1332 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1333 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1334 sizeof(req->r_reassert_version));
Sage Weil2169aea2013-02-25 16:13:08 -08001335
Sage Weil3dd72fc2010-03-22 14:42:30 -07001336 req->r_stamp = jiffies;
Henry C Chang07a27e22010-08-22 21:34:27 -07001337 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001338
1339 ceph_msg_get(req->r_request); /* send consumes a ref */
Alex Elder26be8802013-04-15 11:20:42 -05001340
1341 /* Mark the request unsafe if this is the first timet's being sent. */
1342
1343 if (!req->r_sent && req->r_unsafe_callback)
1344 req->r_unsafe_callback(req, true);
Sage Weilf24e9982009-10-06 11:31:10 -07001345 req->r_sent = req->r_osd->o_incarnation;
Alex Elder26be8802013-04-15 11:20:42 -05001346
1347 ceph_con_send(&req->r_osd->o_con, req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001348}
1349
1350/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001351 * Send any requests in the queue (req_unsent).
1352 */
Alex Elderf9d25192013-02-15 11:42:29 -06001353static void __send_queued(struct ceph_osd_client *osdc)
Sage Weil6f6c7002011-01-17 20:34:08 -08001354{
1355 struct ceph_osd_request *req, *tmp;
1356
Alex Elderf9d25192013-02-15 11:42:29 -06001357 dout("__send_queued\n");
1358 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
Sage Weil6f6c7002011-01-17 20:34:08 -08001359 __send_request(osdc, req);
Sage Weil6f6c7002011-01-17 20:34:08 -08001360}
1361
1362/*
Sage Weilf24e9982009-10-06 11:31:10 -07001363 * Timeout callback, called every N seconds when 1 or more osd
1364 * requests has been active for more than N seconds. When this
1365 * happens, we ping all OSDs with requests who have timed out to
1366 * ensure any communications channel reset is detected. Reset the
1367 * request timeouts another N seconds in the future as we go.
1368 * Reschedule the timeout event another N seconds in future (unless
1369 * there are no open requests).
1370 */
1371static void handle_timeout(struct work_struct *work)
1372{
1373 struct ceph_osd_client *osdc =
1374 container_of(work, struct ceph_osd_client, timeout_work.work);
Sage Weil83aff952012-11-28 12:28:24 -08001375 struct ceph_osd_request *req;
Sage Weilf24e9982009-10-06 11:31:10 -07001376 struct ceph_osd *osd;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001377 unsigned long keepalive =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001378 osdc->client->options->osd_keepalive_timeout * HZ;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001379 struct list_head slow_osds;
Sage Weilf24e9982009-10-06 11:31:10 -07001380 dout("timeout\n");
1381 down_read(&osdc->map_sem);
1382
1383 ceph_monc_request_next_osdmap(&osdc->client->monc);
1384
1385 mutex_lock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001386
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001387 /*
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001388 * ping osds that are a bit slow. this ensures that if there
1389 * is a break in the TCP connection we will notice, and reopen
1390 * a connection with that osd (from the fault callback).
1391 */
1392 INIT_LIST_HEAD(&slow_osds);
1393 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
Sage Weil3dd72fc2010-03-22 14:42:30 -07001394 if (time_before(jiffies, req->r_stamp + keepalive))
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001395 break;
1396
1397 osd = req->r_osd;
1398 BUG_ON(!osd);
1399 dout(" tid %llu is slow, will send keepalive on osd%d\n",
Sage Weilf24e9982009-10-06 11:31:10 -07001400 req->r_tid, osd->o_osd);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001401 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1402 }
1403 while (!list_empty(&slow_osds)) {
1404 osd = list_entry(slow_osds.next, struct ceph_osd,
1405 o_keepalive_item);
1406 list_del_init(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001407 ceph_con_keepalive(&osd->o_con);
1408 }
1409
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001410 __schedule_osd_timeout(osdc);
Alex Elderf9d25192013-02-15 11:42:29 -06001411 __send_queued(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001412 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001413 up_read(&osdc->map_sem);
1414}
1415
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001416static void handle_osds_timeout(struct work_struct *work)
1417{
1418 struct ceph_osd_client *osdc =
1419 container_of(work, struct ceph_osd_client,
1420 osds_timeout_work.work);
1421 unsigned long delay =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001422 osdc->client->options->osd_idle_ttl * HZ >> 2;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001423
1424 dout("osds timeout\n");
1425 down_read(&osdc->map_sem);
Sage Weilaca420b2011-08-31 14:45:53 -07001426 remove_old_osds(osdc);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001427 up_read(&osdc->map_sem);
1428
1429 schedule_delayed_work(&osdc->osds_timeout_work,
1430 round_jiffies_relative(delay));
1431}
1432
Sage Weil25845472011-06-03 09:37:09 -07001433static void complete_request(struct ceph_osd_request *req)
1434{
Alex Elder26be8802013-04-15 11:20:42 -05001435 if (req->r_unsafe_callback)
1436 req->r_unsafe_callback(req, false);
Sage Weil25845472011-06-03 09:37:09 -07001437 complete_all(&req->r_safe_completion); /* fsync waiter */
1438}
1439
Sage Weilf24e9982009-10-06 11:31:10 -07001440/*
1441 * handle osd op reply. either call the callback if it is specified,
1442 * or do the completion to wake up the waiting thread.
1443 */
Sage Weil350b1c32009-12-22 10:45:45 -08001444static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1445 struct ceph_connection *con)
Sage Weilf24e9982009-10-06 11:31:10 -07001446{
Sage Weil1b83bef2013-02-25 16:11:12 -08001447 void *p, *end;
Sage Weilf24e9982009-10-06 11:31:10 -07001448 struct ceph_osd_request *req;
1449 u64 tid;
Sage Weil1b83bef2013-02-25 16:11:12 -08001450 int object_len;
Alex Elder79528732013-04-03 21:32:51 -05001451 unsigned int numops;
1452 int payload_len, flags;
Sage Weil0ceed5d2010-05-11 09:53:18 -07001453 s32 result;
Sage Weil1b83bef2013-02-25 16:11:12 -08001454 s32 retry_attempt;
1455 struct ceph_pg pg;
1456 int err;
1457 u32 reassert_epoch;
1458 u64 reassert_version;
1459 u32 osdmap_epoch;
Alex Elder0d5af162013-02-27 10:26:25 -06001460 int already_completed;
Alex Elder9fc6e062013-04-03 01:28:57 -05001461 u32 bytes;
Alex Elder79528732013-04-03 21:32:51 -05001462 unsigned int i;
Sage Weilf24e9982009-10-06 11:31:10 -07001463
Sage Weil6df058c2009-12-22 11:24:33 -08001464 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil1b83bef2013-02-25 16:11:12 -08001465 dout("handle_reply %p tid %llu\n", msg, tid);
1466
1467 p = msg->front.iov_base;
1468 end = p + msg->front.iov_len;
1469
1470 ceph_decode_need(&p, end, 4, bad);
1471 object_len = ceph_decode_32(&p);
1472 ceph_decode_need(&p, end, object_len, bad);
1473 p += object_len;
1474
Alex Elderef4859d2013-04-01 18:58:26 -05001475 err = ceph_decode_pgid(&p, end, &pg);
Sage Weil1b83bef2013-02-25 16:11:12 -08001476 if (err)
Sage Weilf24e9982009-10-06 11:31:10 -07001477 goto bad;
Sage Weil1b83bef2013-02-25 16:11:12 -08001478
1479 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1480 flags = ceph_decode_64(&p);
1481 result = ceph_decode_32(&p);
1482 reassert_epoch = ceph_decode_32(&p);
1483 reassert_version = ceph_decode_64(&p);
1484 osdmap_epoch = ceph_decode_32(&p);
1485
Sage Weilf24e9982009-10-06 11:31:10 -07001486 /* lookup */
1487 mutex_lock(&osdc->request_mutex);
1488 req = __lookup_request(osdc, tid);
1489 if (req == NULL) {
1490 dout("handle_reply tid %llu dne\n", tid);
Alex Elder8058fd42013-04-01 18:58:26 -05001491 goto bad_mutex;
Sage Weilf24e9982009-10-06 11:31:10 -07001492 }
1493 ceph_osdc_get_request(req);
Sage Weil1b83bef2013-02-25 16:11:12 -08001494
1495 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1496 req, result);
1497
1498 ceph_decode_need(&p, end, 4, bad);
1499 numops = ceph_decode_32(&p);
1500 if (numops > CEPH_OSD_MAX_OP)
1501 goto bad_put;
1502 if (numops != req->r_num_ops)
1503 goto bad_put;
1504 payload_len = 0;
1505 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1506 for (i = 0; i < numops; i++) {
1507 struct ceph_osd_op *op = p;
1508 int len;
1509
1510 len = le32_to_cpu(op->payload_len);
1511 req->r_reply_op_len[i] = len;
1512 dout(" op %d has %d bytes\n", i, len);
1513 payload_len += len;
1514 p += sizeof(*op);
1515 }
Alex Elder9fc6e062013-04-03 01:28:57 -05001516 bytes = le32_to_cpu(msg->hdr.data_len);
1517 if (payload_len != bytes) {
Sage Weil1b83bef2013-02-25 16:11:12 -08001518 pr_warning("sum of op payload lens %d != data_len %d",
Alex Elder9fc6e062013-04-03 01:28:57 -05001519 payload_len, bytes);
Sage Weil1b83bef2013-02-25 16:11:12 -08001520 goto bad_put;
1521 }
1522
1523 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1524 retry_attempt = ceph_decode_32(&p);
1525 for (i = 0; i < numops; i++)
1526 req->r_reply_op_result[i] = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001527
Sage Weilf24e9982009-10-06 11:31:10 -07001528 if (!req->r_got_reply) {
Sage Weilf24e9982009-10-06 11:31:10 -07001529
Sage Weil1b83bef2013-02-25 16:11:12 -08001530 req->r_result = result;
Sage Weilf24e9982009-10-06 11:31:10 -07001531 dout("handle_reply result %d bytes %d\n", req->r_result,
1532 bytes);
1533 if (req->r_result == 0)
1534 req->r_result = bytes;
1535
1536 /* in case this is a write and we need to replay, */
Sage Weil1b83bef2013-02-25 16:11:12 -08001537 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1538 req->r_reassert_version.version = cpu_to_le64(reassert_version);
Sage Weilf24e9982009-10-06 11:31:10 -07001539
1540 req->r_got_reply = 1;
1541 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1542 dout("handle_reply tid %llu dup ack\n", tid);
Sage Weil34b43a52009-12-01 12:23:54 -08001543 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001544 goto done;
1545 }
1546
1547 dout("handle_reply tid %llu flags %d\n", tid, flags);
1548
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001549 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1550 __register_linger_request(osdc, req);
1551
Sage Weilf24e9982009-10-06 11:31:10 -07001552 /* either this is a read, or we got the safe response */
Sage Weil0ceed5d2010-05-11 09:53:18 -07001553 if (result < 0 ||
1554 (flags & CEPH_OSD_FLAG_ONDISK) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001555 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1556 __unregister_request(osdc, req);
1557
Alex Elder0d5af162013-02-27 10:26:25 -06001558 already_completed = req->r_completed;
1559 req->r_completed = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07001560 mutex_unlock(&osdc->request_mutex);
Alex Elder0d5af162013-02-27 10:26:25 -06001561 if (already_completed)
1562 goto done;
Sage Weilf24e9982009-10-06 11:31:10 -07001563
1564 if (req->r_callback)
1565 req->r_callback(req, msg);
1566 else
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001567 complete_all(&req->r_completion);
Sage Weilf24e9982009-10-06 11:31:10 -07001568
Sage Weil25845472011-06-03 09:37:09 -07001569 if (flags & CEPH_OSD_FLAG_ONDISK)
1570 complete_request(req);
Sage Weilf24e9982009-10-06 11:31:10 -07001571
1572done:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001573 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07001574 ceph_osdc_put_request(req);
1575 return;
1576
Sage Weil1b83bef2013-02-25 16:11:12 -08001577bad_put:
1578 ceph_osdc_put_request(req);
Alex Elder8058fd42013-04-01 18:58:26 -05001579bad_mutex:
1580 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001581bad:
Sage Weil1b83bef2013-02-25 16:11:12 -08001582 pr_err("corrupt osd_op_reply got %d %d\n",
1583 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
Sage Weil9ec7cab2009-12-14 15:13:47 -08001584 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001585}
1586
Sage Weil6f6c7002011-01-17 20:34:08 -08001587static void reset_changed_osds(struct ceph_osd_client *osdc)
Sage Weilf24e9982009-10-06 11:31:10 -07001588{
Sage Weilf24e9982009-10-06 11:31:10 -07001589 struct rb_node *p, *n;
Sage Weilf24e9982009-10-06 11:31:10 -07001590
Sage Weil6f6c7002011-01-17 20:34:08 -08001591 for (p = rb_first(&osdc->osds); p; p = n) {
1592 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
Sage Weilf24e9982009-10-06 11:31:10 -07001593
Sage Weil6f6c7002011-01-17 20:34:08 -08001594 n = rb_next(p);
1595 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1596 memcmp(&osd->o_con.peer_addr,
1597 ceph_osd_addr(osdc->osdmap,
1598 osd->o_osd),
1599 sizeof(struct ceph_entity_addr)) != 0)
1600 __reset_osd(osdc, osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001601 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001602}
1603
1604/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001605 * Requeue requests whose mapping to an OSD has changed. If requests map to
1606 * no osd, request a new map.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001607 *
Alex Eldere6d50f62012-12-26 14:31:40 -06001608 * Caller should hold map_sem for read.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001609 */
Sage Weil38d64532011-10-14 13:33:55 -07001610static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001611{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001612 struct ceph_osd_request *req, *nreq;
Sage Weil6f6c7002011-01-17 20:34:08 -08001613 struct rb_node *p;
1614 int needmap = 0;
1615 int err;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001616
Sage Weil38d64532011-10-14 13:33:55 -07001617 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001618 mutex_lock(&osdc->request_mutex);
Sage Weil6194ea82012-07-30 16:19:28 -07001619 for (p = rb_first(&osdc->requests); p; ) {
Sage Weil6f6c7002011-01-17 20:34:08 -08001620 req = rb_entry(p, struct ceph_osd_request, r_node);
Sage Weil6194ea82012-07-30 16:19:28 -07001621 p = rb_next(p);
Alex Elderab60b162012-12-19 15:52:36 -06001622
1623 /*
1624 * For linger requests that have not yet been
1625 * registered, move them to the linger list; they'll
1626 * be sent to the osd in the loop below. Unregister
1627 * the request before re-registering it as a linger
1628 * request to ensure the __map_request() below
1629 * will decide it needs to be sent.
1630 */
1631 if (req->r_linger && list_empty(&req->r_linger_item)) {
1632 dout("%p tid %llu restart on osd%d\n",
1633 req, req->r_tid,
1634 req->r_osd ? req->r_osd->o_osd : -1);
1635 __unregister_request(osdc, req);
1636 __register_linger_request(osdc, req);
1637 continue;
1638 }
1639
Sage Weil38d64532011-10-14 13:33:55 -07001640 err = __map_request(osdc, req, force_resend);
Sage Weil6f6c7002011-01-17 20:34:08 -08001641 if (err < 0)
1642 continue; /* error */
1643 if (req->r_osd == NULL) {
1644 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1645 needmap++; /* request a newer map */
1646 } else if (err > 0) {
Sage Weil6194ea82012-07-30 16:19:28 -07001647 if (!req->r_linger) {
1648 dout("%p tid %llu requeued on osd%d\n", req,
1649 req->r_tid,
1650 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001651 req->r_flags |= CEPH_OSD_FLAG_RETRY;
Sage Weil6194ea82012-07-30 16:19:28 -07001652 }
1653 }
Sage Weil6f6c7002011-01-17 20:34:08 -08001654 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001655
1656 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1657 r_linger_item) {
1658 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1659
Sage Weil38d64532011-10-14 13:33:55 -07001660 err = __map_request(osdc, req, force_resend);
Alex Elderab60b162012-12-19 15:52:36 -06001661 dout("__map_request returned %d\n", err);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001662 if (err == 0)
1663 continue; /* no change and no osd was specified */
1664 if (err < 0)
1665 continue; /* hrm! */
1666 if (req->r_osd == NULL) {
1667 dout("tid %llu maps to no valid osd\n", req->r_tid);
1668 needmap++; /* request a newer map */
1669 continue;
1670 }
1671
1672 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1673 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001674 __register_request(osdc, req);
Alex Elderc89ce052012-12-06 07:22:04 -06001675 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001676 }
Sage Weilf24e9982009-10-06 11:31:10 -07001677 mutex_unlock(&osdc->request_mutex);
1678
1679 if (needmap) {
1680 dout("%d requests for down osds, need new map\n", needmap);
1681 ceph_monc_request_next_osdmap(&osdc->client->monc);
1682 }
Alex Eldere6d50f62012-12-26 14:31:40 -06001683 reset_changed_osds(osdc);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001684}
Sage Weil6f6c7002011-01-17 20:34:08 -08001685
1686
Sage Weilf24e9982009-10-06 11:31:10 -07001687/*
1688 * Process updated osd map.
1689 *
1690 * The message contains any number of incremental and full maps, normally
1691 * indicating some sort of topology change in the cluster. Kick requests
1692 * off to different OSDs as needed.
1693 */
1694void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1695{
1696 void *p, *end, *next;
1697 u32 nr_maps, maplen;
1698 u32 epoch;
1699 struct ceph_osdmap *newmap = NULL, *oldmap;
1700 int err;
1701 struct ceph_fsid fsid;
1702
1703 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1704 p = msg->front.iov_base;
1705 end = p + msg->front.iov_len;
1706
1707 /* verify fsid */
1708 ceph_decode_need(&p, end, sizeof(fsid), bad);
1709 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Sage Weil07433042009-11-18 16:50:41 -08001710 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1711 return;
Sage Weilf24e9982009-10-06 11:31:10 -07001712
1713 down_write(&osdc->map_sem);
1714
1715 /* incremental maps */
1716 ceph_decode_32_safe(&p, end, nr_maps, bad);
1717 dout(" %d inc maps\n", nr_maps);
1718 while (nr_maps > 0) {
1719 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07001720 epoch = ceph_decode_32(&p);
1721 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001722 ceph_decode_need(&p, end, maplen, bad);
1723 next = p + maplen;
1724 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1725 dout("applying incremental map %u len %d\n",
1726 epoch, maplen);
1727 newmap = osdmap_apply_incremental(&p, next,
1728 osdc->osdmap,
Alex Elder15d98822012-05-26 23:26:43 -05001729 &osdc->client->msgr);
Sage Weilf24e9982009-10-06 11:31:10 -07001730 if (IS_ERR(newmap)) {
1731 err = PTR_ERR(newmap);
1732 goto bad;
1733 }
Sage Weil30dc6382009-12-21 14:49:37 -08001734 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07001735 if (newmap != osdc->osdmap) {
1736 ceph_osdmap_destroy(osdc->osdmap);
1737 osdc->osdmap = newmap;
1738 }
Sage Weil38d64532011-10-14 13:33:55 -07001739 kick_requests(osdc, 0);
Sage Weilf24e9982009-10-06 11:31:10 -07001740 } else {
1741 dout("ignoring incremental map %u len %d\n",
1742 epoch, maplen);
1743 }
1744 p = next;
1745 nr_maps--;
1746 }
1747 if (newmap)
1748 goto done;
1749
1750 /* full maps */
1751 ceph_decode_32_safe(&p, end, nr_maps, bad);
1752 dout(" %d full maps\n", nr_maps);
1753 while (nr_maps) {
1754 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07001755 epoch = ceph_decode_32(&p);
1756 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001757 ceph_decode_need(&p, end, maplen, bad);
1758 if (nr_maps > 1) {
1759 dout("skipping non-latest full map %u len %d\n",
1760 epoch, maplen);
1761 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1762 dout("skipping full map %u len %d, "
1763 "older than our %u\n", epoch, maplen,
1764 osdc->osdmap->epoch);
1765 } else {
Sage Weil38d64532011-10-14 13:33:55 -07001766 int skipped_map = 0;
1767
Sage Weilf24e9982009-10-06 11:31:10 -07001768 dout("taking full map %u len %d\n", epoch, maplen);
1769 newmap = osdmap_decode(&p, p+maplen);
1770 if (IS_ERR(newmap)) {
1771 err = PTR_ERR(newmap);
1772 goto bad;
1773 }
Sage Weil30dc6382009-12-21 14:49:37 -08001774 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07001775 oldmap = osdc->osdmap;
1776 osdc->osdmap = newmap;
Sage Weil38d64532011-10-14 13:33:55 -07001777 if (oldmap) {
1778 if (oldmap->epoch + 1 < newmap->epoch)
1779 skipped_map = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07001780 ceph_osdmap_destroy(oldmap);
Sage Weil38d64532011-10-14 13:33:55 -07001781 }
1782 kick_requests(osdc, skipped_map);
Sage Weilf24e9982009-10-06 11:31:10 -07001783 }
1784 p += maplen;
1785 nr_maps--;
1786 }
1787
1788done:
1789 downgrade_write(&osdc->map_sem);
1790 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
Sage Weilcd634fb2011-05-12 09:29:18 -07001791
1792 /*
1793 * subscribe to subsequent osdmap updates if full to ensure
1794 * we find out when we are no longer full and stop returning
1795 * ENOSPC.
1796 */
1797 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1798 ceph_monc_request_next_osdmap(&osdc->client->monc);
1799
Alex Elderf9d25192013-02-15 11:42:29 -06001800 mutex_lock(&osdc->request_mutex);
1801 __send_queued(osdc);
1802 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001803 up_read(&osdc->map_sem);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001804 wake_up_all(&osdc->client->auth_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07001805 return;
1806
1807bad:
1808 pr_err("osdc handle_map corrupt msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08001809 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001810 up_write(&osdc->map_sem);
1811 return;
1812}
1813
Sage Weilf24e9982009-10-06 11:31:10 -07001814/*
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001815 * watch/notify callback event infrastructure
1816 *
1817 * These callbacks are used both for watch and notify operations.
1818 */
1819static void __release_event(struct kref *kref)
1820{
1821 struct ceph_osd_event *event =
1822 container_of(kref, struct ceph_osd_event, kref);
1823
1824 dout("__release_event %p\n", event);
1825 kfree(event);
1826}
1827
1828static void get_event(struct ceph_osd_event *event)
1829{
1830 kref_get(&event->kref);
1831}
1832
1833void ceph_osdc_put_event(struct ceph_osd_event *event)
1834{
1835 kref_put(&event->kref, __release_event);
1836}
1837EXPORT_SYMBOL(ceph_osdc_put_event);
1838
1839static void __insert_event(struct ceph_osd_client *osdc,
1840 struct ceph_osd_event *new)
1841{
1842 struct rb_node **p = &osdc->event_tree.rb_node;
1843 struct rb_node *parent = NULL;
1844 struct ceph_osd_event *event = NULL;
1845
1846 while (*p) {
1847 parent = *p;
1848 event = rb_entry(parent, struct ceph_osd_event, node);
1849 if (new->cookie < event->cookie)
1850 p = &(*p)->rb_left;
1851 else if (new->cookie > event->cookie)
1852 p = &(*p)->rb_right;
1853 else
1854 BUG();
1855 }
1856
1857 rb_link_node(&new->node, parent, p);
1858 rb_insert_color(&new->node, &osdc->event_tree);
1859}
1860
1861static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1862 u64 cookie)
1863{
1864 struct rb_node **p = &osdc->event_tree.rb_node;
1865 struct rb_node *parent = NULL;
1866 struct ceph_osd_event *event = NULL;
1867
1868 while (*p) {
1869 parent = *p;
1870 event = rb_entry(parent, struct ceph_osd_event, node);
1871 if (cookie < event->cookie)
1872 p = &(*p)->rb_left;
1873 else if (cookie > event->cookie)
1874 p = &(*p)->rb_right;
1875 else
1876 return event;
1877 }
1878 return NULL;
1879}
1880
1881static void __remove_event(struct ceph_osd_event *event)
1882{
1883 struct ceph_osd_client *osdc = event->osdc;
1884
1885 if (!RB_EMPTY_NODE(&event->node)) {
1886 dout("__remove_event removed %p\n", event);
1887 rb_erase(&event->node, &osdc->event_tree);
1888 ceph_osdc_put_event(event);
1889 } else {
1890 dout("__remove_event didn't remove %p\n", event);
1891 }
1892}
1893
1894int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1895 void (*event_cb)(u64, u64, u8, void *),
Alex Elder3c663bb2013-02-15 11:42:30 -06001896 void *data, struct ceph_osd_event **pevent)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001897{
1898 struct ceph_osd_event *event;
1899
1900 event = kmalloc(sizeof(*event), GFP_NOIO);
1901 if (!event)
1902 return -ENOMEM;
1903
1904 dout("create_event %p\n", event);
1905 event->cb = event_cb;
Alex Elder3c663bb2013-02-15 11:42:30 -06001906 event->one_shot = 0;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001907 event->data = data;
1908 event->osdc = osdc;
1909 INIT_LIST_HEAD(&event->osd_node);
Alex Elder3ee52342012-12-17 12:23:48 -06001910 RB_CLEAR_NODE(&event->node);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001911 kref_init(&event->kref); /* one ref for us */
1912 kref_get(&event->kref); /* one ref for the caller */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001913
1914 spin_lock(&osdc->event_lock);
1915 event->cookie = ++osdc->event_count;
1916 __insert_event(osdc, event);
1917 spin_unlock(&osdc->event_lock);
1918
1919 *pevent = event;
1920 return 0;
1921}
1922EXPORT_SYMBOL(ceph_osdc_create_event);
1923
1924void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1925{
1926 struct ceph_osd_client *osdc = event->osdc;
1927
1928 dout("cancel_event %p\n", event);
1929 spin_lock(&osdc->event_lock);
1930 __remove_event(event);
1931 spin_unlock(&osdc->event_lock);
1932 ceph_osdc_put_event(event); /* caller's */
1933}
1934EXPORT_SYMBOL(ceph_osdc_cancel_event);
1935
1936
1937static void do_event_work(struct work_struct *work)
1938{
1939 struct ceph_osd_event_work *event_work =
1940 container_of(work, struct ceph_osd_event_work, work);
1941 struct ceph_osd_event *event = event_work->event;
1942 u64 ver = event_work->ver;
1943 u64 notify_id = event_work->notify_id;
1944 u8 opcode = event_work->opcode;
1945
1946 dout("do_event_work completing %p\n", event);
1947 event->cb(ver, notify_id, opcode, event->data);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001948 dout("do_event_work completed %p\n", event);
1949 ceph_osdc_put_event(event);
1950 kfree(event_work);
1951}
1952
1953
1954/*
1955 * Process osd watch notifications
1956 */
Alex Elder3c663bb2013-02-15 11:42:30 -06001957static void handle_watch_notify(struct ceph_osd_client *osdc,
1958 struct ceph_msg *msg)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001959{
1960 void *p, *end;
1961 u8 proto_ver;
1962 u64 cookie, ver, notify_id;
1963 u8 opcode;
1964 struct ceph_osd_event *event;
1965 struct ceph_osd_event_work *event_work;
1966
1967 p = msg->front.iov_base;
1968 end = p + msg->front.iov_len;
1969
1970 ceph_decode_8_safe(&p, end, proto_ver, bad);
1971 ceph_decode_8_safe(&p, end, opcode, bad);
1972 ceph_decode_64_safe(&p, end, cookie, bad);
1973 ceph_decode_64_safe(&p, end, ver, bad);
1974 ceph_decode_64_safe(&p, end, notify_id, bad);
1975
1976 spin_lock(&osdc->event_lock);
1977 event = __find_event(osdc, cookie);
1978 if (event) {
Alex Elder3c663bb2013-02-15 11:42:30 -06001979 BUG_ON(event->one_shot);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001980 get_event(event);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001981 }
1982 spin_unlock(&osdc->event_lock);
1983 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1984 cookie, ver, event);
1985 if (event) {
1986 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001987 if (!event_work) {
1988 dout("ERROR: could not allocate event_work\n");
1989 goto done_err;
1990 }
Mariusz Kozlowski6b0ae402011-03-26 19:29:34 +01001991 INIT_WORK(&event_work->work, do_event_work);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001992 event_work->event = event;
1993 event_work->ver = ver;
1994 event_work->notify_id = notify_id;
1995 event_work->opcode = opcode;
1996 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1997 dout("WARNING: failed to queue notify event work\n");
1998 goto done_err;
1999 }
2000 }
2001
2002 return;
2003
2004done_err:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002005 ceph_osdc_put_event(event);
2006 return;
2007
2008bad:
2009 pr_err("osdc handle_watch_notify corrupt msg\n");
2010 return;
2011}
2012
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002013/*
Alex Eldere65550f2013-04-05 01:27:12 -05002014 * build new request AND message
2015 *
2016 */
2017void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2018 struct ceph_snap_context *snapc, u64 snap_id,
2019 struct timespec *mtime)
2020{
2021 struct ceph_msg *msg = req->r_request;
2022 void *p;
2023 size_t msg_size;
2024 int flags = req->r_flags;
2025 u64 data_len;
2026 unsigned int i;
2027
2028 req->r_snapid = snap_id;
2029 req->r_snapc = ceph_get_snap_context(snapc);
2030
2031 /* encode request */
2032 msg->hdr.version = cpu_to_le16(4);
2033
2034 p = msg->front.iov_base;
2035 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2036 req->r_request_osdmap_epoch = p;
2037 p += 4;
2038 req->r_request_flags = p;
2039 p += 4;
2040 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2041 ceph_encode_timespec(p, mtime);
2042 p += sizeof(struct ceph_timespec);
2043 req->r_request_reassert_version = p;
2044 p += sizeof(struct ceph_eversion); /* will get filled in */
2045
2046 /* oloc */
2047 ceph_encode_8(&p, 4);
2048 ceph_encode_8(&p, 4);
2049 ceph_encode_32(&p, 8 + 4 + 4);
2050 req->r_request_pool = p;
2051 p += 8;
2052 ceph_encode_32(&p, -1); /* preferred */
2053 ceph_encode_32(&p, 0); /* key len */
2054
2055 ceph_encode_8(&p, 1);
2056 req->r_request_pgid = p;
2057 p += 8 + 4;
2058 ceph_encode_32(&p, -1); /* preferred */
2059
2060 /* oid */
2061 ceph_encode_32(&p, req->r_oid_len);
2062 memcpy(p, req->r_oid, req->r_oid_len);
2063 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2064 p += req->r_oid_len;
2065
2066 /* ops--can imply data */
2067 ceph_encode_16(&p, (u16)req->r_num_ops);
2068 data_len = 0;
2069 for (i = 0; i < req->r_num_ops; i++) {
2070 data_len += osd_req_encode_op(req, p, i);
2071 p += sizeof(struct ceph_osd_op);
2072 }
2073
2074 /* snaps */
2075 ceph_encode_64(&p, req->r_snapid);
2076 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2077 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2078 if (req->r_snapc) {
2079 for (i = 0; i < snapc->num_snaps; i++) {
2080 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2081 }
2082 }
2083
2084 req->r_request_attempts = p;
2085 p += 4;
2086
2087 /* data */
2088 if (flags & CEPH_OSD_FLAG_WRITE) {
2089 u16 data_off;
2090
2091 /*
2092 * The header "data_off" is a hint to the receiver
2093 * allowing it to align received data into its
2094 * buffers such that there's no need to re-copy
2095 * it before writing it to disk (direct I/O).
2096 */
2097 data_off = (u16) (off & 0xffff);
2098 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2099 }
2100 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2101
2102 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2103 msg_size = p - msg->front.iov_base;
2104 msg->front.iov_len = msg_size;
2105 msg->hdr.front_len = cpu_to_le32(msg_size);
2106
2107 dout("build_request msg_size was %d\n", (int)msg_size);
2108}
2109EXPORT_SYMBOL(ceph_osdc_build_request);
2110
2111/*
Sage Weilf24e9982009-10-06 11:31:10 -07002112 * Register request, send initial attempt.
2113 */
2114int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2115 struct ceph_osd_request *req,
2116 bool nofail)
2117{
Sage Weilc1ea8822009-10-08 16:55:47 -07002118 int rc = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07002119
Sage Weilf24e9982009-10-06 11:31:10 -07002120 down_read(&osdc->map_sem);
2121 mutex_lock(&osdc->request_mutex);
Alex Elderdc4b8702013-03-25 18:16:11 -05002122 __register_request(osdc, req);
Alex Elder92451b42013-03-25 18:16:11 -05002123 WARN_ON(req->r_sent);
2124 rc = __map_request(osdc, req, 0);
2125 if (rc < 0) {
2126 if (nofail) {
2127 dout("osdc_start_request failed map, "
2128 " will retry %lld\n", req->r_tid);
2129 rc = 0;
Sage Weil9d6fcb02011-05-12 15:48:16 -07002130 }
Alex Elder92451b42013-03-25 18:16:11 -05002131 goto out_unlock;
Sage Weilf24e9982009-10-06 11:31:10 -07002132 }
Alex Elder92451b42013-03-25 18:16:11 -05002133 if (req->r_osd == NULL) {
2134 dout("send_request %p no up osds in pg\n", req);
2135 ceph_monc_request_next_osdmap(&osdc->client->monc);
2136 } else {
Alex Elder7e2766a2013-03-25 18:16:11 -05002137 __send_queued(osdc);
Alex Elder92451b42013-03-25 18:16:11 -05002138 }
2139 rc = 0;
Dan Carpenter234af26f2011-03-29 06:25:59 +03002140out_unlock:
Sage Weilf24e9982009-10-06 11:31:10 -07002141 mutex_unlock(&osdc->request_mutex);
2142 up_read(&osdc->map_sem);
2143 return rc;
2144}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002145EXPORT_SYMBOL(ceph_osdc_start_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002146
2147/*
2148 * wait for a request to complete
2149 */
2150int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2151 struct ceph_osd_request *req)
2152{
2153 int rc;
2154
2155 rc = wait_for_completion_interruptible(&req->r_completion);
2156 if (rc < 0) {
2157 mutex_lock(&osdc->request_mutex);
2158 __cancel_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08002159 __unregister_request(osdc, req);
Sage Weilf24e9982009-10-06 11:31:10 -07002160 mutex_unlock(&osdc->request_mutex);
Sage Weil25845472011-06-03 09:37:09 -07002161 complete_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08002162 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07002163 return rc;
2164 }
2165
2166 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2167 return req->r_result;
2168}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002169EXPORT_SYMBOL(ceph_osdc_wait_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002170
2171/*
2172 * sync - wait for all in-flight requests to flush. avoid starvation.
2173 */
2174void ceph_osdc_sync(struct ceph_osd_client *osdc)
2175{
2176 struct ceph_osd_request *req;
2177 u64 last_tid, next_tid = 0;
2178
2179 mutex_lock(&osdc->request_mutex);
2180 last_tid = osdc->last_tid;
2181 while (1) {
2182 req = __lookup_request_ge(osdc, next_tid);
2183 if (!req)
2184 break;
2185 if (req->r_tid > last_tid)
2186 break;
2187
2188 next_tid = req->r_tid + 1;
2189 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2190 continue;
2191
2192 ceph_osdc_get_request(req);
2193 mutex_unlock(&osdc->request_mutex);
2194 dout("sync waiting on tid %llu (last is %llu)\n",
2195 req->r_tid, last_tid);
2196 wait_for_completion(&req->r_safe_completion);
2197 mutex_lock(&osdc->request_mutex);
2198 ceph_osdc_put_request(req);
2199 }
2200 mutex_unlock(&osdc->request_mutex);
2201 dout("sync done (thru tid %llu)\n", last_tid);
2202}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002203EXPORT_SYMBOL(ceph_osdc_sync);
Sage Weilf24e9982009-10-06 11:31:10 -07002204
2205/*
2206 * init, shutdown
2207 */
2208int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2209{
2210 int err;
2211
2212 dout("init\n");
2213 osdc->client = client;
2214 osdc->osdmap = NULL;
2215 init_rwsem(&osdc->map_sem);
2216 init_completion(&osdc->map_waiters);
2217 osdc->last_requested_map = 0;
2218 mutex_init(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07002219 osdc->last_tid = 0;
2220 osdc->osds = RB_ROOT;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002221 INIT_LIST_HEAD(&osdc->osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07002222 osdc->requests = RB_ROOT;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08002223 INIT_LIST_HEAD(&osdc->req_lru);
Sage Weil6f6c7002011-01-17 20:34:08 -08002224 INIT_LIST_HEAD(&osdc->req_unsent);
2225 INIT_LIST_HEAD(&osdc->req_notarget);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002226 INIT_LIST_HEAD(&osdc->req_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07002227 osdc->num_requests = 0;
2228 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002229 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002230 spin_lock_init(&osdc->event_lock);
2231 osdc->event_tree = RB_ROOT;
2232 osdc->event_count = 0;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002233
2234 schedule_delayed_work(&osdc->osds_timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002235 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
Sage Weilf24e9982009-10-06 11:31:10 -07002236
Sage Weil5f44f142009-11-18 14:52:18 -08002237 err = -ENOMEM;
Sage Weilf24e9982009-10-06 11:31:10 -07002238 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2239 sizeof(struct ceph_osd_request));
2240 if (!osdc->req_mempool)
Sage Weil5f44f142009-11-18 14:52:18 -08002241 goto out;
Sage Weilf24e9982009-10-06 11:31:10 -07002242
Sage Weild50b4092012-07-09 14:22:34 -07002243 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2244 OSD_OP_FRONT_LEN, 10, true,
Sage Weil4f482802010-04-24 09:56:35 -07002245 "osd_op");
Sage Weilf24e9982009-10-06 11:31:10 -07002246 if (err < 0)
Sage Weil5f44f142009-11-18 14:52:18 -08002247 goto out_mempool;
Sage Weild50b4092012-07-09 14:22:34 -07002248 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
Sage Weil4f482802010-04-24 09:56:35 -07002249 OSD_OPREPLY_FRONT_LEN, 10, true,
2250 "osd_op_reply");
Sage Weilc16e7862010-03-01 13:02:00 -08002251 if (err < 0)
2252 goto out_msgpool;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002253
2254 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2255 if (IS_ERR(osdc->notify_wq)) {
2256 err = PTR_ERR(osdc->notify_wq);
2257 osdc->notify_wq = NULL;
2258 goto out_msgpool;
2259 }
Sage Weilf24e9982009-10-06 11:31:10 -07002260 return 0;
Sage Weil5f44f142009-11-18 14:52:18 -08002261
Sage Weilc16e7862010-03-01 13:02:00 -08002262out_msgpool:
2263 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weil5f44f142009-11-18 14:52:18 -08002264out_mempool:
2265 mempool_destroy(osdc->req_mempool);
2266out:
2267 return err;
Sage Weilf24e9982009-10-06 11:31:10 -07002268}
2269
2270void ceph_osdc_stop(struct ceph_osd_client *osdc)
2271{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002272 flush_workqueue(osdc->notify_wq);
2273 destroy_workqueue(osdc->notify_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07002274 cancel_delayed_work_sync(&osdc->timeout_work);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002275 cancel_delayed_work_sync(&osdc->osds_timeout_work);
Sage Weilf24e9982009-10-06 11:31:10 -07002276 if (osdc->osdmap) {
2277 ceph_osdmap_destroy(osdc->osdmap);
2278 osdc->osdmap = NULL;
2279 }
Sage Weilaca420b2011-08-31 14:45:53 -07002280 remove_all_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07002281 mempool_destroy(osdc->req_mempool);
2282 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weilc16e7862010-03-01 13:02:00 -08002283 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
Sage Weilf24e9982009-10-06 11:31:10 -07002284}
2285
2286/*
2287 * Read some contiguous pages. If we cross a stripe boundary, shorten
2288 * *plen. Return number of bytes read, or error.
2289 */
2290int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2291 struct ceph_vino vino, struct ceph_file_layout *layout,
2292 u64 off, u64 *plen,
2293 u32 truncate_seq, u64 truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -08002294 struct page **pages, int num_pages, int page_align)
Sage Weilf24e9982009-10-06 11:31:10 -07002295{
2296 struct ceph_osd_request *req;
2297 int rc = 0;
2298
2299 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2300 vino.snap, off, *plen);
Alex Elder79528732013-04-03 21:32:51 -05002301 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002302 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
Alex Elderacead002013-03-14 14:09:05 -05002303 NULL, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002304 false);
Sage Weil68162822012-09-24 21:01:02 -07002305 if (IS_ERR(req))
2306 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002307
2308 /* it may be a short read due to an object boundary */
Alex Elder0fff87e2013-02-14 12:16:43 -06002309
Alex Elder406e2c92013-04-15 14:50:36 -05002310 osd_req_op_extent_osd_data_pages(req, 0,
Alex Eldera4ce40a2013-04-05 01:27:12 -05002311 pages, *plen, page_align, false, false);
Sage Weilf24e9982009-10-06 11:31:10 -07002312
Alex Eldere0c59482013-03-07 15:38:25 -06002313 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
Alex Elder43bfe5d2013-04-03 01:28:57 -05002314 off, *plen, *plen, page_align);
Sage Weilf24e9982009-10-06 11:31:10 -07002315
Alex Elder79528732013-04-03 21:32:51 -05002316 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
Alex Elder02ee07d2013-03-14 14:09:06 -05002317
Sage Weilf24e9982009-10-06 11:31:10 -07002318 rc = ceph_osdc_start_request(osdc, req, false);
2319 if (!rc)
2320 rc = ceph_osdc_wait_request(osdc, req);
2321
2322 ceph_osdc_put_request(req);
2323 dout("readpages result %d\n", rc);
2324 return rc;
2325}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002326EXPORT_SYMBOL(ceph_osdc_readpages);
Sage Weilf24e9982009-10-06 11:31:10 -07002327
2328/*
2329 * do a synchronous write on N pages
2330 */
2331int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2332 struct ceph_file_layout *layout,
2333 struct ceph_snap_context *snapc,
2334 u64 off, u64 len,
2335 u32 truncate_seq, u64 truncate_size,
2336 struct timespec *mtime,
Alex Elder24808822013-02-15 11:42:29 -06002337 struct page **pages, int num_pages)
Sage Weilf24e9982009-10-06 11:31:10 -07002338{
2339 struct ceph_osd_request *req;
2340 int rc = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -08002341 int page_align = off & ~PAGE_MASK;
Sage Weilf24e9982009-10-06 11:31:10 -07002342
Alex Elderacead002013-03-14 14:09:05 -05002343 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
Alex Elder79528732013-04-03 21:32:51 -05002344 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002345 CEPH_OSD_OP_WRITE,
Alex Elder24808822013-02-15 11:42:29 -06002346 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
Alex Elderacead002013-03-14 14:09:05 -05002347 snapc, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002348 true);
Sage Weil68162822012-09-24 21:01:02 -07002349 if (IS_ERR(req))
2350 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002351
2352 /* it may be a short write due to an object boundary */
Alex Elder406e2c92013-04-15 14:50:36 -05002353 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
Alex Elder43bfe5d2013-04-03 01:28:57 -05002354 false, false);
2355 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
Sage Weilf24e9982009-10-06 11:31:10 -07002356
Alex Elder79528732013-04-03 21:32:51 -05002357 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
Alex Elder02ee07d2013-03-14 14:09:06 -05002358
Alex Elder87f979d2013-02-15 11:42:29 -06002359 rc = ceph_osdc_start_request(osdc, req, true);
Sage Weilf24e9982009-10-06 11:31:10 -07002360 if (!rc)
2361 rc = ceph_osdc_wait_request(osdc, req);
2362
2363 ceph_osdc_put_request(req);
2364 if (rc == 0)
2365 rc = len;
2366 dout("writepages result %d\n", rc);
2367 return rc;
2368}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002369EXPORT_SYMBOL(ceph_osdc_writepages);
Sage Weilf24e9982009-10-06 11:31:10 -07002370
Alex Elder5522ae02013-05-01 12:43:04 -05002371int ceph_osdc_setup(void)
2372{
2373 BUG_ON(ceph_osd_request_cache);
2374 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2375 sizeof (struct ceph_osd_request),
2376 __alignof__(struct ceph_osd_request),
2377 0, NULL);
2378
2379 return ceph_osd_request_cache ? 0 : -ENOMEM;
2380}
2381EXPORT_SYMBOL(ceph_osdc_setup);
2382
2383void ceph_osdc_cleanup(void)
2384{
2385 BUG_ON(!ceph_osd_request_cache);
2386 kmem_cache_destroy(ceph_osd_request_cache);
2387 ceph_osd_request_cache = NULL;
2388}
2389EXPORT_SYMBOL(ceph_osdc_cleanup);
2390
Sage Weilf24e9982009-10-06 11:31:10 -07002391/*
2392 * handle incoming message
2393 */
2394static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2395{
2396 struct ceph_osd *osd = con->private;
Julia Lawall32c895e2009-11-21 16:53:16 +01002397 struct ceph_osd_client *osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002398 int type = le16_to_cpu(msg->hdr.type);
2399
2400 if (!osd)
Sage Weil4a32f932010-06-13 10:27:53 -07002401 goto out;
Julia Lawall32c895e2009-11-21 16:53:16 +01002402 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002403
2404 switch (type) {
2405 case CEPH_MSG_OSD_MAP:
2406 ceph_osdc_handle_map(osdc, msg);
2407 break;
2408 case CEPH_MSG_OSD_OPREPLY:
Sage Weil350b1c32009-12-22 10:45:45 -08002409 handle_reply(osdc, msg, con);
Sage Weilf24e9982009-10-06 11:31:10 -07002410 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002411 case CEPH_MSG_WATCH_NOTIFY:
2412 handle_watch_notify(osdc, msg);
2413 break;
Sage Weilf24e9982009-10-06 11:31:10 -07002414
2415 default:
2416 pr_err("received unknown message type %d %s\n", type,
2417 ceph_msg_type_name(type));
2418 }
Sage Weil4a32f932010-06-13 10:27:53 -07002419out:
Sage Weilf24e9982009-10-06 11:31:10 -07002420 ceph_msg_put(msg);
2421}
2422
Sage Weil5b3a4db2010-02-19 21:43:23 -08002423/*
Sage Weil21b667f2010-03-04 10:22:59 -08002424 * lookup and return message for incoming reply. set up reply message
2425 * pages.
Sage Weil5b3a4db2010-02-19 21:43:23 -08002426 */
2427static struct ceph_msg *get_reply(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002428 struct ceph_msg_header *hdr,
2429 int *skip)
Sage Weilf24e9982009-10-06 11:31:10 -07002430{
2431 struct ceph_osd *osd = con->private;
2432 struct ceph_osd_client *osdc = osd->o_osdc;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002433 struct ceph_msg *m;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002434 struct ceph_osd_request *req;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002435 int front = le32_to_cpu(hdr->front_len);
2436 int data_len = le32_to_cpu(hdr->data_len);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002437 u64 tid;
Sage Weilf24e9982009-10-06 11:31:10 -07002438
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002439 tid = le64_to_cpu(hdr->tid);
2440 mutex_lock(&osdc->request_mutex);
2441 req = __lookup_request(osdc, tid);
2442 if (!req) {
2443 *skip = 1;
2444 m = NULL;
Sage Weil756a16a2012-07-30 16:26:13 -07002445 dout("get_reply unknown tid %llu from osd%d\n", tid,
2446 osd->o_osd);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002447 goto out;
2448 }
Sage Weilc16e7862010-03-01 13:02:00 -08002449
Alex Elderace6d3a2013-04-01 16:12:14 -05002450 if (req->r_reply->con)
Alex Elder8921d112012-06-01 14:56:43 -05002451 dout("%s revoking msg %p from old con %p\n", __func__,
Alex Elderace6d3a2013-04-01 16:12:14 -05002452 req->r_reply, req->r_reply->con);
2453 ceph_msg_revoke_incoming(req->r_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002454
Sage Weilc16e7862010-03-01 13:02:00 -08002455 if (front > req->r_reply->front.iov_len) {
2456 pr_warning("get_reply front %d > preallocated %d\n",
2457 front, (int)req->r_reply->front.iov_len);
Sage Weilb61c2762011-08-09 15:03:46 -07002458 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07002459 if (!m)
Sage Weilc16e7862010-03-01 13:02:00 -08002460 goto out;
2461 ceph_msg_put(req->r_reply);
2462 req->r_reply = m;
2463 }
2464 m = ceph_msg_get(req->r_reply);
2465
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002466 if (data_len > 0) {
Alex Eldera4ce40a2013-04-05 01:27:12 -05002467 struct ceph_osd_data *osd_data;
Alex Elder0fff87e2013-02-14 12:16:43 -06002468
Alex Eldera4ce40a2013-04-05 01:27:12 -05002469 /*
2470 * XXX This is assuming there is only one op containing
2471 * XXX page data. Probably OK for reads, but this
2472 * XXX ought to be done more generally.
2473 */
Alex Elder406e2c92013-04-15 14:50:36 -05002474 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder0fff87e2013-02-14 12:16:43 -06002475 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
Alex Elder0fff87e2013-02-14 12:16:43 -06002476 if (osd_data->pages &&
Alex Eldere0c59482013-03-07 15:38:25 -06002477 unlikely(osd_data->length < data_len)) {
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002478
Alex Eldere0c59482013-03-07 15:38:25 -06002479 pr_warning("tid %lld reply has %d bytes "
2480 "we had only %llu bytes ready\n",
2481 tid, data_len, osd_data->length);
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002482 *skip = 1;
2483 ceph_msg_put(m);
2484 m = NULL;
2485 goto out;
2486 }
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002487 }
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002488 }
Sage Weil5b3a4db2010-02-19 21:43:23 -08002489 *skip = 0;
Sage Weilc16e7862010-03-01 13:02:00 -08002490 dout("get_reply tid %lld %p\n", tid, m);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002491
2492out:
2493 mutex_unlock(&osdc->request_mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002494 return m;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002495
2496}
2497
2498static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2499 struct ceph_msg_header *hdr,
2500 int *skip)
2501{
2502 struct ceph_osd *osd = con->private;
2503 int type = le16_to_cpu(hdr->type);
2504 int front = le32_to_cpu(hdr->front_len);
2505
Alex Elder1c20f2d2012-06-04 14:43:32 -05002506 *skip = 0;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002507 switch (type) {
2508 case CEPH_MSG_OSD_MAP:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002509 case CEPH_MSG_WATCH_NOTIFY:
Sage Weilb61c2762011-08-09 15:03:46 -07002510 return ceph_msg_new(type, front, GFP_NOFS, false);
Sage Weil5b3a4db2010-02-19 21:43:23 -08002511 case CEPH_MSG_OSD_OPREPLY:
2512 return get_reply(con, hdr, skip);
2513 default:
2514 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2515 osd->o_osd);
2516 *skip = 1;
2517 return NULL;
2518 }
Sage Weilf24e9982009-10-06 11:31:10 -07002519}
2520
2521/*
2522 * Wrappers to refcount containing ceph_osd struct
2523 */
2524static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2525{
2526 struct ceph_osd *osd = con->private;
2527 if (get_osd(osd))
2528 return con;
2529 return NULL;
2530}
2531
2532static void put_osd_con(struct ceph_connection *con)
2533{
2534 struct ceph_osd *osd = con->private;
2535 put_osd(osd);
2536}
2537
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002538/*
2539 * authentication
2540 */
Alex Eldera3530df2012-05-16 15:16:39 -05002541/*
2542 * Note: returned pointer is the address of a structure that's
2543 * managed separately. Caller must *not* attempt to free it.
2544 */
2545static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
Alex Elder8f43fb52012-05-16 15:16:39 -05002546 int *proto, int force_new)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002547{
2548 struct ceph_osd *o = con->private;
2549 struct ceph_osd_client *osdc = o->o_osdc;
2550 struct ceph_auth_client *ac = osdc->client->monc.auth;
Alex Elder74f18692012-05-16 15:16:39 -05002551 struct ceph_auth_handshake *auth = &o->o_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002552
Alex Elder74f18692012-05-16 15:16:39 -05002553 if (force_new && auth->authorizer) {
Sage Weil27859f92013-03-25 10:26:14 -07002554 ceph_auth_destroy_authorizer(ac, auth->authorizer);
Alex Elder74f18692012-05-16 15:16:39 -05002555 auth->authorizer = NULL;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002556 }
Sage Weil27859f92013-03-25 10:26:14 -07002557 if (!auth->authorizer) {
2558 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2559 auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002560 if (ret)
Alex Eldera3530df2012-05-16 15:16:39 -05002561 return ERR_PTR(ret);
Sage Weil27859f92013-03-25 10:26:14 -07002562 } else {
2563 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
Sage Weil0bed9b52013-03-25 10:26:01 -07002564 auth);
2565 if (ret)
2566 return ERR_PTR(ret);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002567 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002568 *proto = ac->protocol;
Alex Elder74f18692012-05-16 15:16:39 -05002569
Alex Eldera3530df2012-05-16 15:16:39 -05002570 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002571}
2572
2573
2574static int verify_authorizer_reply(struct ceph_connection *con, int len)
2575{
2576 struct ceph_osd *o = con->private;
2577 struct ceph_osd_client *osdc = o->o_osdc;
2578 struct ceph_auth_client *ac = osdc->client->monc.auth;
2579
Sage Weil27859f92013-03-25 10:26:14 -07002580 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002581}
2582
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002583static int invalidate_authorizer(struct ceph_connection *con)
2584{
2585 struct ceph_osd *o = con->private;
2586 struct ceph_osd_client *osdc = o->o_osdc;
2587 struct ceph_auth_client *ac = osdc->client->monc.auth;
2588
Sage Weil27859f92013-03-25 10:26:14 -07002589 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002590 return ceph_monc_validate_auth(&osdc->client->monc);
2591}
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002592
Tobias Klauser9e327892010-05-20 10:40:19 +02002593static const struct ceph_connection_operations osd_con_ops = {
Sage Weilf24e9982009-10-06 11:31:10 -07002594 .get = get_osd_con,
2595 .put = put_osd_con,
2596 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002597 .get_authorizer = get_authorizer,
2598 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002599 .invalidate_authorizer = invalidate_authorizer,
Sage Weilf24e9982009-10-06 11:31:10 -07002600 .alloc_msg = alloc_msg,
Sage Weil81b024e2009-10-09 10:29:18 -07002601 .fault = osd_reset,
Sage Weilf24e9982009-10-06 11:31:10 -07002602};