blob: 50033677c0fa5134d540fba82cc8e298ce1367a0 [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);
Ilya Dryomov2cc61282014-09-03 14:41:45 +040033static void __unregister_request(struct ceph_osd_client *osdc,
34 struct ceph_osd_request *req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -070035static void __unregister_linger_request(struct ceph_osd_client *osdc,
36 struct ceph_osd_request *req);
Ilya Dryomov2cc61282014-09-03 14:41:45 +040037static void __enqueue_request(struct ceph_osd_request *req);
Sage Weil56e925b2012-01-03 12:34:34 -080038static void __send_request(struct ceph_osd_client *osdc,
39 struct ceph_osd_request *req);
Sage Weilf24e9982009-10-06 11:31:10 -070040
41/*
42 * Implement client access to distributed object storage cluster.
43 *
44 * All data objects are stored within a cluster/cloud of OSDs, or
45 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
46 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
47 * remote daemons serving up and coordinating consistent and safe
48 * access to storage.
49 *
50 * Cluster membership and the mapping of data objects onto storage devices
51 * are described by the osd map.
52 *
53 * We keep track of pending OSD requests (read, write), resubmit
54 * requests to different OSDs when the cluster topology/data layout
55 * change, or retry the affected requests when the communications
56 * channel with an OSD is reset.
57 */
58
59/*
60 * calculate the mapping of a file extent onto an object, and fill out the
61 * request accordingly. shorten extent as necessary if it crosses an
62 * object boundary.
63 *
64 * fill osd op in request message.
65 */
Alex Elderdbe0fc42013-02-15 22:10:17 -060066static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
Alex Eldera19dadf2013-03-13 20:50:01 -050067 u64 *objnum, u64 *objoff, u64 *objlen)
Sage Weilf24e9982009-10-06 11:31:10 -070068{
Alex Elder60e56f12013-02-15 11:42:29 -060069 u64 orig_len = *plen;
Sage Weild63b77f2012-09-24 20:59:48 -070070 int r;
Sage Weilf24e9982009-10-06 11:31:10 -070071
Alex Elder60e56f12013-02-15 11:42:29 -060072 /* object extent? */
Alex Elder75d1c942013-03-13 20:50:00 -050073 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
74 objoff, objlen);
Sage Weild63b77f2012-09-24 20:59:48 -070075 if (r < 0)
76 return r;
Alex Elder75d1c942013-03-13 20:50:00 -050077 if (*objlen < orig_len) {
78 *plen = *objlen;
Alex Elder60e56f12013-02-15 11:42:29 -060079 dout(" skipping last %llu, final file extent %llu~%llu\n",
80 orig_len - *plen, off, *plen);
81 }
82
Alex Elder75d1c942013-03-13 20:50:00 -050083 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
Sage Weilf24e9982009-10-06 11:31:10 -070084
Alex Elder3ff5f382013-02-15 22:10:17 -060085 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -070086}
87
Alex Elderc54d47b2013-04-03 01:28:57 -050088static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
89{
90 memset(osd_data, 0, sizeof (*osd_data));
91 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
92}
93
Alex Eldera4ce40a2013-04-05 01:27:12 -050094static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -050095 struct page **pages, u64 length, u32 alignment,
96 bool pages_from_pool, bool own_pages)
97{
98 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
99 osd_data->pages = pages;
100 osd_data->length = length;
101 osd_data->alignment = alignment;
102 osd_data->pages_from_pool = pages_from_pool;
103 osd_data->own_pages = own_pages;
104}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500105
Alex Eldera4ce40a2013-04-05 01:27:12 -0500106static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -0500107 struct ceph_pagelist *pagelist)
108{
109 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
110 osd_data->pagelist = pagelist;
111}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500112
113#ifdef CONFIG_BLOCK
Alex Eldera4ce40a2013-04-05 01:27:12 -0500114static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -0500115 struct bio *bio, size_t bio_length)
116{
117 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
118 osd_data->bio = bio;
119 osd_data->bio_length = bio_length;
120}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500121#endif /* CONFIG_BLOCK */
122
Alex Elder863c7eb2013-04-15 14:50:36 -0500123#define osd_req_op_data(oreq, whch, typ, fld) \
124 ({ \
125 BUG_ON(whch >= (oreq)->r_num_ops); \
126 &(oreq)->r_ops[whch].typ.fld; \
127 })
128
Alex Elder49719772013-02-11 12:33:24 -0600129static struct ceph_osd_data *
130osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
131{
132 BUG_ON(which >= osd_req->r_num_ops);
133
134 return &osd_req->r_ops[which].raw_data_in;
135}
136
Alex Eldera4ce40a2013-04-05 01:27:12 -0500137struct ceph_osd_data *
138osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500139 unsigned int which)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500140{
Alex Elder863c7eb2013-04-15 14:50:36 -0500141 return osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500142}
143EXPORT_SYMBOL(osd_req_op_extent_osd_data);
144
145struct ceph_osd_data *
Alex Eldera4ce40a2013-04-05 01:27:12 -0500146osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
147 unsigned int which)
148{
Alex Elder863c7eb2013-04-15 14:50:36 -0500149 return osd_req_op_data(osd_req, which, cls, response_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500150}
151EXPORT_SYMBOL(osd_req_op_cls_response_data); /* ??? */
152
Alex Elder49719772013-02-11 12:33:24 -0600153void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
154 unsigned int which, struct page **pages,
155 u64 length, u32 alignment,
156 bool pages_from_pool, bool own_pages)
157{
158 struct ceph_osd_data *osd_data;
159
160 osd_data = osd_req_op_raw_data_in(osd_req, which);
161 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
162 pages_from_pool, own_pages);
163}
164EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
165
Alex Eldera4ce40a2013-04-05 01:27:12 -0500166void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500167 unsigned int which, struct page **pages,
168 u64 length, u32 alignment,
Alex Eldera4ce40a2013-04-05 01:27:12 -0500169 bool pages_from_pool, bool own_pages)
170{
171 struct ceph_osd_data *osd_data;
172
Alex Elder863c7eb2013-04-15 14:50:36 -0500173 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500174 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
175 pages_from_pool, own_pages);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500176}
177EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
178
179void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500180 unsigned int which, struct ceph_pagelist *pagelist)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500181{
182 struct ceph_osd_data *osd_data;
183
Alex Elder863c7eb2013-04-15 14:50:36 -0500184 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500185 ceph_osd_data_pagelist_init(osd_data, pagelist);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500186}
187EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
188
189#ifdef CONFIG_BLOCK
190void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500191 unsigned int which, struct bio *bio, size_t bio_length)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500192{
193 struct ceph_osd_data *osd_data;
Alex Elder863c7eb2013-04-15 14:50:36 -0500194
195 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500196 ceph_osd_data_bio_init(osd_data, bio, bio_length);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500197}
198EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
199#endif /* CONFIG_BLOCK */
200
201static void osd_req_op_cls_request_info_pagelist(
202 struct ceph_osd_request *osd_req,
203 unsigned int which, struct ceph_pagelist *pagelist)
204{
205 struct ceph_osd_data *osd_data;
206
Alex Elder863c7eb2013-04-15 14:50:36 -0500207 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500208 ceph_osd_data_pagelist_init(osd_data, pagelist);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500209}
210
Alex Elder04017e22013-04-05 14:46:02 -0500211void osd_req_op_cls_request_data_pagelist(
212 struct ceph_osd_request *osd_req,
213 unsigned int which, struct ceph_pagelist *pagelist)
214{
215 struct ceph_osd_data *osd_data;
216
Alex Elder863c7eb2013-04-15 14:50:36 -0500217 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
Alex Elder04017e22013-04-05 14:46:02 -0500218 ceph_osd_data_pagelist_init(osd_data, pagelist);
219}
220EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
221
Alex Elder6c57b552013-04-19 15:34:49 -0500222void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
223 unsigned int which, struct page **pages, u64 length,
224 u32 alignment, bool pages_from_pool, bool own_pages)
225{
226 struct ceph_osd_data *osd_data;
227
228 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
229 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
230 pages_from_pool, own_pages);
231}
232EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
233
Alex Eldera4ce40a2013-04-05 01:27:12 -0500234void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
235 unsigned int which, struct page **pages, u64 length,
236 u32 alignment, bool pages_from_pool, bool own_pages)
237{
238 struct ceph_osd_data *osd_data;
239
Alex Elder863c7eb2013-04-15 14:50:36 -0500240 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500241 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
242 pages_from_pool, own_pages);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500243}
244EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
245
Alex Elder23c08a9cb2013-04-03 01:28:58 -0500246static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
247{
248 switch (osd_data->type) {
249 case CEPH_OSD_DATA_TYPE_NONE:
250 return 0;
251 case CEPH_OSD_DATA_TYPE_PAGES:
252 return osd_data->length;
253 case CEPH_OSD_DATA_TYPE_PAGELIST:
254 return (u64)osd_data->pagelist->length;
255#ifdef CONFIG_BLOCK
256 case CEPH_OSD_DATA_TYPE_BIO:
257 return (u64)osd_data->bio_length;
258#endif /* CONFIG_BLOCK */
259 default:
260 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
261 return 0;
262 }
263}
264
Alex Elderc54d47b2013-04-03 01:28:57 -0500265static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
266{
Alex Elder54764922013-04-05 01:27:12 -0500267 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
Alex Elderc54d47b2013-04-03 01:28:57 -0500268 int num_pages;
269
270 num_pages = calc_pages_for((u64)osd_data->alignment,
271 (u64)osd_data->length);
272 ceph_release_page_vector(osd_data->pages, num_pages);
273 }
Alex Elder54764922013-04-05 01:27:12 -0500274 ceph_osd_data_init(osd_data);
275}
276
277static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
278 unsigned int which)
279{
280 struct ceph_osd_req_op *op;
281
282 BUG_ON(which >= osd_req->r_num_ops);
283 op = &osd_req->r_ops[which];
284
285 switch (op->op) {
286 case CEPH_OSD_OP_READ:
287 case CEPH_OSD_OP_WRITE:
288 ceph_osd_data_release(&op->extent.osd_data);
289 break;
290 case CEPH_OSD_OP_CALL:
291 ceph_osd_data_release(&op->cls.request_info);
Alex Elder04017e22013-04-05 14:46:02 -0500292 ceph_osd_data_release(&op->cls.request_data);
Alex Elder54764922013-04-05 01:27:12 -0500293 ceph_osd_data_release(&op->cls.response_data);
294 break;
Yan, Zhengd74b50b2014-11-12 14:00:43 +0800295 case CEPH_OSD_OP_SETXATTR:
296 case CEPH_OSD_OP_CMPXATTR:
297 ceph_osd_data_release(&op->xattr.osd_data);
298 break;
Yan, Zheng66ba6092015-04-27 11:02:35 +0800299 case CEPH_OSD_OP_STAT:
300 ceph_osd_data_release(&op->raw_data_in);
301 break;
Alex Elder54764922013-04-05 01:27:12 -0500302 default:
303 break;
304 }
Alex Elderc54d47b2013-04-03 01:28:57 -0500305}
306
Sage Weilf24e9982009-10-06 11:31:10 -0700307/*
308 * requests
309 */
Ilya Dryomov9e94af22014-06-20 14:14:42 +0400310static void ceph_osdc_release_request(struct kref *kref)
Sage Weilf24e9982009-10-06 11:31:10 -0700311{
Ilya Dryomov9e94af22014-06-20 14:14:42 +0400312 struct ceph_osd_request *req = container_of(kref,
313 struct ceph_osd_request, r_kref);
Alex Elder54764922013-04-05 01:27:12 -0500314 unsigned int which;
Sage Weil415e49a2009-12-07 13:37:03 -0800315
Ilya Dryomov9e94af22014-06-20 14:14:42 +0400316 dout("%s %p (r_request %p r_reply %p)\n", __func__, req,
317 req->r_request, req->r_reply);
Ilya Dryomov6562d662014-06-20 14:14:42 +0400318 WARN_ON(!RB_EMPTY_NODE(&req->r_node));
319 WARN_ON(!list_empty(&req->r_req_lru_item));
320 WARN_ON(!list_empty(&req->r_osd_item));
321 WARN_ON(!list_empty(&req->r_linger_item));
322 WARN_ON(!list_empty(&req->r_linger_osd_item));
323 WARN_ON(req->r_osd);
Ilya Dryomov9e94af22014-06-20 14:14:42 +0400324
Sage Weil415e49a2009-12-07 13:37:03 -0800325 if (req->r_request)
326 ceph_msg_put(req->r_request);
Alex Elderace6d3a2013-04-01 16:12:14 -0500327 if (req->r_reply) {
Alex Elder8921d112012-06-01 14:56:43 -0500328 ceph_msg_revoke_incoming(req->r_reply);
Alex Elderab8cb342012-06-04 14:43:32 -0500329 ceph_msg_put(req->r_reply);
Alex Elderace6d3a2013-04-01 16:12:14 -0500330 }
Alex Elder0fff87e2013-02-14 12:16:43 -0600331
Alex Elder54764922013-04-05 01:27:12 -0500332 for (which = 0; which < req->r_num_ops; which++)
333 osd_req_op_data_release(req, which);
Alex Elder0fff87e2013-02-14 12:16:43 -0600334
Sage Weil415e49a2009-12-07 13:37:03 -0800335 ceph_put_snap_context(req->r_snapc);
336 if (req->r_mempool)
337 mempool_free(req, req->r_osdc->req_mempool);
338 else
Alex Elder5522ae02013-05-01 12:43:04 -0500339 kmem_cache_free(ceph_osd_request_cache, req);
340
Sage Weilf24e9982009-10-06 11:31:10 -0700341}
Ilya Dryomov9e94af22014-06-20 14:14:42 +0400342
343void ceph_osdc_get_request(struct ceph_osd_request *req)
344{
345 dout("%s %p (was %d)\n", __func__, req,
346 atomic_read(&req->r_kref.refcount));
347 kref_get(&req->r_kref);
348}
349EXPORT_SYMBOL(ceph_osdc_get_request);
350
351void ceph_osdc_put_request(struct ceph_osd_request *req)
352{
353 dout("%s %p (was %d)\n", __func__, req,
354 atomic_read(&req->r_kref.refcount));
355 kref_put(&req->r_kref, ceph_osdc_release_request);
356}
357EXPORT_SYMBOL(ceph_osdc_put_request);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700358
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700359struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700360 struct ceph_snap_context *snapc,
Sage Weil1b83bef2013-02-25 16:11:12 -0800361 unsigned int num_ops,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700362 bool use_mempool,
Alex Elder54a54002012-11-13 21:11:15 -0600363 gfp_t gfp_flags)
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700364{
365 struct ceph_osd_request *req;
366 struct ceph_msg *msg;
Sage Weil1b83bef2013-02-25 16:11:12 -0800367 size_t msg_size;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700368
Alex Elder79528732013-04-03 21:32:51 -0500369 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
370 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
371
Sage Weil1b83bef2013-02-25 16:11:12 -0800372 msg_size = 4 + 4 + 8 + 8 + 4+8;
373 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
374 msg_size += 1 + 8 + 4 + 4; /* pg_t */
Ilya Dryomov2d0ebc52014-01-27 17:40:18 +0200375 msg_size += 4 + CEPH_MAX_OID_NAME_LEN; /* oid */
Sage Weil1b83bef2013-02-25 16:11:12 -0800376 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
377 msg_size += 8; /* snapid */
378 msg_size += 8; /* snap_seq */
379 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
380 msg_size += 4;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700381
382 if (use_mempool) {
383 req = mempool_alloc(osdc->req_mempool, gfp_flags);
384 memset(req, 0, sizeof(*req));
385 } else {
Alex Elder5522ae02013-05-01 12:43:04 -0500386 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700387 }
388 if (req == NULL)
389 return NULL;
390
391 req->r_osdc = osdc;
392 req->r_mempool = use_mempool;
Alex Elder79528732013-04-03 21:32:51 -0500393 req->r_num_ops = num_ops;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700394
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700395 kref_init(&req->r_kref);
396 init_completion(&req->r_completion);
397 init_completion(&req->r_safe_completion);
Alex Eldera978fa22012-12-17 12:23:48 -0600398 RB_CLEAR_NODE(&req->r_node);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700399 INIT_LIST_HEAD(&req->r_unsafe_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700400 INIT_LIST_HEAD(&req->r_linger_item);
Ilya Dryomov1d0326b2014-06-20 14:14:41 +0400401 INIT_LIST_HEAD(&req->r_linger_osd_item);
Sage Weil935b6392011-09-16 11:13:17 -0700402 INIT_LIST_HEAD(&req->r_req_lru_item);
Sage Weilcd430452012-07-09 14:31:41 -0700403 INIT_LIST_HEAD(&req->r_osd_item);
404
Ilya Dryomov3c972c92014-01-27 17:40:20 +0200405 req->r_base_oloc.pool = -1;
Ilya Dryomov205ee1182014-01-27 17:40:20 +0200406 req->r_target_oloc.pool = -1;
Ilya Dryomov221165252014-01-27 17:40:18 +0200407
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700408 /* create reply message */
409 if (use_mempool)
410 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
411 else
412 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
Sage Weilb61c2762011-08-09 15:03:46 -0700413 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700414 if (!msg) {
415 ceph_osdc_put_request(req);
416 return NULL;
417 }
418 req->r_reply = msg;
419
420 /* create request message; allow space for oid */
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700421 if (use_mempool)
422 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
423 else
Sage Weilb61c2762011-08-09 15:03:46 -0700424 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700425 if (!msg) {
426 ceph_osdc_put_request(req);
427 return NULL;
428 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700429
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700430 memset(msg->front.iov_base, 0, msg->front.iov_len);
431
432 req->r_request = msg;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700433
434 return req;
435}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700436EXPORT_SYMBOL(ceph_osdc_alloc_request);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700437
Alex Eldera8dd0a32013-03-13 20:50:00 -0500438static bool osd_req_opcode_valid(u16 opcode)
439{
440 switch (opcode) {
Ilya Dryomov70b5bfa2014-10-02 17:22:29 +0400441#define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true;
442__CEPH_FORALL_OSD_OPS(GENERATE_CASE)
443#undef GENERATE_CASE
Alex Eldera8dd0a32013-03-13 20:50:00 -0500444 default:
445 return false;
446 }
447}
448
Alex Elder33803f32013-03-13 20:50:00 -0500449/*
450 * This is an osd op init function for opcodes that have no data or
451 * other information associated with them. It also serves as a
452 * common init routine for all the other init functions, below.
453 */
Alex Elderc99d2d42013-04-05 01:27:11 -0500454static struct ceph_osd_req_op *
Alex Elder49719772013-02-11 12:33:24 -0600455_osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
Yan, Zheng144cba12015-04-27 11:09:54 +0800456 u16 opcode, u32 flags)
Alex Elder33803f32013-03-13 20:50:00 -0500457{
Alex Elderc99d2d42013-04-05 01:27:11 -0500458 struct ceph_osd_req_op *op;
459
460 BUG_ON(which >= osd_req->r_num_ops);
Alex Elder33803f32013-03-13 20:50:00 -0500461 BUG_ON(!osd_req_opcode_valid(opcode));
462
Alex Elderc99d2d42013-04-05 01:27:11 -0500463 op = &osd_req->r_ops[which];
Alex Elder33803f32013-03-13 20:50:00 -0500464 memset(op, 0, sizeof (*op));
Alex Elder33803f32013-03-13 20:50:00 -0500465 op->op = opcode;
Yan, Zheng144cba12015-04-27 11:09:54 +0800466 op->flags = flags;
Alex Elderc99d2d42013-04-05 01:27:11 -0500467
468 return op;
Alex Elder33803f32013-03-13 20:50:00 -0500469}
470
Alex Elder49719772013-02-11 12:33:24 -0600471void osd_req_op_init(struct ceph_osd_request *osd_req,
Yan, Zheng144cba12015-04-27 11:09:54 +0800472 unsigned int which, u16 opcode, u32 flags)
Alex Elder49719772013-02-11 12:33:24 -0600473{
Yan, Zheng144cba12015-04-27 11:09:54 +0800474 (void)_osd_req_op_init(osd_req, which, opcode, flags);
Alex Elder49719772013-02-11 12:33:24 -0600475}
476EXPORT_SYMBOL(osd_req_op_init);
477
Alex Elderc99d2d42013-04-05 01:27:11 -0500478void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
479 unsigned int which, u16 opcode,
Alex Elder33803f32013-03-13 20:50:00 -0500480 u64 offset, u64 length,
481 u64 truncate_size, u32 truncate_seq)
482{
Yan, Zheng144cba12015-04-27 11:09:54 +0800483 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
484 opcode, 0);
Alex Elder33803f32013-03-13 20:50:00 -0500485 size_t payload_len = 0;
486
Li Wangad7a60d2013-08-15 11:51:44 +0800487 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
Yan, Zheng864e9192014-11-13 10:47:25 +0800488 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE);
Alex Elder33803f32013-03-13 20:50:00 -0500489
Alex Elder33803f32013-03-13 20:50:00 -0500490 op->extent.offset = offset;
491 op->extent.length = length;
492 op->extent.truncate_size = truncate_size;
493 op->extent.truncate_seq = truncate_seq;
494 if (opcode == CEPH_OSD_OP_WRITE)
495 payload_len += length;
496
497 op->payload_len = payload_len;
498}
499EXPORT_SYMBOL(osd_req_op_extent_init);
500
Alex Elderc99d2d42013-04-05 01:27:11 -0500501void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
502 unsigned int which, u64 length)
Alex Eldere5975c72013-03-14 14:09:05 -0500503{
Alex Elderc99d2d42013-04-05 01:27:11 -0500504 struct ceph_osd_req_op *op;
505 u64 previous;
506
507 BUG_ON(which >= osd_req->r_num_ops);
508 op = &osd_req->r_ops[which];
509 previous = op->extent.length;
Alex Eldere5975c72013-03-14 14:09:05 -0500510
511 if (length == previous)
512 return; /* Nothing to do */
513 BUG_ON(length > previous);
514
515 op->extent.length = length;
516 op->payload_len -= previous - length;
517}
518EXPORT_SYMBOL(osd_req_op_extent_update);
519
Alex Elderc99d2d42013-04-05 01:27:11 -0500520void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
Alex Elder04017e22013-04-05 14:46:02 -0500521 u16 opcode, const char *class, const char *method)
Alex Elder33803f32013-03-13 20:50:00 -0500522{
Yan, Zheng144cba12015-04-27 11:09:54 +0800523 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
524 opcode, 0);
Alex Elder5f562df2013-04-05 01:27:12 -0500525 struct ceph_pagelist *pagelist;
Alex Elder33803f32013-03-13 20:50:00 -0500526 size_t payload_len = 0;
527 size_t size;
528
529 BUG_ON(opcode != CEPH_OSD_OP_CALL);
530
Alex Elder5f562df2013-04-05 01:27:12 -0500531 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
532 BUG_ON(!pagelist);
533 ceph_pagelist_init(pagelist);
534
Alex Elder33803f32013-03-13 20:50:00 -0500535 op->cls.class_name = class;
536 size = strlen(class);
537 BUG_ON(size > (size_t) U8_MAX);
538 op->cls.class_len = size;
Alex Elder5f562df2013-04-05 01:27:12 -0500539 ceph_pagelist_append(pagelist, class, size);
Alex Elder33803f32013-03-13 20:50:00 -0500540 payload_len += size;
541
542 op->cls.method_name = method;
543 size = strlen(method);
544 BUG_ON(size > (size_t) U8_MAX);
545 op->cls.method_len = size;
Alex Elder5f562df2013-04-05 01:27:12 -0500546 ceph_pagelist_append(pagelist, method, size);
Alex Elder33803f32013-03-13 20:50:00 -0500547 payload_len += size;
548
Alex Eldera4ce40a2013-04-05 01:27:12 -0500549 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
Alex Elder5f562df2013-04-05 01:27:12 -0500550
Alex Elder33803f32013-03-13 20:50:00 -0500551 op->cls.argc = 0; /* currently unused */
552
553 op->payload_len = payload_len;
554}
555EXPORT_SYMBOL(osd_req_op_cls_init);
Alex Elder8c042b02013-04-03 01:28:58 -0500556
Yan, Zhengd74b50b2014-11-12 14:00:43 +0800557int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which,
558 u16 opcode, const char *name, const void *value,
559 size_t size, u8 cmp_op, u8 cmp_mode)
560{
Yan, Zheng144cba12015-04-27 11:09:54 +0800561 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
562 opcode, 0);
Yan, Zhengd74b50b2014-11-12 14:00:43 +0800563 struct ceph_pagelist *pagelist;
564 size_t payload_len;
565
566 BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR);
567
568 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
569 if (!pagelist)
570 return -ENOMEM;
571
572 ceph_pagelist_init(pagelist);
573
574 payload_len = strlen(name);
575 op->xattr.name_len = payload_len;
576 ceph_pagelist_append(pagelist, name, payload_len);
577
578 op->xattr.value_len = size;
579 ceph_pagelist_append(pagelist, value, size);
580 payload_len += size;
581
582 op->xattr.cmp_op = cmp_op;
583 op->xattr.cmp_mode = cmp_mode;
584
585 ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist);
586 op->payload_len = payload_len;
587 return 0;
588}
589EXPORT_SYMBOL(osd_req_op_xattr_init);
590
Alex Elderc99d2d42013-04-05 01:27:11 -0500591void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
592 unsigned int which, u16 opcode,
Alex Elder33803f32013-03-13 20:50:00 -0500593 u64 cookie, u64 version, int flag)
594{
Yan, Zheng144cba12015-04-27 11:09:54 +0800595 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
596 opcode, 0);
Alex Elder33803f32013-03-13 20:50:00 -0500597
Alex Elderc99d2d42013-04-05 01:27:11 -0500598 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
Alex Elder33803f32013-03-13 20:50:00 -0500599
600 op->watch.cookie = cookie;
Alex Elder9ef1ee52013-04-21 16:51:50 -0500601 op->watch.ver = version;
Alex Elder33803f32013-03-13 20:50:00 -0500602 if (opcode == CEPH_OSD_OP_WATCH && flag)
Alex Elderc99d2d42013-04-05 01:27:11 -0500603 op->watch.flag = (u8)1;
Alex Elder33803f32013-03-13 20:50:00 -0500604}
605EXPORT_SYMBOL(osd_req_op_watch_init);
606
Ilya Dryomovc647b8a2014-02-25 16:22:27 +0200607void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
608 unsigned int which,
609 u64 expected_object_size,
610 u64 expected_write_size)
611{
612 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
Yan, Zheng144cba12015-04-27 11:09:54 +0800613 CEPH_OSD_OP_SETALLOCHINT,
614 0);
Ilya Dryomovc647b8a2014-02-25 16:22:27 +0200615
616 op->alloc_hint.expected_object_size = expected_object_size;
617 op->alloc_hint.expected_write_size = expected_write_size;
618
619 /*
620 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
621 * not worth a feature bit. Set FAILOK per-op flag to make
622 * sure older osds don't trip over an unsupported opcode.
623 */
624 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
625}
626EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
627
Alex Elder90af3602013-04-05 14:46:01 -0500628static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
Alex Elderec9123c2013-04-05 01:27:12 -0500629 struct ceph_osd_data *osd_data)
630{
631 u64 length = ceph_osd_data_length(osd_data);
632
633 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
634 BUG_ON(length > (u64) SIZE_MAX);
635 if (length)
Alex Elder90af3602013-04-05 14:46:01 -0500636 ceph_msg_data_add_pages(msg, osd_data->pages,
Alex Elderec9123c2013-04-05 01:27:12 -0500637 length, osd_data->alignment);
638 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
639 BUG_ON(!length);
Alex Elder90af3602013-04-05 14:46:01 -0500640 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
Alex Elderec9123c2013-04-05 01:27:12 -0500641#ifdef CONFIG_BLOCK
642 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
Alex Elder90af3602013-04-05 14:46:01 -0500643 ceph_msg_data_add_bio(msg, osd_data->bio, length);
Alex Elderec9123c2013-04-05 01:27:12 -0500644#endif
645 } else {
646 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
647 }
648}
649
Alex Elder175face2013-03-08 13:35:36 -0600650static u64 osd_req_encode_op(struct ceph_osd_request *req,
Alex Elder79528732013-04-03 21:32:51 -0500651 struct ceph_osd_op *dst, unsigned int which)
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700652{
Alex Elder79528732013-04-03 21:32:51 -0500653 struct ceph_osd_req_op *src;
Alex Elder04017e22013-04-05 14:46:02 -0500654 struct ceph_osd_data *osd_data;
Alex Elder54d50642013-04-03 01:28:58 -0500655 u64 request_data_len = 0;
Alex Elder04017e22013-04-05 14:46:02 -0500656 u64 data_length;
Alex Elder175face2013-03-08 13:35:36 -0600657
Alex Elder79528732013-04-03 21:32:51 -0500658 BUG_ON(which >= req->r_num_ops);
659 src = &req->r_ops[which];
Alex Eldera8dd0a32013-03-13 20:50:00 -0500660 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
661 pr_err("unrecognized osd opcode %d\n", src->op);
662
663 return 0;
664 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700665
Alex Elder065a68f2012-04-20 15:49:43 -0500666 switch (src->op) {
Alex Elderfbfab532013-02-08 09:55:48 -0600667 case CEPH_OSD_OP_STAT:
Alex Elder49719772013-02-11 12:33:24 -0600668 osd_data = &src->raw_data_in;
669 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Alex Elderfbfab532013-02-08 09:55:48 -0600670 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700671 case CEPH_OSD_OP_READ:
672 case CEPH_OSD_OP_WRITE:
Li Wangad7a60d2013-08-15 11:51:44 +0800673 case CEPH_OSD_OP_ZERO:
Li Wangad7a60d2013-08-15 11:51:44 +0800674 case CEPH_OSD_OP_TRUNCATE:
Alex Elder175face2013-03-08 13:35:36 -0600675 if (src->op == CEPH_OSD_OP_WRITE)
Alex Elder54d50642013-04-03 01:28:58 -0500676 request_data_len = src->extent.length;
Alex Elder175face2013-03-08 13:35:36 -0600677 dst->extent.offset = cpu_to_le64(src->extent.offset);
678 dst->extent.length = cpu_to_le64(src->extent.length);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700679 dst->extent.truncate_size =
680 cpu_to_le64(src->extent.truncate_size);
681 dst->extent.truncate_seq =
682 cpu_to_le32(src->extent.truncate_seq);
Alex Elder04017e22013-04-05 14:46:02 -0500683 osd_data = &src->extent.osd_data;
Alex Elder54764922013-04-05 01:27:12 -0500684 if (src->op == CEPH_OSD_OP_WRITE)
Alex Elder04017e22013-04-05 14:46:02 -0500685 ceph_osdc_msg_data_add(req->r_request, osd_data);
Alex Elder54764922013-04-05 01:27:12 -0500686 else
Alex Elder04017e22013-04-05 14:46:02 -0500687 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700688 break;
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700689 case CEPH_OSD_OP_CALL:
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700690 dst->cls.class_len = src->cls.class_len;
691 dst->cls.method_len = src->cls.method_len;
Alex Elder04017e22013-04-05 14:46:02 -0500692 osd_data = &src->cls.request_info;
693 ceph_osdc_msg_data_add(req->r_request, osd_data);
694 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
695 request_data_len = osd_data->pagelist->length;
696
697 osd_data = &src->cls.request_data;
698 data_length = ceph_osd_data_length(osd_data);
699 if (data_length) {
700 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
701 dst->cls.indata_len = cpu_to_le32(data_length);
702 ceph_osdc_msg_data_add(req->r_request, osd_data);
703 src->payload_len += data_length;
704 request_data_len += data_length;
705 }
706 osd_data = &src->cls.response_data;
707 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700708 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700709 case CEPH_OSD_OP_STARTSYNC:
710 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700711 case CEPH_OSD_OP_NOTIFY_ACK:
712 case CEPH_OSD_OP_WATCH:
713 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
714 dst->watch.ver = cpu_to_le64(src->watch.ver);
715 dst->watch.flag = src->watch.flag;
716 break;
Ilya Dryomovc647b8a2014-02-25 16:22:27 +0200717 case CEPH_OSD_OP_SETALLOCHINT:
718 dst->alloc_hint.expected_object_size =
719 cpu_to_le64(src->alloc_hint.expected_object_size);
720 dst->alloc_hint.expected_write_size =
721 cpu_to_le64(src->alloc_hint.expected_write_size);
722 break;
Yan, Zhengd74b50b2014-11-12 14:00:43 +0800723 case CEPH_OSD_OP_SETXATTR:
724 case CEPH_OSD_OP_CMPXATTR:
725 dst->xattr.name_len = cpu_to_le32(src->xattr.name_len);
726 dst->xattr.value_len = cpu_to_le32(src->xattr.value_len);
727 dst->xattr.cmp_op = src->xattr.cmp_op;
728 dst->xattr.cmp_mode = src->xattr.cmp_mode;
729 osd_data = &src->xattr.osd_data;
730 ceph_osdc_msg_data_add(req->r_request, osd_data);
731 request_data_len = osd_data->pagelist->length;
732 break;
Yan, Zheng864e9192014-11-13 10:47:25 +0800733 case CEPH_OSD_OP_CREATE:
734 case CEPH_OSD_OP_DELETE:
735 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700736 default:
Alex Elder4c464592013-02-15 11:42:30 -0600737 pr_err("unsupported osd opcode %s\n",
Alex Elder8f63ca22013-03-04 11:08:29 -0600738 ceph_osd_op_name(src->op));
Alex Elder4c464592013-02-15 11:42:30 -0600739 WARN_ON(1);
Alex Eldera8dd0a32013-03-13 20:50:00 -0500740
741 return 0;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700742 }
Ilya Dryomov7b25bf52014-02-25 16:22:26 +0200743
Alex Eldera8dd0a32013-03-13 20:50:00 -0500744 dst->op = cpu_to_le16(src->op);
Ilya Dryomov7b25bf52014-02-25 16:22:26 +0200745 dst->flags = cpu_to_le32(src->flags);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700746 dst->payload_len = cpu_to_le32(src->payload_len);
Alex Elder175face2013-03-08 13:35:36 -0600747
Alex Elder54d50642013-04-03 01:28:58 -0500748 return request_data_len;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700749}
750
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700751/*
Sage Weilf24e9982009-10-06 11:31:10 -0700752 * build new request AND message, calculate layout, and adjust file
753 * extent as needed.
754 *
755 * if the file was recently truncated, we include information about its
756 * old and new size so that the object can be updated appropriately. (we
757 * avoid synchronously deleting truncated objects because it's slow.)
758 *
759 * if @do_sync, include a 'startsync' command so that the osd will flush
760 * data quickly.
761 */
762struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
763 struct ceph_file_layout *layout,
764 struct ceph_vino vino,
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800765 u64 off, u64 *plen,
766 unsigned int which, int num_ops,
Sage Weilf24e9982009-10-06 11:31:10 -0700767 int opcode, int flags,
768 struct ceph_snap_context *snapc,
Sage Weilf24e9982009-10-06 11:31:10 -0700769 u32 truncate_seq,
770 u64 truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -0600771 bool use_mempool)
Sage Weilf24e9982009-10-06 11:31:10 -0700772{
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700773 struct ceph_osd_request *req;
Alex Elder75d1c942013-03-13 20:50:00 -0500774 u64 objnum = 0;
775 u64 objoff = 0;
776 u64 objlen = 0;
Sage Weil68162822012-09-24 21:01:02 -0700777 int r;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700778
Li Wangad7a60d2013-08-15 11:51:44 +0800779 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
Yan, Zheng864e9192014-11-13 10:47:25 +0800780 opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE &&
781 opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700782
Alex Elderacead002013-03-14 14:09:05 -0500783 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
Alex Elderae7ca4a32012-11-13 21:11:15 -0600784 GFP_NOFS);
Sage Weil4ad12622011-05-03 09:23:36 -0700785 if (!req)
Sage Weil68162822012-09-24 21:01:02 -0700786 return ERR_PTR(-ENOMEM);
Alex Elder79528732013-04-03 21:32:51 -0500787
Alex Elderd178a9e2012-11-13 21:11:15 -0600788 req->r_flags = flags;
Sage Weilf24e9982009-10-06 11:31:10 -0700789
790 /* calculate max write size */
Alex Eldera19dadf2013-03-13 20:50:01 -0500791 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
Alex Elder3ff5f382013-02-15 22:10:17 -0600792 if (r < 0) {
793 ceph_osdc_put_request(req);
Sage Weil68162822012-09-24 21:01:02 -0700794 return ERR_PTR(r);
Alex Elder3ff5f382013-02-15 22:10:17 -0600795 }
Alex Eldera19dadf2013-03-13 20:50:01 -0500796
Yan, Zheng864e9192014-11-13 10:47:25 +0800797 if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) {
Yan, Zheng144cba12015-04-27 11:09:54 +0800798 osd_req_op_init(req, which, opcode, 0);
Yan, Zheng864e9192014-11-13 10:47:25 +0800799 } else {
800 u32 object_size = le32_to_cpu(layout->fl_object_size);
801 u32 object_base = off - objoff;
802 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
803 if (truncate_size <= object_base) {
804 truncate_size = 0;
805 } else {
806 truncate_size -= object_base;
807 if (truncate_size > object_size)
808 truncate_size = object_size;
809 }
Yan, Zhengccca4e32013-06-02 18:40:23 +0800810 }
Yan, Zheng715e4cd2014-11-13 14:40:37 +0800811 osd_req_op_extent_init(req, which, opcode, objoff, objlen,
Yan, Zheng864e9192014-11-13 10:47:25 +0800812 truncate_size, truncate_seq);
Alex Eldera19dadf2013-03-13 20:50:01 -0500813 }
Alex Elderd18d1e22013-03-13 20:50:01 -0500814
Ilya Dryomov3c972c92014-01-27 17:40:20 +0200815 req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
Sage Weilf24e9982009-10-06 11:31:10 -0700816
Ilya Dryomov3c972c92014-01-27 17:40:20 +0200817 snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
Ilya Dryomov4295f222014-01-27 17:40:18 +0200818 "%llx.%08llx", vino.ino, objnum);
Ilya Dryomov3c972c92014-01-27 17:40:20 +0200819 req->r_base_oid.name_len = strlen(req->r_base_oid.name);
Alex Elderdbe0fc42013-02-15 22:10:17 -0600820
Sage Weilf24e9982009-10-06 11:31:10 -0700821 return req;
822}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700823EXPORT_SYMBOL(ceph_osdc_new_request);
Sage Weilf24e9982009-10-06 11:31:10 -0700824
825/*
826 * We keep osd requests in an rbtree, sorted by ->r_tid.
827 */
828static void __insert_request(struct ceph_osd_client *osdc,
829 struct ceph_osd_request *new)
830{
831 struct rb_node **p = &osdc->requests.rb_node;
832 struct rb_node *parent = NULL;
833 struct ceph_osd_request *req = NULL;
834
835 while (*p) {
836 parent = *p;
837 req = rb_entry(parent, struct ceph_osd_request, r_node);
838 if (new->r_tid < req->r_tid)
839 p = &(*p)->rb_left;
840 else if (new->r_tid > req->r_tid)
841 p = &(*p)->rb_right;
842 else
843 BUG();
844 }
845
846 rb_link_node(&new->r_node, parent, p);
847 rb_insert_color(&new->r_node, &osdc->requests);
848}
849
850static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
851 u64 tid)
852{
853 struct ceph_osd_request *req;
854 struct rb_node *n = osdc->requests.rb_node;
855
856 while (n) {
857 req = rb_entry(n, struct ceph_osd_request, r_node);
858 if (tid < req->r_tid)
859 n = n->rb_left;
860 else if (tid > req->r_tid)
861 n = n->rb_right;
862 else
863 return req;
864 }
865 return NULL;
866}
867
868static struct ceph_osd_request *
869__lookup_request_ge(struct ceph_osd_client *osdc,
870 u64 tid)
871{
872 struct ceph_osd_request *req;
873 struct rb_node *n = osdc->requests.rb_node;
874
875 while (n) {
876 req = rb_entry(n, struct ceph_osd_request, r_node);
877 if (tid < req->r_tid) {
878 if (!n->rb_left)
879 return req;
880 n = n->rb_left;
881 } else if (tid > req->r_tid) {
882 n = n->rb_right;
883 } else {
884 return req;
885 }
886 }
887 return NULL;
888}
889
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400890static void __kick_linger_request(struct ceph_osd_request *req)
891{
892 struct ceph_osd_client *osdc = req->r_osdc;
893 struct ceph_osd *osd = req->r_osd;
894
895 /*
896 * Linger requests need to be resent with a new tid to avoid
897 * the dup op detection logic on the OSDs. Achieve this with
898 * a re-register dance instead of open-coding.
899 */
900 ceph_osdc_get_request(req);
901 if (!list_empty(&req->r_linger_item))
902 __unregister_linger_request(osdc, req);
903 else
904 __unregister_request(osdc, req);
905 __register_request(osdc, req);
906 ceph_osdc_put_request(req);
907
908 /*
909 * Unless request has been registered as both normal and
910 * lingering, __unregister{,_linger}_request clears r_osd.
911 * However, here we need to preserve r_osd to make sure we
912 * requeue on the same OSD.
913 */
914 WARN_ON(req->r_osd || !osd);
915 req->r_osd = osd;
916
917 dout("%s requeueing %p tid %llu\n", __func__, req, req->r_tid);
918 __enqueue_request(req);
919}
920
Sage Weil6f6c7002011-01-17 20:34:08 -0800921/*
922 * Resubmit requests pending on the given osd.
923 */
924static void __kick_osd_requests(struct ceph_osd_client *osdc,
925 struct ceph_osd *osd)
926{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700927 struct ceph_osd_request *req, *nreq;
Alex Eldere02493c2013-03-25 18:16:11 -0500928 LIST_HEAD(resend);
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400929 LIST_HEAD(resend_linger);
Sage Weil6f6c7002011-01-17 20:34:08 -0800930 int err;
931
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400932 dout("%s osd%d\n", __func__, osd->o_osd);
Sage Weil6f6c7002011-01-17 20:34:08 -0800933 err = __reset_osd(osdc, osd);
Alex Elder685a7552012-12-07 09:57:58 -0600934 if (err)
Sage Weil6f6c7002011-01-17 20:34:08 -0800935 return;
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400936
Alex Eldere02493c2013-03-25 18:16:11 -0500937 /*
938 * Build up a list of requests to resend by traversing the
939 * osd's list of requests. Requests for a given object are
940 * sent in tid order, and that is also the order they're
941 * kept on this list. Therefore all requests that are in
942 * flight will be found first, followed by all requests that
943 * have not yet been sent. And to resend requests while
944 * preserving this order we will want to put any sent
945 * requests back on the front of the osd client's unsent
946 * list.
947 *
948 * So we build a separate ordered list of already-sent
949 * requests for the affected osd and splice it onto the
950 * front of the osd client's unsent list. Once we've seen a
951 * request that has not yet been sent we're done. Those
952 * requests are already sitting right where they belong.
953 */
Sage Weil6f6c7002011-01-17 20:34:08 -0800954 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
Alex Eldere02493c2013-03-25 18:16:11 -0500955 if (!req->r_sent)
956 break;
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400957
958 if (!req->r_linger) {
959 dout("%s requeueing %p tid %llu\n", __func__, req,
960 req->r_tid);
961 list_move_tail(&req->r_req_lru_item, &resend);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700962 req->r_flags |= CEPH_OSD_FLAG_RETRY;
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400963 } else {
964 list_move_tail(&req->r_req_lru_item, &resend_linger);
965 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700966 }
Alex Eldere02493c2013-03-25 18:16:11 -0500967 list_splice(&resend, &osdc->req_unsent);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700968
Alex Eldere02493c2013-03-25 18:16:11 -0500969 /*
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400970 * Both registered and not yet registered linger requests are
971 * enqueued with a new tid on the same OSD. We add/move them
972 * to req_unsent/o_requests at the end to keep things in tid
973 * order.
Alex Eldere02493c2013-03-25 18:16:11 -0500974 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700975 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
Ilya Dryomov1d0326b2014-06-20 14:14:41 +0400976 r_linger_osd_item) {
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400977 WARN_ON(!list_empty(&req->r_req_lru_item));
978 __kick_linger_request(req);
Sage Weil6f6c7002011-01-17 20:34:08 -0800979 }
Ilya Dryomov2cc61282014-09-03 14:41:45 +0400980
981 list_for_each_entry_safe(req, nreq, &resend_linger, r_req_lru_item)
982 __kick_linger_request(req);
Sage Weil6f6c7002011-01-17 20:34:08 -0800983}
984
Sage Weilf24e9982009-10-06 11:31:10 -0700985/*
Sage Weil81b024e2009-10-09 10:29:18 -0700986 * If the osd connection drops, we need to resubmit all requests.
Sage Weilf24e9982009-10-06 11:31:10 -0700987 */
988static void osd_reset(struct ceph_connection *con)
989{
990 struct ceph_osd *osd = con->private;
991 struct ceph_osd_client *osdc;
992
993 if (!osd)
994 return;
995 dout("osd_reset osd%d\n", osd->o_osd);
996 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -0700997 down_read(&osdc->map_sem);
Sage Weil83aff952012-11-28 12:28:24 -0800998 mutex_lock(&osdc->request_mutex);
999 __kick_osd_requests(osdc, osd);
Alex Elderf9d25192013-02-15 11:42:29 -06001000 __send_queued(osdc);
Sage Weil83aff952012-11-28 12:28:24 -08001001 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001002 up_read(&osdc->map_sem);
1003}
1004
1005/*
1006 * Track open sessions with osds.
1007 */
Alex Eldere10006f2012-05-26 23:26:43 -05001008static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
Sage Weilf24e9982009-10-06 11:31:10 -07001009{
1010 struct ceph_osd *osd;
1011
1012 osd = kzalloc(sizeof(*osd), GFP_NOFS);
1013 if (!osd)
1014 return NULL;
1015
1016 atomic_set(&osd->o_ref, 1);
1017 osd->o_osdc = osdc;
Alex Eldere10006f2012-05-26 23:26:43 -05001018 osd->o_osd = onum;
Alex Elderf4077312012-12-06 07:22:04 -06001019 RB_CLEAR_NODE(&osd->o_node);
Sage Weilf24e9982009-10-06 11:31:10 -07001020 INIT_LIST_HEAD(&osd->o_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001021 INIT_LIST_HEAD(&osd->o_linger_requests);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001022 INIT_LIST_HEAD(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001023 osd->o_incarnation = 1;
1024
Sage Weilb7a9e5d2012-06-27 12:24:08 -07001025 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001026
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001027 INIT_LIST_HEAD(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001028 return osd;
1029}
1030
1031static struct ceph_osd *get_osd(struct ceph_osd *osd)
1032{
1033 if (atomic_inc_not_zero(&osd->o_ref)) {
1034 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
1035 atomic_read(&osd->o_ref));
1036 return osd;
1037 } else {
1038 dout("get_osd %p FAIL\n", osd);
1039 return NULL;
1040 }
1041}
1042
1043static void put_osd(struct ceph_osd *osd)
1044{
1045 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
1046 atomic_read(&osd->o_ref) - 1);
Ilya Dryomovb28ec2f2015-02-16 11:49:42 +03001047 if (atomic_dec_and_test(&osd->o_ref)) {
Sage Weil79494d12010-05-27 14:15:49 -07001048 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
1049
Ilya Dryomovb28ec2f2015-02-16 11:49:42 +03001050 if (osd->o_auth.authorizer)
1051 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
Sage Weilf24e9982009-10-06 11:31:10 -07001052 kfree(osd);
Sage Weil79494d12010-05-27 14:15:49 -07001053 }
Sage Weilf24e9982009-10-06 11:31:10 -07001054}
1055
1056/*
1057 * remove an osd from our map
1058 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001059static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -07001060{
Ilya Dryomov7eb71e02015-02-17 19:37:15 +03001061 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
Ilya Dryomovcc9f1f52014-11-05 19:33:44 +03001062 WARN_ON(!list_empty(&osd->o_requests));
1063 WARN_ON(!list_empty(&osd->o_linger_requests));
Ilya Dryomov7c6e6fc2014-06-18 13:02:12 +04001064
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001065 list_del_init(&osd->o_osd_lru);
Ilya Dryomov7eb71e02015-02-17 19:37:15 +03001066 rb_erase(&osd->o_node, &osdc->osds);
1067 RB_CLEAR_NODE(&osd->o_node);
1068}
1069
1070static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1071{
1072 dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
1073
1074 if (!RB_EMPTY_NODE(&osd->o_node)) {
1075 ceph_con_close(&osd->o_con);
1076 __remove_osd(osdc, osd);
1077 put_osd(osd);
1078 }
Sage Weilf24e9982009-10-06 11:31:10 -07001079}
1080
Sage Weilaca420b2011-08-31 14:45:53 -07001081static void remove_all_osds(struct ceph_osd_client *osdc)
1082{
Jiaju Zhang048a9d22012-07-20 08:18:36 -05001083 dout("%s %p\n", __func__, osdc);
Sage Weilaca420b2011-08-31 14:45:53 -07001084 mutex_lock(&osdc->request_mutex);
1085 while (!RB_EMPTY_ROOT(&osdc->osds)) {
1086 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
1087 struct ceph_osd, o_node);
Ilya Dryomov7eb71e02015-02-17 19:37:15 +03001088 remove_osd(osdc, osd);
Sage Weilaca420b2011-08-31 14:45:53 -07001089 }
1090 mutex_unlock(&osdc->request_mutex);
1091}
1092
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001093static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1094 struct ceph_osd *osd)
1095{
Ilya Dryomovbbf37ec2014-06-20 14:14:41 +04001096 dout("%s %p\n", __func__, osd);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001097 BUG_ON(!list_empty(&osd->o_osd_lru));
Ilya Dryomovbbf37ec2014-06-20 14:14:41 +04001098
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001099 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
Ilya Dryomova319bf52015-05-15 12:02:17 +03001100 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001101}
1102
Ilya Dryomovbbf37ec2014-06-20 14:14:41 +04001103static void maybe_move_osd_to_lru(struct ceph_osd_client *osdc,
1104 struct ceph_osd *osd)
1105{
1106 dout("%s %p\n", __func__, osd);
1107
1108 if (list_empty(&osd->o_requests) &&
1109 list_empty(&osd->o_linger_requests))
1110 __move_osd_to_lru(osdc, osd);
1111}
1112
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001113static void __remove_osd_from_lru(struct ceph_osd *osd)
1114{
1115 dout("__remove_osd_from_lru %p\n", osd);
1116 if (!list_empty(&osd->o_osd_lru))
1117 list_del_init(&osd->o_osd_lru);
1118}
1119
Sage Weilaca420b2011-08-31 14:45:53 -07001120static void remove_old_osds(struct ceph_osd_client *osdc)
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001121{
1122 struct ceph_osd *osd, *nosd;
1123
1124 dout("__remove_old_osds %p\n", osdc);
1125 mutex_lock(&osdc->request_mutex);
1126 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
Sage Weilaca420b2011-08-31 14:45:53 -07001127 if (time_before(jiffies, osd->lru_ttl))
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001128 break;
Ilya Dryomov7eb71e02015-02-17 19:37:15 +03001129 remove_osd(osdc, osd);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001130 }
1131 mutex_unlock(&osdc->request_mutex);
1132}
1133
Sage Weilf24e9982009-10-06 11:31:10 -07001134/*
1135 * reset osd connect
1136 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001137static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -07001138{
Alex Elderc3acb182012-12-07 09:57:58 -06001139 struct ceph_entity_addr *peer_addr;
Sage Weilf24e9982009-10-06 11:31:10 -07001140
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001141 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001142 if (list_empty(&osd->o_requests) &&
1143 list_empty(&osd->o_linger_requests)) {
Ilya Dryomov7eb71e02015-02-17 19:37:15 +03001144 remove_osd(osdc, osd);
Alex Elderc3acb182012-12-07 09:57:58 -06001145 return -ENODEV;
1146 }
1147
1148 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1149 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1150 !ceph_con_opened(&osd->o_con)) {
1151 struct ceph_osd_request *req;
1152
Ilya Dryomov0b4af2e2014-01-16 19:18:27 +02001153 dout("osd addr hasn't changed and connection never opened, "
1154 "letting msgr retry\n");
Sage Weil87b315a2010-03-22 14:51:18 -07001155 /* touch each r_stamp for handle_timeout()'s benfit */
1156 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1157 req->r_stamp = jiffies;
Alex Elderc3acb182012-12-07 09:57:58 -06001158
1159 return -EAGAIN;
Sage Weilf24e9982009-10-06 11:31:10 -07001160 }
Alex Elderc3acb182012-12-07 09:57:58 -06001161
1162 ceph_con_close(&osd->o_con);
1163 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1164 osd->o_incarnation++;
1165
1166 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001167}
1168
1169static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1170{
1171 struct rb_node **p = &osdc->osds.rb_node;
1172 struct rb_node *parent = NULL;
1173 struct ceph_osd *osd = NULL;
1174
Sage Weilaca420b2011-08-31 14:45:53 -07001175 dout("__insert_osd %p osd%d\n", new, new->o_osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001176 while (*p) {
1177 parent = *p;
1178 osd = rb_entry(parent, struct ceph_osd, o_node);
1179 if (new->o_osd < osd->o_osd)
1180 p = &(*p)->rb_left;
1181 else if (new->o_osd > osd->o_osd)
1182 p = &(*p)->rb_right;
1183 else
1184 BUG();
1185 }
1186
1187 rb_link_node(&new->o_node, parent, p);
1188 rb_insert_color(&new->o_node, &osdc->osds);
1189}
1190
1191static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1192{
1193 struct ceph_osd *osd;
1194 struct rb_node *n = osdc->osds.rb_node;
1195
1196 while (n) {
1197 osd = rb_entry(n, struct ceph_osd, o_node);
1198 if (o < osd->o_osd)
1199 n = n->rb_left;
1200 else if (o > osd->o_osd)
1201 n = n->rb_right;
1202 else
1203 return osd;
1204 }
1205 return NULL;
1206}
1207
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001208static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1209{
1210 schedule_delayed_work(&osdc->timeout_work,
Ilya Dryomova319bf52015-05-15 12:02:17 +03001211 osdc->client->options->osd_keepalive_timeout);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001212}
1213
1214static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1215{
1216 cancel_delayed_work(&osdc->timeout_work);
1217}
Sage Weilf24e9982009-10-06 11:31:10 -07001218
1219/*
1220 * Register request, assign tid. If this is the first request, set up
1221 * the timeout event.
1222 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001223static void __register_request(struct ceph_osd_client *osdc,
1224 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001225{
Sage Weilf24e9982009-10-06 11:31:10 -07001226 req->r_tid = ++osdc->last_tid;
Sage Weil6df058c2009-12-22 11:24:33 -08001227 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
Sage Weil77f38e02011-04-06 09:09:16 -07001228 dout("__register_request %p tid %lld\n", req, req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07001229 __insert_request(osdc, req);
1230 ceph_osdc_get_request(req);
1231 osdc->num_requests++;
Sage Weilf24e9982009-10-06 11:31:10 -07001232 if (osdc->num_requests == 1) {
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001233 dout(" first request, scheduling timeout\n");
1234 __schedule_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001235 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001236}
1237
Sage Weilf24e9982009-10-06 11:31:10 -07001238/*
1239 * called under osdc->request_mutex
1240 */
1241static void __unregister_request(struct ceph_osd_client *osdc,
1242 struct ceph_osd_request *req)
1243{
Sage Weil35f9f8a2012-05-16 15:16:38 -05001244 if (RB_EMPTY_NODE(&req->r_node)) {
1245 dout("__unregister_request %p tid %lld not registered\n",
1246 req, req->r_tid);
1247 return;
1248 }
1249
Sage Weilf24e9982009-10-06 11:31:10 -07001250 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1251 rb_erase(&req->r_node, &osdc->requests);
Ilya Dryomov6562d662014-06-20 14:14:42 +04001252 RB_CLEAR_NODE(&req->r_node);
Sage Weilf24e9982009-10-06 11:31:10 -07001253 osdc->num_requests--;
1254
Sage Weil0ba64782009-10-08 16:57:16 -07001255 if (req->r_osd) {
1256 /* make sure the original request isn't in flight. */
Alex Elder6740a842012-06-01 14:56:43 -05001257 ceph_msg_revoke(req->r_request);
Sage Weil0ba64782009-10-08 16:57:16 -07001258
1259 list_del_init(&req->r_osd_item);
Ilya Dryomovbbf37ec2014-06-20 14:14:41 +04001260 maybe_move_osd_to_lru(osdc, req->r_osd);
Ilya Dryomov4f234092014-06-20 18:29:20 +04001261 if (list_empty(&req->r_linger_osd_item))
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001262 req->r_osd = NULL;
Sage Weil0ba64782009-10-08 16:57:16 -07001263 }
Sage Weilf24e9982009-10-06 11:31:10 -07001264
Alex Elder7d5f2482012-11-29 08:37:03 -06001265 list_del_init(&req->r_req_lru_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001266 ceph_osdc_put_request(req);
1267
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001268 if (osdc->num_requests == 0) {
1269 dout(" no requests, canceling timeout\n");
1270 __cancel_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001271 }
1272}
1273
1274/*
1275 * Cancel a previously queued request message
1276 */
1277static void __cancel_request(struct ceph_osd_request *req)
1278{
Sage Weil6bc18872010-09-27 10:18:52 -07001279 if (req->r_sent && req->r_osd) {
Alex Elder6740a842012-06-01 14:56:43 -05001280 ceph_msg_revoke(req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001281 req->r_sent = 0;
1282 }
1283}
1284
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001285static void __register_linger_request(struct ceph_osd_client *osdc,
1286 struct ceph_osd_request *req)
1287{
Ilya Dryomovaf593062014-06-20 18:29:20 +04001288 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
1289 WARN_ON(!req->r_linger);
1290
Alex Elder96e4dac2013-05-22 20:54:25 -05001291 ceph_osdc_get_request(req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001292 list_add_tail(&req->r_linger_item, &osdc->req_linger);
Sage Weil6194ea82012-07-30 16:19:28 -07001293 if (req->r_osd)
Ilya Dryomov1d0326b2014-06-20 14:14:41 +04001294 list_add_tail(&req->r_linger_osd_item,
Sage Weil6194ea82012-07-30 16:19:28 -07001295 &req->r_osd->o_linger_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001296}
1297
1298static void __unregister_linger_request(struct ceph_osd_client *osdc,
1299 struct ceph_osd_request *req)
1300{
Ilya Dryomovaf593062014-06-20 18:29:20 +04001301 WARN_ON(!req->r_linger);
1302
1303 if (list_empty(&req->r_linger_item)) {
1304 dout("%s %p tid %llu not registered\n", __func__, req,
1305 req->r_tid);
1306 return;
1307 }
1308
1309 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
Alex Elder61c74032012-12-06 09:37:23 -06001310 list_del_init(&req->r_linger_item);
Ilya Dryomovaf593062014-06-20 18:29:20 +04001311
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001312 if (req->r_osd) {
Ilya Dryomov1d0326b2014-06-20 14:14:41 +04001313 list_del_init(&req->r_linger_osd_item);
Ilya Dryomovbbf37ec2014-06-20 14:14:41 +04001314 maybe_move_osd_to_lru(osdc, req->r_osd);
Sage Weilfbdb9192011-03-29 12:11:06 -07001315 if (list_empty(&req->r_osd_item))
1316 req->r_osd = NULL;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001317 }
Alex Elder96e4dac2013-05-22 20:54:25 -05001318 ceph_osdc_put_request(req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001319}
1320
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001321void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1322 struct ceph_osd_request *req)
1323{
1324 if (!req->r_linger) {
1325 dout("set_request_linger %p\n", req);
1326 req->r_linger = 1;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001327 }
1328}
1329EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1330
Sage Weilf24e9982009-10-06 11:31:10 -07001331/*
Josh Durgind29adb32013-12-02 19:11:48 -08001332 * Returns whether a request should be blocked from being sent
1333 * based on the current osdmap and osd_client settings.
1334 *
1335 * Caller should hold map_sem for read.
1336 */
1337static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1338 struct ceph_osd_request *req)
1339{
1340 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1341 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1342 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1343 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1344 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1345}
1346
1347/*
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001348 * Calculate mapping of a request to a PG. Takes tiering into account.
1349 */
1350static int __calc_request_pg(struct ceph_osdmap *osdmap,
1351 struct ceph_osd_request *req,
1352 struct ceph_pg *pg_out)
1353{
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001354 bool need_check_tiering;
1355
1356 need_check_tiering = false;
1357 if (req->r_target_oloc.pool == -1) {
1358 req->r_target_oloc = req->r_base_oloc; /* struct */
1359 need_check_tiering = true;
1360 }
1361 if (req->r_target_oid.name_len == 0) {
1362 ceph_oid_copy(&req->r_target_oid, &req->r_base_oid);
1363 need_check_tiering = true;
1364 }
1365
1366 if (need_check_tiering &&
1367 (req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001368 struct ceph_pg_pool_info *pi;
1369
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001370 pi = ceph_pg_pool_by_id(osdmap, req->r_target_oloc.pool);
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001371 if (pi) {
1372 if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1373 pi->read_tier >= 0)
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001374 req->r_target_oloc.pool = pi->read_tier;
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001375 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1376 pi->write_tier >= 0)
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001377 req->r_target_oloc.pool = pi->write_tier;
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001378 }
1379 /* !pi is caught in ceph_oloc_oid_to_pg() */
1380 }
1381
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001382 return ceph_oloc_oid_to_pg(osdmap, &req->r_target_oloc,
1383 &req->r_target_oid, pg_out);
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001384}
1385
Ilya Dryomovf671b582014-09-02 13:40:33 +04001386static void __enqueue_request(struct ceph_osd_request *req)
1387{
1388 struct ceph_osd_client *osdc = req->r_osdc;
1389
1390 dout("%s %p tid %llu to osd%d\n", __func__, req, req->r_tid,
1391 req->r_osd ? req->r_osd->o_osd : -1);
1392
1393 if (req->r_osd) {
1394 __remove_osd_from_lru(req->r_osd);
1395 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1396 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1397 } else {
1398 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1399 }
1400}
1401
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001402/*
Sage Weilf24e9982009-10-06 11:31:10 -07001403 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1404 * (as needed), and set the request r_osd appropriately. If there is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001405 * no up osd, set r_osd to NULL. Move the request to the appropriate list
Sage Weil6f6c7002011-01-17 20:34:08 -08001406 * (unsent, homeless) or leave on in-flight lru.
Sage Weilf24e9982009-10-06 11:31:10 -07001407 *
1408 * Return 0 if unchanged, 1 if changed, or negative on error.
1409 *
1410 * Caller should hold map_sem for read and request_mutex.
1411 */
Sage Weil6f6c7002011-01-17 20:34:08 -08001412static int __map_request(struct ceph_osd_client *osdc,
Sage Weil38d64532011-10-14 13:33:55 -07001413 struct ceph_osd_request *req, int force_resend)
Sage Weilf24e9982009-10-06 11:31:10 -07001414{
Sage Weil5b191d92013-02-23 10:38:16 -08001415 struct ceph_pg pgid;
Sage Weild85b7052010-05-10 10:24:48 -07001416 int acting[CEPH_PG_MAX_SIZE];
Ilya Dryomov8008ab12014-03-24 17:12:48 +02001417 int num, o;
Sage Weilf24e9982009-10-06 11:31:10 -07001418 int err;
Josh Durgind29adb32013-12-02 19:11:48 -08001419 bool was_paused;
Sage Weilf24e9982009-10-06 11:31:10 -07001420
Sage Weil6f6c7002011-01-17 20:34:08 -08001421 dout("map_request %p tid %lld\n", req, req->r_tid);
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001422
1423 err = __calc_request_pg(osdc->osdmap, req, &pgid);
Sage Weil6f6c7002011-01-17 20:34:08 -08001424 if (err) {
1425 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilf24e9982009-10-06 11:31:10 -07001426 return err;
Sage Weil6f6c7002011-01-17 20:34:08 -08001427 }
Sage Weil7740a422010-01-08 15:58:25 -08001428 req->r_pgid = pgid;
1429
Ilya Dryomov8008ab12014-03-24 17:12:48 +02001430 num = ceph_calc_pg_acting(osdc->osdmap, pgid, acting, &o);
1431 if (num < 0)
1432 num = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001433
Josh Durgind29adb32013-12-02 19:11:48 -08001434 was_paused = req->r_paused;
1435 req->r_paused = __req_should_be_paused(osdc, req);
1436 if (was_paused && !req->r_paused)
1437 force_resend = 1;
1438
Sage Weil38d64532011-10-14 13:33:55 -07001439 if ((!force_resend &&
1440 req->r_osd && req->r_osd->o_osd == o &&
Sage Weild85b7052010-05-10 10:24:48 -07001441 req->r_sent >= req->r_osd->o_incarnation &&
1442 req->r_num_pg_osds == num &&
1443 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
Josh Durgind29adb32013-12-02 19:11:48 -08001444 (req->r_osd == NULL && o == -1) ||
1445 req->r_paused)
Sage Weilf24e9982009-10-06 11:31:10 -07001446 return 0; /* no change */
1447
Sage Weil5b191d92013-02-23 10:38:16 -08001448 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1449 req->r_tid, pgid.pool, pgid.seed, o,
Sage Weilf24e9982009-10-06 11:31:10 -07001450 req->r_osd ? req->r_osd->o_osd : -1);
1451
Sage Weild85b7052010-05-10 10:24:48 -07001452 /* record full pg acting set */
1453 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1454 req->r_num_pg_osds = num;
1455
Sage Weilf24e9982009-10-06 11:31:10 -07001456 if (req->r_osd) {
1457 __cancel_request(req);
1458 list_del_init(&req->r_osd_item);
Ilya Dryomova390de02014-11-04 18:32:14 +03001459 list_del_init(&req->r_linger_osd_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001460 req->r_osd = NULL;
1461 }
1462
1463 req->r_osd = __lookup_osd(osdc, o);
1464 if (!req->r_osd && o >= 0) {
Sage Weilc99eb1c2010-02-26 09:37:33 -08001465 err = -ENOMEM;
Alex Eldere10006f2012-05-26 23:26:43 -05001466 req->r_osd = create_osd(osdc, o);
Sage Weil6f6c7002011-01-17 20:34:08 -08001467 if (!req->r_osd) {
1468 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilc99eb1c2010-02-26 09:37:33 -08001469 goto out;
Sage Weil6f6c7002011-01-17 20:34:08 -08001470 }
Sage Weilf24e9982009-10-06 11:31:10 -07001471
Sage Weil6f6c7002011-01-17 20:34:08 -08001472 dout("map_request osd %p is osd%d\n", req->r_osd, o);
Sage Weilf24e9982009-10-06 11:31:10 -07001473 __insert_osd(osdc, req->r_osd);
1474
Sage Weilb7a9e5d2012-06-27 12:24:08 -07001475 ceph_con_open(&req->r_osd->o_con,
1476 CEPH_ENTITY_TYPE_OSD, o,
1477 &osdc->osdmap->osd_addr[o]);
Sage Weilf24e9982009-10-06 11:31:10 -07001478 }
1479
Ilya Dryomovf671b582014-09-02 13:40:33 +04001480 __enqueue_request(req);
Sage Weild85b7052010-05-10 10:24:48 -07001481 err = 1; /* osd or pg changed */
Sage Weilf24e9982009-10-06 11:31:10 -07001482
1483out:
Sage Weilf24e9982009-10-06 11:31:10 -07001484 return err;
1485}
1486
1487/*
1488 * caller should hold map_sem (for read) and request_mutex
1489 */
Sage Weil56e925b2012-01-03 12:34:34 -08001490static void __send_request(struct ceph_osd_client *osdc,
1491 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001492{
Sage Weil1b83bef2013-02-25 16:11:12 -08001493 void *p;
Sage Weilf24e9982009-10-06 11:31:10 -07001494
Sage Weil1b83bef2013-02-25 16:11:12 -08001495 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1496 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1497 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
Sage Weilf24e9982009-10-06 11:31:10 -07001498
Sage Weil1b83bef2013-02-25 16:11:12 -08001499 /* fill in message content that changes each time we send it */
1500 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1501 put_unaligned_le32(req->r_flags, req->r_request_flags);
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001502 put_unaligned_le64(req->r_target_oloc.pool, req->r_request_pool);
Sage Weil1b83bef2013-02-25 16:11:12 -08001503 p = req->r_request_pgid;
1504 ceph_encode_64(&p, req->r_pgid.pool);
1505 ceph_encode_32(&p, req->r_pgid.seed);
1506 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1507 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1508 sizeof(req->r_reassert_version));
Sage Weil2169aea2013-02-25 16:13:08 -08001509
Sage Weil3dd72fc2010-03-22 14:42:30 -07001510 req->r_stamp = jiffies;
Henry C Chang07a27e22010-08-22 21:34:27 -07001511 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001512
1513 ceph_msg_get(req->r_request); /* send consumes a ref */
Alex Elder26be8802013-04-15 11:20:42 -05001514
Sage Weilf24e9982009-10-06 11:31:10 -07001515 req->r_sent = req->r_osd->o_incarnation;
Alex Elder26be8802013-04-15 11:20:42 -05001516
1517 ceph_con_send(&req->r_osd->o_con, req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001518}
1519
1520/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001521 * Send any requests in the queue (req_unsent).
1522 */
Alex Elderf9d25192013-02-15 11:42:29 -06001523static void __send_queued(struct ceph_osd_client *osdc)
Sage Weil6f6c7002011-01-17 20:34:08 -08001524{
1525 struct ceph_osd_request *req, *tmp;
1526
Alex Elderf9d25192013-02-15 11:42:29 -06001527 dout("__send_queued\n");
1528 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
Sage Weil6f6c7002011-01-17 20:34:08 -08001529 __send_request(osdc, req);
Sage Weil6f6c7002011-01-17 20:34:08 -08001530}
1531
1532/*
Ilya Dryomov0bbfdfe2014-01-31 19:33:39 +02001533 * Caller should hold map_sem for read and request_mutex.
1534 */
1535static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1536 struct ceph_osd_request *req,
1537 bool nofail)
1538{
1539 int rc;
1540
1541 __register_request(osdc, req);
1542 req->r_sent = 0;
1543 req->r_got_reply = 0;
1544 rc = __map_request(osdc, req, 0);
1545 if (rc < 0) {
1546 if (nofail) {
1547 dout("osdc_start_request failed map, "
1548 " will retry %lld\n", req->r_tid);
1549 rc = 0;
1550 } else {
1551 __unregister_request(osdc, req);
1552 }
1553 return rc;
1554 }
1555
1556 if (req->r_osd == NULL) {
1557 dout("send_request %p no up osds in pg\n", req);
1558 ceph_monc_request_next_osdmap(&osdc->client->monc);
1559 } else {
1560 __send_queued(osdc);
1561 }
1562
1563 return 0;
1564}
1565
1566/*
Sage Weilf24e9982009-10-06 11:31:10 -07001567 * Timeout callback, called every N seconds when 1 or more osd
1568 * requests has been active for more than N seconds. When this
1569 * happens, we ping all OSDs with requests who have timed out to
1570 * ensure any communications channel reset is detected. Reset the
1571 * request timeouts another N seconds in the future as we go.
1572 * Reschedule the timeout event another N seconds in future (unless
1573 * there are no open requests).
1574 */
1575static void handle_timeout(struct work_struct *work)
1576{
1577 struct ceph_osd_client *osdc =
1578 container_of(work, struct ceph_osd_client, timeout_work.work);
Ilya Dryomova319bf52015-05-15 12:02:17 +03001579 struct ceph_options *opts = osdc->client->options;
Sage Weil83aff952012-11-28 12:28:24 -08001580 struct ceph_osd_request *req;
Sage Weilf24e9982009-10-06 11:31:10 -07001581 struct ceph_osd *osd;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001582 struct list_head slow_osds;
Sage Weilf24e9982009-10-06 11:31:10 -07001583 dout("timeout\n");
1584 down_read(&osdc->map_sem);
1585
1586 ceph_monc_request_next_osdmap(&osdc->client->monc);
1587
1588 mutex_lock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001589
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001590 /*
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001591 * ping osds that are a bit slow. this ensures that if there
1592 * is a break in the TCP connection we will notice, and reopen
1593 * a connection with that osd (from the fault callback).
1594 */
1595 INIT_LIST_HEAD(&slow_osds);
1596 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
Ilya Dryomova319bf52015-05-15 12:02:17 +03001597 if (time_before(jiffies,
1598 req->r_stamp + opts->osd_keepalive_timeout))
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001599 break;
1600
1601 osd = req->r_osd;
1602 BUG_ON(!osd);
1603 dout(" tid %llu is slow, will send keepalive on osd%d\n",
Sage Weilf24e9982009-10-06 11:31:10 -07001604 req->r_tid, osd->o_osd);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001605 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1606 }
1607 while (!list_empty(&slow_osds)) {
1608 osd = list_entry(slow_osds.next, struct ceph_osd,
1609 o_keepalive_item);
1610 list_del_init(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001611 ceph_con_keepalive(&osd->o_con);
1612 }
1613
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001614 __schedule_osd_timeout(osdc);
Alex Elderf9d25192013-02-15 11:42:29 -06001615 __send_queued(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001616 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001617 up_read(&osdc->map_sem);
1618}
1619
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001620static void handle_osds_timeout(struct work_struct *work)
1621{
1622 struct ceph_osd_client *osdc =
1623 container_of(work, struct ceph_osd_client,
1624 osds_timeout_work.work);
Ilya Dryomova319bf52015-05-15 12:02:17 +03001625 unsigned long delay = osdc->client->options->osd_idle_ttl / 4;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001626
1627 dout("osds timeout\n");
1628 down_read(&osdc->map_sem);
Sage Weilaca420b2011-08-31 14:45:53 -07001629 remove_old_osds(osdc);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001630 up_read(&osdc->map_sem);
1631
1632 schedule_delayed_work(&osdc->osds_timeout_work,
1633 round_jiffies_relative(delay));
1634}
1635
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001636static int ceph_oloc_decode(void **p, void *end,
1637 struct ceph_object_locator *oloc)
1638{
1639 u8 struct_v, struct_cv;
1640 u32 len;
1641 void *struct_end;
1642 int ret = 0;
1643
1644 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1645 struct_v = ceph_decode_8(p);
1646 struct_cv = ceph_decode_8(p);
1647 if (struct_v < 3) {
1648 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1649 struct_v, struct_cv);
1650 goto e_inval;
1651 }
1652 if (struct_cv > 6) {
1653 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1654 struct_v, struct_cv);
1655 goto e_inval;
1656 }
1657 len = ceph_decode_32(p);
1658 ceph_decode_need(p, end, len, e_inval);
1659 struct_end = *p + len;
1660
1661 oloc->pool = ceph_decode_64(p);
1662 *p += 4; /* skip preferred */
1663
1664 len = ceph_decode_32(p);
1665 if (len > 0) {
1666 pr_warn("ceph_object_locator::key is set\n");
1667 goto e_inval;
1668 }
1669
1670 if (struct_v >= 5) {
1671 len = ceph_decode_32(p);
1672 if (len > 0) {
1673 pr_warn("ceph_object_locator::nspace is set\n");
1674 goto e_inval;
1675 }
1676 }
1677
1678 if (struct_v >= 6) {
1679 s64 hash = ceph_decode_64(p);
1680 if (hash != -1) {
1681 pr_warn("ceph_object_locator::hash is set\n");
1682 goto e_inval;
1683 }
1684 }
1685
1686 /* skip the rest */
1687 *p = struct_end;
1688out:
1689 return ret;
1690
1691e_inval:
1692 ret = -EINVAL;
1693 goto out;
1694}
1695
1696static int ceph_redirect_decode(void **p, void *end,
1697 struct ceph_request_redirect *redir)
1698{
1699 u8 struct_v, struct_cv;
1700 u32 len;
1701 void *struct_end;
1702 int ret;
1703
1704 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1705 struct_v = ceph_decode_8(p);
1706 struct_cv = ceph_decode_8(p);
1707 if (struct_cv > 1) {
1708 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1709 struct_v, struct_cv);
1710 goto e_inval;
1711 }
1712 len = ceph_decode_32(p);
1713 ceph_decode_need(p, end, len, e_inval);
1714 struct_end = *p + len;
1715
1716 ret = ceph_oloc_decode(p, end, &redir->oloc);
1717 if (ret)
1718 goto out;
1719
1720 len = ceph_decode_32(p);
1721 if (len > 0) {
1722 pr_warn("ceph_request_redirect::object_name is set\n");
1723 goto e_inval;
1724 }
1725
1726 len = ceph_decode_32(p);
1727 *p += len; /* skip osd_instructions */
1728
1729 /* skip the rest */
1730 *p = struct_end;
1731out:
1732 return ret;
1733
1734e_inval:
1735 ret = -EINVAL;
1736 goto out;
1737}
1738
Sage Weil25845472011-06-03 09:37:09 -07001739static void complete_request(struct ceph_osd_request *req)
1740{
Sage Weil25845472011-06-03 09:37:09 -07001741 complete_all(&req->r_safe_completion); /* fsync waiter */
1742}
1743
Sage Weilf24e9982009-10-06 11:31:10 -07001744/*
1745 * handle osd op reply. either call the callback if it is specified,
1746 * or do the completion to wake up the waiting thread.
1747 */
Sage Weil350b1c32009-12-22 10:45:45 -08001748static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1749 struct ceph_connection *con)
Sage Weilf24e9982009-10-06 11:31:10 -07001750{
Sage Weil1b83bef2013-02-25 16:11:12 -08001751 void *p, *end;
Sage Weilf24e9982009-10-06 11:31:10 -07001752 struct ceph_osd_request *req;
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001753 struct ceph_request_redirect redir;
Sage Weilf24e9982009-10-06 11:31:10 -07001754 u64 tid;
Sage Weil1b83bef2013-02-25 16:11:12 -08001755 int object_len;
Alex Elder79528732013-04-03 21:32:51 -05001756 unsigned int numops;
1757 int payload_len, flags;
Sage Weil0ceed5d2010-05-11 09:53:18 -07001758 s32 result;
Sage Weil1b83bef2013-02-25 16:11:12 -08001759 s32 retry_attempt;
1760 struct ceph_pg pg;
1761 int err;
1762 u32 reassert_epoch;
1763 u64 reassert_version;
1764 u32 osdmap_epoch;
Alex Elder0d5af162013-02-27 10:26:25 -06001765 int already_completed;
Alex Elder9fc6e062013-04-03 01:28:57 -05001766 u32 bytes;
Alex Elder79528732013-04-03 21:32:51 -05001767 unsigned int i;
Sage Weilf24e9982009-10-06 11:31:10 -07001768
Sage Weil6df058c2009-12-22 11:24:33 -08001769 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil1b83bef2013-02-25 16:11:12 -08001770 dout("handle_reply %p tid %llu\n", msg, tid);
1771
1772 p = msg->front.iov_base;
1773 end = p + msg->front.iov_len;
1774
1775 ceph_decode_need(&p, end, 4, bad);
1776 object_len = ceph_decode_32(&p);
1777 ceph_decode_need(&p, end, object_len, bad);
1778 p += object_len;
1779
Alex Elderef4859d2013-04-01 18:58:26 -05001780 err = ceph_decode_pgid(&p, end, &pg);
Sage Weil1b83bef2013-02-25 16:11:12 -08001781 if (err)
Sage Weilf24e9982009-10-06 11:31:10 -07001782 goto bad;
Sage Weil1b83bef2013-02-25 16:11:12 -08001783
1784 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1785 flags = ceph_decode_64(&p);
1786 result = ceph_decode_32(&p);
1787 reassert_epoch = ceph_decode_32(&p);
1788 reassert_version = ceph_decode_64(&p);
1789 osdmap_epoch = ceph_decode_32(&p);
1790
Sage Weilf24e9982009-10-06 11:31:10 -07001791 /* lookup */
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001792 down_read(&osdc->map_sem);
Sage Weilf24e9982009-10-06 11:31:10 -07001793 mutex_lock(&osdc->request_mutex);
1794 req = __lookup_request(osdc, tid);
1795 if (req == NULL) {
1796 dout("handle_reply tid %llu dne\n", tid);
Alex Elder8058fd42013-04-01 18:58:26 -05001797 goto bad_mutex;
Sage Weilf24e9982009-10-06 11:31:10 -07001798 }
1799 ceph_osdc_get_request(req);
Sage Weil1b83bef2013-02-25 16:11:12 -08001800
1801 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1802 req, result);
1803
Dan Carpenter18741192013-08-15 08:51:58 +03001804 ceph_decode_need(&p, end, 4, bad_put);
Sage Weil1b83bef2013-02-25 16:11:12 -08001805 numops = ceph_decode_32(&p);
1806 if (numops > CEPH_OSD_MAX_OP)
1807 goto bad_put;
1808 if (numops != req->r_num_ops)
1809 goto bad_put;
1810 payload_len = 0;
Dan Carpenter18741192013-08-15 08:51:58 +03001811 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
Sage Weil1b83bef2013-02-25 16:11:12 -08001812 for (i = 0; i < numops; i++) {
1813 struct ceph_osd_op *op = p;
1814 int len;
1815
1816 len = le32_to_cpu(op->payload_len);
1817 req->r_reply_op_len[i] = len;
1818 dout(" op %d has %d bytes\n", i, len);
1819 payload_len += len;
1820 p += sizeof(*op);
1821 }
Alex Elder9fc6e062013-04-03 01:28:57 -05001822 bytes = le32_to_cpu(msg->hdr.data_len);
1823 if (payload_len != bytes) {
Joe Perchesb9a67892014-09-09 21:17:29 -07001824 pr_warn("sum of op payload lens %d != data_len %d\n",
1825 payload_len, bytes);
Sage Weil1b83bef2013-02-25 16:11:12 -08001826 goto bad_put;
1827 }
1828
Dan Carpenter18741192013-08-15 08:51:58 +03001829 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
Sage Weil1b83bef2013-02-25 16:11:12 -08001830 retry_attempt = ceph_decode_32(&p);
1831 for (i = 0; i < numops; i++)
1832 req->r_reply_op_result[i] = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001833
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001834 if (le16_to_cpu(msg->hdr.version) >= 6) {
1835 p += 8 + 4; /* skip replay_version */
1836 p += 8; /* skip user_version */
1837
1838 err = ceph_redirect_decode(&p, end, &redir);
1839 if (err)
1840 goto bad_put;
1841 } else {
1842 redir.oloc.pool = -1;
1843 }
1844
1845 if (redir.oloc.pool != -1) {
1846 dout("redirect pool %lld\n", redir.oloc.pool);
1847
1848 __unregister_request(osdc, req);
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001849
1850 req->r_target_oloc = redir.oloc; /* struct */
1851
1852 /*
1853 * Start redirect requests with nofail=true. If
1854 * mapping fails, request will end up on the notarget
1855 * list, waiting for the new osdmap (which can take
1856 * a while), even though the original request mapped
1857 * successfully. In the future we might want to follow
1858 * original request's nofail setting here.
1859 */
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001860 err = __ceph_osdc_start_request(osdc, req, true);
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001861 BUG_ON(err);
1862
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001863 goto out_unlock;
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001864 }
1865
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001866 already_completed = req->r_got_reply;
Sage Weilf24e9982009-10-06 11:31:10 -07001867 if (!req->r_got_reply) {
Sage Weil1b83bef2013-02-25 16:11:12 -08001868 req->r_result = result;
Sage Weilf24e9982009-10-06 11:31:10 -07001869 dout("handle_reply result %d bytes %d\n", req->r_result,
1870 bytes);
1871 if (req->r_result == 0)
1872 req->r_result = bytes;
1873
1874 /* in case this is a write and we need to replay, */
Sage Weil1b83bef2013-02-25 16:11:12 -08001875 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1876 req->r_reassert_version.version = cpu_to_le64(reassert_version);
Sage Weilf24e9982009-10-06 11:31:10 -07001877
1878 req->r_got_reply = 1;
1879 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1880 dout("handle_reply tid %llu dup ack\n", tid);
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001881 goto out_unlock;
Sage Weilf24e9982009-10-06 11:31:10 -07001882 }
1883
1884 dout("handle_reply tid %llu flags %d\n", tid, flags);
1885
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001886 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1887 __register_linger_request(osdc, req);
1888
Sage Weilf24e9982009-10-06 11:31:10 -07001889 /* either this is a read, or we got the safe response */
Sage Weil0ceed5d2010-05-11 09:53:18 -07001890 if (result < 0 ||
1891 (flags & CEPH_OSD_FLAG_ONDISK) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001892 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1893 __unregister_request(osdc, req);
1894
1895 mutex_unlock(&osdc->request_mutex);
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001896 up_read(&osdc->map_sem);
Sage Weilf24e9982009-10-06 11:31:10 -07001897
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001898 if (!already_completed) {
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001899 if (req->r_unsafe_callback &&
1900 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1901 req->r_unsafe_callback(req, true);
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001902 if (req->r_callback)
1903 req->r_callback(req, msg);
1904 else
1905 complete_all(&req->r_completion);
1906 }
Sage Weilf24e9982009-10-06 11:31:10 -07001907
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001908 if (flags & CEPH_OSD_FLAG_ONDISK) {
1909 if (req->r_unsafe_callback && already_completed)
1910 req->r_unsafe_callback(req, false);
Sage Weil25845472011-06-03 09:37:09 -07001911 complete_request(req);
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001912 }
Sage Weilf24e9982009-10-06 11:31:10 -07001913
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001914out:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001915 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07001916 ceph_osdc_put_request(req);
1917 return;
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001918out_unlock:
1919 mutex_unlock(&osdc->request_mutex);
1920 up_read(&osdc->map_sem);
1921 goto out;
Sage Weilf24e9982009-10-06 11:31:10 -07001922
Sage Weil1b83bef2013-02-25 16:11:12 -08001923bad_put:
Li Wang37c89bd2013-11-27 22:28:14 +08001924 req->r_result = -EIO;
1925 __unregister_request(osdc, req);
1926 if (req->r_callback)
1927 req->r_callback(req, msg);
1928 else
1929 complete_all(&req->r_completion);
1930 complete_request(req);
Sage Weil1b83bef2013-02-25 16:11:12 -08001931 ceph_osdc_put_request(req);
Alex Elder8058fd42013-04-01 18:58:26 -05001932bad_mutex:
1933 mutex_unlock(&osdc->request_mutex);
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001934 up_read(&osdc->map_sem);
Sage Weilf24e9982009-10-06 11:31:10 -07001935bad:
Sage Weil1b83bef2013-02-25 16:11:12 -08001936 pr_err("corrupt osd_op_reply got %d %d\n",
1937 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
Sage Weil9ec7cab2009-12-14 15:13:47 -08001938 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001939}
1940
Sage Weil6f6c7002011-01-17 20:34:08 -08001941static void reset_changed_osds(struct ceph_osd_client *osdc)
Sage Weilf24e9982009-10-06 11:31:10 -07001942{
Sage Weilf24e9982009-10-06 11:31:10 -07001943 struct rb_node *p, *n;
Sage Weilf24e9982009-10-06 11:31:10 -07001944
Ilya Dryomov7eb71e02015-02-17 19:37:15 +03001945 dout("%s %p\n", __func__, osdc);
Sage Weil6f6c7002011-01-17 20:34:08 -08001946 for (p = rb_first(&osdc->osds); p; p = n) {
1947 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
Sage Weilf24e9982009-10-06 11:31:10 -07001948
Sage Weil6f6c7002011-01-17 20:34:08 -08001949 n = rb_next(p);
1950 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1951 memcmp(&osd->o_con.peer_addr,
1952 ceph_osd_addr(osdc->osdmap,
1953 osd->o_osd),
1954 sizeof(struct ceph_entity_addr)) != 0)
1955 __reset_osd(osdc, osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001956 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001957}
1958
1959/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001960 * Requeue requests whose mapping to an OSD has changed. If requests map to
1961 * no osd, request a new map.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001962 *
Alex Eldere6d50f62012-12-26 14:31:40 -06001963 * Caller should hold map_sem for read.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001964 */
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001965static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1966 bool force_resend_writes)
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001967{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001968 struct ceph_osd_request *req, *nreq;
Sage Weil6f6c7002011-01-17 20:34:08 -08001969 struct rb_node *p;
1970 int needmap = 0;
1971 int err;
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001972 bool force_resend_req;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001973
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001974 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1975 force_resend_writes ? " (force resend writes)" : "");
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001976 mutex_lock(&osdc->request_mutex);
Sage Weil6194ea82012-07-30 16:19:28 -07001977 for (p = rb_first(&osdc->requests); p; ) {
Sage Weil6f6c7002011-01-17 20:34:08 -08001978 req = rb_entry(p, struct ceph_osd_request, r_node);
Sage Weil6194ea82012-07-30 16:19:28 -07001979 p = rb_next(p);
Alex Elderab60b162012-12-19 15:52:36 -06001980
1981 /*
1982 * For linger requests that have not yet been
1983 * registered, move them to the linger list; they'll
1984 * be sent to the osd in the loop below. Unregister
1985 * the request before re-registering it as a linger
1986 * request to ensure the __map_request() below
1987 * will decide it needs to be sent.
1988 */
1989 if (req->r_linger && list_empty(&req->r_linger_item)) {
1990 dout("%p tid %llu restart on osd%d\n",
1991 req, req->r_tid,
1992 req->r_osd ? req->r_osd->o_osd : -1);
Alex Elder96e4dac2013-05-22 20:54:25 -05001993 ceph_osdc_get_request(req);
Alex Elderab60b162012-12-19 15:52:36 -06001994 __unregister_request(osdc, req);
1995 __register_linger_request(osdc, req);
Alex Elder96e4dac2013-05-22 20:54:25 -05001996 ceph_osdc_put_request(req);
Alex Elderab60b162012-12-19 15:52:36 -06001997 continue;
1998 }
1999
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002000 force_resend_req = force_resend ||
2001 (force_resend_writes &&
2002 req->r_flags & CEPH_OSD_FLAG_WRITE);
2003 err = __map_request(osdc, req, force_resend_req);
Sage Weil6f6c7002011-01-17 20:34:08 -08002004 if (err < 0)
2005 continue; /* error */
2006 if (req->r_osd == NULL) {
2007 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
2008 needmap++; /* request a newer map */
2009 } else if (err > 0) {
Sage Weil6194ea82012-07-30 16:19:28 -07002010 if (!req->r_linger) {
2011 dout("%p tid %llu requeued on osd%d\n", req,
2012 req->r_tid,
2013 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002014 req->r_flags |= CEPH_OSD_FLAG_RETRY;
Sage Weil6194ea82012-07-30 16:19:28 -07002015 }
2016 }
Sage Weil6f6c7002011-01-17 20:34:08 -08002017 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002018
2019 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
2020 r_linger_item) {
2021 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
2022
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002023 err = __map_request(osdc, req,
2024 force_resend || force_resend_writes);
Alex Elderab60b162012-12-19 15:52:36 -06002025 dout("__map_request returned %d\n", err);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002026 if (err < 0)
2027 continue; /* hrm! */
Ilya Dryomovb0494532015-05-11 17:53:10 +03002028 if (req->r_osd == NULL || err > 0) {
2029 if (req->r_osd == NULL) {
2030 dout("lingering %p tid %llu maps to no osd\n",
2031 req, req->r_tid);
2032 /*
2033 * A homeless lingering request makes
2034 * no sense, as it's job is to keep
2035 * a particular OSD connection open.
2036 * Request a newer map and kick the
2037 * request, knowing that it won't be
2038 * resent until we actually get a map
2039 * that can tell us where to send it.
2040 */
2041 needmap++;
2042 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002043
Ilya Dryomovb0494532015-05-11 17:53:10 +03002044 dout("kicking lingering %p tid %llu osd%d\n", req,
2045 req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
2046 __register_request(osdc, req);
2047 __unregister_linger_request(osdc, req);
2048 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002049 }
Alex Elder14d2f382013-05-15 16:28:33 -05002050 reset_changed_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07002051 mutex_unlock(&osdc->request_mutex);
2052
2053 if (needmap) {
2054 dout("%d requests for down osds, need new map\n", needmap);
2055 ceph_monc_request_next_osdmap(&osdc->client->monc);
2056 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08002057}
Sage Weil6f6c7002011-01-17 20:34:08 -08002058
2059
Sage Weilf24e9982009-10-06 11:31:10 -07002060/*
2061 * Process updated osd map.
2062 *
2063 * The message contains any number of incremental and full maps, normally
2064 * indicating some sort of topology change in the cluster. Kick requests
2065 * off to different OSDs as needed.
2066 */
2067void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
2068{
2069 void *p, *end, *next;
2070 u32 nr_maps, maplen;
2071 u32 epoch;
2072 struct ceph_osdmap *newmap = NULL, *oldmap;
2073 int err;
2074 struct ceph_fsid fsid;
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002075 bool was_full;
Sage Weilf24e9982009-10-06 11:31:10 -07002076
2077 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
2078 p = msg->front.iov_base;
2079 end = p + msg->front.iov_len;
2080
2081 /* verify fsid */
2082 ceph_decode_need(&p, end, sizeof(fsid), bad);
2083 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Sage Weil07433042009-11-18 16:50:41 -08002084 if (ceph_check_fsid(osdc->client, &fsid) < 0)
2085 return;
Sage Weilf24e9982009-10-06 11:31:10 -07002086
2087 down_write(&osdc->map_sem);
2088
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002089 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2090
Sage Weilf24e9982009-10-06 11:31:10 -07002091 /* incremental maps */
2092 ceph_decode_32_safe(&p, end, nr_maps, bad);
2093 dout(" %d inc maps\n", nr_maps);
2094 while (nr_maps > 0) {
2095 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07002096 epoch = ceph_decode_32(&p);
2097 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07002098 ceph_decode_need(&p, end, maplen, bad);
2099 next = p + maplen;
2100 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
2101 dout("applying incremental map %u len %d\n",
2102 epoch, maplen);
2103 newmap = osdmap_apply_incremental(&p, next,
2104 osdc->osdmap,
Alex Elder15d98822012-05-26 23:26:43 -05002105 &osdc->client->msgr);
Sage Weilf24e9982009-10-06 11:31:10 -07002106 if (IS_ERR(newmap)) {
2107 err = PTR_ERR(newmap);
2108 goto bad;
2109 }
Sage Weil30dc6382009-12-21 14:49:37 -08002110 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07002111 if (newmap != osdc->osdmap) {
2112 ceph_osdmap_destroy(osdc->osdmap);
2113 osdc->osdmap = newmap;
2114 }
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002115 was_full = was_full ||
2116 ceph_osdmap_flag(osdc->osdmap,
2117 CEPH_OSDMAP_FULL);
2118 kick_requests(osdc, 0, was_full);
Sage Weilf24e9982009-10-06 11:31:10 -07002119 } else {
2120 dout("ignoring incremental map %u len %d\n",
2121 epoch, maplen);
2122 }
2123 p = next;
2124 nr_maps--;
2125 }
2126 if (newmap)
2127 goto done;
2128
2129 /* full maps */
2130 ceph_decode_32_safe(&p, end, nr_maps, bad);
2131 dout(" %d full maps\n", nr_maps);
2132 while (nr_maps) {
2133 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07002134 epoch = ceph_decode_32(&p);
2135 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07002136 ceph_decode_need(&p, end, maplen, bad);
2137 if (nr_maps > 1) {
2138 dout("skipping non-latest full map %u len %d\n",
2139 epoch, maplen);
2140 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2141 dout("skipping full map %u len %d, "
2142 "older than our %u\n", epoch, maplen,
2143 osdc->osdmap->epoch);
2144 } else {
Sage Weil38d64532011-10-14 13:33:55 -07002145 int skipped_map = 0;
2146
Sage Weilf24e9982009-10-06 11:31:10 -07002147 dout("taking full map %u len %d\n", epoch, maplen);
Ilya Dryomova2505d62014-03-13 16:36:13 +02002148 newmap = ceph_osdmap_decode(&p, p+maplen);
Sage Weilf24e9982009-10-06 11:31:10 -07002149 if (IS_ERR(newmap)) {
2150 err = PTR_ERR(newmap);
2151 goto bad;
2152 }
Sage Weil30dc6382009-12-21 14:49:37 -08002153 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07002154 oldmap = osdc->osdmap;
2155 osdc->osdmap = newmap;
Sage Weil38d64532011-10-14 13:33:55 -07002156 if (oldmap) {
2157 if (oldmap->epoch + 1 < newmap->epoch)
2158 skipped_map = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07002159 ceph_osdmap_destroy(oldmap);
Sage Weil38d64532011-10-14 13:33:55 -07002160 }
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002161 was_full = was_full ||
2162 ceph_osdmap_flag(osdc->osdmap,
2163 CEPH_OSDMAP_FULL);
2164 kick_requests(osdc, skipped_map, was_full);
Sage Weilf24e9982009-10-06 11:31:10 -07002165 }
2166 p += maplen;
2167 nr_maps--;
2168 }
2169
Dan Carpenterb72e19b2013-08-15 08:52:48 +03002170 if (!osdc->osdmap)
2171 goto bad;
Sage Weilf24e9982009-10-06 11:31:10 -07002172done:
2173 downgrade_write(&osdc->map_sem);
2174 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
Sage Weilcd634fb2011-05-12 09:29:18 -07002175
2176 /*
2177 * subscribe to subsequent osdmap updates if full to ensure
2178 * we find out when we are no longer full and stop returning
2179 * ENOSPC.
2180 */
Josh Durgind29adb32013-12-02 19:11:48 -08002181 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2182 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2183 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
Sage Weilcd634fb2011-05-12 09:29:18 -07002184 ceph_monc_request_next_osdmap(&osdc->client->monc);
2185
Alex Elderf9d25192013-02-15 11:42:29 -06002186 mutex_lock(&osdc->request_mutex);
2187 __send_queued(osdc);
2188 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07002189 up_read(&osdc->map_sem);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002190 wake_up_all(&osdc->client->auth_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07002191 return;
2192
2193bad:
2194 pr_err("osdc handle_map corrupt msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002195 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07002196 up_write(&osdc->map_sem);
Sage Weilf24e9982009-10-06 11:31:10 -07002197}
2198
Sage Weilf24e9982009-10-06 11:31:10 -07002199/*
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002200 * watch/notify callback event infrastructure
2201 *
2202 * These callbacks are used both for watch and notify operations.
2203 */
2204static void __release_event(struct kref *kref)
2205{
2206 struct ceph_osd_event *event =
2207 container_of(kref, struct ceph_osd_event, kref);
2208
2209 dout("__release_event %p\n", event);
2210 kfree(event);
2211}
2212
2213static void get_event(struct ceph_osd_event *event)
2214{
2215 kref_get(&event->kref);
2216}
2217
2218void ceph_osdc_put_event(struct ceph_osd_event *event)
2219{
2220 kref_put(&event->kref, __release_event);
2221}
2222EXPORT_SYMBOL(ceph_osdc_put_event);
2223
2224static void __insert_event(struct ceph_osd_client *osdc,
2225 struct ceph_osd_event *new)
2226{
2227 struct rb_node **p = &osdc->event_tree.rb_node;
2228 struct rb_node *parent = NULL;
2229 struct ceph_osd_event *event = NULL;
2230
2231 while (*p) {
2232 parent = *p;
2233 event = rb_entry(parent, struct ceph_osd_event, node);
2234 if (new->cookie < event->cookie)
2235 p = &(*p)->rb_left;
2236 else if (new->cookie > event->cookie)
2237 p = &(*p)->rb_right;
2238 else
2239 BUG();
2240 }
2241
2242 rb_link_node(&new->node, parent, p);
2243 rb_insert_color(&new->node, &osdc->event_tree);
2244}
2245
2246static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2247 u64 cookie)
2248{
2249 struct rb_node **p = &osdc->event_tree.rb_node;
2250 struct rb_node *parent = NULL;
2251 struct ceph_osd_event *event = NULL;
2252
2253 while (*p) {
2254 parent = *p;
2255 event = rb_entry(parent, struct ceph_osd_event, node);
2256 if (cookie < event->cookie)
2257 p = &(*p)->rb_left;
2258 else if (cookie > event->cookie)
2259 p = &(*p)->rb_right;
2260 else
2261 return event;
2262 }
2263 return NULL;
2264}
2265
2266static void __remove_event(struct ceph_osd_event *event)
2267{
2268 struct ceph_osd_client *osdc = event->osdc;
2269
2270 if (!RB_EMPTY_NODE(&event->node)) {
2271 dout("__remove_event removed %p\n", event);
2272 rb_erase(&event->node, &osdc->event_tree);
2273 ceph_osdc_put_event(event);
2274 } else {
2275 dout("__remove_event didn't remove %p\n", event);
2276 }
2277}
2278
2279int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2280 void (*event_cb)(u64, u64, u8, void *),
Alex Elder3c663bb2013-02-15 11:42:30 -06002281 void *data, struct ceph_osd_event **pevent)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002282{
2283 struct ceph_osd_event *event;
2284
2285 event = kmalloc(sizeof(*event), GFP_NOIO);
2286 if (!event)
2287 return -ENOMEM;
2288
2289 dout("create_event %p\n", event);
2290 event->cb = event_cb;
Alex Elder3c663bb2013-02-15 11:42:30 -06002291 event->one_shot = 0;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002292 event->data = data;
2293 event->osdc = osdc;
2294 INIT_LIST_HEAD(&event->osd_node);
Alex Elder3ee52342012-12-17 12:23:48 -06002295 RB_CLEAR_NODE(&event->node);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002296 kref_init(&event->kref); /* one ref for us */
2297 kref_get(&event->kref); /* one ref for the caller */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002298
2299 spin_lock(&osdc->event_lock);
2300 event->cookie = ++osdc->event_count;
2301 __insert_event(osdc, event);
2302 spin_unlock(&osdc->event_lock);
2303
2304 *pevent = event;
2305 return 0;
2306}
2307EXPORT_SYMBOL(ceph_osdc_create_event);
2308
2309void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2310{
2311 struct ceph_osd_client *osdc = event->osdc;
2312
2313 dout("cancel_event %p\n", event);
2314 spin_lock(&osdc->event_lock);
2315 __remove_event(event);
2316 spin_unlock(&osdc->event_lock);
2317 ceph_osdc_put_event(event); /* caller's */
2318}
2319EXPORT_SYMBOL(ceph_osdc_cancel_event);
2320
2321
2322static void do_event_work(struct work_struct *work)
2323{
2324 struct ceph_osd_event_work *event_work =
2325 container_of(work, struct ceph_osd_event_work, work);
2326 struct ceph_osd_event *event = event_work->event;
2327 u64 ver = event_work->ver;
2328 u64 notify_id = event_work->notify_id;
2329 u8 opcode = event_work->opcode;
2330
2331 dout("do_event_work completing %p\n", event);
2332 event->cb(ver, notify_id, opcode, event->data);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002333 dout("do_event_work completed %p\n", event);
2334 ceph_osdc_put_event(event);
2335 kfree(event_work);
2336}
2337
2338
2339/*
2340 * Process osd watch notifications
2341 */
Alex Elder3c663bb2013-02-15 11:42:30 -06002342static void handle_watch_notify(struct ceph_osd_client *osdc,
2343 struct ceph_msg *msg)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002344{
2345 void *p, *end;
2346 u8 proto_ver;
2347 u64 cookie, ver, notify_id;
2348 u8 opcode;
2349 struct ceph_osd_event *event;
2350 struct ceph_osd_event_work *event_work;
2351
2352 p = msg->front.iov_base;
2353 end = p + msg->front.iov_len;
2354
2355 ceph_decode_8_safe(&p, end, proto_ver, bad);
2356 ceph_decode_8_safe(&p, end, opcode, bad);
2357 ceph_decode_64_safe(&p, end, cookie, bad);
2358 ceph_decode_64_safe(&p, end, ver, bad);
2359 ceph_decode_64_safe(&p, end, notify_id, bad);
2360
2361 spin_lock(&osdc->event_lock);
2362 event = __find_event(osdc, cookie);
2363 if (event) {
Alex Elder3c663bb2013-02-15 11:42:30 -06002364 BUG_ON(event->one_shot);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002365 get_event(event);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002366 }
2367 spin_unlock(&osdc->event_lock);
2368 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2369 cookie, ver, event);
2370 if (event) {
2371 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002372 if (!event_work) {
Ilya Dryomov91883cd2014-09-11 12:18:53 +04002373 pr_err("couldn't allocate event_work\n");
2374 ceph_osdc_put_event(event);
2375 return;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002376 }
Mariusz Kozlowski6b0ae402011-03-26 19:29:34 +01002377 INIT_WORK(&event_work->work, do_event_work);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002378 event_work->event = event;
2379 event_work->ver = ver;
2380 event_work->notify_id = notify_id;
2381 event_work->opcode = opcode;
Ilya Dryomov91883cd2014-09-11 12:18:53 +04002382
2383 queue_work(osdc->notify_wq, &event_work->work);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002384 }
2385
2386 return;
2387
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002388bad:
2389 pr_err("osdc handle_watch_notify corrupt msg\n");
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002390}
2391
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002392/*
Alex Eldere65550f2013-04-05 01:27:12 -05002393 * build new request AND message
2394 *
2395 */
2396void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2397 struct ceph_snap_context *snapc, u64 snap_id,
2398 struct timespec *mtime)
2399{
2400 struct ceph_msg *msg = req->r_request;
2401 void *p;
2402 size_t msg_size;
2403 int flags = req->r_flags;
2404 u64 data_len;
2405 unsigned int i;
2406
2407 req->r_snapid = snap_id;
2408 req->r_snapc = ceph_get_snap_context(snapc);
2409
2410 /* encode request */
2411 msg->hdr.version = cpu_to_le16(4);
2412
2413 p = msg->front.iov_base;
2414 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2415 req->r_request_osdmap_epoch = p;
2416 p += 4;
2417 req->r_request_flags = p;
2418 p += 4;
2419 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2420 ceph_encode_timespec(p, mtime);
2421 p += sizeof(struct ceph_timespec);
2422 req->r_request_reassert_version = p;
2423 p += sizeof(struct ceph_eversion); /* will get filled in */
2424
2425 /* oloc */
2426 ceph_encode_8(&p, 4);
2427 ceph_encode_8(&p, 4);
2428 ceph_encode_32(&p, 8 + 4 + 4);
2429 req->r_request_pool = p;
2430 p += 8;
2431 ceph_encode_32(&p, -1); /* preferred */
2432 ceph_encode_32(&p, 0); /* key len */
2433
2434 ceph_encode_8(&p, 1);
2435 req->r_request_pgid = p;
2436 p += 8 + 4;
2437 ceph_encode_32(&p, -1); /* preferred */
2438
2439 /* oid */
Ilya Dryomov3c972c92014-01-27 17:40:20 +02002440 ceph_encode_32(&p, req->r_base_oid.name_len);
2441 memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2442 dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2443 req->r_base_oid.name, req->r_base_oid.name_len);
2444 p += req->r_base_oid.name_len;
Alex Eldere65550f2013-04-05 01:27:12 -05002445
2446 /* ops--can imply data */
2447 ceph_encode_16(&p, (u16)req->r_num_ops);
2448 data_len = 0;
2449 for (i = 0; i < req->r_num_ops; i++) {
2450 data_len += osd_req_encode_op(req, p, i);
2451 p += sizeof(struct ceph_osd_op);
2452 }
2453
2454 /* snaps */
2455 ceph_encode_64(&p, req->r_snapid);
2456 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2457 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2458 if (req->r_snapc) {
2459 for (i = 0; i < snapc->num_snaps; i++) {
2460 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2461 }
2462 }
2463
2464 req->r_request_attempts = p;
2465 p += 4;
2466
2467 /* data */
2468 if (flags & CEPH_OSD_FLAG_WRITE) {
2469 u16 data_off;
2470
2471 /*
2472 * The header "data_off" is a hint to the receiver
2473 * allowing it to align received data into its
2474 * buffers such that there's no need to re-copy
2475 * it before writing it to disk (direct I/O).
2476 */
2477 data_off = (u16) (off & 0xffff);
2478 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2479 }
2480 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2481
2482 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2483 msg_size = p - msg->front.iov_base;
2484 msg->front.iov_len = msg_size;
2485 msg->hdr.front_len = cpu_to_le32(msg_size);
2486
2487 dout("build_request msg_size was %d\n", (int)msg_size);
2488}
2489EXPORT_SYMBOL(ceph_osdc_build_request);
2490
2491/*
Sage Weilf24e9982009-10-06 11:31:10 -07002492 * Register request, send initial attempt.
2493 */
2494int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2495 struct ceph_osd_request *req,
2496 bool nofail)
2497{
Ilya Dryomov0bbfdfe2014-01-31 19:33:39 +02002498 int rc;
Sage Weilf24e9982009-10-06 11:31:10 -07002499
Sage Weilf24e9982009-10-06 11:31:10 -07002500 down_read(&osdc->map_sem);
2501 mutex_lock(&osdc->request_mutex);
Ilya Dryomov0bbfdfe2014-01-31 19:33:39 +02002502
2503 rc = __ceph_osdc_start_request(osdc, req, nofail);
2504
Sage Weilf24e9982009-10-06 11:31:10 -07002505 mutex_unlock(&osdc->request_mutex);
2506 up_read(&osdc->map_sem);
Ilya Dryomov0bbfdfe2014-01-31 19:33:39 +02002507
Sage Weilf24e9982009-10-06 11:31:10 -07002508 return rc;
2509}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002510EXPORT_SYMBOL(ceph_osdc_start_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002511
2512/*
Ilya Dryomovc9f9b93d2014-06-19 11:38:13 +04002513 * Unregister a registered request. The request is not completed (i.e.
2514 * no callbacks or wakeups) - higher layers are supposed to know what
2515 * they are canceling.
2516 */
2517void ceph_osdc_cancel_request(struct ceph_osd_request *req)
2518{
2519 struct ceph_osd_client *osdc = req->r_osdc;
2520
2521 mutex_lock(&osdc->request_mutex);
2522 if (req->r_linger)
2523 __unregister_linger_request(osdc, req);
2524 __unregister_request(osdc, req);
2525 mutex_unlock(&osdc->request_mutex);
2526
2527 dout("%s %p tid %llu canceled\n", __func__, req, req->r_tid);
2528}
2529EXPORT_SYMBOL(ceph_osdc_cancel_request);
2530
2531/*
Sage Weilf24e9982009-10-06 11:31:10 -07002532 * wait for a request to complete
2533 */
2534int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2535 struct ceph_osd_request *req)
2536{
2537 int rc;
2538
Ilya Dryomovc9f9b93d2014-06-19 11:38:13 +04002539 dout("%s %p tid %llu\n", __func__, req, req->r_tid);
2540
Sage Weilf24e9982009-10-06 11:31:10 -07002541 rc = wait_for_completion_interruptible(&req->r_completion);
2542 if (rc < 0) {
Ilya Dryomovc9f9b93d2014-06-19 11:38:13 +04002543 dout("%s %p tid %llu interrupted\n", __func__, req, req->r_tid);
2544 ceph_osdc_cancel_request(req);
Sage Weil25845472011-06-03 09:37:09 -07002545 complete_request(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002546 return rc;
2547 }
2548
Ilya Dryomovc9f9b93d2014-06-19 11:38:13 +04002549 dout("%s %p tid %llu result %d\n", __func__, req, req->r_tid,
2550 req->r_result);
Sage Weilf24e9982009-10-06 11:31:10 -07002551 return req->r_result;
2552}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002553EXPORT_SYMBOL(ceph_osdc_wait_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002554
2555/*
2556 * sync - wait for all in-flight requests to flush. avoid starvation.
2557 */
2558void ceph_osdc_sync(struct ceph_osd_client *osdc)
2559{
2560 struct ceph_osd_request *req;
2561 u64 last_tid, next_tid = 0;
2562
2563 mutex_lock(&osdc->request_mutex);
2564 last_tid = osdc->last_tid;
2565 while (1) {
2566 req = __lookup_request_ge(osdc, next_tid);
2567 if (!req)
2568 break;
2569 if (req->r_tid > last_tid)
2570 break;
2571
2572 next_tid = req->r_tid + 1;
2573 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2574 continue;
2575
2576 ceph_osdc_get_request(req);
2577 mutex_unlock(&osdc->request_mutex);
2578 dout("sync waiting on tid %llu (last is %llu)\n",
2579 req->r_tid, last_tid);
2580 wait_for_completion(&req->r_safe_completion);
2581 mutex_lock(&osdc->request_mutex);
2582 ceph_osdc_put_request(req);
2583 }
2584 mutex_unlock(&osdc->request_mutex);
2585 dout("sync done (thru tid %llu)\n", last_tid);
2586}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002587EXPORT_SYMBOL(ceph_osdc_sync);
Sage Weilf24e9982009-10-06 11:31:10 -07002588
2589/*
Josh Durgindd935f42013-08-28 21:43:09 -07002590 * Call all pending notify callbacks - for use after a watch is
2591 * unregistered, to make sure no more callbacks for it will be invoked
2592 */
stephen hemmingerf64794492014-06-10 20:30:13 -07002593void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
Josh Durgindd935f42013-08-28 21:43:09 -07002594{
2595 flush_workqueue(osdc->notify_wq);
2596}
2597EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2598
2599
2600/*
Sage Weilf24e9982009-10-06 11:31:10 -07002601 * init, shutdown
2602 */
2603int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2604{
2605 int err;
2606
2607 dout("init\n");
2608 osdc->client = client;
2609 osdc->osdmap = NULL;
2610 init_rwsem(&osdc->map_sem);
2611 init_completion(&osdc->map_waiters);
2612 osdc->last_requested_map = 0;
2613 mutex_init(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07002614 osdc->last_tid = 0;
2615 osdc->osds = RB_ROOT;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002616 INIT_LIST_HEAD(&osdc->osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07002617 osdc->requests = RB_ROOT;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08002618 INIT_LIST_HEAD(&osdc->req_lru);
Sage Weil6f6c7002011-01-17 20:34:08 -08002619 INIT_LIST_HEAD(&osdc->req_unsent);
2620 INIT_LIST_HEAD(&osdc->req_notarget);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002621 INIT_LIST_HEAD(&osdc->req_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07002622 osdc->num_requests = 0;
2623 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002624 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002625 spin_lock_init(&osdc->event_lock);
2626 osdc->event_tree = RB_ROOT;
2627 osdc->event_count = 0;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002628
2629 schedule_delayed_work(&osdc->osds_timeout_work,
Ilya Dryomova319bf52015-05-15 12:02:17 +03002630 round_jiffies_relative(osdc->client->options->osd_idle_ttl));
Sage Weilf24e9982009-10-06 11:31:10 -07002631
Sage Weil5f44f142009-11-18 14:52:18 -08002632 err = -ENOMEM;
Sage Weilf24e9982009-10-06 11:31:10 -07002633 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2634 sizeof(struct ceph_osd_request));
2635 if (!osdc->req_mempool)
Sage Weil5f44f142009-11-18 14:52:18 -08002636 goto out;
Sage Weilf24e9982009-10-06 11:31:10 -07002637
Sage Weild50b4092012-07-09 14:22:34 -07002638 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2639 OSD_OP_FRONT_LEN, 10, true,
Sage Weil4f482802010-04-24 09:56:35 -07002640 "osd_op");
Sage Weilf24e9982009-10-06 11:31:10 -07002641 if (err < 0)
Sage Weil5f44f142009-11-18 14:52:18 -08002642 goto out_mempool;
Sage Weild50b4092012-07-09 14:22:34 -07002643 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
Sage Weil4f482802010-04-24 09:56:35 -07002644 OSD_OPREPLY_FRONT_LEN, 10, true,
2645 "osd_op_reply");
Sage Weilc16e7862010-03-01 13:02:00 -08002646 if (err < 0)
2647 goto out_msgpool;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002648
Dan Carpenterdbcae082013-08-15 08:58:59 +03002649 err = -ENOMEM;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002650 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
Dan Carpenterdbcae082013-08-15 08:58:59 +03002651 if (!osdc->notify_wq)
Ilya Dryomovc172ec52014-01-31 17:49:22 +02002652 goto out_msgpool_reply;
2653
Sage Weilf24e9982009-10-06 11:31:10 -07002654 return 0;
Sage Weil5f44f142009-11-18 14:52:18 -08002655
Ilya Dryomovc172ec52014-01-31 17:49:22 +02002656out_msgpool_reply:
2657 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
Sage Weilc16e7862010-03-01 13:02:00 -08002658out_msgpool:
2659 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weil5f44f142009-11-18 14:52:18 -08002660out_mempool:
2661 mempool_destroy(osdc->req_mempool);
2662out:
2663 return err;
Sage Weilf24e9982009-10-06 11:31:10 -07002664}
2665
2666void ceph_osdc_stop(struct ceph_osd_client *osdc)
2667{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002668 flush_workqueue(osdc->notify_wq);
2669 destroy_workqueue(osdc->notify_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07002670 cancel_delayed_work_sync(&osdc->timeout_work);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002671 cancel_delayed_work_sync(&osdc->osds_timeout_work);
Sage Weilf24e9982009-10-06 11:31:10 -07002672 if (osdc->osdmap) {
2673 ceph_osdmap_destroy(osdc->osdmap);
2674 osdc->osdmap = NULL;
2675 }
Sage Weilaca420b2011-08-31 14:45:53 -07002676 remove_all_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07002677 mempool_destroy(osdc->req_mempool);
2678 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weilc16e7862010-03-01 13:02:00 -08002679 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
Sage Weilf24e9982009-10-06 11:31:10 -07002680}
2681
2682/*
2683 * Read some contiguous pages. If we cross a stripe boundary, shorten
2684 * *plen. Return number of bytes read, or error.
2685 */
2686int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2687 struct ceph_vino vino, struct ceph_file_layout *layout,
2688 u64 off, u64 *plen,
2689 u32 truncate_seq, u64 truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -08002690 struct page **pages, int num_pages, int page_align)
Sage Weilf24e9982009-10-06 11:31:10 -07002691{
2692 struct ceph_osd_request *req;
2693 int rc = 0;
2694
2695 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2696 vino.snap, off, *plen);
Yan, Zheng715e4cd2014-11-13 14:40:37 +08002697 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002698 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
Alex Elderacead002013-03-14 14:09:05 -05002699 NULL, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002700 false);
Sage Weil68162822012-09-24 21:01:02 -07002701 if (IS_ERR(req))
2702 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002703
2704 /* it may be a short read due to an object boundary */
Alex Elder0fff87e2013-02-14 12:16:43 -06002705
Alex Elder406e2c92013-04-15 14:50:36 -05002706 osd_req_op_extent_osd_data_pages(req, 0,
Alex Eldera4ce40a2013-04-05 01:27:12 -05002707 pages, *plen, page_align, false, false);
Sage Weilf24e9982009-10-06 11:31:10 -07002708
Alex Eldere0c59482013-03-07 15:38:25 -06002709 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
Alex Elder43bfe5d2013-04-03 01:28:57 -05002710 off, *plen, *plen, page_align);
Sage Weilf24e9982009-10-06 11:31:10 -07002711
Alex Elder79528732013-04-03 21:32:51 -05002712 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
Alex Elder02ee07d2013-03-14 14:09:06 -05002713
Sage Weilf24e9982009-10-06 11:31:10 -07002714 rc = ceph_osdc_start_request(osdc, req, false);
2715 if (!rc)
2716 rc = ceph_osdc_wait_request(osdc, req);
2717
2718 ceph_osdc_put_request(req);
2719 dout("readpages result %d\n", rc);
2720 return rc;
2721}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002722EXPORT_SYMBOL(ceph_osdc_readpages);
Sage Weilf24e9982009-10-06 11:31:10 -07002723
2724/*
2725 * do a synchronous write on N pages
2726 */
2727int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2728 struct ceph_file_layout *layout,
2729 struct ceph_snap_context *snapc,
2730 u64 off, u64 len,
2731 u32 truncate_seq, u64 truncate_size,
2732 struct timespec *mtime,
Alex Elder24808822013-02-15 11:42:29 -06002733 struct page **pages, int num_pages)
Sage Weilf24e9982009-10-06 11:31:10 -07002734{
2735 struct ceph_osd_request *req;
2736 int rc = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -08002737 int page_align = off & ~PAGE_MASK;
Sage Weilf24e9982009-10-06 11:31:10 -07002738
Alex Elderacead002013-03-14 14:09:05 -05002739 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
Yan, Zheng715e4cd2014-11-13 14:40:37 +08002740 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002741 CEPH_OSD_OP_WRITE,
Alex Elder24808822013-02-15 11:42:29 -06002742 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
Alex Elderacead002013-03-14 14:09:05 -05002743 snapc, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002744 true);
Sage Weil68162822012-09-24 21:01:02 -07002745 if (IS_ERR(req))
2746 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002747
2748 /* it may be a short write due to an object boundary */
Alex Elder406e2c92013-04-15 14:50:36 -05002749 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
Alex Elder43bfe5d2013-04-03 01:28:57 -05002750 false, false);
2751 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
Sage Weilf24e9982009-10-06 11:31:10 -07002752
Alex Elder79528732013-04-03 21:32:51 -05002753 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
Alex Elder02ee07d2013-03-14 14:09:06 -05002754
Alex Elder87f979d2013-02-15 11:42:29 -06002755 rc = ceph_osdc_start_request(osdc, req, true);
Sage Weilf24e9982009-10-06 11:31:10 -07002756 if (!rc)
2757 rc = ceph_osdc_wait_request(osdc, req);
2758
2759 ceph_osdc_put_request(req);
2760 if (rc == 0)
2761 rc = len;
2762 dout("writepages result %d\n", rc);
2763 return rc;
2764}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002765EXPORT_SYMBOL(ceph_osdc_writepages);
Sage Weilf24e9982009-10-06 11:31:10 -07002766
Alex Elder5522ae02013-05-01 12:43:04 -05002767int ceph_osdc_setup(void)
2768{
2769 BUG_ON(ceph_osd_request_cache);
2770 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2771 sizeof (struct ceph_osd_request),
2772 __alignof__(struct ceph_osd_request),
2773 0, NULL);
2774
2775 return ceph_osd_request_cache ? 0 : -ENOMEM;
2776}
2777EXPORT_SYMBOL(ceph_osdc_setup);
2778
2779void ceph_osdc_cleanup(void)
2780{
2781 BUG_ON(!ceph_osd_request_cache);
2782 kmem_cache_destroy(ceph_osd_request_cache);
2783 ceph_osd_request_cache = NULL;
2784}
2785EXPORT_SYMBOL(ceph_osdc_cleanup);
2786
Sage Weilf24e9982009-10-06 11:31:10 -07002787/*
2788 * handle incoming message
2789 */
2790static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2791{
2792 struct ceph_osd *osd = con->private;
Julia Lawall32c895e2009-11-21 16:53:16 +01002793 struct ceph_osd_client *osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002794 int type = le16_to_cpu(msg->hdr.type);
2795
2796 if (!osd)
Sage Weil4a32f932010-06-13 10:27:53 -07002797 goto out;
Julia Lawall32c895e2009-11-21 16:53:16 +01002798 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002799
2800 switch (type) {
2801 case CEPH_MSG_OSD_MAP:
2802 ceph_osdc_handle_map(osdc, msg);
2803 break;
2804 case CEPH_MSG_OSD_OPREPLY:
Sage Weil350b1c32009-12-22 10:45:45 -08002805 handle_reply(osdc, msg, con);
Sage Weilf24e9982009-10-06 11:31:10 -07002806 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002807 case CEPH_MSG_WATCH_NOTIFY:
2808 handle_watch_notify(osdc, msg);
2809 break;
Sage Weilf24e9982009-10-06 11:31:10 -07002810
2811 default:
2812 pr_err("received unknown message type %d %s\n", type,
2813 ceph_msg_type_name(type));
2814 }
Sage Weil4a32f932010-06-13 10:27:53 -07002815out:
Sage Weilf24e9982009-10-06 11:31:10 -07002816 ceph_msg_put(msg);
2817}
2818
Sage Weil5b3a4db2010-02-19 21:43:23 -08002819/*
Sage Weil21b667f2010-03-04 10:22:59 -08002820 * lookup and return message for incoming reply. set up reply message
2821 * pages.
Sage Weil5b3a4db2010-02-19 21:43:23 -08002822 */
2823static struct ceph_msg *get_reply(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002824 struct ceph_msg_header *hdr,
2825 int *skip)
Sage Weilf24e9982009-10-06 11:31:10 -07002826{
2827 struct ceph_osd *osd = con->private;
2828 struct ceph_osd_client *osdc = osd->o_osdc;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002829 struct ceph_msg *m;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002830 struct ceph_osd_request *req;
Ilya Dryomov3f0a4ac2014-01-09 20:08:21 +02002831 int front_len = le32_to_cpu(hdr->front_len);
Sage Weil5b3a4db2010-02-19 21:43:23 -08002832 int data_len = le32_to_cpu(hdr->data_len);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002833 u64 tid;
Sage Weilf24e9982009-10-06 11:31:10 -07002834
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002835 tid = le64_to_cpu(hdr->tid);
2836 mutex_lock(&osdc->request_mutex);
2837 req = __lookup_request(osdc, tid);
2838 if (!req) {
2839 *skip = 1;
2840 m = NULL;
Sage Weil756a16a2012-07-30 16:26:13 -07002841 dout("get_reply unknown tid %llu from osd%d\n", tid,
2842 osd->o_osd);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002843 goto out;
2844 }
Sage Weilc16e7862010-03-01 13:02:00 -08002845
Alex Elderace6d3a2013-04-01 16:12:14 -05002846 if (req->r_reply->con)
Alex Elder8921d112012-06-01 14:56:43 -05002847 dout("%s revoking msg %p from old con %p\n", __func__,
Alex Elderace6d3a2013-04-01 16:12:14 -05002848 req->r_reply, req->r_reply->con);
2849 ceph_msg_revoke_incoming(req->r_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002850
Ilya Dryomovf2be82b2014-01-09 20:08:21 +02002851 if (front_len > req->r_reply->front_alloc_len) {
Joe Perchesb9a67892014-09-09 21:17:29 -07002852 pr_warn("get_reply front %d > preallocated %d (%u#%llu)\n",
2853 front_len, req->r_reply->front_alloc_len,
2854 (unsigned int)con->peer_name.type,
2855 le64_to_cpu(con->peer_name.num));
Ilya Dryomov3f0a4ac2014-01-09 20:08:21 +02002856 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2857 false);
Sage Weila79832f2010-04-01 16:06:19 -07002858 if (!m)
Sage Weilc16e7862010-03-01 13:02:00 -08002859 goto out;
2860 ceph_msg_put(req->r_reply);
2861 req->r_reply = m;
2862 }
2863 m = ceph_msg_get(req->r_reply);
2864
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002865 if (data_len > 0) {
Alex Eldera4ce40a2013-04-05 01:27:12 -05002866 struct ceph_osd_data *osd_data;
Alex Elder0fff87e2013-02-14 12:16:43 -06002867
Alex Eldera4ce40a2013-04-05 01:27:12 -05002868 /*
2869 * XXX This is assuming there is only one op containing
2870 * XXX page data. Probably OK for reads, but this
2871 * XXX ought to be done more generally.
2872 */
Alex Elder406e2c92013-04-15 14:50:36 -05002873 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder0fff87e2013-02-14 12:16:43 -06002874 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
Alex Elder0fff87e2013-02-14 12:16:43 -06002875 if (osd_data->pages &&
Alex Eldere0c59482013-03-07 15:38:25 -06002876 unlikely(osd_data->length < data_len)) {
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002877
Joe Perchesb9a67892014-09-09 21:17:29 -07002878 pr_warn("tid %lld reply has %d bytes we had only %llu bytes ready\n",
Alex Eldere0c59482013-03-07 15:38:25 -06002879 tid, data_len, osd_data->length);
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002880 *skip = 1;
2881 ceph_msg_put(m);
2882 m = NULL;
2883 goto out;
2884 }
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002885 }
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002886 }
Sage Weil5b3a4db2010-02-19 21:43:23 -08002887 *skip = 0;
Sage Weilc16e7862010-03-01 13:02:00 -08002888 dout("get_reply tid %lld %p\n", tid, m);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002889
2890out:
2891 mutex_unlock(&osdc->request_mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002892 return m;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002893
2894}
2895
2896static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2897 struct ceph_msg_header *hdr,
2898 int *skip)
2899{
2900 struct ceph_osd *osd = con->private;
2901 int type = le16_to_cpu(hdr->type);
2902 int front = le32_to_cpu(hdr->front_len);
2903
Alex Elder1c20f2d2012-06-04 14:43:32 -05002904 *skip = 0;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002905 switch (type) {
2906 case CEPH_MSG_OSD_MAP:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002907 case CEPH_MSG_WATCH_NOTIFY:
Sage Weilb61c2762011-08-09 15:03:46 -07002908 return ceph_msg_new(type, front, GFP_NOFS, false);
Sage Weil5b3a4db2010-02-19 21:43:23 -08002909 case CEPH_MSG_OSD_OPREPLY:
2910 return get_reply(con, hdr, skip);
2911 default:
2912 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2913 osd->o_osd);
2914 *skip = 1;
2915 return NULL;
2916 }
Sage Weilf24e9982009-10-06 11:31:10 -07002917}
2918
2919/*
2920 * Wrappers to refcount containing ceph_osd struct
2921 */
2922static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2923{
2924 struct ceph_osd *osd = con->private;
2925 if (get_osd(osd))
2926 return con;
2927 return NULL;
2928}
2929
2930static void put_osd_con(struct ceph_connection *con)
2931{
2932 struct ceph_osd *osd = con->private;
2933 put_osd(osd);
2934}
2935
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002936/*
2937 * authentication
2938 */
Alex Eldera3530df2012-05-16 15:16:39 -05002939/*
2940 * Note: returned pointer is the address of a structure that's
2941 * managed separately. Caller must *not* attempt to free it.
2942 */
2943static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
Alex Elder8f43fb52012-05-16 15:16:39 -05002944 int *proto, int force_new)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002945{
2946 struct ceph_osd *o = con->private;
2947 struct ceph_osd_client *osdc = o->o_osdc;
2948 struct ceph_auth_client *ac = osdc->client->monc.auth;
Alex Elder74f18692012-05-16 15:16:39 -05002949 struct ceph_auth_handshake *auth = &o->o_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002950
Alex Elder74f18692012-05-16 15:16:39 -05002951 if (force_new && auth->authorizer) {
Sage Weil27859f92013-03-25 10:26:14 -07002952 ceph_auth_destroy_authorizer(ac, auth->authorizer);
Alex Elder74f18692012-05-16 15:16:39 -05002953 auth->authorizer = NULL;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002954 }
Sage Weil27859f92013-03-25 10:26:14 -07002955 if (!auth->authorizer) {
2956 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2957 auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002958 if (ret)
Alex Eldera3530df2012-05-16 15:16:39 -05002959 return ERR_PTR(ret);
Sage Weil27859f92013-03-25 10:26:14 -07002960 } else {
2961 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
Sage Weil0bed9b52013-03-25 10:26:01 -07002962 auth);
2963 if (ret)
2964 return ERR_PTR(ret);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002965 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002966 *proto = ac->protocol;
Alex Elder74f18692012-05-16 15:16:39 -05002967
Alex Eldera3530df2012-05-16 15:16:39 -05002968 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002969}
2970
2971
2972static int verify_authorizer_reply(struct ceph_connection *con, int len)
2973{
2974 struct ceph_osd *o = con->private;
2975 struct ceph_osd_client *osdc = o->o_osdc;
2976 struct ceph_auth_client *ac = osdc->client->monc.auth;
2977
Sage Weil27859f92013-03-25 10:26:14 -07002978 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002979}
2980
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002981static int invalidate_authorizer(struct ceph_connection *con)
2982{
2983 struct ceph_osd *o = con->private;
2984 struct ceph_osd_client *osdc = o->o_osdc;
2985 struct ceph_auth_client *ac = osdc->client->monc.auth;
2986
Sage Weil27859f92013-03-25 10:26:14 -07002987 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002988 return ceph_monc_validate_auth(&osdc->client->monc);
2989}
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002990
Yan, Zheng33d07332014-11-04 16:33:37 +08002991static int sign_message(struct ceph_connection *con, struct ceph_msg *msg)
2992{
2993 struct ceph_osd *o = con->private;
2994 struct ceph_auth_handshake *auth = &o->o_auth;
2995 return ceph_auth_sign_message(auth, msg);
2996}
2997
2998static int check_message_signature(struct ceph_connection *con, struct ceph_msg *msg)
2999{
3000 struct ceph_osd *o = con->private;
3001 struct ceph_auth_handshake *auth = &o->o_auth;
3002 return ceph_auth_check_message_signature(auth, msg);
3003}
3004
Tobias Klauser9e327892010-05-20 10:40:19 +02003005static const struct ceph_connection_operations osd_con_ops = {
Sage Weilf24e9982009-10-06 11:31:10 -07003006 .get = get_osd_con,
3007 .put = put_osd_con,
3008 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003009 .get_authorizer = get_authorizer,
3010 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003011 .invalidate_authorizer = invalidate_authorizer,
Sage Weilf24e9982009-10-06 11:31:10 -07003012 .alloc_msg = alloc_msg,
Yan, Zheng33d07332014-11-04 16:33:37 +08003013 .sign_message = sign_message,
3014 .check_message_signature = check_message_signature,
Sage Weil81b024e2009-10-09 10:29:18 -07003015 .fault = osd_reset,
Sage Weilf24e9982009-10-06 11:31:10 -07003016};