blob: b0dfce77656a0ba9c43d6d6616be1d28fb4dc137 [file] [log] [blame]
Alex Eldera4ce40a2013-04-05 01:27:12 -05001
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002#include <linux/ceph/ceph_debug.h>
Sage Weilf24e9982009-10-06 11:31:10 -07003
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07004#include <linux/module.h>
Sage Weilf24e9982009-10-06 11:31:10 -07005#include <linux/err.h>
6#include <linux/highmem.h>
7#include <linux/mm.h>
8#include <linux/pagemap.h>
9#include <linux/slab.h>
10#include <linux/uaccess.h>
Yehuda Sadeh68b44762010-04-06 15:01:27 -070011#ifdef CONFIG_BLOCK
12#include <linux/bio.h>
13#endif
Sage Weilf24e9982009-10-06 11:31:10 -070014
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070015#include <linux/ceph/libceph.h>
16#include <linux/ceph/osd_client.h>
17#include <linux/ceph/messenger.h>
18#include <linux/ceph/decode.h>
19#include <linux/ceph/auth.h>
20#include <linux/ceph/pagelist.h>
Sage Weilf24e9982009-10-06 11:31:10 -070021
Sage Weilc16e7862010-03-01 13:02:00 -080022#define OSD_OP_FRONT_LEN 4096
23#define OSD_OPREPLY_FRONT_LEN 512
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -080024
Alex Elder5522ae02013-05-01 12:43:04 -050025static struct kmem_cache *ceph_osd_request_cache;
26
Tobias Klauser9e327892010-05-20 10:40:19 +020027static const struct ceph_connection_operations osd_con_ops;
Sage Weilf24e9982009-10-06 11:31:10 -070028
Alex Elderf9d25192013-02-15 11:42:29 -060029static void __send_queued(struct ceph_osd_client *osdc);
Sage Weil6f6c7002011-01-17 20:34:08 -080030static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -070031static void __register_request(struct ceph_osd_client *osdc,
32 struct ceph_osd_request *req);
33static void __unregister_linger_request(struct ceph_osd_client *osdc,
34 struct ceph_osd_request *req);
Sage Weil56e925b2012-01-03 12:34:34 -080035static void __send_request(struct ceph_osd_client *osdc,
36 struct ceph_osd_request *req);
Sage Weilf24e9982009-10-06 11:31:10 -070037
38/*
39 * Implement client access to distributed object storage cluster.
40 *
41 * All data objects are stored within a cluster/cloud of OSDs, or
42 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
43 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
44 * remote daemons serving up and coordinating consistent and safe
45 * access to storage.
46 *
47 * Cluster membership and the mapping of data objects onto storage devices
48 * are described by the osd map.
49 *
50 * We keep track of pending OSD requests (read, write), resubmit
51 * requests to different OSDs when the cluster topology/data layout
52 * change, or retry the affected requests when the communications
53 * channel with an OSD is reset.
54 */
55
56/*
57 * calculate the mapping of a file extent onto an object, and fill out the
58 * request accordingly. shorten extent as necessary if it crosses an
59 * object boundary.
60 *
61 * fill osd op in request message.
62 */
Alex Elderdbe0fc42013-02-15 22:10:17 -060063static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
Alex Eldera19dadf2013-03-13 20:50:01 -050064 u64 *objnum, u64 *objoff, u64 *objlen)
Sage Weilf24e9982009-10-06 11:31:10 -070065{
Alex Elder60e56f12013-02-15 11:42:29 -060066 u64 orig_len = *plen;
Sage Weild63b77f2012-09-24 20:59:48 -070067 int r;
Sage Weilf24e9982009-10-06 11:31:10 -070068
Alex Elder60e56f12013-02-15 11:42:29 -060069 /* object extent? */
Alex Elder75d1c942013-03-13 20:50:00 -050070 r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
71 objoff, objlen);
Sage Weild63b77f2012-09-24 20:59:48 -070072 if (r < 0)
73 return r;
Alex Elder75d1c942013-03-13 20:50:00 -050074 if (*objlen < orig_len) {
75 *plen = *objlen;
Alex Elder60e56f12013-02-15 11:42:29 -060076 dout(" skipping last %llu, final file extent %llu~%llu\n",
77 orig_len - *plen, off, *plen);
78 }
79
Alex Elder75d1c942013-03-13 20:50:00 -050080 dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
Sage Weilf24e9982009-10-06 11:31:10 -070081
Alex Elder3ff5f382013-02-15 22:10:17 -060082 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -070083}
84
Alex Elderc54d47b2013-04-03 01:28:57 -050085static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
86{
87 memset(osd_data, 0, sizeof (*osd_data));
88 osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
89}
90
Alex Eldera4ce40a2013-04-05 01:27:12 -050091static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -050092 struct page **pages, u64 length, u32 alignment,
93 bool pages_from_pool, bool own_pages)
94{
95 osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
96 osd_data->pages = pages;
97 osd_data->length = length;
98 osd_data->alignment = alignment;
99 osd_data->pages_from_pool = pages_from_pool;
100 osd_data->own_pages = own_pages;
101}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500102
Alex Eldera4ce40a2013-04-05 01:27:12 -0500103static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -0500104 struct ceph_pagelist *pagelist)
105{
106 osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
107 osd_data->pagelist = pagelist;
108}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500109
110#ifdef CONFIG_BLOCK
Alex Eldera4ce40a2013-04-05 01:27:12 -0500111static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
Alex Elder43bfe5d2013-04-03 01:28:57 -0500112 struct bio *bio, size_t bio_length)
113{
114 osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
115 osd_data->bio = bio;
116 osd_data->bio_length = bio_length;
117}
Alex Elder43bfe5d2013-04-03 01:28:57 -0500118#endif /* CONFIG_BLOCK */
119
Alex Elder863c7eb2013-04-15 14:50:36 -0500120#define osd_req_op_data(oreq, whch, typ, fld) \
121 ({ \
122 BUG_ON(whch >= (oreq)->r_num_ops); \
123 &(oreq)->r_ops[whch].typ.fld; \
124 })
125
Alex Elder49719772013-02-11 12:33:24 -0600126static struct ceph_osd_data *
127osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
128{
129 BUG_ON(which >= osd_req->r_num_ops);
130
131 return &osd_req->r_ops[which].raw_data_in;
132}
133
Alex Eldera4ce40a2013-04-05 01:27:12 -0500134struct ceph_osd_data *
135osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500136 unsigned int which)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500137{
Alex Elder863c7eb2013-04-15 14:50:36 -0500138 return osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500139}
140EXPORT_SYMBOL(osd_req_op_extent_osd_data);
141
142struct ceph_osd_data *
Alex Eldera4ce40a2013-04-05 01:27:12 -0500143osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
144 unsigned int which)
145{
Alex Elder863c7eb2013-04-15 14:50:36 -0500146 return osd_req_op_data(osd_req, which, cls, response_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500147}
148EXPORT_SYMBOL(osd_req_op_cls_response_data); /* ??? */
149
Alex Elder49719772013-02-11 12:33:24 -0600150void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
151 unsigned int which, struct page **pages,
152 u64 length, u32 alignment,
153 bool pages_from_pool, bool own_pages)
154{
155 struct ceph_osd_data *osd_data;
156
157 osd_data = osd_req_op_raw_data_in(osd_req, which);
158 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
159 pages_from_pool, own_pages);
160}
161EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
162
Alex Eldera4ce40a2013-04-05 01:27:12 -0500163void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500164 unsigned int which, struct page **pages,
165 u64 length, u32 alignment,
Alex Eldera4ce40a2013-04-05 01:27:12 -0500166 bool pages_from_pool, bool own_pages)
167{
168 struct ceph_osd_data *osd_data;
169
Alex Elder863c7eb2013-04-15 14:50:36 -0500170 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500171 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
172 pages_from_pool, own_pages);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500173}
174EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
175
176void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500177 unsigned int which, struct ceph_pagelist *pagelist)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500178{
179 struct ceph_osd_data *osd_data;
180
Alex Elder863c7eb2013-04-15 14:50:36 -0500181 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500182 ceph_osd_data_pagelist_init(osd_data, pagelist);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500183}
184EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
185
186#ifdef CONFIG_BLOCK
187void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
Alex Elder406e2c92013-04-15 14:50:36 -0500188 unsigned int which, struct bio *bio, size_t bio_length)
Alex Eldera4ce40a2013-04-05 01:27:12 -0500189{
190 struct ceph_osd_data *osd_data;
Alex Elder863c7eb2013-04-15 14:50:36 -0500191
192 osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500193 ceph_osd_data_bio_init(osd_data, bio, bio_length);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500194}
195EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
196#endif /* CONFIG_BLOCK */
197
198static void osd_req_op_cls_request_info_pagelist(
199 struct ceph_osd_request *osd_req,
200 unsigned int which, struct ceph_pagelist *pagelist)
201{
202 struct ceph_osd_data *osd_data;
203
Alex Elder863c7eb2013-04-15 14:50:36 -0500204 osd_data = osd_req_op_data(osd_req, which, cls, request_info);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500205 ceph_osd_data_pagelist_init(osd_data, pagelist);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500206}
207
Alex Elder04017e22013-04-05 14:46:02 -0500208void osd_req_op_cls_request_data_pagelist(
209 struct ceph_osd_request *osd_req,
210 unsigned int which, struct ceph_pagelist *pagelist)
211{
212 struct ceph_osd_data *osd_data;
213
Alex Elder863c7eb2013-04-15 14:50:36 -0500214 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
Alex Elder04017e22013-04-05 14:46:02 -0500215 ceph_osd_data_pagelist_init(osd_data, pagelist);
216}
217EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
218
Alex Elder6c57b552013-04-19 15:34:49 -0500219void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
220 unsigned int which, struct page **pages, u64 length,
221 u32 alignment, bool pages_from_pool, bool own_pages)
222{
223 struct ceph_osd_data *osd_data;
224
225 osd_data = osd_req_op_data(osd_req, which, cls, request_data);
226 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
227 pages_from_pool, own_pages);
228}
229EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
230
Alex Eldera4ce40a2013-04-05 01:27:12 -0500231void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
232 unsigned int which, struct page **pages, u64 length,
233 u32 alignment, bool pages_from_pool, bool own_pages)
234{
235 struct ceph_osd_data *osd_data;
236
Alex Elder863c7eb2013-04-15 14:50:36 -0500237 osd_data = osd_req_op_data(osd_req, which, cls, response_data);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500238 ceph_osd_data_pages_init(osd_data, pages, length, alignment,
239 pages_from_pool, own_pages);
Alex Eldera4ce40a2013-04-05 01:27:12 -0500240}
241EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
242
Alex Elder23c08a9cb2013-04-03 01:28:58 -0500243static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
244{
245 switch (osd_data->type) {
246 case CEPH_OSD_DATA_TYPE_NONE:
247 return 0;
248 case CEPH_OSD_DATA_TYPE_PAGES:
249 return osd_data->length;
250 case CEPH_OSD_DATA_TYPE_PAGELIST:
251 return (u64)osd_data->pagelist->length;
252#ifdef CONFIG_BLOCK
253 case CEPH_OSD_DATA_TYPE_BIO:
254 return (u64)osd_data->bio_length;
255#endif /* CONFIG_BLOCK */
256 default:
257 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
258 return 0;
259 }
260}
261
Alex Elderc54d47b2013-04-03 01:28:57 -0500262static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
263{
Alex Elder54764922013-04-05 01:27:12 -0500264 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
Alex Elderc54d47b2013-04-03 01:28:57 -0500265 int num_pages;
266
267 num_pages = calc_pages_for((u64)osd_data->alignment,
268 (u64)osd_data->length);
269 ceph_release_page_vector(osd_data->pages, num_pages);
270 }
Alex Elder54764922013-04-05 01:27:12 -0500271 ceph_osd_data_init(osd_data);
272}
273
274static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
275 unsigned int which)
276{
277 struct ceph_osd_req_op *op;
278
279 BUG_ON(which >= osd_req->r_num_ops);
280 op = &osd_req->r_ops[which];
281
282 switch (op->op) {
283 case CEPH_OSD_OP_READ:
284 case CEPH_OSD_OP_WRITE:
285 ceph_osd_data_release(&op->extent.osd_data);
286 break;
287 case CEPH_OSD_OP_CALL:
288 ceph_osd_data_release(&op->cls.request_info);
Alex Elder04017e22013-04-05 14:46:02 -0500289 ceph_osd_data_release(&op->cls.request_data);
Alex Elder54764922013-04-05 01:27:12 -0500290 ceph_osd_data_release(&op->cls.response_data);
291 break;
292 default:
293 break;
294 }
Alex Elderc54d47b2013-04-03 01:28:57 -0500295}
296
Sage Weilf24e9982009-10-06 11:31:10 -0700297/*
298 * requests
299 */
Sage Weil415e49a2009-12-07 13:37:03 -0800300void ceph_osdc_release_request(struct kref *kref)
Sage Weilf24e9982009-10-06 11:31:10 -0700301{
Alex Elderc54d47b2013-04-03 01:28:57 -0500302 struct ceph_osd_request *req;
Alex Elder54764922013-04-05 01:27:12 -0500303 unsigned int which;
Sage Weil415e49a2009-12-07 13:37:03 -0800304
Alex Elderc54d47b2013-04-03 01:28:57 -0500305 req = container_of(kref, struct ceph_osd_request, r_kref);
Sage Weil415e49a2009-12-07 13:37:03 -0800306 if (req->r_request)
307 ceph_msg_put(req->r_request);
Alex Elderace6d3a2013-04-01 16:12:14 -0500308 if (req->r_reply) {
Alex Elder8921d112012-06-01 14:56:43 -0500309 ceph_msg_revoke_incoming(req->r_reply);
Alex Elderab8cb342012-06-04 14:43:32 -0500310 ceph_msg_put(req->r_reply);
Alex Elderace6d3a2013-04-01 16:12:14 -0500311 }
Alex Elder0fff87e2013-02-14 12:16:43 -0600312
Alex Elder54764922013-04-05 01:27:12 -0500313 for (which = 0; which < req->r_num_ops; which++)
314 osd_req_op_data_release(req, which);
Alex Elder0fff87e2013-02-14 12:16:43 -0600315
Sage Weil415e49a2009-12-07 13:37:03 -0800316 ceph_put_snap_context(req->r_snapc);
317 if (req->r_mempool)
318 mempool_free(req, req->r_osdc->req_mempool);
319 else
Alex Elder5522ae02013-05-01 12:43:04 -0500320 kmem_cache_free(ceph_osd_request_cache, req);
321
Sage Weilf24e9982009-10-06 11:31:10 -0700322}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700323EXPORT_SYMBOL(ceph_osdc_release_request);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700324
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700325struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700326 struct ceph_snap_context *snapc,
Sage Weil1b83bef2013-02-25 16:11:12 -0800327 unsigned int num_ops,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700328 bool use_mempool,
Alex Elder54a54002012-11-13 21:11:15 -0600329 gfp_t gfp_flags)
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700330{
331 struct ceph_osd_request *req;
332 struct ceph_msg *msg;
Sage Weil1b83bef2013-02-25 16:11:12 -0800333 size_t msg_size;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700334
Alex Elder79528732013-04-03 21:32:51 -0500335 BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
336 BUG_ON(num_ops > CEPH_OSD_MAX_OP);
337
Sage Weil1b83bef2013-02-25 16:11:12 -0800338 msg_size = 4 + 4 + 8 + 8 + 4+8;
339 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
340 msg_size += 1 + 8 + 4 + 4; /* pg_t */
Ilya Dryomov2d0ebc52014-01-27 17:40:18 +0200341 msg_size += 4 + CEPH_MAX_OID_NAME_LEN; /* oid */
Sage Weil1b83bef2013-02-25 16:11:12 -0800342 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
343 msg_size += 8; /* snapid */
344 msg_size += 8; /* snap_seq */
345 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
346 msg_size += 4;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700347
348 if (use_mempool) {
349 req = mempool_alloc(osdc->req_mempool, gfp_flags);
350 memset(req, 0, sizeof(*req));
351 } else {
Alex Elder5522ae02013-05-01 12:43:04 -0500352 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700353 }
354 if (req == NULL)
355 return NULL;
356
357 req->r_osdc = osdc;
358 req->r_mempool = use_mempool;
Alex Elder79528732013-04-03 21:32:51 -0500359 req->r_num_ops = num_ops;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700360
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700361 kref_init(&req->r_kref);
362 init_completion(&req->r_completion);
363 init_completion(&req->r_safe_completion);
Alex Eldera978fa22012-12-17 12:23:48 -0600364 RB_CLEAR_NODE(&req->r_node);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700365 INIT_LIST_HEAD(&req->r_unsafe_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700366 INIT_LIST_HEAD(&req->r_linger_item);
367 INIT_LIST_HEAD(&req->r_linger_osd);
Sage Weil935b6392011-09-16 11:13:17 -0700368 INIT_LIST_HEAD(&req->r_req_lru_item);
Sage Weilcd430452012-07-09 14:31:41 -0700369 INIT_LIST_HEAD(&req->r_osd_item);
370
Ilya Dryomov3c972c92014-01-27 17:40:20 +0200371 req->r_base_oloc.pool = -1;
Ilya Dryomov205ee1182014-01-27 17:40:20 +0200372 req->r_target_oloc.pool = -1;
Ilya Dryomov221165252014-01-27 17:40:18 +0200373
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700374 /* create reply message */
375 if (use_mempool)
376 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
377 else
378 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
Sage Weilb61c2762011-08-09 15:03:46 -0700379 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700380 if (!msg) {
381 ceph_osdc_put_request(req);
382 return NULL;
383 }
384 req->r_reply = msg;
385
386 /* create request message; allow space for oid */
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700387 if (use_mempool)
388 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
389 else
Sage Weilb61c2762011-08-09 15:03:46 -0700390 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700391 if (!msg) {
392 ceph_osdc_put_request(req);
393 return NULL;
394 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700395
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700396 memset(msg->front.iov_base, 0, msg->front.iov_len);
397
398 req->r_request = msg;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700399
400 return req;
401}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700402EXPORT_SYMBOL(ceph_osdc_alloc_request);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700403
Alex Eldera8dd0a32013-03-13 20:50:00 -0500404static bool osd_req_opcode_valid(u16 opcode)
405{
406 switch (opcode) {
407 case CEPH_OSD_OP_READ:
408 case CEPH_OSD_OP_STAT:
409 case CEPH_OSD_OP_MAPEXT:
410 case CEPH_OSD_OP_MASKTRUNC:
411 case CEPH_OSD_OP_SPARSE_READ:
412 case CEPH_OSD_OP_NOTIFY:
413 case CEPH_OSD_OP_NOTIFY_ACK:
414 case CEPH_OSD_OP_ASSERT_VER:
415 case CEPH_OSD_OP_WRITE:
416 case CEPH_OSD_OP_WRITEFULL:
417 case CEPH_OSD_OP_TRUNCATE:
418 case CEPH_OSD_OP_ZERO:
419 case CEPH_OSD_OP_DELETE:
420 case CEPH_OSD_OP_APPEND:
421 case CEPH_OSD_OP_STARTSYNC:
422 case CEPH_OSD_OP_SETTRUNC:
423 case CEPH_OSD_OP_TRIMTRUNC:
424 case CEPH_OSD_OP_TMAPUP:
425 case CEPH_OSD_OP_TMAPPUT:
426 case CEPH_OSD_OP_TMAPGET:
427 case CEPH_OSD_OP_CREATE:
428 case CEPH_OSD_OP_ROLLBACK:
429 case CEPH_OSD_OP_WATCH:
430 case CEPH_OSD_OP_OMAPGETKEYS:
431 case CEPH_OSD_OP_OMAPGETVALS:
432 case CEPH_OSD_OP_OMAPGETHEADER:
433 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
434 case CEPH_OSD_OP_OMAPSETVALS:
435 case CEPH_OSD_OP_OMAPSETHEADER:
436 case CEPH_OSD_OP_OMAPCLEAR:
437 case CEPH_OSD_OP_OMAPRMKEYS:
438 case CEPH_OSD_OP_OMAP_CMP:
Ilya Dryomovc647b8a2014-02-25 16:22:27 +0200439 case CEPH_OSD_OP_SETALLOCHINT:
Alex Eldera8dd0a32013-03-13 20:50:00 -0500440 case CEPH_OSD_OP_CLONERANGE:
441 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
442 case CEPH_OSD_OP_SRC_CMPXATTR:
443 case CEPH_OSD_OP_GETXATTR:
444 case CEPH_OSD_OP_GETXATTRS:
445 case CEPH_OSD_OP_CMPXATTR:
446 case CEPH_OSD_OP_SETXATTR:
447 case CEPH_OSD_OP_SETXATTRS:
448 case CEPH_OSD_OP_RESETXATTRS:
449 case CEPH_OSD_OP_RMXATTR:
450 case CEPH_OSD_OP_PULL:
451 case CEPH_OSD_OP_PUSH:
452 case CEPH_OSD_OP_BALANCEREADS:
453 case CEPH_OSD_OP_UNBALANCEREADS:
454 case CEPH_OSD_OP_SCRUB:
455 case CEPH_OSD_OP_SCRUB_RESERVE:
456 case CEPH_OSD_OP_SCRUB_UNRESERVE:
457 case CEPH_OSD_OP_SCRUB_STOP:
458 case CEPH_OSD_OP_SCRUB_MAP:
459 case CEPH_OSD_OP_WRLOCK:
460 case CEPH_OSD_OP_WRUNLOCK:
461 case CEPH_OSD_OP_RDLOCK:
462 case CEPH_OSD_OP_RDUNLOCK:
463 case CEPH_OSD_OP_UPLOCK:
464 case CEPH_OSD_OP_DNLOCK:
465 case CEPH_OSD_OP_CALL:
466 case CEPH_OSD_OP_PGLS:
467 case CEPH_OSD_OP_PGLS_FILTER:
468 return true;
469 default:
470 return false;
471 }
472}
473
Alex Elder33803f32013-03-13 20:50:00 -0500474/*
475 * This is an osd op init function for opcodes that have no data or
476 * other information associated with them. It also serves as a
477 * common init routine for all the other init functions, below.
478 */
Alex Elderc99d2d42013-04-05 01:27:11 -0500479static struct ceph_osd_req_op *
Alex Elder49719772013-02-11 12:33:24 -0600480_osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
Alex Elderc99d2d42013-04-05 01:27:11 -0500481 u16 opcode)
Alex Elder33803f32013-03-13 20:50:00 -0500482{
Alex Elderc99d2d42013-04-05 01:27:11 -0500483 struct ceph_osd_req_op *op;
484
485 BUG_ON(which >= osd_req->r_num_ops);
Alex Elder33803f32013-03-13 20:50:00 -0500486 BUG_ON(!osd_req_opcode_valid(opcode));
487
Alex Elderc99d2d42013-04-05 01:27:11 -0500488 op = &osd_req->r_ops[which];
Alex Elder33803f32013-03-13 20:50:00 -0500489 memset(op, 0, sizeof (*op));
Alex Elder33803f32013-03-13 20:50:00 -0500490 op->op = opcode;
Alex Elderc99d2d42013-04-05 01:27:11 -0500491
492 return op;
Alex Elder33803f32013-03-13 20:50:00 -0500493}
494
Alex Elder49719772013-02-11 12:33:24 -0600495void osd_req_op_init(struct ceph_osd_request *osd_req,
496 unsigned int which, u16 opcode)
497{
498 (void)_osd_req_op_init(osd_req, which, opcode);
499}
500EXPORT_SYMBOL(osd_req_op_init);
501
Alex Elderc99d2d42013-04-05 01:27:11 -0500502void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
503 unsigned int which, u16 opcode,
Alex Elder33803f32013-03-13 20:50:00 -0500504 u64 offset, u64 length,
505 u64 truncate_size, u32 truncate_seq)
506{
Alex Elder49719772013-02-11 12:33:24 -0600507 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder33803f32013-03-13 20:50:00 -0500508 size_t payload_len = 0;
509
Li Wangad7a60d2013-08-15 11:51:44 +0800510 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
511 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
512 opcode != CEPH_OSD_OP_TRUNCATE);
Alex Elder33803f32013-03-13 20:50:00 -0500513
Alex Elder33803f32013-03-13 20:50:00 -0500514 op->extent.offset = offset;
515 op->extent.length = length;
516 op->extent.truncate_size = truncate_size;
517 op->extent.truncate_seq = truncate_seq;
518 if (opcode == CEPH_OSD_OP_WRITE)
519 payload_len += length;
520
521 op->payload_len = payload_len;
522}
523EXPORT_SYMBOL(osd_req_op_extent_init);
524
Alex Elderc99d2d42013-04-05 01:27:11 -0500525void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
526 unsigned int which, u64 length)
Alex Eldere5975c72013-03-14 14:09:05 -0500527{
Alex Elderc99d2d42013-04-05 01:27:11 -0500528 struct ceph_osd_req_op *op;
529 u64 previous;
530
531 BUG_ON(which >= osd_req->r_num_ops);
532 op = &osd_req->r_ops[which];
533 previous = op->extent.length;
Alex Eldere5975c72013-03-14 14:09:05 -0500534
535 if (length == previous)
536 return; /* Nothing to do */
537 BUG_ON(length > previous);
538
539 op->extent.length = length;
540 op->payload_len -= previous - length;
541}
542EXPORT_SYMBOL(osd_req_op_extent_update);
543
Alex Elderc99d2d42013-04-05 01:27:11 -0500544void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
Alex Elder04017e22013-04-05 14:46:02 -0500545 u16 opcode, const char *class, const char *method)
Alex Elder33803f32013-03-13 20:50:00 -0500546{
Alex Elder49719772013-02-11 12:33:24 -0600547 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder5f562df2013-04-05 01:27:12 -0500548 struct ceph_pagelist *pagelist;
Alex Elder33803f32013-03-13 20:50:00 -0500549 size_t payload_len = 0;
550 size_t size;
551
552 BUG_ON(opcode != CEPH_OSD_OP_CALL);
553
Alex Elder5f562df2013-04-05 01:27:12 -0500554 pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
555 BUG_ON(!pagelist);
556 ceph_pagelist_init(pagelist);
557
Alex Elder33803f32013-03-13 20:50:00 -0500558 op->cls.class_name = class;
559 size = strlen(class);
560 BUG_ON(size > (size_t) U8_MAX);
561 op->cls.class_len = size;
Alex Elder5f562df2013-04-05 01:27:12 -0500562 ceph_pagelist_append(pagelist, class, size);
Alex Elder33803f32013-03-13 20:50:00 -0500563 payload_len += size;
564
565 op->cls.method_name = method;
566 size = strlen(method);
567 BUG_ON(size > (size_t) U8_MAX);
568 op->cls.method_len = size;
Alex Elder5f562df2013-04-05 01:27:12 -0500569 ceph_pagelist_append(pagelist, method, size);
Alex Elder33803f32013-03-13 20:50:00 -0500570 payload_len += size;
571
Alex Eldera4ce40a2013-04-05 01:27:12 -0500572 osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
Alex Elder5f562df2013-04-05 01:27:12 -0500573
Alex Elder33803f32013-03-13 20:50:00 -0500574 op->cls.argc = 0; /* currently unused */
575
576 op->payload_len = payload_len;
577}
578EXPORT_SYMBOL(osd_req_op_cls_init);
Alex Elder8c042b02013-04-03 01:28:58 -0500579
Alex Elderc99d2d42013-04-05 01:27:11 -0500580void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
581 unsigned int which, u16 opcode,
Alex Elder33803f32013-03-13 20:50:00 -0500582 u64 cookie, u64 version, int flag)
583{
Alex Elder49719772013-02-11 12:33:24 -0600584 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
Alex Elder33803f32013-03-13 20:50:00 -0500585
Alex Elderc99d2d42013-04-05 01:27:11 -0500586 BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
Alex Elder33803f32013-03-13 20:50:00 -0500587
588 op->watch.cookie = cookie;
Alex Elder9ef1ee52013-04-21 16:51:50 -0500589 op->watch.ver = version;
Alex Elder33803f32013-03-13 20:50:00 -0500590 if (opcode == CEPH_OSD_OP_WATCH && flag)
Alex Elderc99d2d42013-04-05 01:27:11 -0500591 op->watch.flag = (u8)1;
Alex Elder33803f32013-03-13 20:50:00 -0500592}
593EXPORT_SYMBOL(osd_req_op_watch_init);
594
Ilya Dryomovc647b8a2014-02-25 16:22:27 +0200595void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req,
596 unsigned int which,
597 u64 expected_object_size,
598 u64 expected_write_size)
599{
600 struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which,
601 CEPH_OSD_OP_SETALLOCHINT);
602
603 op->alloc_hint.expected_object_size = expected_object_size;
604 op->alloc_hint.expected_write_size = expected_write_size;
605
606 /*
607 * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed
608 * not worth a feature bit. Set FAILOK per-op flag to make
609 * sure older osds don't trip over an unsupported opcode.
610 */
611 op->flags |= CEPH_OSD_OP_FLAG_FAILOK;
612}
613EXPORT_SYMBOL(osd_req_op_alloc_hint_init);
614
Alex Elder90af3602013-04-05 14:46:01 -0500615static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
Alex Elderec9123c2013-04-05 01:27:12 -0500616 struct ceph_osd_data *osd_data)
617{
618 u64 length = ceph_osd_data_length(osd_data);
619
620 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
621 BUG_ON(length > (u64) SIZE_MAX);
622 if (length)
Alex Elder90af3602013-04-05 14:46:01 -0500623 ceph_msg_data_add_pages(msg, osd_data->pages,
Alex Elderec9123c2013-04-05 01:27:12 -0500624 length, osd_data->alignment);
625 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
626 BUG_ON(!length);
Alex Elder90af3602013-04-05 14:46:01 -0500627 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
Alex Elderec9123c2013-04-05 01:27:12 -0500628#ifdef CONFIG_BLOCK
629 } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
Alex Elder90af3602013-04-05 14:46:01 -0500630 ceph_msg_data_add_bio(msg, osd_data->bio, length);
Alex Elderec9123c2013-04-05 01:27:12 -0500631#endif
632 } else {
633 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
634 }
635}
636
Alex Elder175face2013-03-08 13:35:36 -0600637static u64 osd_req_encode_op(struct ceph_osd_request *req,
Alex Elder79528732013-04-03 21:32:51 -0500638 struct ceph_osd_op *dst, unsigned int which)
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700639{
Alex Elder79528732013-04-03 21:32:51 -0500640 struct ceph_osd_req_op *src;
Alex Elder04017e22013-04-05 14:46:02 -0500641 struct ceph_osd_data *osd_data;
Alex Elder54d50642013-04-03 01:28:58 -0500642 u64 request_data_len = 0;
Alex Elder04017e22013-04-05 14:46:02 -0500643 u64 data_length;
Alex Elder175face2013-03-08 13:35:36 -0600644
Alex Elder79528732013-04-03 21:32:51 -0500645 BUG_ON(which >= req->r_num_ops);
646 src = &req->r_ops[which];
Alex Eldera8dd0a32013-03-13 20:50:00 -0500647 if (WARN_ON(!osd_req_opcode_valid(src->op))) {
648 pr_err("unrecognized osd opcode %d\n", src->op);
649
650 return 0;
651 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700652
Alex Elder065a68f2012-04-20 15:49:43 -0500653 switch (src->op) {
Alex Elderfbfab532013-02-08 09:55:48 -0600654 case CEPH_OSD_OP_STAT:
Alex Elder49719772013-02-11 12:33:24 -0600655 osd_data = &src->raw_data_in;
656 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Alex Elderfbfab532013-02-08 09:55:48 -0600657 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700658 case CEPH_OSD_OP_READ:
659 case CEPH_OSD_OP_WRITE:
Li Wangad7a60d2013-08-15 11:51:44 +0800660 case CEPH_OSD_OP_ZERO:
661 case CEPH_OSD_OP_DELETE:
662 case CEPH_OSD_OP_TRUNCATE:
Alex Elder175face2013-03-08 13:35:36 -0600663 if (src->op == CEPH_OSD_OP_WRITE)
Alex Elder54d50642013-04-03 01:28:58 -0500664 request_data_len = src->extent.length;
Alex Elder175face2013-03-08 13:35:36 -0600665 dst->extent.offset = cpu_to_le64(src->extent.offset);
666 dst->extent.length = cpu_to_le64(src->extent.length);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700667 dst->extent.truncate_size =
668 cpu_to_le64(src->extent.truncate_size);
669 dst->extent.truncate_seq =
670 cpu_to_le32(src->extent.truncate_seq);
Alex Elder04017e22013-04-05 14:46:02 -0500671 osd_data = &src->extent.osd_data;
Alex Elder54764922013-04-05 01:27:12 -0500672 if (src->op == CEPH_OSD_OP_WRITE)
Alex Elder04017e22013-04-05 14:46:02 -0500673 ceph_osdc_msg_data_add(req->r_request, osd_data);
Alex Elder54764922013-04-05 01:27:12 -0500674 else
Alex Elder04017e22013-04-05 14:46:02 -0500675 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700676 break;
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700677 case CEPH_OSD_OP_CALL:
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700678 dst->cls.class_len = src->cls.class_len;
679 dst->cls.method_len = src->cls.method_len;
Alex Elder04017e22013-04-05 14:46:02 -0500680 osd_data = &src->cls.request_info;
681 ceph_osdc_msg_data_add(req->r_request, osd_data);
682 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
683 request_data_len = osd_data->pagelist->length;
684
685 osd_data = &src->cls.request_data;
686 data_length = ceph_osd_data_length(osd_data);
687 if (data_length) {
688 BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
689 dst->cls.indata_len = cpu_to_le32(data_length);
690 ceph_osdc_msg_data_add(req->r_request, osd_data);
691 src->payload_len += data_length;
692 request_data_len += data_length;
693 }
694 osd_data = &src->cls.response_data;
695 ceph_osdc_msg_data_add(req->r_reply, osd_data);
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700696 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700697 case CEPH_OSD_OP_STARTSYNC:
698 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700699 case CEPH_OSD_OP_NOTIFY_ACK:
700 case CEPH_OSD_OP_WATCH:
701 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
702 dst->watch.ver = cpu_to_le64(src->watch.ver);
703 dst->watch.flag = src->watch.flag;
704 break;
Ilya Dryomovc647b8a2014-02-25 16:22:27 +0200705 case CEPH_OSD_OP_SETALLOCHINT:
706 dst->alloc_hint.expected_object_size =
707 cpu_to_le64(src->alloc_hint.expected_object_size);
708 dst->alloc_hint.expected_write_size =
709 cpu_to_le64(src->alloc_hint.expected_write_size);
710 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700711 default:
Alex Elder4c464592013-02-15 11:42:30 -0600712 pr_err("unsupported osd opcode %s\n",
Alex Elder8f63ca22013-03-04 11:08:29 -0600713 ceph_osd_op_name(src->op));
Alex Elder4c464592013-02-15 11:42:30 -0600714 WARN_ON(1);
Alex Eldera8dd0a32013-03-13 20:50:00 -0500715
716 return 0;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700717 }
Ilya Dryomov7b25bf52014-02-25 16:22:26 +0200718
Alex Eldera8dd0a32013-03-13 20:50:00 -0500719 dst->op = cpu_to_le16(src->op);
Ilya Dryomov7b25bf52014-02-25 16:22:26 +0200720 dst->flags = cpu_to_le32(src->flags);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700721 dst->payload_len = cpu_to_le32(src->payload_len);
Alex Elder175face2013-03-08 13:35:36 -0600722
Alex Elder54d50642013-04-03 01:28:58 -0500723 return request_data_len;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700724}
725
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700726/*
Sage Weilf24e9982009-10-06 11:31:10 -0700727 * build new request AND message, calculate layout, and adjust file
728 * extent as needed.
729 *
730 * if the file was recently truncated, we include information about its
731 * old and new size so that the object can be updated appropriately. (we
732 * avoid synchronously deleting truncated objects because it's slow.)
733 *
734 * if @do_sync, include a 'startsync' command so that the osd will flush
735 * data quickly.
736 */
737struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
738 struct ceph_file_layout *layout,
739 struct ceph_vino vino,
Alex Elderacead002013-03-14 14:09:05 -0500740 u64 off, u64 *plen, int num_ops,
Sage Weilf24e9982009-10-06 11:31:10 -0700741 int opcode, int flags,
742 struct ceph_snap_context *snapc,
Sage Weilf24e9982009-10-06 11:31:10 -0700743 u32 truncate_seq,
744 u64 truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -0600745 bool use_mempool)
Sage Weilf24e9982009-10-06 11:31:10 -0700746{
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700747 struct ceph_osd_request *req;
Alex Elder75d1c942013-03-13 20:50:00 -0500748 u64 objnum = 0;
749 u64 objoff = 0;
750 u64 objlen = 0;
Alex Elderd18d1e22013-03-13 20:50:01 -0500751 u32 object_size;
752 u64 object_base;
Sage Weil68162822012-09-24 21:01:02 -0700753 int r;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700754
Li Wangad7a60d2013-08-15 11:51:44 +0800755 BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
756 opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
757 opcode != CEPH_OSD_OP_TRUNCATE);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700758
Alex Elderacead002013-03-14 14:09:05 -0500759 req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
Alex Elderae7ca4a32012-11-13 21:11:15 -0600760 GFP_NOFS);
Sage Weil4ad12622011-05-03 09:23:36 -0700761 if (!req)
Sage Weil68162822012-09-24 21:01:02 -0700762 return ERR_PTR(-ENOMEM);
Alex Elder79528732013-04-03 21:32:51 -0500763
Alex Elderd178a9e2012-11-13 21:11:15 -0600764 req->r_flags = flags;
Sage Weilf24e9982009-10-06 11:31:10 -0700765
766 /* calculate max write size */
Alex Eldera19dadf2013-03-13 20:50:01 -0500767 r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
Alex Elder3ff5f382013-02-15 22:10:17 -0600768 if (r < 0) {
769 ceph_osdc_put_request(req);
Sage Weil68162822012-09-24 21:01:02 -0700770 return ERR_PTR(r);
Alex Elder3ff5f382013-02-15 22:10:17 -0600771 }
Alex Eldera19dadf2013-03-13 20:50:01 -0500772
Alex Elderd18d1e22013-03-13 20:50:01 -0500773 object_size = le32_to_cpu(layout->fl_object_size);
774 object_base = off - objoff;
Yan, Zhengccca4e32013-06-02 18:40:23 +0800775 if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
776 if (truncate_size <= object_base) {
777 truncate_size = 0;
778 } else {
779 truncate_size -= object_base;
780 if (truncate_size > object_size)
781 truncate_size = object_size;
782 }
Alex Eldera19dadf2013-03-13 20:50:01 -0500783 }
Alex Elderd18d1e22013-03-13 20:50:01 -0500784
Alex Elderc99d2d42013-04-05 01:27:11 -0500785 osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
Alex Elderb0270322013-03-13 20:50:01 -0500786 truncate_size, truncate_seq);
Alex Elder8c042b02013-04-03 01:28:58 -0500787
Alex Elderacead002013-03-14 14:09:05 -0500788 /*
789 * A second op in the ops array means the caller wants to
790 * also issue a include a 'startsync' command so that the
791 * osd will flush data quickly.
792 */
793 if (num_ops > 1)
Alex Elderc99d2d42013-04-05 01:27:11 -0500794 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
Alex Elderd18d1e22013-03-13 20:50:01 -0500795
Ilya Dryomov3c972c92014-01-27 17:40:20 +0200796 req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout);
Sage Weilf24e9982009-10-06 11:31:10 -0700797
Ilya Dryomov3c972c92014-01-27 17:40:20 +0200798 snprintf(req->r_base_oid.name, sizeof(req->r_base_oid.name),
Ilya Dryomov4295f222014-01-27 17:40:18 +0200799 "%llx.%08llx", vino.ino, objnum);
Ilya Dryomov3c972c92014-01-27 17:40:20 +0200800 req->r_base_oid.name_len = strlen(req->r_base_oid.name);
Alex Elderdbe0fc42013-02-15 22:10:17 -0600801
Sage Weilf24e9982009-10-06 11:31:10 -0700802 return req;
803}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700804EXPORT_SYMBOL(ceph_osdc_new_request);
Sage Weilf24e9982009-10-06 11:31:10 -0700805
806/*
807 * We keep osd requests in an rbtree, sorted by ->r_tid.
808 */
809static void __insert_request(struct ceph_osd_client *osdc,
810 struct ceph_osd_request *new)
811{
812 struct rb_node **p = &osdc->requests.rb_node;
813 struct rb_node *parent = NULL;
814 struct ceph_osd_request *req = NULL;
815
816 while (*p) {
817 parent = *p;
818 req = rb_entry(parent, struct ceph_osd_request, r_node);
819 if (new->r_tid < req->r_tid)
820 p = &(*p)->rb_left;
821 else if (new->r_tid > req->r_tid)
822 p = &(*p)->rb_right;
823 else
824 BUG();
825 }
826
827 rb_link_node(&new->r_node, parent, p);
828 rb_insert_color(&new->r_node, &osdc->requests);
829}
830
831static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
832 u64 tid)
833{
834 struct ceph_osd_request *req;
835 struct rb_node *n = osdc->requests.rb_node;
836
837 while (n) {
838 req = rb_entry(n, struct ceph_osd_request, r_node);
839 if (tid < req->r_tid)
840 n = n->rb_left;
841 else if (tid > req->r_tid)
842 n = n->rb_right;
843 else
844 return req;
845 }
846 return NULL;
847}
848
849static struct ceph_osd_request *
850__lookup_request_ge(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 if (!n->rb_left)
860 return req;
861 n = n->rb_left;
862 } else if (tid > req->r_tid) {
863 n = n->rb_right;
864 } else {
865 return req;
866 }
867 }
868 return NULL;
869}
870
Sage Weil6f6c7002011-01-17 20:34:08 -0800871/*
872 * Resubmit requests pending on the given osd.
873 */
874static void __kick_osd_requests(struct ceph_osd_client *osdc,
875 struct ceph_osd *osd)
876{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700877 struct ceph_osd_request *req, *nreq;
Alex Eldere02493c2013-03-25 18:16:11 -0500878 LIST_HEAD(resend);
Sage Weil6f6c7002011-01-17 20:34:08 -0800879 int err;
880
881 dout("__kick_osd_requests osd%d\n", osd->o_osd);
882 err = __reset_osd(osdc, osd);
Alex Elder685a7552012-12-07 09:57:58 -0600883 if (err)
Sage Weil6f6c7002011-01-17 20:34:08 -0800884 return;
Alex Eldere02493c2013-03-25 18:16:11 -0500885 /*
886 * Build up a list of requests to resend by traversing the
887 * osd's list of requests. Requests for a given object are
888 * sent in tid order, and that is also the order they're
889 * kept on this list. Therefore all requests that are in
890 * flight will be found first, followed by all requests that
891 * have not yet been sent. And to resend requests while
892 * preserving this order we will want to put any sent
893 * requests back on the front of the osd client's unsent
894 * list.
895 *
896 * So we build a separate ordered list of already-sent
897 * requests for the affected osd and splice it onto the
898 * front of the osd client's unsent list. Once we've seen a
899 * request that has not yet been sent we're done. Those
900 * requests are already sitting right where they belong.
901 */
Sage Weil6f6c7002011-01-17 20:34:08 -0800902 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
Alex Eldere02493c2013-03-25 18:16:11 -0500903 if (!req->r_sent)
904 break;
905 list_move_tail(&req->r_req_lru_item, &resend);
906 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
Sage Weil6f6c7002011-01-17 20:34:08 -0800907 osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700908 if (!req->r_linger)
909 req->r_flags |= CEPH_OSD_FLAG_RETRY;
910 }
Alex Eldere02493c2013-03-25 18:16:11 -0500911 list_splice(&resend, &osdc->req_unsent);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700912
Alex Eldere02493c2013-03-25 18:16:11 -0500913 /*
914 * Linger requests are re-registered before sending, which
915 * sets up a new tid for each. We add them to the unsent
916 * list at the end to keep things in tid order.
917 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700918 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
919 r_linger_osd) {
Sage Weil77f38e02011-04-06 09:09:16 -0700920 /*
921 * reregister request prior to unregistering linger so
922 * that r_osd is preserved.
923 */
924 BUG_ON(!list_empty(&req->r_req_lru_item));
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700925 __register_request(osdc, req);
Alex Eldere02493c2013-03-25 18:16:11 -0500926 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
Alex Elderad885922013-03-25 18:16:11 -0500927 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
Sage Weil77f38e02011-04-06 09:09:16 -0700928 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700929 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
930 osd->o_osd);
Sage Weil6f6c7002011-01-17 20:34:08 -0800931 }
932}
933
Sage Weilf24e9982009-10-06 11:31:10 -0700934/*
Sage Weil81b024e2009-10-09 10:29:18 -0700935 * If the osd connection drops, we need to resubmit all requests.
Sage Weilf24e9982009-10-06 11:31:10 -0700936 */
937static void osd_reset(struct ceph_connection *con)
938{
939 struct ceph_osd *osd = con->private;
940 struct ceph_osd_client *osdc;
941
942 if (!osd)
943 return;
944 dout("osd_reset osd%d\n", osd->o_osd);
945 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -0700946 down_read(&osdc->map_sem);
Sage Weil83aff952012-11-28 12:28:24 -0800947 mutex_lock(&osdc->request_mutex);
948 __kick_osd_requests(osdc, osd);
Alex Elderf9d25192013-02-15 11:42:29 -0600949 __send_queued(osdc);
Sage Weil83aff952012-11-28 12:28:24 -0800950 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -0700951 up_read(&osdc->map_sem);
952}
953
954/*
955 * Track open sessions with osds.
956 */
Alex Eldere10006f2012-05-26 23:26:43 -0500957static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
Sage Weilf24e9982009-10-06 11:31:10 -0700958{
959 struct ceph_osd *osd;
960
961 osd = kzalloc(sizeof(*osd), GFP_NOFS);
962 if (!osd)
963 return NULL;
964
965 atomic_set(&osd->o_ref, 1);
966 osd->o_osdc = osdc;
Alex Eldere10006f2012-05-26 23:26:43 -0500967 osd->o_osd = onum;
Alex Elderf4077312012-12-06 07:22:04 -0600968 RB_CLEAR_NODE(&osd->o_node);
Sage Weilf24e9982009-10-06 11:31:10 -0700969 INIT_LIST_HEAD(&osd->o_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700970 INIT_LIST_HEAD(&osd->o_linger_requests);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800971 INIT_LIST_HEAD(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700972 osd->o_incarnation = 1;
973
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700974 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800975
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800976 INIT_LIST_HEAD(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -0700977 return osd;
978}
979
980static struct ceph_osd *get_osd(struct ceph_osd *osd)
981{
982 if (atomic_inc_not_zero(&osd->o_ref)) {
983 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
984 atomic_read(&osd->o_ref));
985 return osd;
986 } else {
987 dout("get_osd %p FAIL\n", osd);
988 return NULL;
989 }
990}
991
992static void put_osd(struct ceph_osd *osd)
993{
994 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
995 atomic_read(&osd->o_ref) - 1);
Alex Eldera2556512012-05-16 15:16:39 -0500996 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
Sage Weil79494d12010-05-27 14:15:49 -0700997 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
998
Sage Weil27859f92013-03-25 10:26:14 -0700999 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
Sage Weilf24e9982009-10-06 11:31:10 -07001000 kfree(osd);
Sage Weil79494d12010-05-27 14:15:49 -07001001 }
Sage Weilf24e9982009-10-06 11:31:10 -07001002}
1003
1004/*
1005 * remove an osd from our map
1006 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001007static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -07001008{
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001009 dout("__remove_osd %p\n", osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001010 BUG_ON(!list_empty(&osd->o_requests));
1011 rb_erase(&osd->o_node, &osdc->osds);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001012 list_del_init(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001013 ceph_con_close(&osd->o_con);
1014 put_osd(osd);
1015}
1016
Sage Weilaca420b2011-08-31 14:45:53 -07001017static void remove_all_osds(struct ceph_osd_client *osdc)
1018{
Jiaju Zhang048a9d22012-07-20 08:18:36 -05001019 dout("%s %p\n", __func__, osdc);
Sage Weilaca420b2011-08-31 14:45:53 -07001020 mutex_lock(&osdc->request_mutex);
1021 while (!RB_EMPTY_ROOT(&osdc->osds)) {
1022 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
1023 struct ceph_osd, o_node);
1024 __remove_osd(osdc, osd);
1025 }
1026 mutex_unlock(&osdc->request_mutex);
1027}
1028
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001029static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1030 struct ceph_osd *osd)
1031{
1032 dout("__move_osd_to_lru %p\n", osd);
1033 BUG_ON(!list_empty(&osd->o_osd_lru));
1034 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001035 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001036}
1037
1038static void __remove_osd_from_lru(struct ceph_osd *osd)
1039{
1040 dout("__remove_osd_from_lru %p\n", osd);
1041 if (!list_empty(&osd->o_osd_lru))
1042 list_del_init(&osd->o_osd_lru);
1043}
1044
Sage Weilaca420b2011-08-31 14:45:53 -07001045static void remove_old_osds(struct ceph_osd_client *osdc)
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001046{
1047 struct ceph_osd *osd, *nosd;
1048
1049 dout("__remove_old_osds %p\n", osdc);
1050 mutex_lock(&osdc->request_mutex);
1051 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
Sage Weilaca420b2011-08-31 14:45:53 -07001052 if (time_before(jiffies, osd->lru_ttl))
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001053 break;
1054 __remove_osd(osdc, osd);
1055 }
1056 mutex_unlock(&osdc->request_mutex);
1057}
1058
Sage Weilf24e9982009-10-06 11:31:10 -07001059/*
1060 * reset osd connect
1061 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001062static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -07001063{
Alex Elderc3acb182012-12-07 09:57:58 -06001064 struct ceph_entity_addr *peer_addr;
Sage Weilf24e9982009-10-06 11:31:10 -07001065
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001066 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001067 if (list_empty(&osd->o_requests) &&
1068 list_empty(&osd->o_linger_requests)) {
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001069 __remove_osd(osdc, osd);
Alex Elderc3acb182012-12-07 09:57:58 -06001070
1071 return -ENODEV;
1072 }
1073
1074 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1075 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1076 !ceph_con_opened(&osd->o_con)) {
1077 struct ceph_osd_request *req;
1078
Ilya Dryomov0b4af2e2014-01-16 19:18:27 +02001079 dout("osd addr hasn't changed and connection never opened, "
1080 "letting msgr retry\n");
Sage Weil87b315a2010-03-22 14:51:18 -07001081 /* touch each r_stamp for handle_timeout()'s benfit */
1082 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1083 req->r_stamp = jiffies;
Alex Elderc3acb182012-12-07 09:57:58 -06001084
1085 return -EAGAIN;
Sage Weilf24e9982009-10-06 11:31:10 -07001086 }
Alex Elderc3acb182012-12-07 09:57:58 -06001087
1088 ceph_con_close(&osd->o_con);
1089 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1090 osd->o_incarnation++;
1091
1092 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001093}
1094
1095static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1096{
1097 struct rb_node **p = &osdc->osds.rb_node;
1098 struct rb_node *parent = NULL;
1099 struct ceph_osd *osd = NULL;
1100
Sage Weilaca420b2011-08-31 14:45:53 -07001101 dout("__insert_osd %p osd%d\n", new, new->o_osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001102 while (*p) {
1103 parent = *p;
1104 osd = rb_entry(parent, struct ceph_osd, o_node);
1105 if (new->o_osd < osd->o_osd)
1106 p = &(*p)->rb_left;
1107 else if (new->o_osd > osd->o_osd)
1108 p = &(*p)->rb_right;
1109 else
1110 BUG();
1111 }
1112
1113 rb_link_node(&new->o_node, parent, p);
1114 rb_insert_color(&new->o_node, &osdc->osds);
1115}
1116
1117static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1118{
1119 struct ceph_osd *osd;
1120 struct rb_node *n = osdc->osds.rb_node;
1121
1122 while (n) {
1123 osd = rb_entry(n, struct ceph_osd, o_node);
1124 if (o < osd->o_osd)
1125 n = n->rb_left;
1126 else if (o > osd->o_osd)
1127 n = n->rb_right;
1128 else
1129 return osd;
1130 }
1131 return NULL;
1132}
1133
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001134static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1135{
1136 schedule_delayed_work(&osdc->timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001137 osdc->client->options->osd_keepalive_timeout * HZ);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001138}
1139
1140static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1141{
1142 cancel_delayed_work(&osdc->timeout_work);
1143}
Sage Weilf24e9982009-10-06 11:31:10 -07001144
1145/*
1146 * Register request, assign tid. If this is the first request, set up
1147 * the timeout event.
1148 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001149static void __register_request(struct ceph_osd_client *osdc,
1150 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001151{
Sage Weilf24e9982009-10-06 11:31:10 -07001152 req->r_tid = ++osdc->last_tid;
Sage Weil6df058c2009-12-22 11:24:33 -08001153 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
Sage Weil77f38e02011-04-06 09:09:16 -07001154 dout("__register_request %p tid %lld\n", req, req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07001155 __insert_request(osdc, req);
1156 ceph_osdc_get_request(req);
1157 osdc->num_requests++;
Sage Weilf24e9982009-10-06 11:31:10 -07001158 if (osdc->num_requests == 1) {
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001159 dout(" first request, scheduling timeout\n");
1160 __schedule_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001161 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001162}
1163
Sage Weilf24e9982009-10-06 11:31:10 -07001164/*
1165 * called under osdc->request_mutex
1166 */
1167static void __unregister_request(struct ceph_osd_client *osdc,
1168 struct ceph_osd_request *req)
1169{
Sage Weil35f9f8a2012-05-16 15:16:38 -05001170 if (RB_EMPTY_NODE(&req->r_node)) {
1171 dout("__unregister_request %p tid %lld not registered\n",
1172 req, req->r_tid);
1173 return;
1174 }
1175
Sage Weilf24e9982009-10-06 11:31:10 -07001176 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1177 rb_erase(&req->r_node, &osdc->requests);
1178 osdc->num_requests--;
1179
Sage Weil0ba64782009-10-08 16:57:16 -07001180 if (req->r_osd) {
1181 /* make sure the original request isn't in flight. */
Alex Elder6740a842012-06-01 14:56:43 -05001182 ceph_msg_revoke(req->r_request);
Sage Weil0ba64782009-10-08 16:57:16 -07001183
1184 list_del_init(&req->r_osd_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001185 if (list_empty(&req->r_osd->o_requests) &&
1186 list_empty(&req->r_osd->o_linger_requests)) {
1187 dout("moving osd to %p lru\n", req->r_osd);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001188 __move_osd_to_lru(osdc, req->r_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001189 }
Sage Weilfbdb9192011-03-29 12:11:06 -07001190 if (list_empty(&req->r_linger_item))
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001191 req->r_osd = NULL;
Sage Weil0ba64782009-10-08 16:57:16 -07001192 }
Sage Weilf24e9982009-10-06 11:31:10 -07001193
Alex Elder7d5f2482012-11-29 08:37:03 -06001194 list_del_init(&req->r_req_lru_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001195 ceph_osdc_put_request(req);
1196
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001197 if (osdc->num_requests == 0) {
1198 dout(" no requests, canceling timeout\n");
1199 __cancel_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001200 }
1201}
1202
1203/*
1204 * Cancel a previously queued request message
1205 */
1206static void __cancel_request(struct ceph_osd_request *req)
1207{
Sage Weil6bc18872010-09-27 10:18:52 -07001208 if (req->r_sent && req->r_osd) {
Alex Elder6740a842012-06-01 14:56:43 -05001209 ceph_msg_revoke(req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001210 req->r_sent = 0;
1211 }
1212}
1213
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001214static void __register_linger_request(struct ceph_osd_client *osdc,
1215 struct ceph_osd_request *req)
1216{
1217 dout("__register_linger_request %p\n", req);
Alex Elder96e4dac2013-05-22 20:54:25 -05001218 ceph_osdc_get_request(req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001219 list_add_tail(&req->r_linger_item, &osdc->req_linger);
Sage Weil6194ea82012-07-30 16:19:28 -07001220 if (req->r_osd)
1221 list_add_tail(&req->r_linger_osd,
1222 &req->r_osd->o_linger_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001223}
1224
1225static void __unregister_linger_request(struct ceph_osd_client *osdc,
1226 struct ceph_osd_request *req)
1227{
1228 dout("__unregister_linger_request %p\n", req);
Alex Elder61c74032012-12-06 09:37:23 -06001229 list_del_init(&req->r_linger_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001230 if (req->r_osd) {
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001231 list_del_init(&req->r_linger_osd);
1232
1233 if (list_empty(&req->r_osd->o_requests) &&
1234 list_empty(&req->r_osd->o_linger_requests)) {
1235 dout("moving osd to %p lru\n", req->r_osd);
1236 __move_osd_to_lru(osdc, req->r_osd);
1237 }
Sage Weilfbdb9192011-03-29 12:11:06 -07001238 if (list_empty(&req->r_osd_item))
1239 req->r_osd = NULL;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001240 }
Alex Elder96e4dac2013-05-22 20:54:25 -05001241 ceph_osdc_put_request(req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001242}
1243
1244void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1245 struct ceph_osd_request *req)
1246{
1247 mutex_lock(&osdc->request_mutex);
1248 if (req->r_linger) {
Alex Elderc10ebbf2013-05-09 14:56:32 -05001249 req->r_linger = 0;
Alex Elder96e4dac2013-05-22 20:54:25 -05001250 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001251 }
1252 mutex_unlock(&osdc->request_mutex);
1253}
1254EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1255
1256void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1257 struct ceph_osd_request *req)
1258{
1259 if (!req->r_linger) {
1260 dout("set_request_linger %p\n", req);
1261 req->r_linger = 1;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001262 }
1263}
1264EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1265
Sage Weilf24e9982009-10-06 11:31:10 -07001266/*
Josh Durgind29adb32013-12-02 19:11:48 -08001267 * Returns whether a request should be blocked from being sent
1268 * based on the current osdmap and osd_client settings.
1269 *
1270 * Caller should hold map_sem for read.
1271 */
1272static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1273 struct ceph_osd_request *req)
1274{
1275 bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1276 bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1277 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1278 return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1279 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1280}
1281
1282/*
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001283 * Calculate mapping of a request to a PG. Takes tiering into account.
1284 */
1285static int __calc_request_pg(struct ceph_osdmap *osdmap,
1286 struct ceph_osd_request *req,
1287 struct ceph_pg *pg_out)
1288{
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001289 bool need_check_tiering;
1290
1291 need_check_tiering = false;
1292 if (req->r_target_oloc.pool == -1) {
1293 req->r_target_oloc = req->r_base_oloc; /* struct */
1294 need_check_tiering = true;
1295 }
1296 if (req->r_target_oid.name_len == 0) {
1297 ceph_oid_copy(&req->r_target_oid, &req->r_base_oid);
1298 need_check_tiering = true;
1299 }
1300
1301 if (need_check_tiering &&
1302 (req->r_flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) {
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001303 struct ceph_pg_pool_info *pi;
1304
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001305 pi = ceph_pg_pool_by_id(osdmap, req->r_target_oloc.pool);
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001306 if (pi) {
1307 if ((req->r_flags & CEPH_OSD_FLAG_READ) &&
1308 pi->read_tier >= 0)
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001309 req->r_target_oloc.pool = pi->read_tier;
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001310 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) &&
1311 pi->write_tier >= 0)
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001312 req->r_target_oloc.pool = pi->write_tier;
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001313 }
1314 /* !pi is caught in ceph_oloc_oid_to_pg() */
1315 }
1316
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001317 return ceph_oloc_oid_to_pg(osdmap, &req->r_target_oloc,
1318 &req->r_target_oid, pg_out);
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001319}
1320
1321/*
Sage Weilf24e9982009-10-06 11:31:10 -07001322 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1323 * (as needed), and set the request r_osd appropriately. If there is
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001324 * no up osd, set r_osd to NULL. Move the request to the appropriate list
Sage Weil6f6c7002011-01-17 20:34:08 -08001325 * (unsent, homeless) or leave on in-flight lru.
Sage Weilf24e9982009-10-06 11:31:10 -07001326 *
1327 * Return 0 if unchanged, 1 if changed, or negative on error.
1328 *
1329 * Caller should hold map_sem for read and request_mutex.
1330 */
Sage Weil6f6c7002011-01-17 20:34:08 -08001331static int __map_request(struct ceph_osd_client *osdc,
Sage Weil38d64532011-10-14 13:33:55 -07001332 struct ceph_osd_request *req, int force_resend)
Sage Weilf24e9982009-10-06 11:31:10 -07001333{
Sage Weil5b191d92013-02-23 10:38:16 -08001334 struct ceph_pg pgid;
Sage Weild85b7052010-05-10 10:24:48 -07001335 int acting[CEPH_PG_MAX_SIZE];
Ilya Dryomov8008ab12014-03-24 17:12:48 +02001336 int num, o;
Sage Weilf24e9982009-10-06 11:31:10 -07001337 int err;
Josh Durgind29adb32013-12-02 19:11:48 -08001338 bool was_paused;
Sage Weilf24e9982009-10-06 11:31:10 -07001339
Sage Weil6f6c7002011-01-17 20:34:08 -08001340 dout("map_request %p tid %lld\n", req, req->r_tid);
Ilya Dryomov17a13e42014-01-27 17:40:19 +02001341
1342 err = __calc_request_pg(osdc->osdmap, req, &pgid);
Sage Weil6f6c7002011-01-17 20:34:08 -08001343 if (err) {
1344 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilf24e9982009-10-06 11:31:10 -07001345 return err;
Sage Weil6f6c7002011-01-17 20:34:08 -08001346 }
Sage Weil7740a422010-01-08 15:58:25 -08001347 req->r_pgid = pgid;
1348
Ilya Dryomov8008ab12014-03-24 17:12:48 +02001349 num = ceph_calc_pg_acting(osdc->osdmap, pgid, acting, &o);
1350 if (num < 0)
1351 num = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001352
Josh Durgind29adb32013-12-02 19:11:48 -08001353 was_paused = req->r_paused;
1354 req->r_paused = __req_should_be_paused(osdc, req);
1355 if (was_paused && !req->r_paused)
1356 force_resend = 1;
1357
Sage Weil38d64532011-10-14 13:33:55 -07001358 if ((!force_resend &&
1359 req->r_osd && req->r_osd->o_osd == o &&
Sage Weild85b7052010-05-10 10:24:48 -07001360 req->r_sent >= req->r_osd->o_incarnation &&
1361 req->r_num_pg_osds == num &&
1362 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
Josh Durgind29adb32013-12-02 19:11:48 -08001363 (req->r_osd == NULL && o == -1) ||
1364 req->r_paused)
Sage Weilf24e9982009-10-06 11:31:10 -07001365 return 0; /* no change */
1366
Sage Weil5b191d92013-02-23 10:38:16 -08001367 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1368 req->r_tid, pgid.pool, pgid.seed, o,
Sage Weilf24e9982009-10-06 11:31:10 -07001369 req->r_osd ? req->r_osd->o_osd : -1);
1370
Sage Weild85b7052010-05-10 10:24:48 -07001371 /* record full pg acting set */
1372 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1373 req->r_num_pg_osds = num;
1374
Sage Weilf24e9982009-10-06 11:31:10 -07001375 if (req->r_osd) {
1376 __cancel_request(req);
1377 list_del_init(&req->r_osd_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001378 req->r_osd = NULL;
1379 }
1380
1381 req->r_osd = __lookup_osd(osdc, o);
1382 if (!req->r_osd && o >= 0) {
Sage Weilc99eb1c2010-02-26 09:37:33 -08001383 err = -ENOMEM;
Alex Eldere10006f2012-05-26 23:26:43 -05001384 req->r_osd = create_osd(osdc, o);
Sage Weil6f6c7002011-01-17 20:34:08 -08001385 if (!req->r_osd) {
1386 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilc99eb1c2010-02-26 09:37:33 -08001387 goto out;
Sage Weil6f6c7002011-01-17 20:34:08 -08001388 }
Sage Weilf24e9982009-10-06 11:31:10 -07001389
Sage Weil6f6c7002011-01-17 20:34:08 -08001390 dout("map_request osd %p is osd%d\n", req->r_osd, o);
Sage Weilf24e9982009-10-06 11:31:10 -07001391 __insert_osd(osdc, req->r_osd);
1392
Sage Weilb7a9e5d2012-06-27 12:24:08 -07001393 ceph_con_open(&req->r_osd->o_con,
1394 CEPH_ENTITY_TYPE_OSD, o,
1395 &osdc->osdmap->osd_addr[o]);
Sage Weilf24e9982009-10-06 11:31:10 -07001396 }
1397
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001398 if (req->r_osd) {
1399 __remove_osd_from_lru(req->r_osd);
Alex Elderad885922013-03-25 18:16:11 -05001400 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1401 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
Sage Weil6f6c7002011-01-17 20:34:08 -08001402 } else {
Alex Elderad885922013-03-25 18:16:11 -05001403 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001404 }
Sage Weild85b7052010-05-10 10:24:48 -07001405 err = 1; /* osd or pg changed */
Sage Weilf24e9982009-10-06 11:31:10 -07001406
1407out:
Sage Weilf24e9982009-10-06 11:31:10 -07001408 return err;
1409}
1410
1411/*
1412 * caller should hold map_sem (for read) and request_mutex
1413 */
Sage Weil56e925b2012-01-03 12:34:34 -08001414static void __send_request(struct ceph_osd_client *osdc,
1415 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001416{
Sage Weil1b83bef2013-02-25 16:11:12 -08001417 void *p;
Sage Weilf24e9982009-10-06 11:31:10 -07001418
Sage Weil1b83bef2013-02-25 16:11:12 -08001419 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1420 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1421 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
Sage Weilf24e9982009-10-06 11:31:10 -07001422
Sage Weil1b83bef2013-02-25 16:11:12 -08001423 /* fill in message content that changes each time we send it */
1424 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1425 put_unaligned_le32(req->r_flags, req->r_request_flags);
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001426 put_unaligned_le64(req->r_target_oloc.pool, req->r_request_pool);
Sage Weil1b83bef2013-02-25 16:11:12 -08001427 p = req->r_request_pgid;
1428 ceph_encode_64(&p, req->r_pgid.pool);
1429 ceph_encode_32(&p, req->r_pgid.seed);
1430 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1431 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1432 sizeof(req->r_reassert_version));
Sage Weil2169aea2013-02-25 16:13:08 -08001433
Sage Weil3dd72fc2010-03-22 14:42:30 -07001434 req->r_stamp = jiffies;
Henry C Chang07a27e22010-08-22 21:34:27 -07001435 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001436
1437 ceph_msg_get(req->r_request); /* send consumes a ref */
Alex Elder26be8802013-04-15 11:20:42 -05001438
Sage Weilf24e9982009-10-06 11:31:10 -07001439 req->r_sent = req->r_osd->o_incarnation;
Alex Elder26be8802013-04-15 11:20:42 -05001440
1441 ceph_con_send(&req->r_osd->o_con, req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001442}
1443
1444/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001445 * Send any requests in the queue (req_unsent).
1446 */
Alex Elderf9d25192013-02-15 11:42:29 -06001447static void __send_queued(struct ceph_osd_client *osdc)
Sage Weil6f6c7002011-01-17 20:34:08 -08001448{
1449 struct ceph_osd_request *req, *tmp;
1450
Alex Elderf9d25192013-02-15 11:42:29 -06001451 dout("__send_queued\n");
1452 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
Sage Weil6f6c7002011-01-17 20:34:08 -08001453 __send_request(osdc, req);
Sage Weil6f6c7002011-01-17 20:34:08 -08001454}
1455
1456/*
Ilya Dryomov0bbfdfe2014-01-31 19:33:39 +02001457 * Caller should hold map_sem for read and request_mutex.
1458 */
1459static int __ceph_osdc_start_request(struct ceph_osd_client *osdc,
1460 struct ceph_osd_request *req,
1461 bool nofail)
1462{
1463 int rc;
1464
1465 __register_request(osdc, req);
1466 req->r_sent = 0;
1467 req->r_got_reply = 0;
1468 rc = __map_request(osdc, req, 0);
1469 if (rc < 0) {
1470 if (nofail) {
1471 dout("osdc_start_request failed map, "
1472 " will retry %lld\n", req->r_tid);
1473 rc = 0;
1474 } else {
1475 __unregister_request(osdc, req);
1476 }
1477 return rc;
1478 }
1479
1480 if (req->r_osd == NULL) {
1481 dout("send_request %p no up osds in pg\n", req);
1482 ceph_monc_request_next_osdmap(&osdc->client->monc);
1483 } else {
1484 __send_queued(osdc);
1485 }
1486
1487 return 0;
1488}
1489
1490/*
Sage Weilf24e9982009-10-06 11:31:10 -07001491 * Timeout callback, called every N seconds when 1 or more osd
1492 * requests has been active for more than N seconds. When this
1493 * happens, we ping all OSDs with requests who have timed out to
1494 * ensure any communications channel reset is detected. Reset the
1495 * request timeouts another N seconds in the future as we go.
1496 * Reschedule the timeout event another N seconds in future (unless
1497 * there are no open requests).
1498 */
1499static void handle_timeout(struct work_struct *work)
1500{
1501 struct ceph_osd_client *osdc =
1502 container_of(work, struct ceph_osd_client, timeout_work.work);
Sage Weil83aff952012-11-28 12:28:24 -08001503 struct ceph_osd_request *req;
Sage Weilf24e9982009-10-06 11:31:10 -07001504 struct ceph_osd *osd;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001505 unsigned long keepalive =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001506 osdc->client->options->osd_keepalive_timeout * HZ;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001507 struct list_head slow_osds;
Sage Weilf24e9982009-10-06 11:31:10 -07001508 dout("timeout\n");
1509 down_read(&osdc->map_sem);
1510
1511 ceph_monc_request_next_osdmap(&osdc->client->monc);
1512
1513 mutex_lock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001514
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001515 /*
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001516 * ping osds that are a bit slow. this ensures that if there
1517 * is a break in the TCP connection we will notice, and reopen
1518 * a connection with that osd (from the fault callback).
1519 */
1520 INIT_LIST_HEAD(&slow_osds);
1521 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
Sage Weil3dd72fc2010-03-22 14:42:30 -07001522 if (time_before(jiffies, req->r_stamp + keepalive))
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001523 break;
1524
1525 osd = req->r_osd;
1526 BUG_ON(!osd);
1527 dout(" tid %llu is slow, will send keepalive on osd%d\n",
Sage Weilf24e9982009-10-06 11:31:10 -07001528 req->r_tid, osd->o_osd);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001529 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1530 }
1531 while (!list_empty(&slow_osds)) {
1532 osd = list_entry(slow_osds.next, struct ceph_osd,
1533 o_keepalive_item);
1534 list_del_init(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001535 ceph_con_keepalive(&osd->o_con);
1536 }
1537
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001538 __schedule_osd_timeout(osdc);
Alex Elderf9d25192013-02-15 11:42:29 -06001539 __send_queued(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001540 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001541 up_read(&osdc->map_sem);
1542}
1543
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001544static void handle_osds_timeout(struct work_struct *work)
1545{
1546 struct ceph_osd_client *osdc =
1547 container_of(work, struct ceph_osd_client,
1548 osds_timeout_work.work);
1549 unsigned long delay =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001550 osdc->client->options->osd_idle_ttl * HZ >> 2;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001551
1552 dout("osds timeout\n");
1553 down_read(&osdc->map_sem);
Sage Weilaca420b2011-08-31 14:45:53 -07001554 remove_old_osds(osdc);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001555 up_read(&osdc->map_sem);
1556
1557 schedule_delayed_work(&osdc->osds_timeout_work,
1558 round_jiffies_relative(delay));
1559}
1560
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001561static int ceph_oloc_decode(void **p, void *end,
1562 struct ceph_object_locator *oloc)
1563{
1564 u8 struct_v, struct_cv;
1565 u32 len;
1566 void *struct_end;
1567 int ret = 0;
1568
1569 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1570 struct_v = ceph_decode_8(p);
1571 struct_cv = ceph_decode_8(p);
1572 if (struct_v < 3) {
1573 pr_warn("got v %d < 3 cv %d of ceph_object_locator\n",
1574 struct_v, struct_cv);
1575 goto e_inval;
1576 }
1577 if (struct_cv > 6) {
1578 pr_warn("got v %d cv %d > 6 of ceph_object_locator\n",
1579 struct_v, struct_cv);
1580 goto e_inval;
1581 }
1582 len = ceph_decode_32(p);
1583 ceph_decode_need(p, end, len, e_inval);
1584 struct_end = *p + len;
1585
1586 oloc->pool = ceph_decode_64(p);
1587 *p += 4; /* skip preferred */
1588
1589 len = ceph_decode_32(p);
1590 if (len > 0) {
1591 pr_warn("ceph_object_locator::key is set\n");
1592 goto e_inval;
1593 }
1594
1595 if (struct_v >= 5) {
1596 len = ceph_decode_32(p);
1597 if (len > 0) {
1598 pr_warn("ceph_object_locator::nspace is set\n");
1599 goto e_inval;
1600 }
1601 }
1602
1603 if (struct_v >= 6) {
1604 s64 hash = ceph_decode_64(p);
1605 if (hash != -1) {
1606 pr_warn("ceph_object_locator::hash is set\n");
1607 goto e_inval;
1608 }
1609 }
1610
1611 /* skip the rest */
1612 *p = struct_end;
1613out:
1614 return ret;
1615
1616e_inval:
1617 ret = -EINVAL;
1618 goto out;
1619}
1620
1621static int ceph_redirect_decode(void **p, void *end,
1622 struct ceph_request_redirect *redir)
1623{
1624 u8 struct_v, struct_cv;
1625 u32 len;
1626 void *struct_end;
1627 int ret;
1628
1629 ceph_decode_need(p, end, 1 + 1 + 4, e_inval);
1630 struct_v = ceph_decode_8(p);
1631 struct_cv = ceph_decode_8(p);
1632 if (struct_cv > 1) {
1633 pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n",
1634 struct_v, struct_cv);
1635 goto e_inval;
1636 }
1637 len = ceph_decode_32(p);
1638 ceph_decode_need(p, end, len, e_inval);
1639 struct_end = *p + len;
1640
1641 ret = ceph_oloc_decode(p, end, &redir->oloc);
1642 if (ret)
1643 goto out;
1644
1645 len = ceph_decode_32(p);
1646 if (len > 0) {
1647 pr_warn("ceph_request_redirect::object_name is set\n");
1648 goto e_inval;
1649 }
1650
1651 len = ceph_decode_32(p);
1652 *p += len; /* skip osd_instructions */
1653
1654 /* skip the rest */
1655 *p = struct_end;
1656out:
1657 return ret;
1658
1659e_inval:
1660 ret = -EINVAL;
1661 goto out;
1662}
1663
Sage Weil25845472011-06-03 09:37:09 -07001664static void complete_request(struct ceph_osd_request *req)
1665{
Sage Weil25845472011-06-03 09:37:09 -07001666 complete_all(&req->r_safe_completion); /* fsync waiter */
1667}
1668
Sage Weilf24e9982009-10-06 11:31:10 -07001669/*
1670 * handle osd op reply. either call the callback if it is specified,
1671 * or do the completion to wake up the waiting thread.
1672 */
Sage Weil350b1c32009-12-22 10:45:45 -08001673static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1674 struct ceph_connection *con)
Sage Weilf24e9982009-10-06 11:31:10 -07001675{
Sage Weil1b83bef2013-02-25 16:11:12 -08001676 void *p, *end;
Sage Weilf24e9982009-10-06 11:31:10 -07001677 struct ceph_osd_request *req;
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001678 struct ceph_request_redirect redir;
Sage Weilf24e9982009-10-06 11:31:10 -07001679 u64 tid;
Sage Weil1b83bef2013-02-25 16:11:12 -08001680 int object_len;
Alex Elder79528732013-04-03 21:32:51 -05001681 unsigned int numops;
1682 int payload_len, flags;
Sage Weil0ceed5d2010-05-11 09:53:18 -07001683 s32 result;
Sage Weil1b83bef2013-02-25 16:11:12 -08001684 s32 retry_attempt;
1685 struct ceph_pg pg;
1686 int err;
1687 u32 reassert_epoch;
1688 u64 reassert_version;
1689 u32 osdmap_epoch;
Alex Elder0d5af162013-02-27 10:26:25 -06001690 int already_completed;
Alex Elder9fc6e062013-04-03 01:28:57 -05001691 u32 bytes;
Alex Elder79528732013-04-03 21:32:51 -05001692 unsigned int i;
Sage Weilf24e9982009-10-06 11:31:10 -07001693
Sage Weil6df058c2009-12-22 11:24:33 -08001694 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil1b83bef2013-02-25 16:11:12 -08001695 dout("handle_reply %p tid %llu\n", msg, tid);
1696
1697 p = msg->front.iov_base;
1698 end = p + msg->front.iov_len;
1699
1700 ceph_decode_need(&p, end, 4, bad);
1701 object_len = ceph_decode_32(&p);
1702 ceph_decode_need(&p, end, object_len, bad);
1703 p += object_len;
1704
Alex Elderef4859d2013-04-01 18:58:26 -05001705 err = ceph_decode_pgid(&p, end, &pg);
Sage Weil1b83bef2013-02-25 16:11:12 -08001706 if (err)
Sage Weilf24e9982009-10-06 11:31:10 -07001707 goto bad;
Sage Weil1b83bef2013-02-25 16:11:12 -08001708
1709 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1710 flags = ceph_decode_64(&p);
1711 result = ceph_decode_32(&p);
1712 reassert_epoch = ceph_decode_32(&p);
1713 reassert_version = ceph_decode_64(&p);
1714 osdmap_epoch = ceph_decode_32(&p);
1715
Sage Weilf24e9982009-10-06 11:31:10 -07001716 /* lookup */
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001717 down_read(&osdc->map_sem);
Sage Weilf24e9982009-10-06 11:31:10 -07001718 mutex_lock(&osdc->request_mutex);
1719 req = __lookup_request(osdc, tid);
1720 if (req == NULL) {
1721 dout("handle_reply tid %llu dne\n", tid);
Alex Elder8058fd42013-04-01 18:58:26 -05001722 goto bad_mutex;
Sage Weilf24e9982009-10-06 11:31:10 -07001723 }
1724 ceph_osdc_get_request(req);
Sage Weil1b83bef2013-02-25 16:11:12 -08001725
1726 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1727 req, result);
1728
Dan Carpenter18741192013-08-15 08:51:58 +03001729 ceph_decode_need(&p, end, 4, bad_put);
Sage Weil1b83bef2013-02-25 16:11:12 -08001730 numops = ceph_decode_32(&p);
1731 if (numops > CEPH_OSD_MAX_OP)
1732 goto bad_put;
1733 if (numops != req->r_num_ops)
1734 goto bad_put;
1735 payload_len = 0;
Dan Carpenter18741192013-08-15 08:51:58 +03001736 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
Sage Weil1b83bef2013-02-25 16:11:12 -08001737 for (i = 0; i < numops; i++) {
1738 struct ceph_osd_op *op = p;
1739 int len;
1740
1741 len = le32_to_cpu(op->payload_len);
1742 req->r_reply_op_len[i] = len;
1743 dout(" op %d has %d bytes\n", i, len);
1744 payload_len += len;
1745 p += sizeof(*op);
1746 }
Alex Elder9fc6e062013-04-03 01:28:57 -05001747 bytes = le32_to_cpu(msg->hdr.data_len);
1748 if (payload_len != bytes) {
Sage Weil1b83bef2013-02-25 16:11:12 -08001749 pr_warning("sum of op payload lens %d != data_len %d",
Alex Elder9fc6e062013-04-03 01:28:57 -05001750 payload_len, bytes);
Sage Weil1b83bef2013-02-25 16:11:12 -08001751 goto bad_put;
1752 }
1753
Dan Carpenter18741192013-08-15 08:51:58 +03001754 ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
Sage Weil1b83bef2013-02-25 16:11:12 -08001755 retry_attempt = ceph_decode_32(&p);
1756 for (i = 0; i < numops; i++)
1757 req->r_reply_op_result[i] = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001758
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001759 if (le16_to_cpu(msg->hdr.version) >= 6) {
1760 p += 8 + 4; /* skip replay_version */
1761 p += 8; /* skip user_version */
1762
1763 err = ceph_redirect_decode(&p, end, &redir);
1764 if (err)
1765 goto bad_put;
1766 } else {
1767 redir.oloc.pool = -1;
1768 }
1769
1770 if (redir.oloc.pool != -1) {
1771 dout("redirect pool %lld\n", redir.oloc.pool);
1772
1773 __unregister_request(osdc, req);
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001774
1775 req->r_target_oloc = redir.oloc; /* struct */
1776
1777 /*
1778 * Start redirect requests with nofail=true. If
1779 * mapping fails, request will end up on the notarget
1780 * list, waiting for the new osdmap (which can take
1781 * a while), even though the original request mapped
1782 * successfully. In the future we might want to follow
1783 * original request's nofail setting here.
1784 */
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001785 err = __ceph_osdc_start_request(osdc, req, true);
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001786 BUG_ON(err);
1787
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001788 goto out_unlock;
Ilya Dryomov205ee1182014-01-27 17:40:20 +02001789 }
1790
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001791 already_completed = req->r_got_reply;
Sage Weilf24e9982009-10-06 11:31:10 -07001792 if (!req->r_got_reply) {
Sage Weil1b83bef2013-02-25 16:11:12 -08001793 req->r_result = result;
Sage Weilf24e9982009-10-06 11:31:10 -07001794 dout("handle_reply result %d bytes %d\n", req->r_result,
1795 bytes);
1796 if (req->r_result == 0)
1797 req->r_result = bytes;
1798
1799 /* in case this is a write and we need to replay, */
Sage Weil1b83bef2013-02-25 16:11:12 -08001800 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1801 req->r_reassert_version.version = cpu_to_le64(reassert_version);
Sage Weilf24e9982009-10-06 11:31:10 -07001802
1803 req->r_got_reply = 1;
1804 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1805 dout("handle_reply tid %llu dup ack\n", tid);
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001806 goto out_unlock;
Sage Weilf24e9982009-10-06 11:31:10 -07001807 }
1808
1809 dout("handle_reply tid %llu flags %d\n", tid, flags);
1810
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001811 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1812 __register_linger_request(osdc, req);
1813
Sage Weilf24e9982009-10-06 11:31:10 -07001814 /* either this is a read, or we got the safe response */
Sage Weil0ceed5d2010-05-11 09:53:18 -07001815 if (result < 0 ||
1816 (flags & CEPH_OSD_FLAG_ONDISK) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001817 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1818 __unregister_request(osdc, req);
1819
1820 mutex_unlock(&osdc->request_mutex);
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001821 up_read(&osdc->map_sem);
Sage Weilf24e9982009-10-06 11:31:10 -07001822
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001823 if (!already_completed) {
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001824 if (req->r_unsafe_callback &&
1825 result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1826 req->r_unsafe_callback(req, true);
Yan, Zhengeb845ff2013-05-31 15:54:44 +08001827 if (req->r_callback)
1828 req->r_callback(req, msg);
1829 else
1830 complete_all(&req->r_completion);
1831 }
Sage Weilf24e9982009-10-06 11:31:10 -07001832
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001833 if (flags & CEPH_OSD_FLAG_ONDISK) {
1834 if (req->r_unsafe_callback && already_completed)
1835 req->r_unsafe_callback(req, false);
Sage Weil25845472011-06-03 09:37:09 -07001836 complete_request(req);
Yan, Zheng61c5d6b2013-06-24 14:41:27 +08001837 }
Sage Weilf24e9982009-10-06 11:31:10 -07001838
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001839out:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001840 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07001841 ceph_osdc_put_request(req);
1842 return;
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001843out_unlock:
1844 mutex_unlock(&osdc->request_mutex);
1845 up_read(&osdc->map_sem);
1846 goto out;
Sage Weilf24e9982009-10-06 11:31:10 -07001847
Sage Weil1b83bef2013-02-25 16:11:12 -08001848bad_put:
Li Wang37c89bd2013-11-27 22:28:14 +08001849 req->r_result = -EIO;
1850 __unregister_request(osdc, req);
1851 if (req->r_callback)
1852 req->r_callback(req, msg);
1853 else
1854 complete_all(&req->r_completion);
1855 complete_request(req);
Sage Weil1b83bef2013-02-25 16:11:12 -08001856 ceph_osdc_put_request(req);
Alex Elder8058fd42013-04-01 18:58:26 -05001857bad_mutex:
1858 mutex_unlock(&osdc->request_mutex);
Ilya Dryomovff513ac2014-02-03 13:56:33 +02001859 up_read(&osdc->map_sem);
Sage Weilf24e9982009-10-06 11:31:10 -07001860bad:
Sage Weil1b83bef2013-02-25 16:11:12 -08001861 pr_err("corrupt osd_op_reply got %d %d\n",
1862 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
Sage Weil9ec7cab2009-12-14 15:13:47 -08001863 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001864}
1865
Sage Weil6f6c7002011-01-17 20:34:08 -08001866static void reset_changed_osds(struct ceph_osd_client *osdc)
Sage Weilf24e9982009-10-06 11:31:10 -07001867{
Sage Weilf24e9982009-10-06 11:31:10 -07001868 struct rb_node *p, *n;
Sage Weilf24e9982009-10-06 11:31:10 -07001869
Sage Weil6f6c7002011-01-17 20:34:08 -08001870 for (p = rb_first(&osdc->osds); p; p = n) {
1871 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
Sage Weilf24e9982009-10-06 11:31:10 -07001872
Sage Weil6f6c7002011-01-17 20:34:08 -08001873 n = rb_next(p);
1874 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1875 memcmp(&osd->o_con.peer_addr,
1876 ceph_osd_addr(osdc->osdmap,
1877 osd->o_osd),
1878 sizeof(struct ceph_entity_addr)) != 0)
1879 __reset_osd(osdc, osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001880 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001881}
1882
1883/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001884 * Requeue requests whose mapping to an OSD has changed. If requests map to
1885 * no osd, request a new map.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001886 *
Alex Eldere6d50f62012-12-26 14:31:40 -06001887 * Caller should hold map_sem for read.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001888 */
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001889static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1890 bool force_resend_writes)
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001891{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001892 struct ceph_osd_request *req, *nreq;
Sage Weil6f6c7002011-01-17 20:34:08 -08001893 struct rb_node *p;
1894 int needmap = 0;
1895 int err;
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001896 bool force_resend_req;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001897
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001898 dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1899 force_resend_writes ? " (force resend writes)" : "");
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001900 mutex_lock(&osdc->request_mutex);
Sage Weil6194ea82012-07-30 16:19:28 -07001901 for (p = rb_first(&osdc->requests); p; ) {
Sage Weil6f6c7002011-01-17 20:34:08 -08001902 req = rb_entry(p, struct ceph_osd_request, r_node);
Sage Weil6194ea82012-07-30 16:19:28 -07001903 p = rb_next(p);
Alex Elderab60b162012-12-19 15:52:36 -06001904
1905 /*
1906 * For linger requests that have not yet been
1907 * registered, move them to the linger list; they'll
1908 * be sent to the osd in the loop below. Unregister
1909 * the request before re-registering it as a linger
1910 * request to ensure the __map_request() below
1911 * will decide it needs to be sent.
1912 */
1913 if (req->r_linger && list_empty(&req->r_linger_item)) {
1914 dout("%p tid %llu restart on osd%d\n",
1915 req, req->r_tid,
1916 req->r_osd ? req->r_osd->o_osd : -1);
Alex Elder96e4dac2013-05-22 20:54:25 -05001917 ceph_osdc_get_request(req);
Alex Elderab60b162012-12-19 15:52:36 -06001918 __unregister_request(osdc, req);
1919 __register_linger_request(osdc, req);
Alex Elder96e4dac2013-05-22 20:54:25 -05001920 ceph_osdc_put_request(req);
Alex Elderab60b162012-12-19 15:52:36 -06001921 continue;
1922 }
1923
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001924 force_resend_req = force_resend ||
1925 (force_resend_writes &&
1926 req->r_flags & CEPH_OSD_FLAG_WRITE);
1927 err = __map_request(osdc, req, force_resend_req);
Sage Weil6f6c7002011-01-17 20:34:08 -08001928 if (err < 0)
1929 continue; /* error */
1930 if (req->r_osd == NULL) {
1931 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1932 needmap++; /* request a newer map */
1933 } else if (err > 0) {
Sage Weil6194ea82012-07-30 16:19:28 -07001934 if (!req->r_linger) {
1935 dout("%p tid %llu requeued on osd%d\n", req,
1936 req->r_tid,
1937 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001938 req->r_flags |= CEPH_OSD_FLAG_RETRY;
Sage Weil6194ea82012-07-30 16:19:28 -07001939 }
1940 }
Sage Weil6f6c7002011-01-17 20:34:08 -08001941 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001942
1943 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1944 r_linger_item) {
1945 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1946
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001947 err = __map_request(osdc, req,
1948 force_resend || force_resend_writes);
Alex Elderab60b162012-12-19 15:52:36 -06001949 dout("__map_request returned %d\n", err);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001950 if (err == 0)
1951 continue; /* no change and no osd was specified */
1952 if (err < 0)
1953 continue; /* hrm! */
1954 if (req->r_osd == NULL) {
1955 dout("tid %llu maps to no valid osd\n", req->r_tid);
1956 needmap++; /* request a newer map */
1957 continue;
1958 }
1959
1960 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1961 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001962 __register_request(osdc, req);
Alex Elderc89ce052012-12-06 07:22:04 -06001963 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001964 }
Alex Elder14d2f382013-05-15 16:28:33 -05001965 reset_changed_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001966 mutex_unlock(&osdc->request_mutex);
1967
1968 if (needmap) {
1969 dout("%d requests for down osds, need new map\n", needmap);
1970 ceph_monc_request_next_osdmap(&osdc->client->monc);
1971 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001972}
Sage Weil6f6c7002011-01-17 20:34:08 -08001973
1974
Sage Weilf24e9982009-10-06 11:31:10 -07001975/*
1976 * Process updated osd map.
1977 *
1978 * The message contains any number of incremental and full maps, normally
1979 * indicating some sort of topology change in the cluster. Kick requests
1980 * off to different OSDs as needed.
1981 */
1982void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1983{
1984 void *p, *end, *next;
1985 u32 nr_maps, maplen;
1986 u32 epoch;
1987 struct ceph_osdmap *newmap = NULL, *oldmap;
1988 int err;
1989 struct ceph_fsid fsid;
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08001990 bool was_full;
Sage Weilf24e9982009-10-06 11:31:10 -07001991
1992 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1993 p = msg->front.iov_base;
1994 end = p + msg->front.iov_len;
1995
1996 /* verify fsid */
1997 ceph_decode_need(&p, end, sizeof(fsid), bad);
1998 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Sage Weil07433042009-11-18 16:50:41 -08001999 if (ceph_check_fsid(osdc->client, &fsid) < 0)
2000 return;
Sage Weilf24e9982009-10-06 11:31:10 -07002001
2002 down_write(&osdc->map_sem);
2003
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002004 was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
2005
Sage Weilf24e9982009-10-06 11:31:10 -07002006 /* incremental maps */
2007 ceph_decode_32_safe(&p, end, nr_maps, bad);
2008 dout(" %d inc maps\n", nr_maps);
2009 while (nr_maps > 0) {
2010 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07002011 epoch = ceph_decode_32(&p);
2012 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07002013 ceph_decode_need(&p, end, maplen, bad);
2014 next = p + maplen;
2015 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
2016 dout("applying incremental map %u len %d\n",
2017 epoch, maplen);
2018 newmap = osdmap_apply_incremental(&p, next,
2019 osdc->osdmap,
Alex Elder15d98822012-05-26 23:26:43 -05002020 &osdc->client->msgr);
Sage Weilf24e9982009-10-06 11:31:10 -07002021 if (IS_ERR(newmap)) {
2022 err = PTR_ERR(newmap);
2023 goto bad;
2024 }
Sage Weil30dc6382009-12-21 14:49:37 -08002025 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07002026 if (newmap != osdc->osdmap) {
2027 ceph_osdmap_destroy(osdc->osdmap);
2028 osdc->osdmap = newmap;
2029 }
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002030 was_full = was_full ||
2031 ceph_osdmap_flag(osdc->osdmap,
2032 CEPH_OSDMAP_FULL);
2033 kick_requests(osdc, 0, was_full);
Sage Weilf24e9982009-10-06 11:31:10 -07002034 } else {
2035 dout("ignoring incremental map %u len %d\n",
2036 epoch, maplen);
2037 }
2038 p = next;
2039 nr_maps--;
2040 }
2041 if (newmap)
2042 goto done;
2043
2044 /* full maps */
2045 ceph_decode_32_safe(&p, end, nr_maps, bad);
2046 dout(" %d full maps\n", nr_maps);
2047 while (nr_maps) {
2048 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07002049 epoch = ceph_decode_32(&p);
2050 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07002051 ceph_decode_need(&p, end, maplen, bad);
2052 if (nr_maps > 1) {
2053 dout("skipping non-latest full map %u len %d\n",
2054 epoch, maplen);
2055 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
2056 dout("skipping full map %u len %d, "
2057 "older than our %u\n", epoch, maplen,
2058 osdc->osdmap->epoch);
2059 } else {
Sage Weil38d64532011-10-14 13:33:55 -07002060 int skipped_map = 0;
2061
Sage Weilf24e9982009-10-06 11:31:10 -07002062 dout("taking full map %u len %d\n", epoch, maplen);
Ilya Dryomova2505d62014-03-13 16:36:13 +02002063 newmap = ceph_osdmap_decode(&p, p+maplen);
Sage Weilf24e9982009-10-06 11:31:10 -07002064 if (IS_ERR(newmap)) {
2065 err = PTR_ERR(newmap);
2066 goto bad;
2067 }
Sage Weil30dc6382009-12-21 14:49:37 -08002068 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07002069 oldmap = osdc->osdmap;
2070 osdc->osdmap = newmap;
Sage Weil38d64532011-10-14 13:33:55 -07002071 if (oldmap) {
2072 if (oldmap->epoch + 1 < newmap->epoch)
2073 skipped_map = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07002074 ceph_osdmap_destroy(oldmap);
Sage Weil38d64532011-10-14 13:33:55 -07002075 }
Josh Durgin9a1ea2d2013-12-10 09:35:13 -08002076 was_full = was_full ||
2077 ceph_osdmap_flag(osdc->osdmap,
2078 CEPH_OSDMAP_FULL);
2079 kick_requests(osdc, skipped_map, was_full);
Sage Weilf24e9982009-10-06 11:31:10 -07002080 }
2081 p += maplen;
2082 nr_maps--;
2083 }
2084
Dan Carpenterb72e19b2013-08-15 08:52:48 +03002085 if (!osdc->osdmap)
2086 goto bad;
Sage Weilf24e9982009-10-06 11:31:10 -07002087done:
2088 downgrade_write(&osdc->map_sem);
2089 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
Sage Weilcd634fb2011-05-12 09:29:18 -07002090
2091 /*
2092 * subscribe to subsequent osdmap updates if full to ensure
2093 * we find out when we are no longer full and stop returning
2094 * ENOSPC.
2095 */
Josh Durgind29adb32013-12-02 19:11:48 -08002096 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
2097 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
2098 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
Sage Weilcd634fb2011-05-12 09:29:18 -07002099 ceph_monc_request_next_osdmap(&osdc->client->monc);
2100
Alex Elderf9d25192013-02-15 11:42:29 -06002101 mutex_lock(&osdc->request_mutex);
2102 __send_queued(osdc);
2103 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07002104 up_read(&osdc->map_sem);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07002105 wake_up_all(&osdc->client->auth_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07002106 return;
2107
2108bad:
2109 pr_err("osdc handle_map corrupt msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002110 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07002111 up_write(&osdc->map_sem);
Sage Weilf24e9982009-10-06 11:31:10 -07002112}
2113
Sage Weilf24e9982009-10-06 11:31:10 -07002114/*
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002115 * watch/notify callback event infrastructure
2116 *
2117 * These callbacks are used both for watch and notify operations.
2118 */
2119static void __release_event(struct kref *kref)
2120{
2121 struct ceph_osd_event *event =
2122 container_of(kref, struct ceph_osd_event, kref);
2123
2124 dout("__release_event %p\n", event);
2125 kfree(event);
2126}
2127
2128static void get_event(struct ceph_osd_event *event)
2129{
2130 kref_get(&event->kref);
2131}
2132
2133void ceph_osdc_put_event(struct ceph_osd_event *event)
2134{
2135 kref_put(&event->kref, __release_event);
2136}
2137EXPORT_SYMBOL(ceph_osdc_put_event);
2138
2139static void __insert_event(struct ceph_osd_client *osdc,
2140 struct ceph_osd_event *new)
2141{
2142 struct rb_node **p = &osdc->event_tree.rb_node;
2143 struct rb_node *parent = NULL;
2144 struct ceph_osd_event *event = NULL;
2145
2146 while (*p) {
2147 parent = *p;
2148 event = rb_entry(parent, struct ceph_osd_event, node);
2149 if (new->cookie < event->cookie)
2150 p = &(*p)->rb_left;
2151 else if (new->cookie > event->cookie)
2152 p = &(*p)->rb_right;
2153 else
2154 BUG();
2155 }
2156
2157 rb_link_node(&new->node, parent, p);
2158 rb_insert_color(&new->node, &osdc->event_tree);
2159}
2160
2161static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
2162 u64 cookie)
2163{
2164 struct rb_node **p = &osdc->event_tree.rb_node;
2165 struct rb_node *parent = NULL;
2166 struct ceph_osd_event *event = NULL;
2167
2168 while (*p) {
2169 parent = *p;
2170 event = rb_entry(parent, struct ceph_osd_event, node);
2171 if (cookie < event->cookie)
2172 p = &(*p)->rb_left;
2173 else if (cookie > event->cookie)
2174 p = &(*p)->rb_right;
2175 else
2176 return event;
2177 }
2178 return NULL;
2179}
2180
2181static void __remove_event(struct ceph_osd_event *event)
2182{
2183 struct ceph_osd_client *osdc = event->osdc;
2184
2185 if (!RB_EMPTY_NODE(&event->node)) {
2186 dout("__remove_event removed %p\n", event);
2187 rb_erase(&event->node, &osdc->event_tree);
2188 ceph_osdc_put_event(event);
2189 } else {
2190 dout("__remove_event didn't remove %p\n", event);
2191 }
2192}
2193
2194int ceph_osdc_create_event(struct ceph_osd_client *osdc,
2195 void (*event_cb)(u64, u64, u8, void *),
Alex Elder3c663bb2013-02-15 11:42:30 -06002196 void *data, struct ceph_osd_event **pevent)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002197{
2198 struct ceph_osd_event *event;
2199
2200 event = kmalloc(sizeof(*event), GFP_NOIO);
2201 if (!event)
2202 return -ENOMEM;
2203
2204 dout("create_event %p\n", event);
2205 event->cb = event_cb;
Alex Elder3c663bb2013-02-15 11:42:30 -06002206 event->one_shot = 0;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002207 event->data = data;
2208 event->osdc = osdc;
2209 INIT_LIST_HEAD(&event->osd_node);
Alex Elder3ee52342012-12-17 12:23:48 -06002210 RB_CLEAR_NODE(&event->node);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002211 kref_init(&event->kref); /* one ref for us */
2212 kref_get(&event->kref); /* one ref for the caller */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002213
2214 spin_lock(&osdc->event_lock);
2215 event->cookie = ++osdc->event_count;
2216 __insert_event(osdc, event);
2217 spin_unlock(&osdc->event_lock);
2218
2219 *pevent = event;
2220 return 0;
2221}
2222EXPORT_SYMBOL(ceph_osdc_create_event);
2223
2224void ceph_osdc_cancel_event(struct ceph_osd_event *event)
2225{
2226 struct ceph_osd_client *osdc = event->osdc;
2227
2228 dout("cancel_event %p\n", event);
2229 spin_lock(&osdc->event_lock);
2230 __remove_event(event);
2231 spin_unlock(&osdc->event_lock);
2232 ceph_osdc_put_event(event); /* caller's */
2233}
2234EXPORT_SYMBOL(ceph_osdc_cancel_event);
2235
2236
2237static void do_event_work(struct work_struct *work)
2238{
2239 struct ceph_osd_event_work *event_work =
2240 container_of(work, struct ceph_osd_event_work, work);
2241 struct ceph_osd_event *event = event_work->event;
2242 u64 ver = event_work->ver;
2243 u64 notify_id = event_work->notify_id;
2244 u8 opcode = event_work->opcode;
2245
2246 dout("do_event_work completing %p\n", event);
2247 event->cb(ver, notify_id, opcode, event->data);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002248 dout("do_event_work completed %p\n", event);
2249 ceph_osdc_put_event(event);
2250 kfree(event_work);
2251}
2252
2253
2254/*
2255 * Process osd watch notifications
2256 */
Alex Elder3c663bb2013-02-15 11:42:30 -06002257static void handle_watch_notify(struct ceph_osd_client *osdc,
2258 struct ceph_msg *msg)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002259{
2260 void *p, *end;
2261 u8 proto_ver;
2262 u64 cookie, ver, notify_id;
2263 u8 opcode;
2264 struct ceph_osd_event *event;
2265 struct ceph_osd_event_work *event_work;
2266
2267 p = msg->front.iov_base;
2268 end = p + msg->front.iov_len;
2269
2270 ceph_decode_8_safe(&p, end, proto_ver, bad);
2271 ceph_decode_8_safe(&p, end, opcode, bad);
2272 ceph_decode_64_safe(&p, end, cookie, bad);
2273 ceph_decode_64_safe(&p, end, ver, bad);
2274 ceph_decode_64_safe(&p, end, notify_id, bad);
2275
2276 spin_lock(&osdc->event_lock);
2277 event = __find_event(osdc, cookie);
2278 if (event) {
Alex Elder3c663bb2013-02-15 11:42:30 -06002279 BUG_ON(event->one_shot);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002280 get_event(event);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002281 }
2282 spin_unlock(&osdc->event_lock);
2283 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2284 cookie, ver, event);
2285 if (event) {
2286 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002287 if (!event_work) {
2288 dout("ERROR: could not allocate event_work\n");
2289 goto done_err;
2290 }
Mariusz Kozlowski6b0ae402011-03-26 19:29:34 +01002291 INIT_WORK(&event_work->work, do_event_work);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002292 event_work->event = event;
2293 event_work->ver = ver;
2294 event_work->notify_id = notify_id;
2295 event_work->opcode = opcode;
2296 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2297 dout("WARNING: failed to queue notify event work\n");
2298 goto done_err;
2299 }
2300 }
2301
2302 return;
2303
2304done_err:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002305 ceph_osdc_put_event(event);
2306 return;
2307
2308bad:
2309 pr_err("osdc handle_watch_notify corrupt msg\n");
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002310}
2311
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002312/*
Alex Eldere65550f2013-04-05 01:27:12 -05002313 * build new request AND message
2314 *
2315 */
2316void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2317 struct ceph_snap_context *snapc, u64 snap_id,
2318 struct timespec *mtime)
2319{
2320 struct ceph_msg *msg = req->r_request;
2321 void *p;
2322 size_t msg_size;
2323 int flags = req->r_flags;
2324 u64 data_len;
2325 unsigned int i;
2326
2327 req->r_snapid = snap_id;
2328 req->r_snapc = ceph_get_snap_context(snapc);
2329
2330 /* encode request */
2331 msg->hdr.version = cpu_to_le16(4);
2332
2333 p = msg->front.iov_base;
2334 ceph_encode_32(&p, 1); /* client_inc is always 1 */
2335 req->r_request_osdmap_epoch = p;
2336 p += 4;
2337 req->r_request_flags = p;
2338 p += 4;
2339 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2340 ceph_encode_timespec(p, mtime);
2341 p += sizeof(struct ceph_timespec);
2342 req->r_request_reassert_version = p;
2343 p += sizeof(struct ceph_eversion); /* will get filled in */
2344
2345 /* oloc */
2346 ceph_encode_8(&p, 4);
2347 ceph_encode_8(&p, 4);
2348 ceph_encode_32(&p, 8 + 4 + 4);
2349 req->r_request_pool = p;
2350 p += 8;
2351 ceph_encode_32(&p, -1); /* preferred */
2352 ceph_encode_32(&p, 0); /* key len */
2353
2354 ceph_encode_8(&p, 1);
2355 req->r_request_pgid = p;
2356 p += 8 + 4;
2357 ceph_encode_32(&p, -1); /* preferred */
2358
2359 /* oid */
Ilya Dryomov3c972c92014-01-27 17:40:20 +02002360 ceph_encode_32(&p, req->r_base_oid.name_len);
2361 memcpy(p, req->r_base_oid.name, req->r_base_oid.name_len);
2362 dout("oid '%.*s' len %d\n", req->r_base_oid.name_len,
2363 req->r_base_oid.name, req->r_base_oid.name_len);
2364 p += req->r_base_oid.name_len;
Alex Eldere65550f2013-04-05 01:27:12 -05002365
2366 /* ops--can imply data */
2367 ceph_encode_16(&p, (u16)req->r_num_ops);
2368 data_len = 0;
2369 for (i = 0; i < req->r_num_ops; i++) {
2370 data_len += osd_req_encode_op(req, p, i);
2371 p += sizeof(struct ceph_osd_op);
2372 }
2373
2374 /* snaps */
2375 ceph_encode_64(&p, req->r_snapid);
2376 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2377 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2378 if (req->r_snapc) {
2379 for (i = 0; i < snapc->num_snaps; i++) {
2380 ceph_encode_64(&p, req->r_snapc->snaps[i]);
2381 }
2382 }
2383
2384 req->r_request_attempts = p;
2385 p += 4;
2386
2387 /* data */
2388 if (flags & CEPH_OSD_FLAG_WRITE) {
2389 u16 data_off;
2390
2391 /*
2392 * The header "data_off" is a hint to the receiver
2393 * allowing it to align received data into its
2394 * buffers such that there's no need to re-copy
2395 * it before writing it to disk (direct I/O).
2396 */
2397 data_off = (u16) (off & 0xffff);
2398 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2399 }
2400 req->r_request->hdr.data_len = cpu_to_le32(data_len);
2401
2402 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2403 msg_size = p - msg->front.iov_base;
2404 msg->front.iov_len = msg_size;
2405 msg->hdr.front_len = cpu_to_le32(msg_size);
2406
2407 dout("build_request msg_size was %d\n", (int)msg_size);
2408}
2409EXPORT_SYMBOL(ceph_osdc_build_request);
2410
2411/*
Sage Weilf24e9982009-10-06 11:31:10 -07002412 * Register request, send initial attempt.
2413 */
2414int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2415 struct ceph_osd_request *req,
2416 bool nofail)
2417{
Ilya Dryomov0bbfdfe2014-01-31 19:33:39 +02002418 int rc;
Sage Weilf24e9982009-10-06 11:31:10 -07002419
Sage Weilf24e9982009-10-06 11:31:10 -07002420 down_read(&osdc->map_sem);
2421 mutex_lock(&osdc->request_mutex);
Ilya Dryomov0bbfdfe2014-01-31 19:33:39 +02002422
2423 rc = __ceph_osdc_start_request(osdc, req, nofail);
2424
Sage Weilf24e9982009-10-06 11:31:10 -07002425 mutex_unlock(&osdc->request_mutex);
2426 up_read(&osdc->map_sem);
Ilya Dryomov0bbfdfe2014-01-31 19:33:39 +02002427
Sage Weilf24e9982009-10-06 11:31:10 -07002428 return rc;
2429}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002430EXPORT_SYMBOL(ceph_osdc_start_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002431
2432/*
2433 * wait for a request to complete
2434 */
2435int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2436 struct ceph_osd_request *req)
2437{
2438 int rc;
2439
2440 rc = wait_for_completion_interruptible(&req->r_completion);
2441 if (rc < 0) {
2442 mutex_lock(&osdc->request_mutex);
2443 __cancel_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08002444 __unregister_request(osdc, req);
Sage Weilf24e9982009-10-06 11:31:10 -07002445 mutex_unlock(&osdc->request_mutex);
Sage Weil25845472011-06-03 09:37:09 -07002446 complete_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08002447 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07002448 return rc;
2449 }
2450
2451 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2452 return req->r_result;
2453}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002454EXPORT_SYMBOL(ceph_osdc_wait_request);
Sage Weilf24e9982009-10-06 11:31:10 -07002455
2456/*
2457 * sync - wait for all in-flight requests to flush. avoid starvation.
2458 */
2459void ceph_osdc_sync(struct ceph_osd_client *osdc)
2460{
2461 struct ceph_osd_request *req;
2462 u64 last_tid, next_tid = 0;
2463
2464 mutex_lock(&osdc->request_mutex);
2465 last_tid = osdc->last_tid;
2466 while (1) {
2467 req = __lookup_request_ge(osdc, next_tid);
2468 if (!req)
2469 break;
2470 if (req->r_tid > last_tid)
2471 break;
2472
2473 next_tid = req->r_tid + 1;
2474 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2475 continue;
2476
2477 ceph_osdc_get_request(req);
2478 mutex_unlock(&osdc->request_mutex);
2479 dout("sync waiting on tid %llu (last is %llu)\n",
2480 req->r_tid, last_tid);
2481 wait_for_completion(&req->r_safe_completion);
2482 mutex_lock(&osdc->request_mutex);
2483 ceph_osdc_put_request(req);
2484 }
2485 mutex_unlock(&osdc->request_mutex);
2486 dout("sync done (thru tid %llu)\n", last_tid);
2487}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002488EXPORT_SYMBOL(ceph_osdc_sync);
Sage Weilf24e9982009-10-06 11:31:10 -07002489
2490/*
Josh Durgindd935f42013-08-28 21:43:09 -07002491 * Call all pending notify callbacks - for use after a watch is
2492 * unregistered, to make sure no more callbacks for it will be invoked
2493 */
2494extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2495{
2496 flush_workqueue(osdc->notify_wq);
2497}
2498EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2499
2500
2501/*
Sage Weilf24e9982009-10-06 11:31:10 -07002502 * init, shutdown
2503 */
2504int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2505{
2506 int err;
2507
2508 dout("init\n");
2509 osdc->client = client;
2510 osdc->osdmap = NULL;
2511 init_rwsem(&osdc->map_sem);
2512 init_completion(&osdc->map_waiters);
2513 osdc->last_requested_map = 0;
2514 mutex_init(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07002515 osdc->last_tid = 0;
2516 osdc->osds = RB_ROOT;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002517 INIT_LIST_HEAD(&osdc->osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07002518 osdc->requests = RB_ROOT;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08002519 INIT_LIST_HEAD(&osdc->req_lru);
Sage Weil6f6c7002011-01-17 20:34:08 -08002520 INIT_LIST_HEAD(&osdc->req_unsent);
2521 INIT_LIST_HEAD(&osdc->req_notarget);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002522 INIT_LIST_HEAD(&osdc->req_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07002523 osdc->num_requests = 0;
2524 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002525 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002526 spin_lock_init(&osdc->event_lock);
2527 osdc->event_tree = RB_ROOT;
2528 osdc->event_count = 0;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002529
2530 schedule_delayed_work(&osdc->osds_timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002531 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
Sage Weilf24e9982009-10-06 11:31:10 -07002532
Sage Weil5f44f142009-11-18 14:52:18 -08002533 err = -ENOMEM;
Sage Weilf24e9982009-10-06 11:31:10 -07002534 osdc->req_mempool = mempool_create_kmalloc_pool(10,
2535 sizeof(struct ceph_osd_request));
2536 if (!osdc->req_mempool)
Sage Weil5f44f142009-11-18 14:52:18 -08002537 goto out;
Sage Weilf24e9982009-10-06 11:31:10 -07002538
Sage Weild50b4092012-07-09 14:22:34 -07002539 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2540 OSD_OP_FRONT_LEN, 10, true,
Sage Weil4f482802010-04-24 09:56:35 -07002541 "osd_op");
Sage Weilf24e9982009-10-06 11:31:10 -07002542 if (err < 0)
Sage Weil5f44f142009-11-18 14:52:18 -08002543 goto out_mempool;
Sage Weild50b4092012-07-09 14:22:34 -07002544 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
Sage Weil4f482802010-04-24 09:56:35 -07002545 OSD_OPREPLY_FRONT_LEN, 10, true,
2546 "osd_op_reply");
Sage Weilc16e7862010-03-01 13:02:00 -08002547 if (err < 0)
2548 goto out_msgpool;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002549
Dan Carpenterdbcae082013-08-15 08:58:59 +03002550 err = -ENOMEM;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002551 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
Dan Carpenterdbcae082013-08-15 08:58:59 +03002552 if (!osdc->notify_wq)
Ilya Dryomovc172ec52014-01-31 17:49:22 +02002553 goto out_msgpool_reply;
2554
Sage Weilf24e9982009-10-06 11:31:10 -07002555 return 0;
Sage Weil5f44f142009-11-18 14:52:18 -08002556
Ilya Dryomovc172ec52014-01-31 17:49:22 +02002557out_msgpool_reply:
2558 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
Sage Weilc16e7862010-03-01 13:02:00 -08002559out_msgpool:
2560 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weil5f44f142009-11-18 14:52:18 -08002561out_mempool:
2562 mempool_destroy(osdc->req_mempool);
2563out:
2564 return err;
Sage Weilf24e9982009-10-06 11:31:10 -07002565}
2566
2567void ceph_osdc_stop(struct ceph_osd_client *osdc)
2568{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002569 flush_workqueue(osdc->notify_wq);
2570 destroy_workqueue(osdc->notify_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07002571 cancel_delayed_work_sync(&osdc->timeout_work);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08002572 cancel_delayed_work_sync(&osdc->osds_timeout_work);
Sage Weilf24e9982009-10-06 11:31:10 -07002573 if (osdc->osdmap) {
2574 ceph_osdmap_destroy(osdc->osdmap);
2575 osdc->osdmap = NULL;
2576 }
Sage Weilaca420b2011-08-31 14:45:53 -07002577 remove_all_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07002578 mempool_destroy(osdc->req_mempool);
2579 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weilc16e7862010-03-01 13:02:00 -08002580 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
Sage Weilf24e9982009-10-06 11:31:10 -07002581}
2582
2583/*
2584 * Read some contiguous pages. If we cross a stripe boundary, shorten
2585 * *plen. Return number of bytes read, or error.
2586 */
2587int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2588 struct ceph_vino vino, struct ceph_file_layout *layout,
2589 u64 off, u64 *plen,
2590 u32 truncate_seq, u64 truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -08002591 struct page **pages, int num_pages, int page_align)
Sage Weilf24e9982009-10-06 11:31:10 -07002592{
2593 struct ceph_osd_request *req;
2594 int rc = 0;
2595
2596 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2597 vino.snap, off, *plen);
Alex Elder79528732013-04-03 21:32:51 -05002598 req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002599 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
Alex Elderacead002013-03-14 14:09:05 -05002600 NULL, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002601 false);
Sage Weil68162822012-09-24 21:01:02 -07002602 if (IS_ERR(req))
2603 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002604
2605 /* it may be a short read due to an object boundary */
Alex Elder0fff87e2013-02-14 12:16:43 -06002606
Alex Elder406e2c92013-04-15 14:50:36 -05002607 osd_req_op_extent_osd_data_pages(req, 0,
Alex Eldera4ce40a2013-04-05 01:27:12 -05002608 pages, *plen, page_align, false, false);
Sage Weilf24e9982009-10-06 11:31:10 -07002609
Alex Eldere0c59482013-03-07 15:38:25 -06002610 dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n",
Alex Elder43bfe5d2013-04-03 01:28:57 -05002611 off, *plen, *plen, page_align);
Sage Weilf24e9982009-10-06 11:31:10 -07002612
Alex Elder79528732013-04-03 21:32:51 -05002613 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
Alex Elder02ee07d2013-03-14 14:09:06 -05002614
Sage Weilf24e9982009-10-06 11:31:10 -07002615 rc = ceph_osdc_start_request(osdc, req, false);
2616 if (!rc)
2617 rc = ceph_osdc_wait_request(osdc, req);
2618
2619 ceph_osdc_put_request(req);
2620 dout("readpages result %d\n", rc);
2621 return rc;
2622}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002623EXPORT_SYMBOL(ceph_osdc_readpages);
Sage Weilf24e9982009-10-06 11:31:10 -07002624
2625/*
2626 * do a synchronous write on N pages
2627 */
2628int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2629 struct ceph_file_layout *layout,
2630 struct ceph_snap_context *snapc,
2631 u64 off, u64 len,
2632 u32 truncate_seq, u64 truncate_size,
2633 struct timespec *mtime,
Alex Elder24808822013-02-15 11:42:29 -06002634 struct page **pages, int num_pages)
Sage Weilf24e9982009-10-06 11:31:10 -07002635{
2636 struct ceph_osd_request *req;
2637 int rc = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -08002638 int page_align = off & ~PAGE_MASK;
Sage Weilf24e9982009-10-06 11:31:10 -07002639
Alex Elderacead002013-03-14 14:09:05 -05002640 BUG_ON(vino.snap != CEPH_NOSNAP); /* snapshots aren't writeable */
Alex Elder79528732013-04-03 21:32:51 -05002641 req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
Sage Weilf24e9982009-10-06 11:31:10 -07002642 CEPH_OSD_OP_WRITE,
Alex Elder24808822013-02-15 11:42:29 -06002643 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
Alex Elderacead002013-03-14 14:09:05 -05002644 snapc, truncate_seq, truncate_size,
Alex Elder153e5162013-03-01 18:00:15 -06002645 true);
Sage Weil68162822012-09-24 21:01:02 -07002646 if (IS_ERR(req))
2647 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07002648
2649 /* it may be a short write due to an object boundary */
Alex Elder406e2c92013-04-15 14:50:36 -05002650 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
Alex Elder43bfe5d2013-04-03 01:28:57 -05002651 false, false);
2652 dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
Sage Weilf24e9982009-10-06 11:31:10 -07002653
Alex Elder79528732013-04-03 21:32:51 -05002654 ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
Alex Elder02ee07d2013-03-14 14:09:06 -05002655
Alex Elder87f979d2013-02-15 11:42:29 -06002656 rc = ceph_osdc_start_request(osdc, req, true);
Sage Weilf24e9982009-10-06 11:31:10 -07002657 if (!rc)
2658 rc = ceph_osdc_wait_request(osdc, req);
2659
2660 ceph_osdc_put_request(req);
2661 if (rc == 0)
2662 rc = len;
2663 dout("writepages result %d\n", rc);
2664 return rc;
2665}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002666EXPORT_SYMBOL(ceph_osdc_writepages);
Sage Weilf24e9982009-10-06 11:31:10 -07002667
Alex Elder5522ae02013-05-01 12:43:04 -05002668int ceph_osdc_setup(void)
2669{
2670 BUG_ON(ceph_osd_request_cache);
2671 ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2672 sizeof (struct ceph_osd_request),
2673 __alignof__(struct ceph_osd_request),
2674 0, NULL);
2675
2676 return ceph_osd_request_cache ? 0 : -ENOMEM;
2677}
2678EXPORT_SYMBOL(ceph_osdc_setup);
2679
2680void ceph_osdc_cleanup(void)
2681{
2682 BUG_ON(!ceph_osd_request_cache);
2683 kmem_cache_destroy(ceph_osd_request_cache);
2684 ceph_osd_request_cache = NULL;
2685}
2686EXPORT_SYMBOL(ceph_osdc_cleanup);
2687
Sage Weilf24e9982009-10-06 11:31:10 -07002688/*
2689 * handle incoming message
2690 */
2691static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2692{
2693 struct ceph_osd *osd = con->private;
Julia Lawall32c895e2009-11-21 16:53:16 +01002694 struct ceph_osd_client *osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002695 int type = le16_to_cpu(msg->hdr.type);
2696
2697 if (!osd)
Sage Weil4a32f932010-06-13 10:27:53 -07002698 goto out;
Julia Lawall32c895e2009-11-21 16:53:16 +01002699 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002700
2701 switch (type) {
2702 case CEPH_MSG_OSD_MAP:
2703 ceph_osdc_handle_map(osdc, msg);
2704 break;
2705 case CEPH_MSG_OSD_OPREPLY:
Sage Weil350b1c32009-12-22 10:45:45 -08002706 handle_reply(osdc, msg, con);
Sage Weilf24e9982009-10-06 11:31:10 -07002707 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002708 case CEPH_MSG_WATCH_NOTIFY:
2709 handle_watch_notify(osdc, msg);
2710 break;
Sage Weilf24e9982009-10-06 11:31:10 -07002711
2712 default:
2713 pr_err("received unknown message type %d %s\n", type,
2714 ceph_msg_type_name(type));
2715 }
Sage Weil4a32f932010-06-13 10:27:53 -07002716out:
Sage Weilf24e9982009-10-06 11:31:10 -07002717 ceph_msg_put(msg);
2718}
2719
Sage Weil5b3a4db2010-02-19 21:43:23 -08002720/*
Sage Weil21b667f2010-03-04 10:22:59 -08002721 * lookup and return message for incoming reply. set up reply message
2722 * pages.
Sage Weil5b3a4db2010-02-19 21:43:23 -08002723 */
2724static struct ceph_msg *get_reply(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002725 struct ceph_msg_header *hdr,
2726 int *skip)
Sage Weilf24e9982009-10-06 11:31:10 -07002727{
2728 struct ceph_osd *osd = con->private;
2729 struct ceph_osd_client *osdc = osd->o_osdc;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002730 struct ceph_msg *m;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002731 struct ceph_osd_request *req;
Ilya Dryomov3f0a4ac2014-01-09 20:08:21 +02002732 int front_len = le32_to_cpu(hdr->front_len);
Sage Weil5b3a4db2010-02-19 21:43:23 -08002733 int data_len = le32_to_cpu(hdr->data_len);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002734 u64 tid;
Sage Weilf24e9982009-10-06 11:31:10 -07002735
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002736 tid = le64_to_cpu(hdr->tid);
2737 mutex_lock(&osdc->request_mutex);
2738 req = __lookup_request(osdc, tid);
2739 if (!req) {
2740 *skip = 1;
2741 m = NULL;
Sage Weil756a16a2012-07-30 16:26:13 -07002742 dout("get_reply unknown tid %llu from osd%d\n", tid,
2743 osd->o_osd);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002744 goto out;
2745 }
Sage Weilc16e7862010-03-01 13:02:00 -08002746
Alex Elderace6d3a2013-04-01 16:12:14 -05002747 if (req->r_reply->con)
Alex Elder8921d112012-06-01 14:56:43 -05002748 dout("%s revoking msg %p from old con %p\n", __func__,
Alex Elderace6d3a2013-04-01 16:12:14 -05002749 req->r_reply, req->r_reply->con);
2750 ceph_msg_revoke_incoming(req->r_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002751
Ilya Dryomovf2be82b2014-01-09 20:08:21 +02002752 if (front_len > req->r_reply->front_alloc_len) {
Alex Elder49743412013-05-29 11:19:00 -05002753 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
Ilya Dryomovf2be82b2014-01-09 20:08:21 +02002754 front_len, req->r_reply->front_alloc_len,
Alex Elder49743412013-05-29 11:19:00 -05002755 (unsigned int)con->peer_name.type,
2756 le64_to_cpu(con->peer_name.num));
Ilya Dryomov3f0a4ac2014-01-09 20:08:21 +02002757 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2758 false);
Sage Weila79832f2010-04-01 16:06:19 -07002759 if (!m)
Sage Weilc16e7862010-03-01 13:02:00 -08002760 goto out;
2761 ceph_msg_put(req->r_reply);
2762 req->r_reply = m;
2763 }
2764 m = ceph_msg_get(req->r_reply);
2765
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002766 if (data_len > 0) {
Alex Eldera4ce40a2013-04-05 01:27:12 -05002767 struct ceph_osd_data *osd_data;
Alex Elder0fff87e2013-02-14 12:16:43 -06002768
Alex Eldera4ce40a2013-04-05 01:27:12 -05002769 /*
2770 * XXX This is assuming there is only one op containing
2771 * XXX page data. Probably OK for reads, but this
2772 * XXX ought to be done more generally.
2773 */
Alex Elder406e2c92013-04-15 14:50:36 -05002774 osd_data = osd_req_op_extent_osd_data(req, 0);
Alex Elder0fff87e2013-02-14 12:16:43 -06002775 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
Alex Elder0fff87e2013-02-14 12:16:43 -06002776 if (osd_data->pages &&
Alex Eldere0c59482013-03-07 15:38:25 -06002777 unlikely(osd_data->length < data_len)) {
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002778
Alex Eldere0c59482013-03-07 15:38:25 -06002779 pr_warning("tid %lld reply has %d bytes "
2780 "we had only %llu bytes ready\n",
2781 tid, data_len, osd_data->length);
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002782 *skip = 1;
2783 ceph_msg_put(m);
2784 m = NULL;
2785 goto out;
2786 }
Alex Elder2ac2b7a62013-02-14 12:16:43 -06002787 }
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002788 }
Sage Weil5b3a4db2010-02-19 21:43:23 -08002789 *skip = 0;
Sage Weilc16e7862010-03-01 13:02:00 -08002790 dout("get_reply tid %lld %p\n", tid, m);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002791
2792out:
2793 mutex_unlock(&osdc->request_mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002794 return m;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002795
2796}
2797
2798static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2799 struct ceph_msg_header *hdr,
2800 int *skip)
2801{
2802 struct ceph_osd *osd = con->private;
2803 int type = le16_to_cpu(hdr->type);
2804 int front = le32_to_cpu(hdr->front_len);
2805
Alex Elder1c20f2d2012-06-04 14:43:32 -05002806 *skip = 0;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002807 switch (type) {
2808 case CEPH_MSG_OSD_MAP:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002809 case CEPH_MSG_WATCH_NOTIFY:
Sage Weilb61c2762011-08-09 15:03:46 -07002810 return ceph_msg_new(type, front, GFP_NOFS, false);
Sage Weil5b3a4db2010-02-19 21:43:23 -08002811 case CEPH_MSG_OSD_OPREPLY:
2812 return get_reply(con, hdr, skip);
2813 default:
2814 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2815 osd->o_osd);
2816 *skip = 1;
2817 return NULL;
2818 }
Sage Weilf24e9982009-10-06 11:31:10 -07002819}
2820
2821/*
2822 * Wrappers to refcount containing ceph_osd struct
2823 */
2824static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2825{
2826 struct ceph_osd *osd = con->private;
2827 if (get_osd(osd))
2828 return con;
2829 return NULL;
2830}
2831
2832static void put_osd_con(struct ceph_connection *con)
2833{
2834 struct ceph_osd *osd = con->private;
2835 put_osd(osd);
2836}
2837
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002838/*
2839 * authentication
2840 */
Alex Eldera3530df2012-05-16 15:16:39 -05002841/*
2842 * Note: returned pointer is the address of a structure that's
2843 * managed separately. Caller must *not* attempt to free it.
2844 */
2845static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
Alex Elder8f43fb52012-05-16 15:16:39 -05002846 int *proto, int force_new)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002847{
2848 struct ceph_osd *o = con->private;
2849 struct ceph_osd_client *osdc = o->o_osdc;
2850 struct ceph_auth_client *ac = osdc->client->monc.auth;
Alex Elder74f18692012-05-16 15:16:39 -05002851 struct ceph_auth_handshake *auth = &o->o_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002852
Alex Elder74f18692012-05-16 15:16:39 -05002853 if (force_new && auth->authorizer) {
Sage Weil27859f92013-03-25 10:26:14 -07002854 ceph_auth_destroy_authorizer(ac, auth->authorizer);
Alex Elder74f18692012-05-16 15:16:39 -05002855 auth->authorizer = NULL;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002856 }
Sage Weil27859f92013-03-25 10:26:14 -07002857 if (!auth->authorizer) {
2858 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2859 auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002860 if (ret)
Alex Eldera3530df2012-05-16 15:16:39 -05002861 return ERR_PTR(ret);
Sage Weil27859f92013-03-25 10:26:14 -07002862 } else {
2863 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
Sage Weil0bed9b52013-03-25 10:26:01 -07002864 auth);
2865 if (ret)
2866 return ERR_PTR(ret);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002867 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002868 *proto = ac->protocol;
Alex Elder74f18692012-05-16 15:16:39 -05002869
Alex Eldera3530df2012-05-16 15:16:39 -05002870 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002871}
2872
2873
2874static int verify_authorizer_reply(struct ceph_connection *con, int len)
2875{
2876 struct ceph_osd *o = con->private;
2877 struct ceph_osd_client *osdc = o->o_osdc;
2878 struct ceph_auth_client *ac = osdc->client->monc.auth;
2879
Sage Weil27859f92013-03-25 10:26:14 -07002880 return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002881}
2882
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002883static int invalidate_authorizer(struct ceph_connection *con)
2884{
2885 struct ceph_osd *o = con->private;
2886 struct ceph_osd_client *osdc = o->o_osdc;
2887 struct ceph_auth_client *ac = osdc->client->monc.auth;
2888
Sage Weil27859f92013-03-25 10:26:14 -07002889 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002890 return ceph_monc_validate_auth(&osdc->client->monc);
2891}
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002892
Tobias Klauser9e327892010-05-20 10:40:19 +02002893static const struct ceph_connection_operations osd_con_ops = {
Sage Weilf24e9982009-10-06 11:31:10 -07002894 .get = get_osd_con,
2895 .put = put_osd_con,
2896 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002897 .get_authorizer = get_authorizer,
2898 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002899 .invalidate_authorizer = invalidate_authorizer,
Sage Weilf24e9982009-10-06 11:31:10 -07002900 .alloc_msg = alloc_msg,
Sage Weil81b024e2009-10-09 10:29:18 -07002901 .fault = osd_reset,
Sage Weilf24e9982009-10-06 11:31:10 -07002902};