blob: a28c976ae3aebdfceefbe7d288a5e821d43123bf [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weilf24e9982009-10-06 11:31:10 -07002
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003#include <linux/module.h>
Sage Weilf24e9982009-10-06 11:31:10 -07004#include <linux/err.h>
5#include <linux/highmem.h>
6#include <linux/mm.h>
7#include <linux/pagemap.h>
8#include <linux/slab.h>
9#include <linux/uaccess.h>
Yehuda Sadeh68b44762010-04-06 15:01:27 -070010#ifdef CONFIG_BLOCK
11#include <linux/bio.h>
12#endif
Sage Weilf24e9982009-10-06 11:31:10 -070013
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070014#include <linux/ceph/libceph.h>
15#include <linux/ceph/osd_client.h>
16#include <linux/ceph/messenger.h>
17#include <linux/ceph/decode.h>
18#include <linux/ceph/auth.h>
19#include <linux/ceph/pagelist.h>
Sage Weilf24e9982009-10-06 11:31:10 -070020
Sage Weilc16e7862010-03-01 13:02:00 -080021#define OSD_OP_FRONT_LEN 4096
22#define OSD_OPREPLY_FRONT_LEN 512
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -080023
Tobias Klauser9e327892010-05-20 10:40:19 +020024static const struct ceph_connection_operations osd_con_ops;
Sage Weilf24e9982009-10-06 11:31:10 -070025
Alex Elderf9d25192013-02-15 11:42:29 -060026static void __send_queued(struct ceph_osd_client *osdc);
Sage Weil6f6c7002011-01-17 20:34:08 -080027static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -070028static void __register_request(struct ceph_osd_client *osdc,
29 struct ceph_osd_request *req);
30static void __unregister_linger_request(struct ceph_osd_client *osdc,
31 struct ceph_osd_request *req);
Sage Weil56e925b2012-01-03 12:34:34 -080032static void __send_request(struct ceph_osd_client *osdc,
33 struct ceph_osd_request *req);
Sage Weilf24e9982009-10-06 11:31:10 -070034
Yehuda Sadeh68b44762010-04-06 15:01:27 -070035static int op_has_extent(int op)
36{
37 return (op == CEPH_OSD_OP_READ ||
38 op == CEPH_OSD_OP_WRITE);
39}
40
Sage Weilf24e9982009-10-06 11:31:10 -070041/*
42 * Implement client access to distributed object storage cluster.
43 *
44 * All data objects are stored within a cluster/cloud of OSDs, or
45 * "object storage devices." (Note that Ceph OSDs have _nothing_ to
46 * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply
47 * remote daemons serving up and coordinating consistent and safe
48 * access to storage.
49 *
50 * Cluster membership and the mapping of data objects onto storage devices
51 * are described by the osd map.
52 *
53 * We keep track of pending OSD requests (read, write), resubmit
54 * requests to different OSDs when the cluster topology/data layout
55 * change, or retry the affected requests when the communications
56 * channel with an OSD is reset.
57 */
58
59/*
60 * calculate the mapping of a file extent onto an object, and fill out the
61 * request accordingly. shorten extent as necessary if it crosses an
62 * object boundary.
63 *
64 * fill osd op in request message.
65 */
Alex Elderdbe0fc42013-02-15 22:10:17 -060066static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
Alex Elder47a05812013-02-15 22:10:17 -060067 struct ceph_osd_req_op *op, u64 *bno)
Sage Weilf24e9982009-10-06 11:31:10 -070068{
Alex Elder60e56f12013-02-15 11:42:29 -060069 u64 orig_len = *plen;
Alex Elder60e56f12013-02-15 11:42:29 -060070 u64 objoff = 0;
71 u64 objlen = 0;
Sage Weild63b77f2012-09-24 20:59:48 -070072 int r;
Sage Weilf24e9982009-10-06 11:31:10 -070073
Alex Elder60e56f12013-02-15 11:42:29 -060074 /* object extent? */
Alex Elder47a05812013-02-15 22:10:17 -060075 r = ceph_calc_file_object_mapping(layout, off, orig_len, bno,
Alex Elder60e56f12013-02-15 11:42:29 -060076 &objoff, &objlen);
Sage Weild63b77f2012-09-24 20:59:48 -070077 if (r < 0)
78 return r;
Alex Elder60e56f12013-02-15 11:42:29 -060079 if (objlen < orig_len) {
80 *plen = objlen;
81 dout(" skipping last %llu, final file extent %llu~%llu\n",
82 orig_len - *plen, off, *plen);
83 }
84
85 if (op_has_extent(op->op)) {
86 u32 osize = le32_to_cpu(layout->fl_object_size);
87 op->extent.offset = objoff;
88 op->extent.length = objlen;
89 if (op->extent.truncate_size <= off - objoff) {
90 op->extent.truncate_size = 0;
91 } else {
92 op->extent.truncate_size -= off - objoff;
93 if (op->extent.truncate_size > osize)
94 op->extent.truncate_size = osize;
95 }
96 }
Alex Elder60e56f12013-02-15 11:42:29 -060097 if (op->op == CEPH_OSD_OP_WRITE)
98 op->payload_len = *plen;
99
Alex Elder60cf5992013-02-15 22:10:17 -0600100 dout("calc_layout bno=%llx %llu~%llu\n", *bno, objoff, objlen);
Sage Weilf24e9982009-10-06 11:31:10 -0700101
Alex Elder3ff5f382013-02-15 22:10:17 -0600102 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -0700103}
104
Sage Weilf24e9982009-10-06 11:31:10 -0700105/*
106 * requests
107 */
Sage Weil415e49a2009-12-07 13:37:03 -0800108void ceph_osdc_release_request(struct kref *kref)
Sage Weilf24e9982009-10-06 11:31:10 -0700109{
Sage Weil415e49a2009-12-07 13:37:03 -0800110 struct ceph_osd_request *req = container_of(kref,
111 struct ceph_osd_request,
112 r_kref);
113
114 if (req->r_request)
115 ceph_msg_put(req->r_request);
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -0800116 if (req->r_con_filling_msg) {
Alex Elder9cbb1d72013-01-31 16:02:00 -0600117 dout("%s revoking msg %p from con %p\n", __func__,
118 req->r_reply, req->r_con_filling_msg);
Alex Elder8921d112012-06-01 14:56:43 -0500119 ceph_msg_revoke_incoming(req->r_reply);
Sage Weil0d477662012-05-31 20:22:18 -0700120 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
Alex Elder9cbb1d72013-01-31 16:02:00 -0600121 req->r_con_filling_msg = NULL;
Sage Weil350b1c32009-12-22 10:45:45 -0800122 }
Alex Elderab8cb342012-06-04 14:43:32 -0500123 if (req->r_reply)
124 ceph_msg_put(req->r_reply);
Sage Weil415e49a2009-12-07 13:37:03 -0800125 if (req->r_own_pages)
126 ceph_release_page_vector(req->r_pages,
127 req->r_num_pages);
128 ceph_put_snap_context(req->r_snapc);
Alex Elderc8858372012-11-13 21:11:15 -0600129 ceph_pagelist_release(&req->r_trail);
Sage Weil415e49a2009-12-07 13:37:03 -0800130 if (req->r_mempool)
131 mempool_free(req, req->r_osdc->req_mempool);
132 else
133 kfree(req);
Sage Weilf24e9982009-10-06 11:31:10 -0700134}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700135EXPORT_SYMBOL(ceph_osdc_release_request);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700136
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700137struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700138 struct ceph_snap_context *snapc,
Sage Weil1b83bef2013-02-25 16:11:12 -0800139 unsigned int num_ops,
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700140 bool use_mempool,
Alex Elder54a54002012-11-13 21:11:15 -0600141 gfp_t gfp_flags)
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700142{
143 struct ceph_osd_request *req;
144 struct ceph_msg *msg;
Sage Weil1b83bef2013-02-25 16:11:12 -0800145 size_t msg_size;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700146
Sage Weil1b83bef2013-02-25 16:11:12 -0800147 msg_size = 4 + 4 + 8 + 8 + 4+8;
148 msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
149 msg_size += 1 + 8 + 4 + 4; /* pg_t */
150 msg_size += 4 + MAX_OBJ_NAME_SIZE;
151 msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
152 msg_size += 8; /* snapid */
153 msg_size += 8; /* snap_seq */
154 msg_size += 8 * (snapc ? snapc->num_snaps : 0); /* snaps */
155 msg_size += 4;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700156
157 if (use_mempool) {
158 req = mempool_alloc(osdc->req_mempool, gfp_flags);
159 memset(req, 0, sizeof(*req));
160 } else {
161 req = kzalloc(sizeof(*req), gfp_flags);
162 }
163 if (req == NULL)
164 return NULL;
165
166 req->r_osdc = osdc;
167 req->r_mempool = use_mempool;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700168
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700169 kref_init(&req->r_kref);
170 init_completion(&req->r_completion);
171 init_completion(&req->r_safe_completion);
Alex Eldera978fa22012-12-17 12:23:48 -0600172 RB_CLEAR_NODE(&req->r_node);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700173 INIT_LIST_HEAD(&req->r_unsafe_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700174 INIT_LIST_HEAD(&req->r_linger_item);
175 INIT_LIST_HEAD(&req->r_linger_osd);
Sage Weil935b6392011-09-16 11:13:17 -0700176 INIT_LIST_HEAD(&req->r_req_lru_item);
Sage Weilcd430452012-07-09 14:31:41 -0700177 INIT_LIST_HEAD(&req->r_osd_item);
178
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700179 /* create reply message */
180 if (use_mempool)
181 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
182 else
183 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
Sage Weilb61c2762011-08-09 15:03:46 -0700184 OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700185 if (!msg) {
186 ceph_osdc_put_request(req);
187 return NULL;
188 }
189 req->r_reply = msg;
190
Alex Elderc8858372012-11-13 21:11:15 -0600191 ceph_pagelist_init(&req->r_trail);
Sage Weild50b4092012-07-09 14:22:34 -0700192
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700193 /* create request message; allow space for oid */
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700194 if (use_mempool)
195 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
196 else
Sage Weilb61c2762011-08-09 15:03:46 -0700197 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700198 if (!msg) {
199 ceph_osdc_put_request(req);
200 return NULL;
201 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700202
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700203 memset(msg->front.iov_base, 0, msg->front.iov_len);
204
205 req->r_request = msg;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700206
207 return req;
208}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700209EXPORT_SYMBOL(ceph_osdc_alloc_request);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700210
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700211static void osd_req_encode_op(struct ceph_osd_request *req,
212 struct ceph_osd_op *dst,
213 struct ceph_osd_req_op *src)
214{
215 dst->op = cpu_to_le16(src->op);
216
Alex Elder065a68f2012-04-20 15:49:43 -0500217 switch (src->op) {
Alex Elderfbfab532013-02-08 09:55:48 -0600218 case CEPH_OSD_OP_STAT:
219 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700220 case CEPH_OSD_OP_READ:
221 case CEPH_OSD_OP_WRITE:
222 dst->extent.offset =
223 cpu_to_le64(src->extent.offset);
224 dst->extent.length =
225 cpu_to_le64(src->extent.length);
226 dst->extent.truncate_size =
227 cpu_to_le64(src->extent.truncate_size);
228 dst->extent.truncate_seq =
229 cpu_to_le32(src->extent.truncate_seq);
230 break;
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700231 case CEPH_OSD_OP_CALL:
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700232 dst->cls.class_len = src->cls.class_len;
233 dst->cls.method_len = src->cls.method_len;
234 dst->cls.indata_len = cpu_to_le32(src->cls.indata_len);
235
Alex Elderc8858372012-11-13 21:11:15 -0600236 ceph_pagelist_append(&req->r_trail, src->cls.class_name,
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700237 src->cls.class_len);
Alex Elderc8858372012-11-13 21:11:15 -0600238 ceph_pagelist_append(&req->r_trail, src->cls.method_name,
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700239 src->cls.method_len);
Alex Elderc8858372012-11-13 21:11:15 -0600240 ceph_pagelist_append(&req->r_trail, src->cls.indata,
Yehuda Sadehae1533b2010-05-18 16:38:08 -0700241 src->cls.indata_len);
242 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700243 case CEPH_OSD_OP_STARTSYNC:
244 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700245 case CEPH_OSD_OP_NOTIFY_ACK:
246 case CEPH_OSD_OP_WATCH:
247 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
248 dst->watch.ver = cpu_to_le64(src->watch.ver);
249 dst->watch.flag = src->watch.flag;
250 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700251 default:
252 pr_err("unrecognized osd opcode %d\n", dst->op);
253 WARN_ON(1);
254 break;
Alex Elder4c464592013-02-15 11:42:30 -0600255 case CEPH_OSD_OP_MAPEXT:
256 case CEPH_OSD_OP_MASKTRUNC:
257 case CEPH_OSD_OP_SPARSE_READ:
Alex Eldera9f36c32013-02-15 11:42:30 -0600258 case CEPH_OSD_OP_NOTIFY:
Alex Elder4c464592013-02-15 11:42:30 -0600259 case CEPH_OSD_OP_ASSERT_VER:
260 case CEPH_OSD_OP_WRITEFULL:
261 case CEPH_OSD_OP_TRUNCATE:
262 case CEPH_OSD_OP_ZERO:
263 case CEPH_OSD_OP_DELETE:
264 case CEPH_OSD_OP_APPEND:
265 case CEPH_OSD_OP_SETTRUNC:
266 case CEPH_OSD_OP_TRIMTRUNC:
267 case CEPH_OSD_OP_TMAPUP:
268 case CEPH_OSD_OP_TMAPPUT:
269 case CEPH_OSD_OP_TMAPGET:
270 case CEPH_OSD_OP_CREATE:
Alex Eldera9f36c32013-02-15 11:42:30 -0600271 case CEPH_OSD_OP_ROLLBACK:
Alex Elder4c464592013-02-15 11:42:30 -0600272 case CEPH_OSD_OP_OMAPGETKEYS:
273 case CEPH_OSD_OP_OMAPGETVALS:
274 case CEPH_OSD_OP_OMAPGETHEADER:
275 case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
276 case CEPH_OSD_OP_MODE_RD:
277 case CEPH_OSD_OP_OMAPSETVALS:
278 case CEPH_OSD_OP_OMAPSETHEADER:
279 case CEPH_OSD_OP_OMAPCLEAR:
280 case CEPH_OSD_OP_OMAPRMKEYS:
281 case CEPH_OSD_OP_OMAP_CMP:
282 case CEPH_OSD_OP_CLONERANGE:
283 case CEPH_OSD_OP_ASSERT_SRC_VERSION:
284 case CEPH_OSD_OP_SRC_CMPXATTR:
Alex Eldera9f36c32013-02-15 11:42:30 -0600285 case CEPH_OSD_OP_GETXATTR:
Alex Elder4c464592013-02-15 11:42:30 -0600286 case CEPH_OSD_OP_GETXATTRS:
Alex Eldera9f36c32013-02-15 11:42:30 -0600287 case CEPH_OSD_OP_CMPXATTR:
288 case CEPH_OSD_OP_SETXATTR:
Alex Elder4c464592013-02-15 11:42:30 -0600289 case CEPH_OSD_OP_SETXATTRS:
290 case CEPH_OSD_OP_RESETXATTRS:
291 case CEPH_OSD_OP_RMXATTR:
292 case CEPH_OSD_OP_PULL:
293 case CEPH_OSD_OP_PUSH:
294 case CEPH_OSD_OP_BALANCEREADS:
295 case CEPH_OSD_OP_UNBALANCEREADS:
296 case CEPH_OSD_OP_SCRUB:
297 case CEPH_OSD_OP_SCRUB_RESERVE:
298 case CEPH_OSD_OP_SCRUB_UNRESERVE:
299 case CEPH_OSD_OP_SCRUB_STOP:
300 case CEPH_OSD_OP_SCRUB_MAP:
301 case CEPH_OSD_OP_WRLOCK:
302 case CEPH_OSD_OP_WRUNLOCK:
303 case CEPH_OSD_OP_RDLOCK:
304 case CEPH_OSD_OP_RDUNLOCK:
305 case CEPH_OSD_OP_UPLOCK:
306 case CEPH_OSD_OP_DNLOCK:
307 case CEPH_OSD_OP_PGLS:
308 case CEPH_OSD_OP_PGLS_FILTER:
309 pr_err("unsupported osd opcode %s\n",
310 ceph_osd_op_name(dst->op));
311 WARN_ON(1);
312 break;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700313 }
314 dst->payload_len = cpu_to_le32(src->payload_len);
315}
316
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700317/*
318 * build new request AND message
319 *
320 */
321void ceph_osdc_build_request(struct ceph_osd_request *req,
Sage Weil1b83bef2013-02-25 16:11:12 -0800322 u64 off, u64 len, unsigned int num_ops,
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700323 struct ceph_osd_req_op *src_ops,
Alex Elder4d6b2502012-11-13 21:11:15 -0600324 struct ceph_snap_context *snapc, u64 snap_id,
Alex Elderaf77f262012-11-09 08:43:15 -0600325 struct timespec *mtime)
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700326{
327 struct ceph_msg *msg = req->r_request;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700328 struct ceph_osd_req_op *src_op;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700329 void *p;
Sage Weil1b83bef2013-02-25 16:11:12 -0800330 size_t msg_size;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700331 int flags = req->r_flags;
Alex Elderf44246e2013-02-14 12:16:43 -0600332 u64 data_len;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700333 int i;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700334
Sage Weil1b83bef2013-02-25 16:11:12 -0800335 req->r_num_ops = num_ops;
336 req->r_snapid = snap_id;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700337 req->r_snapc = ceph_get_snap_context(snapc);
338
Sage Weil1b83bef2013-02-25 16:11:12 -0800339 /* encode request */
340 msg->hdr.version = cpu_to_le16(4);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700341
Sage Weil1b83bef2013-02-25 16:11:12 -0800342 p = msg->front.iov_base;
343 ceph_encode_32(&p, 1); /* client_inc is always 1 */
344 req->r_request_osdmap_epoch = p;
345 p += 4;
346 req->r_request_flags = p;
347 p += 4;
348 if (req->r_flags & CEPH_OSD_FLAG_WRITE)
349 ceph_encode_timespec(p, mtime);
350 p += sizeof(struct ceph_timespec);
351 req->r_request_reassert_version = p;
352 p += sizeof(struct ceph_eversion); /* will get filled in */
353
354 /* oloc */
355 ceph_encode_8(&p, 4);
356 ceph_encode_8(&p, 4);
357 ceph_encode_32(&p, 8 + 4 + 4);
358 req->r_request_pool = p;
359 p += 8;
360 ceph_encode_32(&p, -1); /* preferred */
361 ceph_encode_32(&p, 0); /* key len */
362
363 ceph_encode_8(&p, 1);
364 req->r_request_pgid = p;
365 p += 8 + 4;
366 ceph_encode_32(&p, -1); /* preferred */
367
368 /* oid */
369 ceph_encode_32(&p, req->r_oid_len);
Alex Elderaf77f262012-11-09 08:43:15 -0600370 memcpy(p, req->r_oid, req->r_oid_len);
Sage Weil1b83bef2013-02-25 16:11:12 -0800371 dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
Alex Elderaf77f262012-11-09 08:43:15 -0600372 p += req->r_oid_len;
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700373
Sage Weil1b83bef2013-02-25 16:11:12 -0800374 /* ops */
375 ceph_encode_16(&p, num_ops);
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700376 src_op = src_ops;
Sage Weil1b83bef2013-02-25 16:11:12 -0800377 req->r_request_ops = p;
378 for (i = 0; i < num_ops; i++, src_op++) {
379 osd_req_encode_op(req, p, src_op);
380 p += sizeof(struct ceph_osd_op);
381 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700382
Sage Weil1b83bef2013-02-25 16:11:12 -0800383 /* snaps */
384 ceph_encode_64(&p, req->r_snapid);
385 ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
386 ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
387 if (req->r_snapc) {
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700388 for (i = 0; i < snapc->num_snaps; i++) {
Sage Weil1b83bef2013-02-25 16:11:12 -0800389 ceph_encode_64(&p, req->r_snapc->snaps[i]);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700390 }
391 }
392
Sage Weil1b83bef2013-02-25 16:11:12 -0800393 req->r_request_attempts = p;
394 p += 4;
395
Alex Elderf44246e2013-02-14 12:16:43 -0600396 data_len = req->r_trail.length;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700397 if (flags & CEPH_OSD_FLAG_WRITE) {
398 req->r_request->hdr.data_off = cpu_to_le16(off);
Alex Elderf44246e2013-02-14 12:16:43 -0600399 data_len += len;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700400 }
Alex Elderf44246e2013-02-14 12:16:43 -0600401 req->r_request->hdr.data_len = cpu_to_le32(data_len);
Sage Weilc5c6b192010-11-09 12:40:00 -0800402
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700403 BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
404 msg_size = p - msg->front.iov_base;
405 msg->front.iov_len = msg_size;
406 msg->hdr.front_len = cpu_to_le32(msg_size);
Sage Weil1b83bef2013-02-25 16:11:12 -0800407
408 dout("build_request msg_size was %d num_ops %d\n", (int)msg_size,
409 num_ops);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700410 return;
411}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700412EXPORT_SYMBOL(ceph_osdc_build_request);
Yehuda Sadeh3499e8a2010-04-06 14:51:47 -0700413
Sage Weilf24e9982009-10-06 11:31:10 -0700414/*
415 * build new request AND message, calculate layout, and adjust file
416 * extent as needed.
417 *
418 * if the file was recently truncated, we include information about its
419 * old and new size so that the object can be updated appropriately. (we
420 * avoid synchronously deleting truncated objects because it's slow.)
421 *
422 * if @do_sync, include a 'startsync' command so that the osd will flush
423 * data quickly.
424 */
425struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
426 struct ceph_file_layout *layout,
427 struct ceph_vino vino,
428 u64 off, u64 *plen,
429 int opcode, int flags,
430 struct ceph_snap_context *snapc,
431 int do_sync,
432 u32 truncate_seq,
433 u64 truncate_size,
434 struct timespec *mtime,
Alex Eldera3bea472013-02-15 11:42:29 -0600435 bool use_mempool,
Sage Weilb7495fc2010-11-09 12:43:12 -0800436 int page_align)
Sage Weilf24e9982009-10-06 11:31:10 -0700437{
Alex Elderae7ca4a32012-11-13 21:11:15 -0600438 struct ceph_osd_req_op ops[2];
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700439 struct ceph_osd_request *req;
Alex Elderae7ca4a32012-11-13 21:11:15 -0600440 unsigned int num_op = 1;
Alex Elder47a05812013-02-15 22:10:17 -0600441 u64 bno = 0;
Sage Weil68162822012-09-24 21:01:02 -0700442 int r;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700443
Alex Elderae7ca4a32012-11-13 21:11:15 -0600444 memset(&ops, 0, sizeof ops);
445
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700446 ops[0].op = opcode;
447 ops[0].extent.truncate_seq = truncate_seq;
448 ops[0].extent.truncate_size = truncate_size;
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700449
450 if (do_sync) {
451 ops[1].op = CEPH_OSD_OP_STARTSYNC;
Alex Elderae7ca4a32012-11-13 21:11:15 -0600452 num_op++;
453 }
Yehuda Sadeh68b44762010-04-06 15:01:27 -0700454
Alex Elderae7ca4a32012-11-13 21:11:15 -0600455 req = ceph_osdc_alloc_request(osdc, snapc, num_op, use_mempool,
456 GFP_NOFS);
Sage Weil4ad12622011-05-03 09:23:36 -0700457 if (!req)
Sage Weil68162822012-09-24 21:01:02 -0700458 return ERR_PTR(-ENOMEM);
Alex Elderd178a9e2012-11-13 21:11:15 -0600459 req->r_flags = flags;
Sage Weilf24e9982009-10-06 11:31:10 -0700460
461 /* calculate max write size */
Alex Elder60cf5992013-02-15 22:10:17 -0600462 r = calc_layout(layout, off, plen, ops, &bno);
Alex Elder3ff5f382013-02-15 22:10:17 -0600463 if (r < 0) {
464 ceph_osdc_put_request(req);
Sage Weil68162822012-09-24 21:01:02 -0700465 return ERR_PTR(r);
Alex Elder3ff5f382013-02-15 22:10:17 -0600466 }
Alex Elder47a05812013-02-15 22:10:17 -0600467
Sage Weilf24e9982009-10-06 11:31:10 -0700468 req->r_file_layout = *layout; /* keep a copy */
469
Alex Elderdbe0fc42013-02-15 22:10:17 -0600470 snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx", vino.ino, bno);
471 req->r_oid_len = strlen(req->r_oid);
472
Alex Elder60cf5992013-02-15 22:10:17 -0600473 /* The alignment may differ from the natural (file) alignment */
474
Sage Weil9bb0ce22011-06-13 16:20:18 -0700475 req->r_num_pages = calc_pages_for(page_align, *plen);
Sage Weilb7495fc2010-11-09 12:43:12 -0800476 req->r_page_alignment = page_align;
477
Alex Elderae7ca4a32012-11-13 21:11:15 -0600478 ceph_osdc_build_request(req, off, *plen, num_op, ops,
479 snapc, vino.snap, mtime);
Sage Weilf24e9982009-10-06 11:31:10 -0700480
Sage Weilf24e9982009-10-06 11:31:10 -0700481 return req;
482}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700483EXPORT_SYMBOL(ceph_osdc_new_request);
Sage Weilf24e9982009-10-06 11:31:10 -0700484
485/*
486 * We keep osd requests in an rbtree, sorted by ->r_tid.
487 */
488static void __insert_request(struct ceph_osd_client *osdc,
489 struct ceph_osd_request *new)
490{
491 struct rb_node **p = &osdc->requests.rb_node;
492 struct rb_node *parent = NULL;
493 struct ceph_osd_request *req = NULL;
494
495 while (*p) {
496 parent = *p;
497 req = rb_entry(parent, struct ceph_osd_request, r_node);
498 if (new->r_tid < req->r_tid)
499 p = &(*p)->rb_left;
500 else if (new->r_tid > req->r_tid)
501 p = &(*p)->rb_right;
502 else
503 BUG();
504 }
505
506 rb_link_node(&new->r_node, parent, p);
507 rb_insert_color(&new->r_node, &osdc->requests);
508}
509
510static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
511 u64 tid)
512{
513 struct ceph_osd_request *req;
514 struct rb_node *n = osdc->requests.rb_node;
515
516 while (n) {
517 req = rb_entry(n, struct ceph_osd_request, r_node);
518 if (tid < req->r_tid)
519 n = n->rb_left;
520 else if (tid > req->r_tid)
521 n = n->rb_right;
522 else
523 return req;
524 }
525 return NULL;
526}
527
528static struct ceph_osd_request *
529__lookup_request_ge(struct ceph_osd_client *osdc,
530 u64 tid)
531{
532 struct ceph_osd_request *req;
533 struct rb_node *n = osdc->requests.rb_node;
534
535 while (n) {
536 req = rb_entry(n, struct ceph_osd_request, r_node);
537 if (tid < req->r_tid) {
538 if (!n->rb_left)
539 return req;
540 n = n->rb_left;
541 } else if (tid > req->r_tid) {
542 n = n->rb_right;
543 } else {
544 return req;
545 }
546 }
547 return NULL;
548}
549
Sage Weil6f6c7002011-01-17 20:34:08 -0800550/*
551 * Resubmit requests pending on the given osd.
552 */
553static void __kick_osd_requests(struct ceph_osd_client *osdc,
554 struct ceph_osd *osd)
555{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700556 struct ceph_osd_request *req, *nreq;
Sage Weil6f6c7002011-01-17 20:34:08 -0800557 int err;
558
559 dout("__kick_osd_requests osd%d\n", osd->o_osd);
560 err = __reset_osd(osdc, osd);
Alex Elder685a7552012-12-07 09:57:58 -0600561 if (err)
Sage Weil6f6c7002011-01-17 20:34:08 -0800562 return;
563
564 list_for_each_entry(req, &osd->o_requests, r_osd_item) {
565 list_move(&req->r_req_lru_item, &osdc->req_unsent);
566 dout("requeued %p tid %llu osd%d\n", req, req->r_tid,
567 osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700568 if (!req->r_linger)
569 req->r_flags |= CEPH_OSD_FLAG_RETRY;
570 }
571
572 list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
573 r_linger_osd) {
Sage Weil77f38e02011-04-06 09:09:16 -0700574 /*
575 * reregister request prior to unregistering linger so
576 * that r_osd is preserved.
577 */
578 BUG_ON(!list_empty(&req->r_req_lru_item));
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700579 __register_request(osdc, req);
Sage Weil77f38e02011-04-06 09:09:16 -0700580 list_add(&req->r_req_lru_item, &osdc->req_unsent);
581 list_add(&req->r_osd_item, &req->r_osd->o_requests);
582 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700583 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
584 osd->o_osd);
Sage Weil6f6c7002011-01-17 20:34:08 -0800585 }
586}
587
Sage Weilf24e9982009-10-06 11:31:10 -0700588/*
Sage Weil81b024e2009-10-09 10:29:18 -0700589 * If the osd connection drops, we need to resubmit all requests.
Sage Weilf24e9982009-10-06 11:31:10 -0700590 */
591static void osd_reset(struct ceph_connection *con)
592{
593 struct ceph_osd *osd = con->private;
594 struct ceph_osd_client *osdc;
595
596 if (!osd)
597 return;
598 dout("osd_reset osd%d\n", osd->o_osd);
599 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -0700600 down_read(&osdc->map_sem);
Sage Weil83aff952012-11-28 12:28:24 -0800601 mutex_lock(&osdc->request_mutex);
602 __kick_osd_requests(osdc, osd);
Alex Elderf9d25192013-02-15 11:42:29 -0600603 __send_queued(osdc);
Sage Weil83aff952012-11-28 12:28:24 -0800604 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -0700605 up_read(&osdc->map_sem);
606}
607
608/*
609 * Track open sessions with osds.
610 */
Alex Eldere10006f2012-05-26 23:26:43 -0500611static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
Sage Weilf24e9982009-10-06 11:31:10 -0700612{
613 struct ceph_osd *osd;
614
615 osd = kzalloc(sizeof(*osd), GFP_NOFS);
616 if (!osd)
617 return NULL;
618
619 atomic_set(&osd->o_ref, 1);
620 osd->o_osdc = osdc;
Alex Eldere10006f2012-05-26 23:26:43 -0500621 osd->o_osd = onum;
Alex Elderf4077312012-12-06 07:22:04 -0600622 RB_CLEAR_NODE(&osd->o_node);
Sage Weilf24e9982009-10-06 11:31:10 -0700623 INIT_LIST_HEAD(&osd->o_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700624 INIT_LIST_HEAD(&osd->o_linger_requests);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800625 INIT_LIST_HEAD(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700626 osd->o_incarnation = 1;
627
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700628 ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800629
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800630 INIT_LIST_HEAD(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -0700631 return osd;
632}
633
634static struct ceph_osd *get_osd(struct ceph_osd *osd)
635{
636 if (atomic_inc_not_zero(&osd->o_ref)) {
637 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
638 atomic_read(&osd->o_ref));
639 return osd;
640 } else {
641 dout("get_osd %p FAIL\n", osd);
642 return NULL;
643 }
644}
645
646static void put_osd(struct ceph_osd *osd)
647{
648 dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
649 atomic_read(&osd->o_ref) - 1);
Alex Eldera2556512012-05-16 15:16:39 -0500650 if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
Sage Weil79494d12010-05-27 14:15:49 -0700651 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
652
Alex Eldera2556512012-05-16 15:16:39 -0500653 if (ac->ops && ac->ops->destroy_authorizer)
Alex Elder6c4a1912012-05-16 15:16:38 -0500654 ac->ops->destroy_authorizer(ac, osd->o_auth.authorizer);
Sage Weilf24e9982009-10-06 11:31:10 -0700655 kfree(osd);
Sage Weil79494d12010-05-27 14:15:49 -0700656 }
Sage Weilf24e9982009-10-06 11:31:10 -0700657}
658
659/*
660 * remove an osd from our map
661 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800662static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -0700663{
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800664 dout("__remove_osd %p\n", osd);
Sage Weilf24e9982009-10-06 11:31:10 -0700665 BUG_ON(!list_empty(&osd->o_requests));
666 rb_erase(&osd->o_node, &osdc->osds);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800667 list_del_init(&osd->o_osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -0700668 ceph_con_close(&osd->o_con);
669 put_osd(osd);
670}
671
Sage Weilaca420b2011-08-31 14:45:53 -0700672static void remove_all_osds(struct ceph_osd_client *osdc)
673{
Jiaju Zhang048a9d22012-07-20 08:18:36 -0500674 dout("%s %p\n", __func__, osdc);
Sage Weilaca420b2011-08-31 14:45:53 -0700675 mutex_lock(&osdc->request_mutex);
676 while (!RB_EMPTY_ROOT(&osdc->osds)) {
677 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
678 struct ceph_osd, o_node);
679 __remove_osd(osdc, osd);
680 }
681 mutex_unlock(&osdc->request_mutex);
682}
683
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800684static void __move_osd_to_lru(struct ceph_osd_client *osdc,
685 struct ceph_osd *osd)
686{
687 dout("__move_osd_to_lru %p\n", osd);
688 BUG_ON(!list_empty(&osd->o_osd_lru));
689 list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700690 osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800691}
692
693static void __remove_osd_from_lru(struct ceph_osd *osd)
694{
695 dout("__remove_osd_from_lru %p\n", osd);
696 if (!list_empty(&osd->o_osd_lru))
697 list_del_init(&osd->o_osd_lru);
698}
699
Sage Weilaca420b2011-08-31 14:45:53 -0700700static void remove_old_osds(struct ceph_osd_client *osdc)
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800701{
702 struct ceph_osd *osd, *nosd;
703
704 dout("__remove_old_osds %p\n", osdc);
705 mutex_lock(&osdc->request_mutex);
706 list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
Sage Weilaca420b2011-08-31 14:45:53 -0700707 if (time_before(jiffies, osd->lru_ttl))
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800708 break;
709 __remove_osd(osdc, osd);
710 }
711 mutex_unlock(&osdc->request_mutex);
712}
713
Sage Weilf24e9982009-10-06 11:31:10 -0700714/*
715 * reset osd connect
716 */
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800717static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
Sage Weilf24e9982009-10-06 11:31:10 -0700718{
Alex Elderc3acb182012-12-07 09:57:58 -0600719 struct ceph_entity_addr *peer_addr;
Sage Weilf24e9982009-10-06 11:31:10 -0700720
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800721 dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700722 if (list_empty(&osd->o_requests) &&
723 list_empty(&osd->o_linger_requests)) {
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800724 __remove_osd(osdc, osd);
Alex Elderc3acb182012-12-07 09:57:58 -0600725
726 return -ENODEV;
727 }
728
729 peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
730 if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
731 !ceph_con_opened(&osd->o_con)) {
732 struct ceph_osd_request *req;
733
Sage Weil87b315a2010-03-22 14:51:18 -0700734 dout(" osd addr hasn't changed and connection never opened,"
735 " letting msgr retry");
736 /* touch each r_stamp for handle_timeout()'s benfit */
737 list_for_each_entry(req, &osd->o_requests, r_osd_item)
738 req->r_stamp = jiffies;
Alex Elderc3acb182012-12-07 09:57:58 -0600739
740 return -EAGAIN;
Sage Weilf24e9982009-10-06 11:31:10 -0700741 }
Alex Elderc3acb182012-12-07 09:57:58 -0600742
743 ceph_con_close(&osd->o_con);
744 ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
745 osd->o_incarnation++;
746
747 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -0700748}
749
750static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
751{
752 struct rb_node **p = &osdc->osds.rb_node;
753 struct rb_node *parent = NULL;
754 struct ceph_osd *osd = NULL;
755
Sage Weilaca420b2011-08-31 14:45:53 -0700756 dout("__insert_osd %p osd%d\n", new, new->o_osd);
Sage Weilf24e9982009-10-06 11:31:10 -0700757 while (*p) {
758 parent = *p;
759 osd = rb_entry(parent, struct ceph_osd, o_node);
760 if (new->o_osd < osd->o_osd)
761 p = &(*p)->rb_left;
762 else if (new->o_osd > osd->o_osd)
763 p = &(*p)->rb_right;
764 else
765 BUG();
766 }
767
768 rb_link_node(&new->o_node, parent, p);
769 rb_insert_color(&new->o_node, &osdc->osds);
770}
771
772static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
773{
774 struct ceph_osd *osd;
775 struct rb_node *n = osdc->osds.rb_node;
776
777 while (n) {
778 osd = rb_entry(n, struct ceph_osd, o_node);
779 if (o < osd->o_osd)
780 n = n->rb_left;
781 else if (o > osd->o_osd)
782 n = n->rb_right;
783 else
784 return osd;
785 }
786 return NULL;
787}
788
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800789static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
790{
791 schedule_delayed_work(&osdc->timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700792 osdc->client->options->osd_keepalive_timeout * HZ);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800793}
794
795static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
796{
797 cancel_delayed_work(&osdc->timeout_work);
798}
Sage Weilf24e9982009-10-06 11:31:10 -0700799
800/*
801 * Register request, assign tid. If this is the first request, set up
802 * the timeout event.
803 */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700804static void __register_request(struct ceph_osd_client *osdc,
805 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -0700806{
Sage Weilf24e9982009-10-06 11:31:10 -0700807 req->r_tid = ++osdc->last_tid;
Sage Weil6df058c2009-12-22 11:24:33 -0800808 req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
Sage Weil77f38e02011-04-06 09:09:16 -0700809 dout("__register_request %p tid %lld\n", req, req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -0700810 __insert_request(osdc, req);
811 ceph_osdc_get_request(req);
812 osdc->num_requests++;
Sage Weilf24e9982009-10-06 11:31:10 -0700813 if (osdc->num_requests == 1) {
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800814 dout(" first request, scheduling timeout\n");
815 __schedule_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -0700816 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700817}
818
819static void register_request(struct ceph_osd_client *osdc,
820 struct ceph_osd_request *req)
821{
822 mutex_lock(&osdc->request_mutex);
823 __register_request(osdc, req);
Sage Weilf24e9982009-10-06 11:31:10 -0700824 mutex_unlock(&osdc->request_mutex);
825}
826
827/*
828 * called under osdc->request_mutex
829 */
830static void __unregister_request(struct ceph_osd_client *osdc,
831 struct ceph_osd_request *req)
832{
Sage Weil35f9f8a2012-05-16 15:16:38 -0500833 if (RB_EMPTY_NODE(&req->r_node)) {
834 dout("__unregister_request %p tid %lld not registered\n",
835 req, req->r_tid);
836 return;
837 }
838
Sage Weilf24e9982009-10-06 11:31:10 -0700839 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
840 rb_erase(&req->r_node, &osdc->requests);
841 osdc->num_requests--;
842
Sage Weil0ba64782009-10-08 16:57:16 -0700843 if (req->r_osd) {
844 /* make sure the original request isn't in flight. */
Alex Elder6740a842012-06-01 14:56:43 -0500845 ceph_msg_revoke(req->r_request);
Sage Weil0ba64782009-10-08 16:57:16 -0700846
847 list_del_init(&req->r_osd_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700848 if (list_empty(&req->r_osd->o_requests) &&
849 list_empty(&req->r_osd->o_linger_requests)) {
850 dout("moving osd to %p lru\n", req->r_osd);
Yehuda Sadehf5a20412010-02-03 11:00:26 -0800851 __move_osd_to_lru(osdc, req->r_osd);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700852 }
Sage Weilfbdb9192011-03-29 12:11:06 -0700853 if (list_empty(&req->r_linger_item))
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700854 req->r_osd = NULL;
Sage Weil0ba64782009-10-08 16:57:16 -0700855 }
Sage Weilf24e9982009-10-06 11:31:10 -0700856
Alex Elder7d5f2482012-11-29 08:37:03 -0600857 list_del_init(&req->r_req_lru_item);
Sage Weilf24e9982009-10-06 11:31:10 -0700858 ceph_osdc_put_request(req);
859
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -0800860 if (osdc->num_requests == 0) {
861 dout(" no requests, canceling timeout\n");
862 __cancel_osd_timeout(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -0700863 }
864}
865
866/*
867 * Cancel a previously queued request message
868 */
869static void __cancel_request(struct ceph_osd_request *req)
870{
Sage Weil6bc18872010-09-27 10:18:52 -0700871 if (req->r_sent && req->r_osd) {
Alex Elder6740a842012-06-01 14:56:43 -0500872 ceph_msg_revoke(req->r_request);
Sage Weilf24e9982009-10-06 11:31:10 -0700873 req->r_sent = 0;
874 }
875}
876
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700877static void __register_linger_request(struct ceph_osd_client *osdc,
878 struct ceph_osd_request *req)
879{
880 dout("__register_linger_request %p\n", req);
881 list_add_tail(&req->r_linger_item, &osdc->req_linger);
Sage Weil6194ea82012-07-30 16:19:28 -0700882 if (req->r_osd)
883 list_add_tail(&req->r_linger_osd,
884 &req->r_osd->o_linger_requests);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700885}
886
887static void __unregister_linger_request(struct ceph_osd_client *osdc,
888 struct ceph_osd_request *req)
889{
890 dout("__unregister_linger_request %p\n", req);
Alex Elder61c74032012-12-06 09:37:23 -0600891 list_del_init(&req->r_linger_item);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700892 if (req->r_osd) {
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700893 list_del_init(&req->r_linger_osd);
894
895 if (list_empty(&req->r_osd->o_requests) &&
896 list_empty(&req->r_osd->o_linger_requests)) {
897 dout("moving osd to %p lru\n", req->r_osd);
898 __move_osd_to_lru(osdc, req->r_osd);
899 }
Sage Weilfbdb9192011-03-29 12:11:06 -0700900 if (list_empty(&req->r_osd_item))
901 req->r_osd = NULL;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -0700902 }
903}
904
905void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
906 struct ceph_osd_request *req)
907{
908 mutex_lock(&osdc->request_mutex);
909 if (req->r_linger) {
910 __unregister_linger_request(osdc, req);
911 ceph_osdc_put_request(req);
912 }
913 mutex_unlock(&osdc->request_mutex);
914}
915EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
916
917void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
918 struct ceph_osd_request *req)
919{
920 if (!req->r_linger) {
921 dout("set_request_linger %p\n", req);
922 req->r_linger = 1;
923 /*
924 * caller is now responsible for calling
925 * unregister_linger_request
926 */
927 ceph_osdc_get_request(req);
928 }
929}
930EXPORT_SYMBOL(ceph_osdc_set_request_linger);
931
Sage Weilf24e9982009-10-06 11:31:10 -0700932/*
933 * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
934 * (as needed), and set the request r_osd appropriately. If there is
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300935 * no up osd, set r_osd to NULL. Move the request to the appropriate list
Sage Weil6f6c7002011-01-17 20:34:08 -0800936 * (unsent, homeless) or leave on in-flight lru.
Sage Weilf24e9982009-10-06 11:31:10 -0700937 *
938 * Return 0 if unchanged, 1 if changed, or negative on error.
939 *
940 * Caller should hold map_sem for read and request_mutex.
941 */
Sage Weil6f6c7002011-01-17 20:34:08 -0800942static int __map_request(struct ceph_osd_client *osdc,
Sage Weil38d64532011-10-14 13:33:55 -0700943 struct ceph_osd_request *req, int force_resend)
Sage Weilf24e9982009-10-06 11:31:10 -0700944{
Sage Weil5b191d92013-02-23 10:38:16 -0800945 struct ceph_pg pgid;
Sage Weild85b7052010-05-10 10:24:48 -0700946 int acting[CEPH_PG_MAX_SIZE];
947 int o = -1, num = 0;
Sage Weilf24e9982009-10-06 11:31:10 -0700948 int err;
Sage Weilf24e9982009-10-06 11:31:10 -0700949
Sage Weil6f6c7002011-01-17 20:34:08 -0800950 dout("map_request %p tid %lld\n", req, req->r_tid);
Sage Weil2169aea2013-02-25 16:13:08 -0800951 err = ceph_calc_object_layout(&pgid, req->r_oid,
Sage Weilf24e9982009-10-06 11:31:10 -0700952 &req->r_file_layout, osdc->osdmap);
Sage Weil6f6c7002011-01-17 20:34:08 -0800953 if (err) {
954 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilf24e9982009-10-06 11:31:10 -0700955 return err;
Sage Weil6f6c7002011-01-17 20:34:08 -0800956 }
Sage Weil7740a422010-01-08 15:58:25 -0800957 req->r_pgid = pgid;
958
Sage Weild85b7052010-05-10 10:24:48 -0700959 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
960 if (err > 0) {
961 o = acting[0];
962 num = err;
963 }
Sage Weilf24e9982009-10-06 11:31:10 -0700964
Sage Weil38d64532011-10-14 13:33:55 -0700965 if ((!force_resend &&
966 req->r_osd && req->r_osd->o_osd == o &&
Sage Weild85b7052010-05-10 10:24:48 -0700967 req->r_sent >= req->r_osd->o_incarnation &&
968 req->r_num_pg_osds == num &&
969 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
Sage Weilf24e9982009-10-06 11:31:10 -0700970 (req->r_osd == NULL && o == -1))
971 return 0; /* no change */
972
Sage Weil5b191d92013-02-23 10:38:16 -0800973 dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
974 req->r_tid, pgid.pool, pgid.seed, o,
Sage Weilf24e9982009-10-06 11:31:10 -0700975 req->r_osd ? req->r_osd->o_osd : -1);
976
Sage Weild85b7052010-05-10 10:24:48 -0700977 /* record full pg acting set */
978 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
979 req->r_num_pg_osds = num;
980
Sage Weilf24e9982009-10-06 11:31:10 -0700981 if (req->r_osd) {
982 __cancel_request(req);
983 list_del_init(&req->r_osd_item);
Sage Weilf24e9982009-10-06 11:31:10 -0700984 req->r_osd = NULL;
985 }
986
987 req->r_osd = __lookup_osd(osdc, o);
988 if (!req->r_osd && o >= 0) {
Sage Weilc99eb1c2010-02-26 09:37:33 -0800989 err = -ENOMEM;
Alex Eldere10006f2012-05-26 23:26:43 -0500990 req->r_osd = create_osd(osdc, o);
Sage Weil6f6c7002011-01-17 20:34:08 -0800991 if (!req->r_osd) {
992 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Sage Weilc99eb1c2010-02-26 09:37:33 -0800993 goto out;
Sage Weil6f6c7002011-01-17 20:34:08 -0800994 }
Sage Weilf24e9982009-10-06 11:31:10 -0700995
Sage Weil6f6c7002011-01-17 20:34:08 -0800996 dout("map_request osd %p is osd%d\n", req->r_osd, o);
Sage Weilf24e9982009-10-06 11:31:10 -0700997 __insert_osd(osdc, req->r_osd);
998
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700999 ceph_con_open(&req->r_osd->o_con,
1000 CEPH_ENTITY_TYPE_OSD, o,
1001 &osdc->osdmap->osd_addr[o]);
Sage Weilf24e9982009-10-06 11:31:10 -07001002 }
1003
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001004 if (req->r_osd) {
1005 __remove_osd_from_lru(req->r_osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001006 list_add(&req->r_osd_item, &req->r_osd->o_requests);
Sage Weil6f6c7002011-01-17 20:34:08 -08001007 list_move(&req->r_req_lru_item, &osdc->req_unsent);
1008 } else {
1009 list_move(&req->r_req_lru_item, &osdc->req_notarget);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001010 }
Sage Weild85b7052010-05-10 10:24:48 -07001011 err = 1; /* osd or pg changed */
Sage Weilf24e9982009-10-06 11:31:10 -07001012
1013out:
Sage Weilf24e9982009-10-06 11:31:10 -07001014 return err;
1015}
1016
1017/*
1018 * caller should hold map_sem (for read) and request_mutex
1019 */
Sage Weil56e925b2012-01-03 12:34:34 -08001020static void __send_request(struct ceph_osd_client *osdc,
1021 struct ceph_osd_request *req)
Sage Weilf24e9982009-10-06 11:31:10 -07001022{
Sage Weil1b83bef2013-02-25 16:11:12 -08001023 void *p;
Sage Weilf24e9982009-10-06 11:31:10 -07001024
Sage Weil1b83bef2013-02-25 16:11:12 -08001025 dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1026 req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1027 (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
Sage Weilf24e9982009-10-06 11:31:10 -07001028
Sage Weil1b83bef2013-02-25 16:11:12 -08001029 /* fill in message content that changes each time we send it */
1030 put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1031 put_unaligned_le32(req->r_flags, req->r_request_flags);
1032 put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1033 p = req->r_request_pgid;
1034 ceph_encode_64(&p, req->r_pgid.pool);
1035 ceph_encode_32(&p, req->r_pgid.seed);
1036 put_unaligned_le64(1, req->r_request_attempts); /* FIXME */
1037 memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1038 sizeof(req->r_reassert_version));
Sage Weil2169aea2013-02-25 16:13:08 -08001039
Sage Weil3dd72fc2010-03-22 14:42:30 -07001040 req->r_stamp = jiffies;
Henry C Chang07a27e22010-08-22 21:34:27 -07001041 list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001042
1043 ceph_msg_get(req->r_request); /* send consumes a ref */
1044 ceph_con_send(&req->r_osd->o_con, req->r_request);
1045 req->r_sent = req->r_osd->o_incarnation;
Sage Weilf24e9982009-10-06 11:31:10 -07001046}
1047
1048/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001049 * Send any requests in the queue (req_unsent).
1050 */
Alex Elderf9d25192013-02-15 11:42:29 -06001051static void __send_queued(struct ceph_osd_client *osdc)
Sage Weil6f6c7002011-01-17 20:34:08 -08001052{
1053 struct ceph_osd_request *req, *tmp;
1054
Alex Elderf9d25192013-02-15 11:42:29 -06001055 dout("__send_queued\n");
1056 list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
Sage Weil6f6c7002011-01-17 20:34:08 -08001057 __send_request(osdc, req);
Sage Weil6f6c7002011-01-17 20:34:08 -08001058}
1059
1060/*
Sage Weilf24e9982009-10-06 11:31:10 -07001061 * Timeout callback, called every N seconds when 1 or more osd
1062 * requests has been active for more than N seconds. When this
1063 * happens, we ping all OSDs with requests who have timed out to
1064 * ensure any communications channel reset is detected. Reset the
1065 * request timeouts another N seconds in the future as we go.
1066 * Reschedule the timeout event another N seconds in future (unless
1067 * there are no open requests).
1068 */
1069static void handle_timeout(struct work_struct *work)
1070{
1071 struct ceph_osd_client *osdc =
1072 container_of(work, struct ceph_osd_client, timeout_work.work);
Sage Weil83aff952012-11-28 12:28:24 -08001073 struct ceph_osd_request *req;
Sage Weilf24e9982009-10-06 11:31:10 -07001074 struct ceph_osd *osd;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001075 unsigned long keepalive =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001076 osdc->client->options->osd_keepalive_timeout * HZ;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001077 struct list_head slow_osds;
Sage Weilf24e9982009-10-06 11:31:10 -07001078 dout("timeout\n");
1079 down_read(&osdc->map_sem);
1080
1081 ceph_monc_request_next_osdmap(&osdc->client->monc);
1082
1083 mutex_lock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001084
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001085 /*
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001086 * ping osds that are a bit slow. this ensures that if there
1087 * is a break in the TCP connection we will notice, and reopen
1088 * a connection with that osd (from the fault callback).
1089 */
1090 INIT_LIST_HEAD(&slow_osds);
1091 list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
Sage Weil3dd72fc2010-03-22 14:42:30 -07001092 if (time_before(jiffies, req->r_stamp + keepalive))
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001093 break;
1094
1095 osd = req->r_osd;
1096 BUG_ON(!osd);
1097 dout(" tid %llu is slow, will send keepalive on osd%d\n",
Sage Weilf24e9982009-10-06 11:31:10 -07001098 req->r_tid, osd->o_osd);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001099 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1100 }
1101 while (!list_empty(&slow_osds)) {
1102 osd = list_entry(slow_osds.next, struct ceph_osd,
1103 o_keepalive_item);
1104 list_del_init(&osd->o_keepalive_item);
Sage Weilf24e9982009-10-06 11:31:10 -07001105 ceph_con_keepalive(&osd->o_con);
1106 }
1107
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001108 __schedule_osd_timeout(osdc);
Alex Elderf9d25192013-02-15 11:42:29 -06001109 __send_queued(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001110 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001111 up_read(&osdc->map_sem);
1112}
1113
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001114static void handle_osds_timeout(struct work_struct *work)
1115{
1116 struct ceph_osd_client *osdc =
1117 container_of(work, struct ceph_osd_client,
1118 osds_timeout_work.work);
1119 unsigned long delay =
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001120 osdc->client->options->osd_idle_ttl * HZ >> 2;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001121
1122 dout("osds timeout\n");
1123 down_read(&osdc->map_sem);
Sage Weilaca420b2011-08-31 14:45:53 -07001124 remove_old_osds(osdc);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001125 up_read(&osdc->map_sem);
1126
1127 schedule_delayed_work(&osdc->osds_timeout_work,
1128 round_jiffies_relative(delay));
1129}
1130
Sage Weil25845472011-06-03 09:37:09 -07001131static void complete_request(struct ceph_osd_request *req)
1132{
1133 if (req->r_safe_callback)
1134 req->r_safe_callback(req, NULL);
1135 complete_all(&req->r_safe_completion); /* fsync waiter */
1136}
1137
Sage Weil1b83bef2013-02-25 16:11:12 -08001138static int __decode_pgid(void **p, void *end, struct ceph_pg *pgid)
1139{
1140 __u8 v;
1141
1142 ceph_decode_need(p, end, 1 + 8 + 4 + 4, bad);
1143 v = ceph_decode_8(p);
1144 if (v > 1) {
1145 pr_warning("do not understand pg encoding %d > 1", v);
1146 return -EINVAL;
1147 }
1148 pgid->pool = ceph_decode_64(p);
1149 pgid->seed = ceph_decode_32(p);
1150 *p += 4;
1151 return 0;
1152
1153bad:
1154 pr_warning("incomplete pg encoding");
1155 return -EINVAL;
1156}
1157
Sage Weilf24e9982009-10-06 11:31:10 -07001158/*
1159 * handle osd op reply. either call the callback if it is specified,
1160 * or do the completion to wake up the waiting thread.
1161 */
Sage Weil350b1c32009-12-22 10:45:45 -08001162static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1163 struct ceph_connection *con)
Sage Weilf24e9982009-10-06 11:31:10 -07001164{
Sage Weil1b83bef2013-02-25 16:11:12 -08001165 void *p, *end;
Sage Weilf24e9982009-10-06 11:31:10 -07001166 struct ceph_osd_request *req;
1167 u64 tid;
Sage Weil1b83bef2013-02-25 16:11:12 -08001168 int object_len;
1169 int numops, payload_len, flags;
Sage Weil0ceed5d2010-05-11 09:53:18 -07001170 s32 result;
Sage Weil1b83bef2013-02-25 16:11:12 -08001171 s32 retry_attempt;
1172 struct ceph_pg pg;
1173 int err;
1174 u32 reassert_epoch;
1175 u64 reassert_version;
1176 u32 osdmap_epoch;
Alex Elder0d5af162013-02-27 10:26:25 -06001177 int already_completed;
Sage Weil1b83bef2013-02-25 16:11:12 -08001178 int i;
Sage Weilf24e9982009-10-06 11:31:10 -07001179
Sage Weil6df058c2009-12-22 11:24:33 -08001180 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil1b83bef2013-02-25 16:11:12 -08001181 dout("handle_reply %p tid %llu\n", msg, tid);
1182
1183 p = msg->front.iov_base;
1184 end = p + msg->front.iov_len;
1185
1186 ceph_decode_need(&p, end, 4, bad);
1187 object_len = ceph_decode_32(&p);
1188 ceph_decode_need(&p, end, object_len, bad);
1189 p += object_len;
1190
1191 err = __decode_pgid(&p, end, &pg);
1192 if (err)
Sage Weilf24e9982009-10-06 11:31:10 -07001193 goto bad;
Sage Weil1b83bef2013-02-25 16:11:12 -08001194
1195 ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1196 flags = ceph_decode_64(&p);
1197 result = ceph_decode_32(&p);
1198 reassert_epoch = ceph_decode_32(&p);
1199 reassert_version = ceph_decode_64(&p);
1200 osdmap_epoch = ceph_decode_32(&p);
1201
Sage Weilf24e9982009-10-06 11:31:10 -07001202 /* lookup */
1203 mutex_lock(&osdc->request_mutex);
1204 req = __lookup_request(osdc, tid);
1205 if (req == NULL) {
1206 dout("handle_reply tid %llu dne\n", tid);
1207 mutex_unlock(&osdc->request_mutex);
1208 return;
1209 }
1210 ceph_osdc_get_request(req);
Sage Weil1b83bef2013-02-25 16:11:12 -08001211
1212 dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1213 req, result);
1214
1215 ceph_decode_need(&p, end, 4, bad);
1216 numops = ceph_decode_32(&p);
1217 if (numops > CEPH_OSD_MAX_OP)
1218 goto bad_put;
1219 if (numops != req->r_num_ops)
1220 goto bad_put;
1221 payload_len = 0;
1222 ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad);
1223 for (i = 0; i < numops; i++) {
1224 struct ceph_osd_op *op = p;
1225 int len;
1226
1227 len = le32_to_cpu(op->payload_len);
1228 req->r_reply_op_len[i] = len;
1229 dout(" op %d has %d bytes\n", i, len);
1230 payload_len += len;
1231 p += sizeof(*op);
1232 }
1233 if (payload_len != le32_to_cpu(msg->hdr.data_len)) {
1234 pr_warning("sum of op payload lens %d != data_len %d",
1235 payload_len, le32_to_cpu(msg->hdr.data_len));
1236 goto bad_put;
1237 }
1238
1239 ceph_decode_need(&p, end, 4 + numops * 4, bad);
1240 retry_attempt = ceph_decode_32(&p);
1241 for (i = 0; i < numops; i++)
1242 req->r_reply_op_result[i] = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001243
Sage Weil350b1c32009-12-22 10:45:45 -08001244 /*
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08001245 * if this connection filled our message, drop our reference now, to
Sage Weil350b1c32009-12-22 10:45:45 -08001246 * avoid a (safe but slower) revoke later.
1247 */
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08001248 if (req->r_con_filling_msg == con && req->r_reply == msg) {
Sage Weilc16e7862010-03-01 13:02:00 -08001249 dout(" dropping con_filling_msg ref %p\n", con);
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08001250 req->r_con_filling_msg = NULL;
Sage Weil0d477662012-05-31 20:22:18 -07001251 con->ops->put(con);
Sage Weil350b1c32009-12-22 10:45:45 -08001252 }
1253
Sage Weilf24e9982009-10-06 11:31:10 -07001254 if (!req->r_got_reply) {
Eric Dumazet95c96172012-04-15 05:58:06 +00001255 unsigned int bytes;
Sage Weilf24e9982009-10-06 11:31:10 -07001256
Sage Weil1b83bef2013-02-25 16:11:12 -08001257 req->r_result = result;
Sage Weilf24e9982009-10-06 11:31:10 -07001258 bytes = le32_to_cpu(msg->hdr.data_len);
1259 dout("handle_reply result %d bytes %d\n", req->r_result,
1260 bytes);
1261 if (req->r_result == 0)
1262 req->r_result = bytes;
1263
1264 /* in case this is a write and we need to replay, */
Sage Weil1b83bef2013-02-25 16:11:12 -08001265 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1266 req->r_reassert_version.version = cpu_to_le64(reassert_version);
Sage Weilf24e9982009-10-06 11:31:10 -07001267
1268 req->r_got_reply = 1;
1269 } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1270 dout("handle_reply tid %llu dup ack\n", tid);
Sage Weil34b43a52009-12-01 12:23:54 -08001271 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001272 goto done;
1273 }
1274
1275 dout("handle_reply tid %llu flags %d\n", tid, flags);
1276
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001277 if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1278 __register_linger_request(osdc, req);
1279
Sage Weilf24e9982009-10-06 11:31:10 -07001280 /* either this is a read, or we got the safe response */
Sage Weil0ceed5d2010-05-11 09:53:18 -07001281 if (result < 0 ||
1282 (flags & CEPH_OSD_FLAG_ONDISK) ||
Sage Weilf24e9982009-10-06 11:31:10 -07001283 ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1284 __unregister_request(osdc, req);
1285
Alex Elder0d5af162013-02-27 10:26:25 -06001286 already_completed = req->r_completed;
1287 req->r_completed = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07001288 mutex_unlock(&osdc->request_mutex);
Alex Elder0d5af162013-02-27 10:26:25 -06001289 if (already_completed)
1290 goto done;
Sage Weilf24e9982009-10-06 11:31:10 -07001291
1292 if (req->r_callback)
1293 req->r_callback(req, msg);
1294 else
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001295 complete_all(&req->r_completion);
Sage Weilf24e9982009-10-06 11:31:10 -07001296
Sage Weil25845472011-06-03 09:37:09 -07001297 if (flags & CEPH_OSD_FLAG_ONDISK)
1298 complete_request(req);
Sage Weilf24e9982009-10-06 11:31:10 -07001299
1300done:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001301 dout("req=%p req->r_linger=%d\n", req, req->r_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07001302 ceph_osdc_put_request(req);
1303 return;
1304
Sage Weil1b83bef2013-02-25 16:11:12 -08001305bad_put:
1306 ceph_osdc_put_request(req);
Sage Weilf24e9982009-10-06 11:31:10 -07001307bad:
Sage Weil1b83bef2013-02-25 16:11:12 -08001308 pr_err("corrupt osd_op_reply got %d %d\n",
1309 (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
Sage Weil9ec7cab2009-12-14 15:13:47 -08001310 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001311}
1312
Sage Weil6f6c7002011-01-17 20:34:08 -08001313static void reset_changed_osds(struct ceph_osd_client *osdc)
Sage Weilf24e9982009-10-06 11:31:10 -07001314{
Sage Weilf24e9982009-10-06 11:31:10 -07001315 struct rb_node *p, *n;
Sage Weilf24e9982009-10-06 11:31:10 -07001316
Sage Weil6f6c7002011-01-17 20:34:08 -08001317 for (p = rb_first(&osdc->osds); p; p = n) {
1318 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
Sage Weilf24e9982009-10-06 11:31:10 -07001319
Sage Weil6f6c7002011-01-17 20:34:08 -08001320 n = rb_next(p);
1321 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1322 memcmp(&osd->o_con.peer_addr,
1323 ceph_osd_addr(osdc->osdmap,
1324 osd->o_osd),
1325 sizeof(struct ceph_entity_addr)) != 0)
1326 __reset_osd(osdc, osd);
Sage Weilf24e9982009-10-06 11:31:10 -07001327 }
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001328}
1329
1330/*
Sage Weil6f6c7002011-01-17 20:34:08 -08001331 * Requeue requests whose mapping to an OSD has changed. If requests map to
1332 * no osd, request a new map.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001333 *
Alex Eldere6d50f62012-12-26 14:31:40 -06001334 * Caller should hold map_sem for read.
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001335 */
Sage Weil38d64532011-10-14 13:33:55 -07001336static void kick_requests(struct ceph_osd_client *osdc, int force_resend)
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001337{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001338 struct ceph_osd_request *req, *nreq;
Sage Weil6f6c7002011-01-17 20:34:08 -08001339 struct rb_node *p;
1340 int needmap = 0;
1341 int err;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001342
Sage Weil38d64532011-10-14 13:33:55 -07001343 dout("kick_requests %s\n", force_resend ? " (force resend)" : "");
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001344 mutex_lock(&osdc->request_mutex);
Sage Weil6194ea82012-07-30 16:19:28 -07001345 for (p = rb_first(&osdc->requests); p; ) {
Sage Weil6f6c7002011-01-17 20:34:08 -08001346 req = rb_entry(p, struct ceph_osd_request, r_node);
Sage Weil6194ea82012-07-30 16:19:28 -07001347 p = rb_next(p);
Alex Elderab60b162012-12-19 15:52:36 -06001348
1349 /*
1350 * For linger requests that have not yet been
1351 * registered, move them to the linger list; they'll
1352 * be sent to the osd in the loop below. Unregister
1353 * the request before re-registering it as a linger
1354 * request to ensure the __map_request() below
1355 * will decide it needs to be sent.
1356 */
1357 if (req->r_linger && list_empty(&req->r_linger_item)) {
1358 dout("%p tid %llu restart on osd%d\n",
1359 req, req->r_tid,
1360 req->r_osd ? req->r_osd->o_osd : -1);
1361 __unregister_request(osdc, req);
1362 __register_linger_request(osdc, req);
1363 continue;
1364 }
1365
Sage Weil38d64532011-10-14 13:33:55 -07001366 err = __map_request(osdc, req, force_resend);
Sage Weil6f6c7002011-01-17 20:34:08 -08001367 if (err < 0)
1368 continue; /* error */
1369 if (req->r_osd == NULL) {
1370 dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1371 needmap++; /* request a newer map */
1372 } else if (err > 0) {
Sage Weil6194ea82012-07-30 16:19:28 -07001373 if (!req->r_linger) {
1374 dout("%p tid %llu requeued on osd%d\n", req,
1375 req->r_tid,
1376 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001377 req->r_flags |= CEPH_OSD_FLAG_RETRY;
Sage Weil6194ea82012-07-30 16:19:28 -07001378 }
1379 }
Sage Weil6f6c7002011-01-17 20:34:08 -08001380 }
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001381
1382 list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1383 r_linger_item) {
1384 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1385
Sage Weil38d64532011-10-14 13:33:55 -07001386 err = __map_request(osdc, req, force_resend);
Alex Elderab60b162012-12-19 15:52:36 -06001387 dout("__map_request returned %d\n", err);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001388 if (err == 0)
1389 continue; /* no change and no osd was specified */
1390 if (err < 0)
1391 continue; /* hrm! */
1392 if (req->r_osd == NULL) {
1393 dout("tid %llu maps to no valid osd\n", req->r_tid);
1394 needmap++; /* request a newer map */
1395 continue;
1396 }
1397
1398 dout("kicking lingering %p tid %llu osd%d\n", req, req->r_tid,
1399 req->r_osd ? req->r_osd->o_osd : -1);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001400 __register_request(osdc, req);
Alex Elderc89ce052012-12-06 07:22:04 -06001401 __unregister_linger_request(osdc, req);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001402 }
Sage Weilf24e9982009-10-06 11:31:10 -07001403 mutex_unlock(&osdc->request_mutex);
1404
1405 if (needmap) {
1406 dout("%d requests for down osds, need new map\n", needmap);
1407 ceph_monc_request_next_osdmap(&osdc->client->monc);
1408 }
Alex Eldere6d50f62012-12-26 14:31:40 -06001409 reset_changed_osds(osdc);
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001410}
Sage Weil6f6c7002011-01-17 20:34:08 -08001411
1412
Sage Weilf24e9982009-10-06 11:31:10 -07001413/*
1414 * Process updated osd map.
1415 *
1416 * The message contains any number of incremental and full maps, normally
1417 * indicating some sort of topology change in the cluster. Kick requests
1418 * off to different OSDs as needed.
1419 */
1420void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1421{
1422 void *p, *end, *next;
1423 u32 nr_maps, maplen;
1424 u32 epoch;
1425 struct ceph_osdmap *newmap = NULL, *oldmap;
1426 int err;
1427 struct ceph_fsid fsid;
1428
1429 dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1430 p = msg->front.iov_base;
1431 end = p + msg->front.iov_len;
1432
1433 /* verify fsid */
1434 ceph_decode_need(&p, end, sizeof(fsid), bad);
1435 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Sage Weil07433042009-11-18 16:50:41 -08001436 if (ceph_check_fsid(osdc->client, &fsid) < 0)
1437 return;
Sage Weilf24e9982009-10-06 11:31:10 -07001438
1439 down_write(&osdc->map_sem);
1440
1441 /* incremental maps */
1442 ceph_decode_32_safe(&p, end, nr_maps, bad);
1443 dout(" %d inc maps\n", nr_maps);
1444 while (nr_maps > 0) {
1445 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07001446 epoch = ceph_decode_32(&p);
1447 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001448 ceph_decode_need(&p, end, maplen, bad);
1449 next = p + maplen;
1450 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1451 dout("applying incremental map %u len %d\n",
1452 epoch, maplen);
1453 newmap = osdmap_apply_incremental(&p, next,
1454 osdc->osdmap,
Alex Elder15d98822012-05-26 23:26:43 -05001455 &osdc->client->msgr);
Sage Weilf24e9982009-10-06 11:31:10 -07001456 if (IS_ERR(newmap)) {
1457 err = PTR_ERR(newmap);
1458 goto bad;
1459 }
Sage Weil30dc6382009-12-21 14:49:37 -08001460 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07001461 if (newmap != osdc->osdmap) {
1462 ceph_osdmap_destroy(osdc->osdmap);
1463 osdc->osdmap = newmap;
1464 }
Sage Weil38d64532011-10-14 13:33:55 -07001465 kick_requests(osdc, 0);
Sage Weilf24e9982009-10-06 11:31:10 -07001466 } else {
1467 dout("ignoring incremental map %u len %d\n",
1468 epoch, maplen);
1469 }
1470 p = next;
1471 nr_maps--;
1472 }
1473 if (newmap)
1474 goto done;
1475
1476 /* full maps */
1477 ceph_decode_32_safe(&p, end, nr_maps, bad);
1478 dout(" %d full maps\n", nr_maps);
1479 while (nr_maps) {
1480 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07001481 epoch = ceph_decode_32(&p);
1482 maplen = ceph_decode_32(&p);
Sage Weilf24e9982009-10-06 11:31:10 -07001483 ceph_decode_need(&p, end, maplen, bad);
1484 if (nr_maps > 1) {
1485 dout("skipping non-latest full map %u len %d\n",
1486 epoch, maplen);
1487 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1488 dout("skipping full map %u len %d, "
1489 "older than our %u\n", epoch, maplen,
1490 osdc->osdmap->epoch);
1491 } else {
Sage Weil38d64532011-10-14 13:33:55 -07001492 int skipped_map = 0;
1493
Sage Weilf24e9982009-10-06 11:31:10 -07001494 dout("taking full map %u len %d\n", epoch, maplen);
1495 newmap = osdmap_decode(&p, p+maplen);
1496 if (IS_ERR(newmap)) {
1497 err = PTR_ERR(newmap);
1498 goto bad;
1499 }
Sage Weil30dc6382009-12-21 14:49:37 -08001500 BUG_ON(!newmap);
Sage Weilf24e9982009-10-06 11:31:10 -07001501 oldmap = osdc->osdmap;
1502 osdc->osdmap = newmap;
Sage Weil38d64532011-10-14 13:33:55 -07001503 if (oldmap) {
1504 if (oldmap->epoch + 1 < newmap->epoch)
1505 skipped_map = 1;
Sage Weilf24e9982009-10-06 11:31:10 -07001506 ceph_osdmap_destroy(oldmap);
Sage Weil38d64532011-10-14 13:33:55 -07001507 }
1508 kick_requests(osdc, skipped_map);
Sage Weilf24e9982009-10-06 11:31:10 -07001509 }
1510 p += maplen;
1511 nr_maps--;
1512 }
1513
1514done:
1515 downgrade_write(&osdc->map_sem);
1516 ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
Sage Weilcd634fb2011-05-12 09:29:18 -07001517
1518 /*
1519 * subscribe to subsequent osdmap updates if full to ensure
1520 * we find out when we are no longer full and stop returning
1521 * ENOSPC.
1522 */
1523 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
1524 ceph_monc_request_next_osdmap(&osdc->client->monc);
1525
Alex Elderf9d25192013-02-15 11:42:29 -06001526 mutex_lock(&osdc->request_mutex);
1527 __send_queued(osdc);
1528 mutex_unlock(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001529 up_read(&osdc->map_sem);
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001530 wake_up_all(&osdc->client->auth_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07001531 return;
1532
1533bad:
1534 pr_err("osdc handle_map corrupt msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08001535 ceph_msg_dump(msg);
Sage Weilf24e9982009-10-06 11:31:10 -07001536 up_write(&osdc->map_sem);
1537 return;
1538}
1539
Sage Weilf24e9982009-10-06 11:31:10 -07001540/*
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001541 * watch/notify callback event infrastructure
1542 *
1543 * These callbacks are used both for watch and notify operations.
1544 */
1545static void __release_event(struct kref *kref)
1546{
1547 struct ceph_osd_event *event =
1548 container_of(kref, struct ceph_osd_event, kref);
1549
1550 dout("__release_event %p\n", event);
1551 kfree(event);
1552}
1553
1554static void get_event(struct ceph_osd_event *event)
1555{
1556 kref_get(&event->kref);
1557}
1558
1559void ceph_osdc_put_event(struct ceph_osd_event *event)
1560{
1561 kref_put(&event->kref, __release_event);
1562}
1563EXPORT_SYMBOL(ceph_osdc_put_event);
1564
1565static void __insert_event(struct ceph_osd_client *osdc,
1566 struct ceph_osd_event *new)
1567{
1568 struct rb_node **p = &osdc->event_tree.rb_node;
1569 struct rb_node *parent = NULL;
1570 struct ceph_osd_event *event = NULL;
1571
1572 while (*p) {
1573 parent = *p;
1574 event = rb_entry(parent, struct ceph_osd_event, node);
1575 if (new->cookie < event->cookie)
1576 p = &(*p)->rb_left;
1577 else if (new->cookie > event->cookie)
1578 p = &(*p)->rb_right;
1579 else
1580 BUG();
1581 }
1582
1583 rb_link_node(&new->node, parent, p);
1584 rb_insert_color(&new->node, &osdc->event_tree);
1585}
1586
1587static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1588 u64 cookie)
1589{
1590 struct rb_node **p = &osdc->event_tree.rb_node;
1591 struct rb_node *parent = NULL;
1592 struct ceph_osd_event *event = NULL;
1593
1594 while (*p) {
1595 parent = *p;
1596 event = rb_entry(parent, struct ceph_osd_event, node);
1597 if (cookie < event->cookie)
1598 p = &(*p)->rb_left;
1599 else if (cookie > event->cookie)
1600 p = &(*p)->rb_right;
1601 else
1602 return event;
1603 }
1604 return NULL;
1605}
1606
1607static void __remove_event(struct ceph_osd_event *event)
1608{
1609 struct ceph_osd_client *osdc = event->osdc;
1610
1611 if (!RB_EMPTY_NODE(&event->node)) {
1612 dout("__remove_event removed %p\n", event);
1613 rb_erase(&event->node, &osdc->event_tree);
1614 ceph_osdc_put_event(event);
1615 } else {
1616 dout("__remove_event didn't remove %p\n", event);
1617 }
1618}
1619
1620int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1621 void (*event_cb)(u64, u64, u8, void *),
Alex Elder3c663bb2013-02-15 11:42:30 -06001622 void *data, struct ceph_osd_event **pevent)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001623{
1624 struct ceph_osd_event *event;
1625
1626 event = kmalloc(sizeof(*event), GFP_NOIO);
1627 if (!event)
1628 return -ENOMEM;
1629
1630 dout("create_event %p\n", event);
1631 event->cb = event_cb;
Alex Elder3c663bb2013-02-15 11:42:30 -06001632 event->one_shot = 0;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001633 event->data = data;
1634 event->osdc = osdc;
1635 INIT_LIST_HEAD(&event->osd_node);
Alex Elder3ee52342012-12-17 12:23:48 -06001636 RB_CLEAR_NODE(&event->node);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001637 kref_init(&event->kref); /* one ref for us */
1638 kref_get(&event->kref); /* one ref for the caller */
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001639
1640 spin_lock(&osdc->event_lock);
1641 event->cookie = ++osdc->event_count;
1642 __insert_event(osdc, event);
1643 spin_unlock(&osdc->event_lock);
1644
1645 *pevent = event;
1646 return 0;
1647}
1648EXPORT_SYMBOL(ceph_osdc_create_event);
1649
1650void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1651{
1652 struct ceph_osd_client *osdc = event->osdc;
1653
1654 dout("cancel_event %p\n", event);
1655 spin_lock(&osdc->event_lock);
1656 __remove_event(event);
1657 spin_unlock(&osdc->event_lock);
1658 ceph_osdc_put_event(event); /* caller's */
1659}
1660EXPORT_SYMBOL(ceph_osdc_cancel_event);
1661
1662
1663static void do_event_work(struct work_struct *work)
1664{
1665 struct ceph_osd_event_work *event_work =
1666 container_of(work, struct ceph_osd_event_work, work);
1667 struct ceph_osd_event *event = event_work->event;
1668 u64 ver = event_work->ver;
1669 u64 notify_id = event_work->notify_id;
1670 u8 opcode = event_work->opcode;
1671
1672 dout("do_event_work completing %p\n", event);
1673 event->cb(ver, notify_id, opcode, event->data);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001674 dout("do_event_work completed %p\n", event);
1675 ceph_osdc_put_event(event);
1676 kfree(event_work);
1677}
1678
1679
1680/*
1681 * Process osd watch notifications
1682 */
Alex Elder3c663bb2013-02-15 11:42:30 -06001683static void handle_watch_notify(struct ceph_osd_client *osdc,
1684 struct ceph_msg *msg)
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001685{
1686 void *p, *end;
1687 u8 proto_ver;
1688 u64 cookie, ver, notify_id;
1689 u8 opcode;
1690 struct ceph_osd_event *event;
1691 struct ceph_osd_event_work *event_work;
1692
1693 p = msg->front.iov_base;
1694 end = p + msg->front.iov_len;
1695
1696 ceph_decode_8_safe(&p, end, proto_ver, bad);
1697 ceph_decode_8_safe(&p, end, opcode, bad);
1698 ceph_decode_64_safe(&p, end, cookie, bad);
1699 ceph_decode_64_safe(&p, end, ver, bad);
1700 ceph_decode_64_safe(&p, end, notify_id, bad);
1701
1702 spin_lock(&osdc->event_lock);
1703 event = __find_event(osdc, cookie);
1704 if (event) {
Alex Elder3c663bb2013-02-15 11:42:30 -06001705 BUG_ON(event->one_shot);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001706 get_event(event);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001707 }
1708 spin_unlock(&osdc->event_lock);
1709 dout("handle_watch_notify cookie %lld ver %lld event %p\n",
1710 cookie, ver, event);
1711 if (event) {
1712 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001713 if (!event_work) {
1714 dout("ERROR: could not allocate event_work\n");
1715 goto done_err;
1716 }
Mariusz Kozlowski6b0ae402011-03-26 19:29:34 +01001717 INIT_WORK(&event_work->work, do_event_work);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001718 event_work->event = event;
1719 event_work->ver = ver;
1720 event_work->notify_id = notify_id;
1721 event_work->opcode = opcode;
1722 if (!queue_work(osdc->notify_wq, &event_work->work)) {
1723 dout("WARNING: failed to queue notify event work\n");
1724 goto done_err;
1725 }
1726 }
1727
1728 return;
1729
1730done_err:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001731 ceph_osdc_put_event(event);
1732 return;
1733
1734bad:
1735 pr_err("osdc handle_watch_notify corrupt msg\n");
1736 return;
1737}
1738
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001739/*
Sage Weilf24e9982009-10-06 11:31:10 -07001740 * Register request, send initial attempt.
1741 */
1742int ceph_osdc_start_request(struct ceph_osd_client *osdc,
1743 struct ceph_osd_request *req,
1744 bool nofail)
1745{
Sage Weilc1ea8822009-10-08 16:55:47 -07001746 int rc = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001747
1748 req->r_request->pages = req->r_pages;
Alex Elderd4b515f2013-02-25 17:35:46 -06001749 req->r_request->page_count = req->r_num_pages;
Alex Elderf51a8222013-02-14 12:16:43 -06001750 req->r_request->page_alignment = req->r_page_alignment;
Yehuda Sadeh68b44762010-04-06 15:01:27 -07001751#ifdef CONFIG_BLOCK
1752 req->r_request->bio = req->r_bio;
1753#endif
Alex Elderc8858372012-11-13 21:11:15 -06001754 req->r_request->trail = &req->r_trail;
Sage Weilf24e9982009-10-06 11:31:10 -07001755
1756 register_request(osdc, req);
1757
1758 down_read(&osdc->map_sem);
1759 mutex_lock(&osdc->request_mutex);
Sage Weilc1ea8822009-10-08 16:55:47 -07001760 /*
1761 * a racing kick_requests() may have sent the message for us
1762 * while we dropped request_mutex above, so only send now if
1763 * the request still han't been touched yet.
1764 */
1765 if (req->r_sent == 0) {
Sage Weil38d64532011-10-14 13:33:55 -07001766 rc = __map_request(osdc, req, 0);
Sage Weil9d6fcb02011-05-12 15:48:16 -07001767 if (rc < 0) {
1768 if (nofail) {
1769 dout("osdc_start_request failed map, "
1770 " will retry %lld\n", req->r_tid);
1771 rc = 0;
1772 }
Dan Carpenter234af26f2011-03-29 06:25:59 +03001773 goto out_unlock;
Sage Weil9d6fcb02011-05-12 15:48:16 -07001774 }
Sage Weil6f6c7002011-01-17 20:34:08 -08001775 if (req->r_osd == NULL) {
1776 dout("send_request %p no up osds in pg\n", req);
1777 ceph_monc_request_next_osdmap(&osdc->client->monc);
1778 } else {
Sage Weil56e925b2012-01-03 12:34:34 -08001779 __send_request(osdc, req);
Sage Weilf24e9982009-10-06 11:31:10 -07001780 }
Sage Weil56e925b2012-01-03 12:34:34 -08001781 rc = 0;
Sage Weilf24e9982009-10-06 11:31:10 -07001782 }
Dan Carpenter234af26f2011-03-29 06:25:59 +03001783
1784out_unlock:
Sage Weilf24e9982009-10-06 11:31:10 -07001785 mutex_unlock(&osdc->request_mutex);
1786 up_read(&osdc->map_sem);
1787 return rc;
1788}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001789EXPORT_SYMBOL(ceph_osdc_start_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001790
1791/*
1792 * wait for a request to complete
1793 */
1794int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
1795 struct ceph_osd_request *req)
1796{
1797 int rc;
1798
1799 rc = wait_for_completion_interruptible(&req->r_completion);
1800 if (rc < 0) {
1801 mutex_lock(&osdc->request_mutex);
1802 __cancel_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08001803 __unregister_request(osdc, req);
Sage Weilf24e9982009-10-06 11:31:10 -07001804 mutex_unlock(&osdc->request_mutex);
Sage Weil25845472011-06-03 09:37:09 -07001805 complete_request(req);
Sage Weil529cfcc2009-12-22 10:29:39 -08001806 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
Sage Weilf24e9982009-10-06 11:31:10 -07001807 return rc;
1808 }
1809
1810 dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
1811 return req->r_result;
1812}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001813EXPORT_SYMBOL(ceph_osdc_wait_request);
Sage Weilf24e9982009-10-06 11:31:10 -07001814
1815/*
1816 * sync - wait for all in-flight requests to flush. avoid starvation.
1817 */
1818void ceph_osdc_sync(struct ceph_osd_client *osdc)
1819{
1820 struct ceph_osd_request *req;
1821 u64 last_tid, next_tid = 0;
1822
1823 mutex_lock(&osdc->request_mutex);
1824 last_tid = osdc->last_tid;
1825 while (1) {
1826 req = __lookup_request_ge(osdc, next_tid);
1827 if (!req)
1828 break;
1829 if (req->r_tid > last_tid)
1830 break;
1831
1832 next_tid = req->r_tid + 1;
1833 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
1834 continue;
1835
1836 ceph_osdc_get_request(req);
1837 mutex_unlock(&osdc->request_mutex);
1838 dout("sync waiting on tid %llu (last is %llu)\n",
1839 req->r_tid, last_tid);
1840 wait_for_completion(&req->r_safe_completion);
1841 mutex_lock(&osdc->request_mutex);
1842 ceph_osdc_put_request(req);
1843 }
1844 mutex_unlock(&osdc->request_mutex);
1845 dout("sync done (thru tid %llu)\n", last_tid);
1846}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001847EXPORT_SYMBOL(ceph_osdc_sync);
Sage Weilf24e9982009-10-06 11:31:10 -07001848
1849/*
1850 * init, shutdown
1851 */
1852int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
1853{
1854 int err;
1855
1856 dout("init\n");
1857 osdc->client = client;
1858 osdc->osdmap = NULL;
1859 init_rwsem(&osdc->map_sem);
1860 init_completion(&osdc->map_waiters);
1861 osdc->last_requested_map = 0;
1862 mutex_init(&osdc->request_mutex);
Sage Weilf24e9982009-10-06 11:31:10 -07001863 osdc->last_tid = 0;
1864 osdc->osds = RB_ROOT;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001865 INIT_LIST_HEAD(&osdc->osd_lru);
Sage Weilf24e9982009-10-06 11:31:10 -07001866 osdc->requests = RB_ROOT;
Yehuda Sadeh422d2cb2010-02-26 15:32:31 -08001867 INIT_LIST_HEAD(&osdc->req_lru);
Sage Weil6f6c7002011-01-17 20:34:08 -08001868 INIT_LIST_HEAD(&osdc->req_unsent);
1869 INIT_LIST_HEAD(&osdc->req_notarget);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001870 INIT_LIST_HEAD(&osdc->req_linger);
Sage Weilf24e9982009-10-06 11:31:10 -07001871 osdc->num_requests = 0;
1872 INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001873 INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001874 spin_lock_init(&osdc->event_lock);
1875 osdc->event_tree = RB_ROOT;
1876 osdc->event_count = 0;
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001877
1878 schedule_delayed_work(&osdc->osds_timeout_work,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001879 round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
Sage Weilf24e9982009-10-06 11:31:10 -07001880
Sage Weil5f44f142009-11-18 14:52:18 -08001881 err = -ENOMEM;
Sage Weilf24e9982009-10-06 11:31:10 -07001882 osdc->req_mempool = mempool_create_kmalloc_pool(10,
1883 sizeof(struct ceph_osd_request));
1884 if (!osdc->req_mempool)
Sage Weil5f44f142009-11-18 14:52:18 -08001885 goto out;
Sage Weilf24e9982009-10-06 11:31:10 -07001886
Sage Weild50b4092012-07-09 14:22:34 -07001887 err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
1888 OSD_OP_FRONT_LEN, 10, true,
Sage Weil4f482802010-04-24 09:56:35 -07001889 "osd_op");
Sage Weilf24e9982009-10-06 11:31:10 -07001890 if (err < 0)
Sage Weil5f44f142009-11-18 14:52:18 -08001891 goto out_mempool;
Sage Weild50b4092012-07-09 14:22:34 -07001892 err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
Sage Weil4f482802010-04-24 09:56:35 -07001893 OSD_OPREPLY_FRONT_LEN, 10, true,
1894 "osd_op_reply");
Sage Weilc16e7862010-03-01 13:02:00 -08001895 if (err < 0)
1896 goto out_msgpool;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001897
1898 osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
1899 if (IS_ERR(osdc->notify_wq)) {
1900 err = PTR_ERR(osdc->notify_wq);
1901 osdc->notify_wq = NULL;
1902 goto out_msgpool;
1903 }
Sage Weilf24e9982009-10-06 11:31:10 -07001904 return 0;
Sage Weil5f44f142009-11-18 14:52:18 -08001905
Sage Weilc16e7862010-03-01 13:02:00 -08001906out_msgpool:
1907 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weil5f44f142009-11-18 14:52:18 -08001908out_mempool:
1909 mempool_destroy(osdc->req_mempool);
1910out:
1911 return err;
Sage Weilf24e9982009-10-06 11:31:10 -07001912}
1913
1914void ceph_osdc_stop(struct ceph_osd_client *osdc)
1915{
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07001916 flush_workqueue(osdc->notify_wq);
1917 destroy_workqueue(osdc->notify_wq);
Sage Weilf24e9982009-10-06 11:31:10 -07001918 cancel_delayed_work_sync(&osdc->timeout_work);
Yehuda Sadehf5a20412010-02-03 11:00:26 -08001919 cancel_delayed_work_sync(&osdc->osds_timeout_work);
Sage Weilf24e9982009-10-06 11:31:10 -07001920 if (osdc->osdmap) {
1921 ceph_osdmap_destroy(osdc->osdmap);
1922 osdc->osdmap = NULL;
1923 }
Sage Weilaca420b2011-08-31 14:45:53 -07001924 remove_all_osds(osdc);
Sage Weilf24e9982009-10-06 11:31:10 -07001925 mempool_destroy(osdc->req_mempool);
1926 ceph_msgpool_destroy(&osdc->msgpool_op);
Sage Weilc16e7862010-03-01 13:02:00 -08001927 ceph_msgpool_destroy(&osdc->msgpool_op_reply);
Sage Weilf24e9982009-10-06 11:31:10 -07001928}
1929
1930/*
1931 * Read some contiguous pages. If we cross a stripe boundary, shorten
1932 * *plen. Return number of bytes read, or error.
1933 */
1934int ceph_osdc_readpages(struct ceph_osd_client *osdc,
1935 struct ceph_vino vino, struct ceph_file_layout *layout,
1936 u64 off, u64 *plen,
1937 u32 truncate_seq, u64 truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -08001938 struct page **pages, int num_pages, int page_align)
Sage Weilf24e9982009-10-06 11:31:10 -07001939{
1940 struct ceph_osd_request *req;
1941 int rc = 0;
1942
1943 dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
1944 vino.snap, off, *plen);
1945 req = ceph_osdc_new_request(osdc, layout, vino, off, plen,
1946 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
1947 NULL, 0, truncate_seq, truncate_size, NULL,
Alex Eldera3bea472013-02-15 11:42:29 -06001948 false, page_align);
Sage Weil68162822012-09-24 21:01:02 -07001949 if (IS_ERR(req))
1950 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07001951
1952 /* it may be a short read due to an object boundary */
1953 req->r_pages = pages;
Sage Weilf24e9982009-10-06 11:31:10 -07001954
Sage Weilb7495fc2010-11-09 12:43:12 -08001955 dout("readpages final extent is %llu~%llu (%d pages align %d)\n",
1956 off, *plen, req->r_num_pages, page_align);
Sage Weilf24e9982009-10-06 11:31:10 -07001957
1958 rc = ceph_osdc_start_request(osdc, req, false);
1959 if (!rc)
1960 rc = ceph_osdc_wait_request(osdc, req);
1961
1962 ceph_osdc_put_request(req);
1963 dout("readpages result %d\n", rc);
1964 return rc;
1965}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001966EXPORT_SYMBOL(ceph_osdc_readpages);
Sage Weilf24e9982009-10-06 11:31:10 -07001967
1968/*
1969 * do a synchronous write on N pages
1970 */
1971int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
1972 struct ceph_file_layout *layout,
1973 struct ceph_snap_context *snapc,
1974 u64 off, u64 len,
1975 u32 truncate_seq, u64 truncate_size,
1976 struct timespec *mtime,
Alex Elder24808822013-02-15 11:42:29 -06001977 struct page **pages, int num_pages)
Sage Weilf24e9982009-10-06 11:31:10 -07001978{
1979 struct ceph_osd_request *req;
1980 int rc = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -08001981 int page_align = off & ~PAGE_MASK;
Sage Weilf24e9982009-10-06 11:31:10 -07001982
1983 BUG_ON(vino.snap != CEPH_NOSNAP);
1984 req = ceph_osdc_new_request(osdc, layout, vino, off, &len,
1985 CEPH_OSD_OP_WRITE,
Alex Elder24808822013-02-15 11:42:29 -06001986 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
Alex Elderfbf86852013-02-15 11:42:29 -06001987 snapc, 0,
Sage Weilf24e9982009-10-06 11:31:10 -07001988 truncate_seq, truncate_size, mtime,
Alex Eldera3bea472013-02-15 11:42:29 -06001989 true, page_align);
Sage Weil68162822012-09-24 21:01:02 -07001990 if (IS_ERR(req))
1991 return PTR_ERR(req);
Sage Weilf24e9982009-10-06 11:31:10 -07001992
1993 /* it may be a short write due to an object boundary */
1994 req->r_pages = pages;
Sage Weilf24e9982009-10-06 11:31:10 -07001995 dout("writepages %llu~%llu (%d pages)\n", off, len,
1996 req->r_num_pages);
1997
Alex Elder87f979d2013-02-15 11:42:29 -06001998 rc = ceph_osdc_start_request(osdc, req, true);
Sage Weilf24e9982009-10-06 11:31:10 -07001999 if (!rc)
2000 rc = ceph_osdc_wait_request(osdc, req);
2001
2002 ceph_osdc_put_request(req);
2003 if (rc == 0)
2004 rc = len;
2005 dout("writepages result %d\n", rc);
2006 return rc;
2007}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002008EXPORT_SYMBOL(ceph_osdc_writepages);
Sage Weilf24e9982009-10-06 11:31:10 -07002009
2010/*
2011 * handle incoming message
2012 */
2013static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2014{
2015 struct ceph_osd *osd = con->private;
Julia Lawall32c895e2009-11-21 16:53:16 +01002016 struct ceph_osd_client *osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002017 int type = le16_to_cpu(msg->hdr.type);
2018
2019 if (!osd)
Sage Weil4a32f932010-06-13 10:27:53 -07002020 goto out;
Julia Lawall32c895e2009-11-21 16:53:16 +01002021 osdc = osd->o_osdc;
Sage Weilf24e9982009-10-06 11:31:10 -07002022
2023 switch (type) {
2024 case CEPH_MSG_OSD_MAP:
2025 ceph_osdc_handle_map(osdc, msg);
2026 break;
2027 case CEPH_MSG_OSD_OPREPLY:
Sage Weil350b1c32009-12-22 10:45:45 -08002028 handle_reply(osdc, msg, con);
Sage Weilf24e9982009-10-06 11:31:10 -07002029 break;
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002030 case CEPH_MSG_WATCH_NOTIFY:
2031 handle_watch_notify(osdc, msg);
2032 break;
Sage Weilf24e9982009-10-06 11:31:10 -07002033
2034 default:
2035 pr_err("received unknown message type %d %s\n", type,
2036 ceph_msg_type_name(type));
2037 }
Sage Weil4a32f932010-06-13 10:27:53 -07002038out:
Sage Weilf24e9982009-10-06 11:31:10 -07002039 ceph_msg_put(msg);
2040}
2041
Sage Weil5b3a4db2010-02-19 21:43:23 -08002042/*
Sage Weil21b667f2010-03-04 10:22:59 -08002043 * lookup and return message for incoming reply. set up reply message
2044 * pages.
Sage Weil5b3a4db2010-02-19 21:43:23 -08002045 */
2046static struct ceph_msg *get_reply(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08002047 struct ceph_msg_header *hdr,
2048 int *skip)
Sage Weilf24e9982009-10-06 11:31:10 -07002049{
2050 struct ceph_osd *osd = con->private;
2051 struct ceph_osd_client *osdc = osd->o_osdc;
Yehuda Sadeh24504182010-01-08 13:58:34 -08002052 struct ceph_msg *m;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002053 struct ceph_osd_request *req;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002054 int front = le32_to_cpu(hdr->front_len);
2055 int data_len = le32_to_cpu(hdr->data_len);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002056 u64 tid;
Sage Weilf24e9982009-10-06 11:31:10 -07002057
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002058 tid = le64_to_cpu(hdr->tid);
2059 mutex_lock(&osdc->request_mutex);
2060 req = __lookup_request(osdc, tid);
2061 if (!req) {
2062 *skip = 1;
2063 m = NULL;
Sage Weil756a16a2012-07-30 16:26:13 -07002064 dout("get_reply unknown tid %llu from osd%d\n", tid,
2065 osd->o_osd);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002066 goto out;
2067 }
Sage Weilc16e7862010-03-01 13:02:00 -08002068
2069 if (req->r_con_filling_msg) {
Alex Elder8921d112012-06-01 14:56:43 -05002070 dout("%s revoking msg %p from old con %p\n", __func__,
Sage Weilc16e7862010-03-01 13:02:00 -08002071 req->r_reply, req->r_con_filling_msg);
Alex Elder8921d112012-06-01 14:56:43 -05002072 ceph_msg_revoke_incoming(req->r_reply);
Sage Weil0d477662012-05-31 20:22:18 -07002073 req->r_con_filling_msg->ops->put(req->r_con_filling_msg);
Sage Weil6f46cb22010-03-24 21:30:19 -07002074 req->r_con_filling_msg = NULL;
Sage Weilf24e9982009-10-06 11:31:10 -07002075 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08002076
Sage Weilc16e7862010-03-01 13:02:00 -08002077 if (front > req->r_reply->front.iov_len) {
2078 pr_warning("get_reply front %d > preallocated %d\n",
2079 front, (int)req->r_reply->front.iov_len);
Sage Weilb61c2762011-08-09 15:03:46 -07002080 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front, GFP_NOFS, false);
Sage Weila79832f2010-04-01 16:06:19 -07002081 if (!m)
Sage Weilc16e7862010-03-01 13:02:00 -08002082 goto out;
2083 ceph_msg_put(req->r_reply);
2084 req->r_reply = m;
2085 }
2086 m = ceph_msg_get(req->r_reply);
2087
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002088 if (data_len > 0) {
Sage Weilb7495fc2010-11-09 12:43:12 -08002089 int want = calc_pages_for(req->r_page_alignment, data_len);
Sage Weil21b667f2010-03-04 10:22:59 -08002090
Alex Elder9cbb1d72013-01-31 16:02:00 -06002091 if (req->r_pages && unlikely(req->r_num_pages < want)) {
Sage Weil9bb0ce22011-06-13 16:20:18 -07002092 pr_warning("tid %lld reply has %d bytes %d pages, we"
2093 " had only %d pages ready\n", tid, data_len,
2094 want, req->r_num_pages);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002095 *skip = 1;
2096 ceph_msg_put(m);
Sage Weila79832f2010-04-01 16:06:19 -07002097 m = NULL;
Sage Weil21b667f2010-03-04 10:22:59 -08002098 goto out;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002099 }
Sage Weil21b667f2010-03-04 10:22:59 -08002100 m->pages = req->r_pages;
Alex Elderd4b515f2013-02-25 17:35:46 -06002101 m->page_count = req->r_num_pages;
Sage Weilc5c6b192010-11-09 12:40:00 -08002102 m->page_alignment = req->r_page_alignment;
Yehuda Sadeh68b44762010-04-06 15:01:27 -07002103#ifdef CONFIG_BLOCK
2104 m->bio = req->r_bio;
2105#endif
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002106 }
Sage Weil5b3a4db2010-02-19 21:43:23 -08002107 *skip = 0;
Sage Weil0d477662012-05-31 20:22:18 -07002108 req->r_con_filling_msg = con->ops->get(con);
Sage Weilc16e7862010-03-01 13:02:00 -08002109 dout("get_reply tid %lld %p\n", tid, m);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002110
2111out:
2112 mutex_unlock(&osdc->request_mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002113 return m;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002114
2115}
2116
2117static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2118 struct ceph_msg_header *hdr,
2119 int *skip)
2120{
2121 struct ceph_osd *osd = con->private;
2122 int type = le16_to_cpu(hdr->type);
2123 int front = le32_to_cpu(hdr->front_len);
2124
Alex Elder1c20f2d2012-06-04 14:43:32 -05002125 *skip = 0;
Sage Weil5b3a4db2010-02-19 21:43:23 -08002126 switch (type) {
2127 case CEPH_MSG_OSD_MAP:
Yehuda Sadeha40c4f12011-03-21 15:07:16 -07002128 case CEPH_MSG_WATCH_NOTIFY:
Sage Weilb61c2762011-08-09 15:03:46 -07002129 return ceph_msg_new(type, front, GFP_NOFS, false);
Sage Weil5b3a4db2010-02-19 21:43:23 -08002130 case CEPH_MSG_OSD_OPREPLY:
2131 return get_reply(con, hdr, skip);
2132 default:
2133 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2134 osd->o_osd);
2135 *skip = 1;
2136 return NULL;
2137 }
Sage Weilf24e9982009-10-06 11:31:10 -07002138}
2139
2140/*
2141 * Wrappers to refcount containing ceph_osd struct
2142 */
2143static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2144{
2145 struct ceph_osd *osd = con->private;
2146 if (get_osd(osd))
2147 return con;
2148 return NULL;
2149}
2150
2151static void put_osd_con(struct ceph_connection *con)
2152{
2153 struct ceph_osd *osd = con->private;
2154 put_osd(osd);
2155}
2156
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002157/*
2158 * authentication
2159 */
Alex Eldera3530df2012-05-16 15:16:39 -05002160/*
2161 * Note: returned pointer is the address of a structure that's
2162 * managed separately. Caller must *not* attempt to free it.
2163 */
2164static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
Alex Elder8f43fb52012-05-16 15:16:39 -05002165 int *proto, int force_new)
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002166{
2167 struct ceph_osd *o = con->private;
2168 struct ceph_osd_client *osdc = o->o_osdc;
2169 struct ceph_auth_client *ac = osdc->client->monc.auth;
Alex Elder74f18692012-05-16 15:16:39 -05002170 struct ceph_auth_handshake *auth = &o->o_auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002171
Alex Elder74f18692012-05-16 15:16:39 -05002172 if (force_new && auth->authorizer) {
Alex Eldera2556512012-05-16 15:16:39 -05002173 if (ac->ops && ac->ops->destroy_authorizer)
2174 ac->ops->destroy_authorizer(ac, auth->authorizer);
Alex Elder74f18692012-05-16 15:16:39 -05002175 auth->authorizer = NULL;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002176 }
Alex Eldera2556512012-05-16 15:16:39 -05002177 if (!auth->authorizer && ac->ops && ac->ops->create_authorizer) {
Alex Eldera3530df2012-05-16 15:16:39 -05002178 int ret = ac->ops->create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2179 auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002180 if (ret)
Alex Eldera3530df2012-05-16 15:16:39 -05002181 return ERR_PTR(ret);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002182 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002183 *proto = ac->protocol;
Alex Elder74f18692012-05-16 15:16:39 -05002184
Alex Eldera3530df2012-05-16 15:16:39 -05002185 return auth;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002186}
2187
2188
2189static int verify_authorizer_reply(struct ceph_connection *con, int len)
2190{
2191 struct ceph_osd *o = con->private;
2192 struct ceph_osd_client *osdc = o->o_osdc;
2193 struct ceph_auth_client *ac = osdc->client->monc.auth;
2194
Alex Eldera2556512012-05-16 15:16:39 -05002195 /*
2196 * XXX If ac->ops or ac->ops->verify_authorizer_reply is null,
2197 * XXX which do we do: succeed or fail?
2198 */
Alex Elder6c4a1912012-05-16 15:16:38 -05002199 return ac->ops->verify_authorizer_reply(ac, o->o_auth.authorizer, len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002200}
2201
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002202static int invalidate_authorizer(struct ceph_connection *con)
2203{
2204 struct ceph_osd *o = con->private;
2205 struct ceph_osd_client *osdc = o->o_osdc;
2206 struct ceph_auth_client *ac = osdc->client->monc.auth;
2207
Alex Eldera2556512012-05-16 15:16:39 -05002208 if (ac->ops && ac->ops->invalidate_authorizer)
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002209 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2210
2211 return ceph_monc_validate_auth(&osdc->client->monc);
2212}
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002213
Tobias Klauser9e327892010-05-20 10:40:19 +02002214static const struct ceph_connection_operations osd_con_ops = {
Sage Weilf24e9982009-10-06 11:31:10 -07002215 .get = get_osd_con,
2216 .put = put_osd_con,
2217 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08002218 .get_authorizer = get_authorizer,
2219 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08002220 .invalidate_authorizer = invalidate_authorizer,
Sage Weilf24e9982009-10-06 11:31:10 -07002221 .alloc_msg = alloc_msg,
Sage Weil81b024e2009-10-09 10:29:18 -07002222 .fault = osd_reset,
Sage Weilf24e9982009-10-06 11:31:10 -07002223};