Commit | Line | Data |
---|---|---|
a4ce40a9 | 1 | |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
f24e9980 | 3 | |
3d14c5d2 | 4 | #include <linux/module.h> |
f24e9980 SW |
5 | #include <linux/err.h> |
6 | #include <linux/highmem.h> | |
7 | #include <linux/mm.h> | |
8 | #include <linux/pagemap.h> | |
9 | #include <linux/slab.h> | |
10 | #include <linux/uaccess.h> | |
68b4476b YS |
11 | #ifdef CONFIG_BLOCK |
12 | #include <linux/bio.h> | |
13 | #endif | |
f24e9980 | 14 | |
3d14c5d2 YS |
15 | #include <linux/ceph/libceph.h> |
16 | #include <linux/ceph/osd_client.h> | |
17 | #include <linux/ceph/messenger.h> | |
18 | #include <linux/ceph/decode.h> | |
19 | #include <linux/ceph/auth.h> | |
20 | #include <linux/ceph/pagelist.h> | |
f24e9980 | 21 | |
c16e7869 | 22 | #define OSD_OPREPLY_FRONT_LEN 512 |
0d59ab81 | 23 | |
5522ae0b AE |
24 | static struct kmem_cache *ceph_osd_request_cache; |
25 | ||
9e32789f | 26 | static const struct ceph_connection_operations osd_con_ops; |
f24e9980 | 27 | |
f9d25199 | 28 | static void __send_queued(struct ceph_osd_client *osdc); |
6f6c7006 | 29 | static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd); |
a40c4f10 YS |
30 | static void __register_request(struct ceph_osd_client *osdc, |
31 | struct ceph_osd_request *req); | |
2cc6128a ID |
32 | static void __unregister_request(struct ceph_osd_client *osdc, |
33 | struct ceph_osd_request *req); | |
a40c4f10 YS |
34 | static void __unregister_linger_request(struct ceph_osd_client *osdc, |
35 | struct ceph_osd_request *req); | |
2cc6128a | 36 | static void __enqueue_request(struct ceph_osd_request *req); |
f24e9980 SW |
37 | |
38 | /* | |
39 | * Implement client access to distributed object storage cluster. | |
40 | * | |
41 | * All data objects are stored within a cluster/cloud of OSDs, or | |
42 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to | |
43 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply | |
44 | * remote daemons serving up and coordinating consistent and safe | |
45 | * access to storage. | |
46 | * | |
47 | * Cluster membership and the mapping of data objects onto storage devices | |
48 | * are described by the osd map. | |
49 | * | |
50 | * We keep track of pending OSD requests (read, write), resubmit | |
51 | * requests to different OSDs when the cluster topology/data layout | |
52 | * change, or retry the affected requests when the communications | |
53 | * channel with an OSD is reset. | |
54 | */ | |
55 | ||
56 | /* | |
57 | * calculate the mapping of a file extent onto an object, and fill out the | |
58 | * request accordingly. shorten extent as necessary if it crosses an | |
59 | * object boundary. | |
60 | * | |
61 | * fill osd op in request message. | |
62 | */ | |
dbe0fc41 | 63 | static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen, |
a19dadfb | 64 | u64 *objnum, u64 *objoff, u64 *objlen) |
f24e9980 | 65 | { |
60e56f13 | 66 | u64 orig_len = *plen; |
d63b77f4 | 67 | int r; |
f24e9980 | 68 | |
60e56f13 | 69 | /* object extent? */ |
75d1c941 AE |
70 | r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum, |
71 | objoff, objlen); | |
d63b77f4 SW |
72 | if (r < 0) |
73 | return r; | |
75d1c941 AE |
74 | if (*objlen < orig_len) { |
75 | *plen = *objlen; | |
60e56f13 AE |
76 | dout(" skipping last %llu, final file extent %llu~%llu\n", |
77 | orig_len - *plen, off, *plen); | |
78 | } | |
79 | ||
75d1c941 | 80 | dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen); |
f24e9980 | 81 | |
3ff5f385 | 82 | return 0; |
f24e9980 SW |
83 | } |
84 | ||
c54d47bf AE |
85 | static void ceph_osd_data_init(struct ceph_osd_data *osd_data) |
86 | { | |
87 | memset(osd_data, 0, sizeof (*osd_data)); | |
88 | osd_data->type = CEPH_OSD_DATA_TYPE_NONE; | |
89 | } | |
90 | ||
a4ce40a9 | 91 | static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
92 | struct page **pages, u64 length, u32 alignment, |
93 | bool pages_from_pool, bool own_pages) | |
94 | { | |
95 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; | |
96 | osd_data->pages = pages; | |
97 | osd_data->length = length; | |
98 | osd_data->alignment = alignment; | |
99 | osd_data->pages_from_pool = pages_from_pool; | |
100 | osd_data->own_pages = own_pages; | |
101 | } | |
43bfe5de | 102 | |
a4ce40a9 | 103 | static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
104 | struct ceph_pagelist *pagelist) |
105 | { | |
106 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST; | |
107 | osd_data->pagelist = pagelist; | |
108 | } | |
43bfe5de AE |
109 | |
110 | #ifdef CONFIG_BLOCK | |
a4ce40a9 | 111 | static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
112 | struct bio *bio, size_t bio_length) |
113 | { | |
114 | osd_data->type = CEPH_OSD_DATA_TYPE_BIO; | |
115 | osd_data->bio = bio; | |
116 | osd_data->bio_length = bio_length; | |
117 | } | |
43bfe5de AE |
118 | #endif /* CONFIG_BLOCK */ |
119 | ||
8a703a38 IC |
120 | #define osd_req_op_data(oreq, whch, typ, fld) \ |
121 | ({ \ | |
122 | struct ceph_osd_request *__oreq = (oreq); \ | |
123 | unsigned int __whch = (whch); \ | |
124 | BUG_ON(__whch >= __oreq->r_num_ops); \ | |
125 | &__oreq->r_ops[__whch].typ.fld; \ | |
126 | }) | |
863c7eb5 | 127 | |
49719778 AE |
128 | static struct ceph_osd_data * |
129 | osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which) | |
130 | { | |
131 | BUG_ON(which >= osd_req->r_num_ops); | |
132 | ||
133 | return &osd_req->r_ops[which].raw_data_in; | |
134 | } | |
135 | ||
a4ce40a9 AE |
136 | struct ceph_osd_data * |
137 | osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req, | |
406e2c9f | 138 | unsigned int which) |
a4ce40a9 | 139 | { |
863c7eb5 | 140 | return osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 AE |
141 | } |
142 | EXPORT_SYMBOL(osd_req_op_extent_osd_data); | |
143 | ||
49719778 AE |
144 | void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req, |
145 | unsigned int which, struct page **pages, | |
146 | u64 length, u32 alignment, | |
147 | bool pages_from_pool, bool own_pages) | |
148 | { | |
149 | struct ceph_osd_data *osd_data; | |
150 | ||
151 | osd_data = osd_req_op_raw_data_in(osd_req, which); | |
152 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, | |
153 | pages_from_pool, own_pages); | |
154 | } | |
155 | EXPORT_SYMBOL(osd_req_op_raw_data_in_pages); | |
156 | ||
a4ce40a9 | 157 | void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req, |
406e2c9f AE |
158 | unsigned int which, struct page **pages, |
159 | u64 length, u32 alignment, | |
a4ce40a9 AE |
160 | bool pages_from_pool, bool own_pages) |
161 | { | |
162 | struct ceph_osd_data *osd_data; | |
163 | ||
863c7eb5 | 164 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 AE |
165 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, |
166 | pages_from_pool, own_pages); | |
a4ce40a9 AE |
167 | } |
168 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages); | |
169 | ||
170 | void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req, | |
406e2c9f | 171 | unsigned int which, struct ceph_pagelist *pagelist) |
a4ce40a9 AE |
172 | { |
173 | struct ceph_osd_data *osd_data; | |
174 | ||
863c7eb5 | 175 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 | 176 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
a4ce40a9 AE |
177 | } |
178 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist); | |
179 | ||
180 | #ifdef CONFIG_BLOCK | |
181 | void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req, | |
406e2c9f | 182 | unsigned int which, struct bio *bio, size_t bio_length) |
a4ce40a9 AE |
183 | { |
184 | struct ceph_osd_data *osd_data; | |
863c7eb5 AE |
185 | |
186 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); | |
a4ce40a9 | 187 | ceph_osd_data_bio_init(osd_data, bio, bio_length); |
a4ce40a9 AE |
188 | } |
189 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio); | |
190 | #endif /* CONFIG_BLOCK */ | |
191 | ||
192 | static void osd_req_op_cls_request_info_pagelist( | |
193 | struct ceph_osd_request *osd_req, | |
194 | unsigned int which, struct ceph_pagelist *pagelist) | |
195 | { | |
196 | struct ceph_osd_data *osd_data; | |
197 | ||
863c7eb5 | 198 | osd_data = osd_req_op_data(osd_req, which, cls, request_info); |
a4ce40a9 | 199 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
a4ce40a9 AE |
200 | } |
201 | ||
04017e29 AE |
202 | void osd_req_op_cls_request_data_pagelist( |
203 | struct ceph_osd_request *osd_req, | |
204 | unsigned int which, struct ceph_pagelist *pagelist) | |
205 | { | |
206 | struct ceph_osd_data *osd_data; | |
207 | ||
863c7eb5 | 208 | osd_data = osd_req_op_data(osd_req, which, cls, request_data); |
04017e29 | 209 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
bb873b53 ID |
210 | osd_req->r_ops[which].cls.indata_len += pagelist->length; |
211 | osd_req->r_ops[which].indata_len += pagelist->length; | |
04017e29 AE |
212 | } |
213 | EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist); | |
214 | ||
6c57b554 AE |
215 | void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req, |
216 | unsigned int which, struct page **pages, u64 length, | |
217 | u32 alignment, bool pages_from_pool, bool own_pages) | |
218 | { | |
219 | struct ceph_osd_data *osd_data; | |
220 | ||
221 | osd_data = osd_req_op_data(osd_req, which, cls, request_data); | |
222 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, | |
223 | pages_from_pool, own_pages); | |
bb873b53 ID |
224 | osd_req->r_ops[which].cls.indata_len += length; |
225 | osd_req->r_ops[which].indata_len += length; | |
6c57b554 AE |
226 | } |
227 | EXPORT_SYMBOL(osd_req_op_cls_request_data_pages); | |
228 | ||
a4ce40a9 AE |
229 | void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req, |
230 | unsigned int which, struct page **pages, u64 length, | |
231 | u32 alignment, bool pages_from_pool, bool own_pages) | |
232 | { | |
233 | struct ceph_osd_data *osd_data; | |
234 | ||
863c7eb5 | 235 | osd_data = osd_req_op_data(osd_req, which, cls, response_data); |
a4ce40a9 AE |
236 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, |
237 | pages_from_pool, own_pages); | |
a4ce40a9 AE |
238 | } |
239 | EXPORT_SYMBOL(osd_req_op_cls_response_data_pages); | |
240 | ||
23c08a9c AE |
241 | static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data) |
242 | { | |
243 | switch (osd_data->type) { | |
244 | case CEPH_OSD_DATA_TYPE_NONE: | |
245 | return 0; | |
246 | case CEPH_OSD_DATA_TYPE_PAGES: | |
247 | return osd_data->length; | |
248 | case CEPH_OSD_DATA_TYPE_PAGELIST: | |
249 | return (u64)osd_data->pagelist->length; | |
250 | #ifdef CONFIG_BLOCK | |
251 | case CEPH_OSD_DATA_TYPE_BIO: | |
252 | return (u64)osd_data->bio_length; | |
253 | #endif /* CONFIG_BLOCK */ | |
254 | default: | |
255 | WARN(true, "unrecognized data type %d\n", (int)osd_data->type); | |
256 | return 0; | |
257 | } | |
258 | } | |
259 | ||
c54d47bf AE |
260 | static void ceph_osd_data_release(struct ceph_osd_data *osd_data) |
261 | { | |
5476492f | 262 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) { |
c54d47bf AE |
263 | int num_pages; |
264 | ||
265 | num_pages = calc_pages_for((u64)osd_data->alignment, | |
266 | (u64)osd_data->length); | |
267 | ceph_release_page_vector(osd_data->pages, num_pages); | |
268 | } | |
5476492f AE |
269 | ceph_osd_data_init(osd_data); |
270 | } | |
271 | ||
272 | static void osd_req_op_data_release(struct ceph_osd_request *osd_req, | |
273 | unsigned int which) | |
274 | { | |
275 | struct ceph_osd_req_op *op; | |
276 | ||
277 | BUG_ON(which >= osd_req->r_num_ops); | |
278 | op = &osd_req->r_ops[which]; | |
279 | ||
280 | switch (op->op) { | |
281 | case CEPH_OSD_OP_READ: | |
282 | case CEPH_OSD_OP_WRITE: | |
e30b7577 | 283 | case CEPH_OSD_OP_WRITEFULL: |
5476492f AE |
284 | ceph_osd_data_release(&op->extent.osd_data); |
285 | break; | |
286 | case CEPH_OSD_OP_CALL: | |
287 | ceph_osd_data_release(&op->cls.request_info); | |
04017e29 | 288 | ceph_osd_data_release(&op->cls.request_data); |
5476492f AE |
289 | ceph_osd_data_release(&op->cls.response_data); |
290 | break; | |
d74b50be YZ |
291 | case CEPH_OSD_OP_SETXATTR: |
292 | case CEPH_OSD_OP_CMPXATTR: | |
293 | ceph_osd_data_release(&op->xattr.osd_data); | |
294 | break; | |
66ba609f YZ |
295 | case CEPH_OSD_OP_STAT: |
296 | ceph_osd_data_release(&op->raw_data_in); | |
297 | break; | |
5476492f AE |
298 | default: |
299 | break; | |
300 | } | |
c54d47bf AE |
301 | } |
302 | ||
63244fa1 ID |
303 | /* |
304 | * Assumes @t is zero-initialized. | |
305 | */ | |
306 | static void target_init(struct ceph_osd_request_target *t) | |
307 | { | |
308 | ceph_oid_init(&t->base_oid); | |
309 | ceph_oloc_init(&t->base_oloc); | |
310 | ceph_oid_init(&t->target_oid); | |
311 | ceph_oloc_init(&t->target_oloc); | |
312 | ||
313 | ceph_osds_init(&t->acting); | |
314 | ceph_osds_init(&t->up); | |
315 | t->size = -1; | |
316 | t->min_size = -1; | |
317 | ||
318 | t->osd = CEPH_HOMELESS_OSD; | |
319 | } | |
320 | ||
321 | static void target_destroy(struct ceph_osd_request_target *t) | |
322 | { | |
323 | ceph_oid_destroy(&t->base_oid); | |
324 | ceph_oid_destroy(&t->target_oid); | |
325 | } | |
326 | ||
f24e9980 SW |
327 | /* |
328 | * requests | |
329 | */ | |
9e94af20 | 330 | static void ceph_osdc_release_request(struct kref *kref) |
f24e9980 | 331 | { |
9e94af20 ID |
332 | struct ceph_osd_request *req = container_of(kref, |
333 | struct ceph_osd_request, r_kref); | |
5476492f | 334 | unsigned int which; |
415e49a9 | 335 | |
9e94af20 ID |
336 | dout("%s %p (r_request %p r_reply %p)\n", __func__, req, |
337 | req->r_request, req->r_reply); | |
6562d661 ID |
338 | WARN_ON(!RB_EMPTY_NODE(&req->r_node)); |
339 | WARN_ON(!list_empty(&req->r_req_lru_item)); | |
340 | WARN_ON(!list_empty(&req->r_osd_item)); | |
341 | WARN_ON(!list_empty(&req->r_linger_item)); | |
342 | WARN_ON(!list_empty(&req->r_linger_osd_item)); | |
343 | WARN_ON(req->r_osd); | |
9e94af20 | 344 | |
415e49a9 SW |
345 | if (req->r_request) |
346 | ceph_msg_put(req->r_request); | |
ace6d3a9 | 347 | if (req->r_reply) { |
8921d114 | 348 | ceph_msg_revoke_incoming(req->r_reply); |
ab8cb34a | 349 | ceph_msg_put(req->r_reply); |
ace6d3a9 | 350 | } |
0fff87ec | 351 | |
5476492f AE |
352 | for (which = 0; which < req->r_num_ops; which++) |
353 | osd_req_op_data_release(req, which); | |
0fff87ec | 354 | |
a66dd383 | 355 | target_destroy(&req->r_t); |
415e49a9 | 356 | ceph_put_snap_context(req->r_snapc); |
d30291b9 | 357 | |
415e49a9 SW |
358 | if (req->r_mempool) |
359 | mempool_free(req, req->r_osdc->req_mempool); | |
3f1af42a | 360 | else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS) |
5522ae0b | 361 | kmem_cache_free(ceph_osd_request_cache, req); |
3f1af42a ID |
362 | else |
363 | kfree(req); | |
f24e9980 | 364 | } |
9e94af20 ID |
365 | |
366 | void ceph_osdc_get_request(struct ceph_osd_request *req) | |
367 | { | |
368 | dout("%s %p (was %d)\n", __func__, req, | |
369 | atomic_read(&req->r_kref.refcount)); | |
370 | kref_get(&req->r_kref); | |
371 | } | |
372 | EXPORT_SYMBOL(ceph_osdc_get_request); | |
373 | ||
374 | void ceph_osdc_put_request(struct ceph_osd_request *req) | |
375 | { | |
3ed97d63 ID |
376 | if (req) { |
377 | dout("%s %p (was %d)\n", __func__, req, | |
378 | atomic_read(&req->r_kref.refcount)); | |
379 | kref_put(&req->r_kref, ceph_osdc_release_request); | |
380 | } | |
9e94af20 ID |
381 | } |
382 | EXPORT_SYMBOL(ceph_osdc_put_request); | |
68b4476b | 383 | |
3499e8a5 | 384 | struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, |
f24e9980 | 385 | struct ceph_snap_context *snapc, |
1b83bef2 | 386 | unsigned int num_ops, |
3499e8a5 | 387 | bool use_mempool, |
54a54007 | 388 | gfp_t gfp_flags) |
f24e9980 SW |
389 | { |
390 | struct ceph_osd_request *req; | |
1b83bef2 | 391 | |
f24e9980 | 392 | if (use_mempool) { |
3f1af42a | 393 | BUG_ON(num_ops > CEPH_OSD_SLAB_OPS); |
3499e8a5 | 394 | req = mempool_alloc(osdc->req_mempool, gfp_flags); |
3f1af42a ID |
395 | } else if (num_ops <= CEPH_OSD_SLAB_OPS) { |
396 | req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags); | |
f24e9980 | 397 | } else { |
3f1af42a ID |
398 | BUG_ON(num_ops > CEPH_OSD_MAX_OPS); |
399 | req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]), | |
400 | gfp_flags); | |
f24e9980 | 401 | } |
3f1af42a | 402 | if (unlikely(!req)) |
a79832f2 | 403 | return NULL; |
f24e9980 | 404 | |
3f1af42a ID |
405 | /* req only, each op is zeroed in _osd_req_op_init() */ |
406 | memset(req, 0, sizeof(*req)); | |
407 | ||
f24e9980 SW |
408 | req->r_osdc = osdc; |
409 | req->r_mempool = use_mempool; | |
79528734 | 410 | req->r_num_ops = num_ops; |
84127282 ID |
411 | req->r_snapid = CEPH_NOSNAP; |
412 | req->r_snapc = ceph_get_snap_context(snapc); | |
68b4476b | 413 | |
415e49a9 | 414 | kref_init(&req->r_kref); |
f24e9980 SW |
415 | init_completion(&req->r_completion); |
416 | init_completion(&req->r_safe_completion); | |
a978fa20 | 417 | RB_CLEAR_NODE(&req->r_node); |
f24e9980 | 418 | INIT_LIST_HEAD(&req->r_unsafe_item); |
a40c4f10 | 419 | INIT_LIST_HEAD(&req->r_linger_item); |
1d0326b1 | 420 | INIT_LIST_HEAD(&req->r_linger_osd_item); |
935b639a | 421 | INIT_LIST_HEAD(&req->r_req_lru_item); |
cd43045c SW |
422 | INIT_LIST_HEAD(&req->r_osd_item); |
423 | ||
a66dd383 | 424 | target_init(&req->r_t); |
22116525 | 425 | |
13d1ad16 ID |
426 | dout("%s req %p\n", __func__, req); |
427 | return req; | |
428 | } | |
429 | EXPORT_SYMBOL(ceph_osdc_alloc_request); | |
3f1af42a | 430 | |
13d1ad16 ID |
431 | int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp) |
432 | { | |
433 | struct ceph_osd_client *osdc = req->r_osdc; | |
434 | struct ceph_msg *msg; | |
435 | int msg_size; | |
c16e7869 | 436 | |
d30291b9 ID |
437 | WARN_ON(ceph_oid_empty(&req->r_base_oid)); |
438 | ||
13d1ad16 | 439 | /* create request message */ |
ae458f5a ID |
440 | msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */ |
441 | msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */ | |
442 | msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */ | |
443 | msg_size += 1 + 8 + 4 + 4; /* pgid */ | |
13d1ad16 ID |
444 | msg_size += 4 + req->r_base_oid.name_len; /* oid */ |
445 | msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op); | |
ae458f5a ID |
446 | msg_size += 8; /* snapid */ |
447 | msg_size += 8; /* snap_seq */ | |
13d1ad16 | 448 | msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0); |
ae458f5a ID |
449 | msg_size += 4; /* retry_attempt */ |
450 | ||
13d1ad16 | 451 | if (req->r_mempool) |
8f3bc053 | 452 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
f24e9980 | 453 | else |
13d1ad16 ID |
454 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true); |
455 | if (!msg) | |
456 | return -ENOMEM; | |
68b4476b | 457 | |
f24e9980 | 458 | memset(msg->front.iov_base, 0, msg->front.iov_len); |
3499e8a5 | 459 | req->r_request = msg; |
3499e8a5 | 460 | |
13d1ad16 ID |
461 | /* create reply message */ |
462 | msg_size = OSD_OPREPLY_FRONT_LEN; | |
711da55d ID |
463 | msg_size += req->r_base_oid.name_len; |
464 | msg_size += req->r_num_ops * sizeof(struct ceph_osd_op); | |
13d1ad16 ID |
465 | |
466 | if (req->r_mempool) | |
467 | msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); | |
468 | else | |
469 | msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true); | |
470 | if (!msg) | |
471 | return -ENOMEM; | |
472 | ||
473 | req->r_reply = msg; | |
474 | ||
475 | return 0; | |
3499e8a5 | 476 | } |
13d1ad16 | 477 | EXPORT_SYMBOL(ceph_osdc_alloc_messages); |
3499e8a5 | 478 | |
a8dd0a37 | 479 | static bool osd_req_opcode_valid(u16 opcode) |
68b4476b | 480 | { |
a8dd0a37 | 481 | switch (opcode) { |
70b5bfa3 ID |
482 | #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true; |
483 | __CEPH_FORALL_OSD_OPS(GENERATE_CASE) | |
484 | #undef GENERATE_CASE | |
a8dd0a37 AE |
485 | default: |
486 | return false; | |
487 | } | |
488 | } | |
489 | ||
33803f33 AE |
490 | /* |
491 | * This is an osd op init function for opcodes that have no data or | |
492 | * other information associated with them. It also serves as a | |
493 | * common init routine for all the other init functions, below. | |
494 | */ | |
c99d2d4a | 495 | static struct ceph_osd_req_op * |
49719778 | 496 | _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which, |
144cba14 | 497 | u16 opcode, u32 flags) |
33803f33 | 498 | { |
c99d2d4a AE |
499 | struct ceph_osd_req_op *op; |
500 | ||
501 | BUG_ON(which >= osd_req->r_num_ops); | |
33803f33 AE |
502 | BUG_ON(!osd_req_opcode_valid(opcode)); |
503 | ||
c99d2d4a | 504 | op = &osd_req->r_ops[which]; |
33803f33 | 505 | memset(op, 0, sizeof (*op)); |
33803f33 | 506 | op->op = opcode; |
144cba14 | 507 | op->flags = flags; |
c99d2d4a AE |
508 | |
509 | return op; | |
33803f33 AE |
510 | } |
511 | ||
49719778 | 512 | void osd_req_op_init(struct ceph_osd_request *osd_req, |
144cba14 | 513 | unsigned int which, u16 opcode, u32 flags) |
49719778 | 514 | { |
144cba14 | 515 | (void)_osd_req_op_init(osd_req, which, opcode, flags); |
49719778 AE |
516 | } |
517 | EXPORT_SYMBOL(osd_req_op_init); | |
518 | ||
c99d2d4a AE |
519 | void osd_req_op_extent_init(struct ceph_osd_request *osd_req, |
520 | unsigned int which, u16 opcode, | |
33803f33 AE |
521 | u64 offset, u64 length, |
522 | u64 truncate_size, u32 truncate_seq) | |
523 | { | |
144cba14 YZ |
524 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
525 | opcode, 0); | |
33803f33 AE |
526 | size_t payload_len = 0; |
527 | ||
ad7a60de | 528 | BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && |
e30b7577 ID |
529 | opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO && |
530 | opcode != CEPH_OSD_OP_TRUNCATE); | |
33803f33 | 531 | |
33803f33 AE |
532 | op->extent.offset = offset; |
533 | op->extent.length = length; | |
534 | op->extent.truncate_size = truncate_size; | |
535 | op->extent.truncate_seq = truncate_seq; | |
e30b7577 | 536 | if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL) |
33803f33 AE |
537 | payload_len += length; |
538 | ||
de2aa102 | 539 | op->indata_len = payload_len; |
33803f33 AE |
540 | } |
541 | EXPORT_SYMBOL(osd_req_op_extent_init); | |
542 | ||
c99d2d4a AE |
543 | void osd_req_op_extent_update(struct ceph_osd_request *osd_req, |
544 | unsigned int which, u64 length) | |
e5975c7c | 545 | { |
c99d2d4a AE |
546 | struct ceph_osd_req_op *op; |
547 | u64 previous; | |
548 | ||
549 | BUG_ON(which >= osd_req->r_num_ops); | |
550 | op = &osd_req->r_ops[which]; | |
551 | previous = op->extent.length; | |
e5975c7c AE |
552 | |
553 | if (length == previous) | |
554 | return; /* Nothing to do */ | |
555 | BUG_ON(length > previous); | |
556 | ||
557 | op->extent.length = length; | |
de2aa102 | 558 | op->indata_len -= previous - length; |
e5975c7c AE |
559 | } |
560 | EXPORT_SYMBOL(osd_req_op_extent_update); | |
561 | ||
2c63f49a YZ |
562 | void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req, |
563 | unsigned int which, u64 offset_inc) | |
564 | { | |
565 | struct ceph_osd_req_op *op, *prev_op; | |
566 | ||
567 | BUG_ON(which + 1 >= osd_req->r_num_ops); | |
568 | ||
569 | prev_op = &osd_req->r_ops[which]; | |
570 | op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags); | |
571 | /* dup previous one */ | |
572 | op->indata_len = prev_op->indata_len; | |
573 | op->outdata_len = prev_op->outdata_len; | |
574 | op->extent = prev_op->extent; | |
575 | /* adjust offset */ | |
576 | op->extent.offset += offset_inc; | |
577 | op->extent.length -= offset_inc; | |
578 | ||
579 | if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) | |
580 | op->indata_len -= offset_inc; | |
581 | } | |
582 | EXPORT_SYMBOL(osd_req_op_extent_dup_last); | |
583 | ||
c99d2d4a | 584 | void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which, |
04017e29 | 585 | u16 opcode, const char *class, const char *method) |
33803f33 | 586 | { |
144cba14 YZ |
587 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
588 | opcode, 0); | |
5f562df5 | 589 | struct ceph_pagelist *pagelist; |
33803f33 AE |
590 | size_t payload_len = 0; |
591 | size_t size; | |
592 | ||
593 | BUG_ON(opcode != CEPH_OSD_OP_CALL); | |
594 | ||
5f562df5 AE |
595 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); |
596 | BUG_ON(!pagelist); | |
597 | ceph_pagelist_init(pagelist); | |
598 | ||
33803f33 AE |
599 | op->cls.class_name = class; |
600 | size = strlen(class); | |
601 | BUG_ON(size > (size_t) U8_MAX); | |
602 | op->cls.class_len = size; | |
5f562df5 | 603 | ceph_pagelist_append(pagelist, class, size); |
33803f33 AE |
604 | payload_len += size; |
605 | ||
606 | op->cls.method_name = method; | |
607 | size = strlen(method); | |
608 | BUG_ON(size > (size_t) U8_MAX); | |
609 | op->cls.method_len = size; | |
5f562df5 | 610 | ceph_pagelist_append(pagelist, method, size); |
33803f33 AE |
611 | payload_len += size; |
612 | ||
a4ce40a9 | 613 | osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist); |
5f562df5 | 614 | |
de2aa102 | 615 | op->indata_len = payload_len; |
33803f33 AE |
616 | } |
617 | EXPORT_SYMBOL(osd_req_op_cls_init); | |
8c042b0d | 618 | |
d74b50be YZ |
619 | int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which, |
620 | u16 opcode, const char *name, const void *value, | |
621 | size_t size, u8 cmp_op, u8 cmp_mode) | |
622 | { | |
144cba14 YZ |
623 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
624 | opcode, 0); | |
d74b50be YZ |
625 | struct ceph_pagelist *pagelist; |
626 | size_t payload_len; | |
627 | ||
628 | BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR); | |
629 | ||
630 | pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS); | |
631 | if (!pagelist) | |
632 | return -ENOMEM; | |
633 | ||
634 | ceph_pagelist_init(pagelist); | |
635 | ||
636 | payload_len = strlen(name); | |
637 | op->xattr.name_len = payload_len; | |
638 | ceph_pagelist_append(pagelist, name, payload_len); | |
639 | ||
640 | op->xattr.value_len = size; | |
641 | ceph_pagelist_append(pagelist, value, size); | |
642 | payload_len += size; | |
643 | ||
644 | op->xattr.cmp_op = cmp_op; | |
645 | op->xattr.cmp_mode = cmp_mode; | |
646 | ||
647 | ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist); | |
de2aa102 | 648 | op->indata_len = payload_len; |
d74b50be YZ |
649 | return 0; |
650 | } | |
651 | EXPORT_SYMBOL(osd_req_op_xattr_init); | |
652 | ||
c99d2d4a AE |
653 | void osd_req_op_watch_init(struct ceph_osd_request *osd_req, |
654 | unsigned int which, u16 opcode, | |
33803f33 AE |
655 | u64 cookie, u64 version, int flag) |
656 | { | |
144cba14 YZ |
657 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
658 | opcode, 0); | |
33803f33 | 659 | |
c99d2d4a | 660 | BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH); |
33803f33 AE |
661 | |
662 | op->watch.cookie = cookie; | |
9ef1ee5a | 663 | op->watch.ver = version; |
33803f33 | 664 | if (opcode == CEPH_OSD_OP_WATCH && flag) |
c99d2d4a | 665 | op->watch.flag = (u8)1; |
33803f33 AE |
666 | } |
667 | EXPORT_SYMBOL(osd_req_op_watch_init); | |
668 | ||
c647b8a8 ID |
669 | void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req, |
670 | unsigned int which, | |
671 | u64 expected_object_size, | |
672 | u64 expected_write_size) | |
673 | { | |
674 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, | |
144cba14 YZ |
675 | CEPH_OSD_OP_SETALLOCHINT, |
676 | 0); | |
c647b8a8 ID |
677 | |
678 | op->alloc_hint.expected_object_size = expected_object_size; | |
679 | op->alloc_hint.expected_write_size = expected_write_size; | |
680 | ||
681 | /* | |
682 | * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed | |
683 | * not worth a feature bit. Set FAILOK per-op flag to make | |
684 | * sure older osds don't trip over an unsupported opcode. | |
685 | */ | |
686 | op->flags |= CEPH_OSD_OP_FLAG_FAILOK; | |
687 | } | |
688 | EXPORT_SYMBOL(osd_req_op_alloc_hint_init); | |
689 | ||
90af3602 | 690 | static void ceph_osdc_msg_data_add(struct ceph_msg *msg, |
ec9123c5 AE |
691 | struct ceph_osd_data *osd_data) |
692 | { | |
693 | u64 length = ceph_osd_data_length(osd_data); | |
694 | ||
695 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { | |
696 | BUG_ON(length > (u64) SIZE_MAX); | |
697 | if (length) | |
90af3602 | 698 | ceph_msg_data_add_pages(msg, osd_data->pages, |
ec9123c5 AE |
699 | length, osd_data->alignment); |
700 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { | |
701 | BUG_ON(!length); | |
90af3602 | 702 | ceph_msg_data_add_pagelist(msg, osd_data->pagelist); |
ec9123c5 AE |
703 | #ifdef CONFIG_BLOCK |
704 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { | |
90af3602 | 705 | ceph_msg_data_add_bio(msg, osd_data->bio, length); |
ec9123c5 AE |
706 | #endif |
707 | } else { | |
708 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); | |
709 | } | |
710 | } | |
711 | ||
bb873b53 ID |
712 | static u32 osd_req_encode_op(struct ceph_osd_op *dst, |
713 | const struct ceph_osd_req_op *src) | |
a8dd0a37 | 714 | { |
a8dd0a37 AE |
715 | if (WARN_ON(!osd_req_opcode_valid(src->op))) { |
716 | pr_err("unrecognized osd opcode %d\n", src->op); | |
717 | ||
718 | return 0; | |
719 | } | |
720 | ||
721 | switch (src->op) { | |
722 | case CEPH_OSD_OP_STAT: | |
723 | break; | |
724 | case CEPH_OSD_OP_READ: | |
725 | case CEPH_OSD_OP_WRITE: | |
e30b7577 | 726 | case CEPH_OSD_OP_WRITEFULL: |
ad7a60de | 727 | case CEPH_OSD_OP_ZERO: |
ad7a60de | 728 | case CEPH_OSD_OP_TRUNCATE: |
a8dd0a37 AE |
729 | dst->extent.offset = cpu_to_le64(src->extent.offset); |
730 | dst->extent.length = cpu_to_le64(src->extent.length); | |
731 | dst->extent.truncate_size = | |
732 | cpu_to_le64(src->extent.truncate_size); | |
733 | dst->extent.truncate_seq = | |
734 | cpu_to_le32(src->extent.truncate_seq); | |
735 | break; | |
736 | case CEPH_OSD_OP_CALL: | |
a8dd0a37 AE |
737 | dst->cls.class_len = src->cls.class_len; |
738 | dst->cls.method_len = src->cls.method_len; | |
bb873b53 | 739 | dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); |
a8dd0a37 AE |
740 | break; |
741 | case CEPH_OSD_OP_STARTSYNC: | |
742 | break; | |
743 | case CEPH_OSD_OP_NOTIFY_ACK: | |
744 | case CEPH_OSD_OP_WATCH: | |
745 | dst->watch.cookie = cpu_to_le64(src->watch.cookie); | |
746 | dst->watch.ver = cpu_to_le64(src->watch.ver); | |
747 | dst->watch.flag = src->watch.flag; | |
748 | break; | |
c647b8a8 ID |
749 | case CEPH_OSD_OP_SETALLOCHINT: |
750 | dst->alloc_hint.expected_object_size = | |
751 | cpu_to_le64(src->alloc_hint.expected_object_size); | |
752 | dst->alloc_hint.expected_write_size = | |
753 | cpu_to_le64(src->alloc_hint.expected_write_size); | |
754 | break; | |
d74b50be YZ |
755 | case CEPH_OSD_OP_SETXATTR: |
756 | case CEPH_OSD_OP_CMPXATTR: | |
757 | dst->xattr.name_len = cpu_to_le32(src->xattr.name_len); | |
758 | dst->xattr.value_len = cpu_to_le32(src->xattr.value_len); | |
759 | dst->xattr.cmp_op = src->xattr.cmp_op; | |
760 | dst->xattr.cmp_mode = src->xattr.cmp_mode; | |
d74b50be | 761 | break; |
864e9197 YZ |
762 | case CEPH_OSD_OP_CREATE: |
763 | case CEPH_OSD_OP_DELETE: | |
764 | break; | |
a8dd0a37 | 765 | default: |
4c46459c | 766 | pr_err("unsupported osd opcode %s\n", |
8f63ca2d | 767 | ceph_osd_op_name(src->op)); |
4c46459c | 768 | WARN_ON(1); |
a8dd0a37 AE |
769 | |
770 | return 0; | |
68b4476b | 771 | } |
7b25bf5f | 772 | |
a8dd0a37 | 773 | dst->op = cpu_to_le16(src->op); |
7b25bf5f | 774 | dst->flags = cpu_to_le32(src->flags); |
de2aa102 | 775 | dst->payload_len = cpu_to_le32(src->indata_len); |
175face2 | 776 | |
bb873b53 | 777 | return src->indata_len; |
68b4476b YS |
778 | } |
779 | ||
3499e8a5 YS |
780 | /* |
781 | * build new request AND message, calculate layout, and adjust file | |
782 | * extent as needed. | |
783 | * | |
784 | * if the file was recently truncated, we include information about its | |
785 | * old and new size so that the object can be updated appropriately. (we | |
786 | * avoid synchronously deleting truncated objects because it's slow.) | |
787 | * | |
788 | * if @do_sync, include a 'startsync' command so that the osd will flush | |
789 | * data quickly. | |
790 | */ | |
791 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, | |
792 | struct ceph_file_layout *layout, | |
793 | struct ceph_vino vino, | |
715e4cd4 YZ |
794 | u64 off, u64 *plen, |
795 | unsigned int which, int num_ops, | |
3499e8a5 YS |
796 | int opcode, int flags, |
797 | struct ceph_snap_context *snapc, | |
3499e8a5 YS |
798 | u32 truncate_seq, |
799 | u64 truncate_size, | |
153e5167 | 800 | bool use_mempool) |
3499e8a5 | 801 | { |
68b4476b | 802 | struct ceph_osd_request *req; |
75d1c941 AE |
803 | u64 objnum = 0; |
804 | u64 objoff = 0; | |
805 | u64 objlen = 0; | |
6816282d | 806 | int r; |
68b4476b | 807 | |
ad7a60de | 808 | BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && |
864e9197 YZ |
809 | opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE && |
810 | opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE); | |
68b4476b | 811 | |
acead002 | 812 | req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool, |
ae7ca4a3 | 813 | GFP_NOFS); |
13d1ad16 ID |
814 | if (!req) { |
815 | r = -ENOMEM; | |
816 | goto fail; | |
817 | } | |
79528734 | 818 | |
3499e8a5 | 819 | /* calculate max write size */ |
a19dadfb | 820 | r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen); |
13d1ad16 ID |
821 | if (r) |
822 | goto fail; | |
a19dadfb | 823 | |
864e9197 | 824 | if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) { |
144cba14 | 825 | osd_req_op_init(req, which, opcode, 0); |
864e9197 YZ |
826 | } else { |
827 | u32 object_size = le32_to_cpu(layout->fl_object_size); | |
828 | u32 object_base = off - objoff; | |
829 | if (!(truncate_seq == 1 && truncate_size == -1ULL)) { | |
830 | if (truncate_size <= object_base) { | |
831 | truncate_size = 0; | |
832 | } else { | |
833 | truncate_size -= object_base; | |
834 | if (truncate_size > object_size) | |
835 | truncate_size = object_size; | |
836 | } | |
ccca4e37 | 837 | } |
715e4cd4 | 838 | osd_req_op_extent_init(req, which, opcode, objoff, objlen, |
864e9197 YZ |
839 | truncate_size, truncate_seq); |
840 | } | |
d18d1e28 | 841 | |
bb873b53 | 842 | req->r_flags = flags; |
3c972c95 | 843 | req->r_base_oloc.pool = ceph_file_layout_pg_pool(*layout); |
d30291b9 | 844 | ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum); |
dbe0fc41 | 845 | |
bb873b53 ID |
846 | req->r_snapid = vino.snap; |
847 | if (flags & CEPH_OSD_FLAG_WRITE) | |
848 | req->r_data_offset = off; | |
849 | ||
13d1ad16 ID |
850 | r = ceph_osdc_alloc_messages(req, GFP_NOFS); |
851 | if (r) | |
852 | goto fail; | |
853 | ||
f24e9980 | 854 | return req; |
13d1ad16 ID |
855 | |
856 | fail: | |
857 | ceph_osdc_put_request(req); | |
858 | return ERR_PTR(r); | |
f24e9980 | 859 | } |
3d14c5d2 | 860 | EXPORT_SYMBOL(ceph_osdc_new_request); |
f24e9980 SW |
861 | |
862 | /* | |
863 | * We keep osd requests in an rbtree, sorted by ->r_tid. | |
864 | */ | |
fcd00b68 | 865 | DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node) |
f24e9980 | 866 | |
0247a0cf ID |
867 | static bool osd_homeless(struct ceph_osd *osd) |
868 | { | |
869 | return osd->o_osd == CEPH_HOMELESS_OSD; | |
870 | } | |
871 | ||
f24e9980 SW |
872 | static struct ceph_osd_request * |
873 | __lookup_request_ge(struct ceph_osd_client *osdc, | |
874 | u64 tid) | |
875 | { | |
876 | struct ceph_osd_request *req; | |
877 | struct rb_node *n = osdc->requests.rb_node; | |
878 | ||
879 | while (n) { | |
880 | req = rb_entry(n, struct ceph_osd_request, r_node); | |
881 | if (tid < req->r_tid) { | |
882 | if (!n->rb_left) | |
883 | return req; | |
884 | n = n->rb_left; | |
885 | } else if (tid > req->r_tid) { | |
886 | n = n->rb_right; | |
887 | } else { | |
888 | return req; | |
889 | } | |
890 | } | |
891 | return NULL; | |
892 | } | |
893 | ||
2cc6128a ID |
894 | static void __kick_linger_request(struct ceph_osd_request *req) |
895 | { | |
896 | struct ceph_osd_client *osdc = req->r_osdc; | |
897 | struct ceph_osd *osd = req->r_osd; | |
898 | ||
899 | /* | |
900 | * Linger requests need to be resent with a new tid to avoid | |
901 | * the dup op detection logic on the OSDs. Achieve this with | |
902 | * a re-register dance instead of open-coding. | |
903 | */ | |
904 | ceph_osdc_get_request(req); | |
905 | if (!list_empty(&req->r_linger_item)) | |
906 | __unregister_linger_request(osdc, req); | |
907 | else | |
908 | __unregister_request(osdc, req); | |
909 | __register_request(osdc, req); | |
910 | ceph_osdc_put_request(req); | |
911 | ||
912 | /* | |
913 | * Unless request has been registered as both normal and | |
914 | * lingering, __unregister{,_linger}_request clears r_osd. | |
915 | * However, here we need to preserve r_osd to make sure we | |
916 | * requeue on the same OSD. | |
917 | */ | |
918 | WARN_ON(req->r_osd || !osd); | |
919 | req->r_osd = osd; | |
920 | ||
921 | dout("%s requeueing %p tid %llu\n", __func__, req, req->r_tid); | |
922 | __enqueue_request(req); | |
923 | } | |
924 | ||
6f6c7006 SW |
925 | /* |
926 | * Resubmit requests pending on the given osd. | |
927 | */ | |
928 | static void __kick_osd_requests(struct ceph_osd_client *osdc, | |
929 | struct ceph_osd *osd) | |
930 | { | |
a40c4f10 | 931 | struct ceph_osd_request *req, *nreq; |
e02493c0 | 932 | LIST_HEAD(resend); |
2cc6128a | 933 | LIST_HEAD(resend_linger); |
6f6c7006 SW |
934 | int err; |
935 | ||
2cc6128a | 936 | dout("%s osd%d\n", __func__, osd->o_osd); |
6f6c7006 | 937 | err = __reset_osd(osdc, osd); |
685a7555 | 938 | if (err) |
6f6c7006 | 939 | return; |
2cc6128a | 940 | |
e02493c0 AE |
941 | /* |
942 | * Build up a list of requests to resend by traversing the | |
943 | * osd's list of requests. Requests for a given object are | |
944 | * sent in tid order, and that is also the order they're | |
945 | * kept on this list. Therefore all requests that are in | |
946 | * flight will be found first, followed by all requests that | |
947 | * have not yet been sent. And to resend requests while | |
948 | * preserving this order we will want to put any sent | |
949 | * requests back on the front of the osd client's unsent | |
950 | * list. | |
951 | * | |
952 | * So we build a separate ordered list of already-sent | |
953 | * requests for the affected osd and splice it onto the | |
954 | * front of the osd client's unsent list. Once we've seen a | |
955 | * request that has not yet been sent we're done. Those | |
956 | * requests are already sitting right where they belong. | |
957 | */ | |
6f6c7006 | 958 | list_for_each_entry(req, &osd->o_requests, r_osd_item) { |
e02493c0 AE |
959 | if (!req->r_sent) |
960 | break; | |
2cc6128a ID |
961 | |
962 | if (!req->r_linger) { | |
963 | dout("%s requeueing %p tid %llu\n", __func__, req, | |
964 | req->r_tid); | |
965 | list_move_tail(&req->r_req_lru_item, &resend); | |
a40c4f10 | 966 | req->r_flags |= CEPH_OSD_FLAG_RETRY; |
2cc6128a ID |
967 | } else { |
968 | list_move_tail(&req->r_req_lru_item, &resend_linger); | |
969 | } | |
a40c4f10 | 970 | } |
e02493c0 | 971 | list_splice(&resend, &osdc->req_unsent); |
a40c4f10 | 972 | |
e02493c0 | 973 | /* |
2cc6128a ID |
974 | * Both registered and not yet registered linger requests are |
975 | * enqueued with a new tid on the same OSD. We add/move them | |
976 | * to req_unsent/o_requests at the end to keep things in tid | |
977 | * order. | |
e02493c0 | 978 | */ |
a40c4f10 | 979 | list_for_each_entry_safe(req, nreq, &osd->o_linger_requests, |
1d0326b1 | 980 | r_linger_osd_item) { |
2cc6128a ID |
981 | WARN_ON(!list_empty(&req->r_req_lru_item)); |
982 | __kick_linger_request(req); | |
6f6c7006 | 983 | } |
2cc6128a ID |
984 | |
985 | list_for_each_entry_safe(req, nreq, &resend_linger, r_req_lru_item) | |
986 | __kick_linger_request(req); | |
6f6c7006 SW |
987 | } |
988 | ||
f24e9980 | 989 | /* |
81b024e7 | 990 | * If the osd connection drops, we need to resubmit all requests. |
f24e9980 SW |
991 | */ |
992 | static void osd_reset(struct ceph_connection *con) | |
993 | { | |
994 | struct ceph_osd *osd = con->private; | |
995 | struct ceph_osd_client *osdc; | |
996 | ||
997 | if (!osd) | |
998 | return; | |
999 | dout("osd_reset osd%d\n", osd->o_osd); | |
1000 | osdc = osd->o_osdc; | |
f24e9980 | 1001 | down_read(&osdc->map_sem); |
83aff95e SW |
1002 | mutex_lock(&osdc->request_mutex); |
1003 | __kick_osd_requests(osdc, osd); | |
f9d25199 | 1004 | __send_queued(osdc); |
83aff95e | 1005 | mutex_unlock(&osdc->request_mutex); |
f24e9980 SW |
1006 | up_read(&osdc->map_sem); |
1007 | } | |
1008 | ||
0247a0cf ID |
1009 | /* |
1010 | * Assumes @osd is zero-initialized. | |
1011 | */ | |
1012 | static void osd_init(struct ceph_osd *osd) | |
1013 | { | |
1014 | atomic_set(&osd->o_ref, 1); | |
1015 | RB_CLEAR_NODE(&osd->o_node); | |
1016 | INIT_LIST_HEAD(&osd->o_requests); | |
1017 | INIT_LIST_HEAD(&osd->o_linger_requests); | |
1018 | INIT_LIST_HEAD(&osd->o_osd_lru); | |
1019 | INIT_LIST_HEAD(&osd->o_keepalive_item); | |
1020 | osd->o_incarnation = 1; | |
1021 | } | |
1022 | ||
1023 | static void osd_cleanup(struct ceph_osd *osd) | |
1024 | { | |
1025 | WARN_ON(!RB_EMPTY_NODE(&osd->o_node)); | |
1026 | WARN_ON(!list_empty(&osd->o_requests)); | |
1027 | WARN_ON(!list_empty(&osd->o_linger_requests)); | |
1028 | WARN_ON(!list_empty(&osd->o_osd_lru)); | |
1029 | WARN_ON(!list_empty(&osd->o_keepalive_item)); | |
1030 | ||
1031 | if (osd->o_auth.authorizer) { | |
1032 | WARN_ON(osd_homeless(osd)); | |
1033 | ceph_auth_destroy_authorizer(osd->o_auth.authorizer); | |
1034 | } | |
1035 | } | |
1036 | ||
f24e9980 SW |
1037 | /* |
1038 | * Track open sessions with osds. | |
1039 | */ | |
e10006f8 | 1040 | static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) |
f24e9980 SW |
1041 | { |
1042 | struct ceph_osd *osd; | |
1043 | ||
0247a0cf ID |
1044 | WARN_ON(onum == CEPH_HOMELESS_OSD); |
1045 | ||
7a28f59b | 1046 | osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL); |
0247a0cf | 1047 | osd_init(osd); |
f24e9980 | 1048 | osd->o_osdc = osdc; |
e10006f8 | 1049 | osd->o_osd = onum; |
f24e9980 | 1050 | |
b7a9e5dd | 1051 | ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); |
4e7a5dcd | 1052 | |
f24e9980 SW |
1053 | return osd; |
1054 | } | |
1055 | ||
1056 | static struct ceph_osd *get_osd(struct ceph_osd *osd) | |
1057 | { | |
1058 | if (atomic_inc_not_zero(&osd->o_ref)) { | |
1059 | dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1, | |
1060 | atomic_read(&osd->o_ref)); | |
1061 | return osd; | |
1062 | } else { | |
1063 | dout("get_osd %p FAIL\n", osd); | |
1064 | return NULL; | |
1065 | } | |
1066 | } | |
1067 | ||
1068 | static void put_osd(struct ceph_osd *osd) | |
1069 | { | |
1070 | dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref), | |
1071 | atomic_read(&osd->o_ref) - 1); | |
b28ec2f3 | 1072 | if (atomic_dec_and_test(&osd->o_ref)) { |
0247a0cf | 1073 | osd_cleanup(osd); |
f24e9980 | 1074 | kfree(osd); |
79494d1b | 1075 | } |
f24e9980 SW |
1076 | } |
1077 | ||
fcd00b68 ID |
1078 | DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node) |
1079 | ||
f24e9980 SW |
1080 | /* |
1081 | * remove an osd from our map | |
1082 | */ | |
f5a2041b | 1083 | static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
f24e9980 | 1084 | { |
7eb71e03 | 1085 | dout("%s %p osd%d\n", __func__, osd, osd->o_osd); |
cc9f1f51 ID |
1086 | WARN_ON(!list_empty(&osd->o_requests)); |
1087 | WARN_ON(!list_empty(&osd->o_linger_requests)); | |
7c6e6fc5 | 1088 | |
f5a2041b | 1089 | list_del_init(&osd->o_osd_lru); |
fcd00b68 | 1090 | erase_osd(&osdc->osds, osd); |
7eb71e03 ID |
1091 | } |
1092 | ||
1093 | static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) | |
1094 | { | |
1095 | dout("%s %p osd%d\n", __func__, osd, osd->o_osd); | |
1096 | ||
1097 | if (!RB_EMPTY_NODE(&osd->o_node)) { | |
1098 | ceph_con_close(&osd->o_con); | |
1099 | __remove_osd(osdc, osd); | |
1100 | put_osd(osd); | |
1101 | } | |
f24e9980 SW |
1102 | } |
1103 | ||
9dd2845c | 1104 | static void __move_osd_to_lru(struct ceph_osd *osd) |
f5a2041b | 1105 | { |
9dd2845c ID |
1106 | struct ceph_osd_client *osdc = osd->o_osdc; |
1107 | ||
1108 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
f5a2041b | 1109 | BUG_ON(!list_empty(&osd->o_osd_lru)); |
bbf37ec3 | 1110 | |
9dd2845c | 1111 | spin_lock(&osdc->osd_lru_lock); |
f5a2041b | 1112 | list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); |
9dd2845c ID |
1113 | spin_unlock(&osdc->osd_lru_lock); |
1114 | ||
a319bf56 | 1115 | osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl; |
f5a2041b YS |
1116 | } |
1117 | ||
9dd2845c | 1118 | static void maybe_move_osd_to_lru(struct ceph_osd *osd) |
bbf37ec3 | 1119 | { |
bbf37ec3 ID |
1120 | if (list_empty(&osd->o_requests) && |
1121 | list_empty(&osd->o_linger_requests)) | |
9dd2845c | 1122 | __move_osd_to_lru(osd); |
bbf37ec3 ID |
1123 | } |
1124 | ||
f5a2041b YS |
1125 | static void __remove_osd_from_lru(struct ceph_osd *osd) |
1126 | { | |
9dd2845c ID |
1127 | struct ceph_osd_client *osdc = osd->o_osdc; |
1128 | ||
1129 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
1130 | ||
1131 | spin_lock(&osdc->osd_lru_lock); | |
f5a2041b YS |
1132 | if (!list_empty(&osd->o_osd_lru)) |
1133 | list_del_init(&osd->o_osd_lru); | |
9dd2845c | 1134 | spin_unlock(&osdc->osd_lru_lock); |
f5a2041b YS |
1135 | } |
1136 | ||
f24e9980 SW |
1137 | /* |
1138 | * reset osd connect | |
1139 | */ | |
f5a2041b | 1140 | static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd) |
f24e9980 | 1141 | { |
c3acb181 | 1142 | struct ceph_entity_addr *peer_addr; |
f24e9980 | 1143 | |
f5a2041b | 1144 | dout("__reset_osd %p osd%d\n", osd, osd->o_osd); |
a40c4f10 YS |
1145 | if (list_empty(&osd->o_requests) && |
1146 | list_empty(&osd->o_linger_requests)) { | |
7eb71e03 | 1147 | remove_osd(osdc, osd); |
c3acb181 AE |
1148 | return -ENODEV; |
1149 | } | |
1150 | ||
1151 | peer_addr = &osdc->osdmap->osd_addr[osd->o_osd]; | |
1152 | if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && | |
1153 | !ceph_con_opened(&osd->o_con)) { | |
1154 | struct ceph_osd_request *req; | |
1155 | ||
0b4af2e8 ID |
1156 | dout("osd addr hasn't changed and connection never opened, " |
1157 | "letting msgr retry\n"); | |
87b315a5 SW |
1158 | /* touch each r_stamp for handle_timeout()'s benfit */ |
1159 | list_for_each_entry(req, &osd->o_requests, r_osd_item) | |
1160 | req->r_stamp = jiffies; | |
c3acb181 AE |
1161 | |
1162 | return -EAGAIN; | |
f24e9980 | 1163 | } |
c3acb181 AE |
1164 | |
1165 | ceph_con_close(&osd->o_con); | |
1166 | ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); | |
1167 | osd->o_incarnation++; | |
1168 | ||
1169 | return 0; | |
f24e9980 SW |
1170 | } |
1171 | ||
f24e9980 SW |
1172 | /* |
1173 | * Register request, assign tid. If this is the first request, set up | |
1174 | * the timeout event. | |
1175 | */ | |
a40c4f10 YS |
1176 | static void __register_request(struct ceph_osd_client *osdc, |
1177 | struct ceph_osd_request *req) | |
f24e9980 | 1178 | { |
f24e9980 | 1179 | req->r_tid = ++osdc->last_tid; |
6df058c0 | 1180 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); |
77f38e0e | 1181 | dout("__register_request %p tid %lld\n", req, req->r_tid); |
fcd00b68 | 1182 | insert_request(&osdc->requests, req); |
f24e9980 SW |
1183 | ceph_osdc_get_request(req); |
1184 | osdc->num_requests++; | |
a40c4f10 YS |
1185 | } |
1186 | ||
f24e9980 SW |
1187 | /* |
1188 | * called under osdc->request_mutex | |
1189 | */ | |
1190 | static void __unregister_request(struct ceph_osd_client *osdc, | |
1191 | struct ceph_osd_request *req) | |
1192 | { | |
35f9f8a0 SW |
1193 | if (RB_EMPTY_NODE(&req->r_node)) { |
1194 | dout("__unregister_request %p tid %lld not registered\n", | |
1195 | req, req->r_tid); | |
1196 | return; | |
1197 | } | |
1198 | ||
f24e9980 | 1199 | dout("__unregister_request %p tid %lld\n", req, req->r_tid); |
fcd00b68 | 1200 | erase_request(&osdc->requests, req); |
f24e9980 SW |
1201 | osdc->num_requests--; |
1202 | ||
0ba6478d SW |
1203 | if (req->r_osd) { |
1204 | /* make sure the original request isn't in flight. */ | |
6740a845 | 1205 | ceph_msg_revoke(req->r_request); |
0ba6478d SW |
1206 | |
1207 | list_del_init(&req->r_osd_item); | |
9dd2845c | 1208 | maybe_move_osd_to_lru(req->r_osd); |
4f23409e | 1209 | if (list_empty(&req->r_linger_osd_item)) |
a40c4f10 | 1210 | req->r_osd = NULL; |
0ba6478d | 1211 | } |
f24e9980 | 1212 | |
7d5f2481 | 1213 | list_del_init(&req->r_req_lru_item); |
f24e9980 | 1214 | ceph_osdc_put_request(req); |
f24e9980 SW |
1215 | } |
1216 | ||
1217 | /* | |
1218 | * Cancel a previously queued request message | |
1219 | */ | |
1220 | static void __cancel_request(struct ceph_osd_request *req) | |
1221 | { | |
6bc18876 | 1222 | if (req->r_sent && req->r_osd) { |
6740a845 | 1223 | ceph_msg_revoke(req->r_request); |
f24e9980 SW |
1224 | req->r_sent = 0; |
1225 | } | |
1226 | } | |
1227 | ||
a40c4f10 YS |
1228 | static void __register_linger_request(struct ceph_osd_client *osdc, |
1229 | struct ceph_osd_request *req) | |
1230 | { | |
af593064 ID |
1231 | dout("%s %p tid %llu\n", __func__, req, req->r_tid); |
1232 | WARN_ON(!req->r_linger); | |
1233 | ||
96e4dac6 | 1234 | ceph_osdc_get_request(req); |
a40c4f10 | 1235 | list_add_tail(&req->r_linger_item, &osdc->req_linger); |
6194ea89 | 1236 | if (req->r_osd) |
1d0326b1 | 1237 | list_add_tail(&req->r_linger_osd_item, |
6194ea89 | 1238 | &req->r_osd->o_linger_requests); |
a40c4f10 YS |
1239 | } |
1240 | ||
1241 | static void __unregister_linger_request(struct ceph_osd_client *osdc, | |
1242 | struct ceph_osd_request *req) | |
1243 | { | |
af593064 ID |
1244 | WARN_ON(!req->r_linger); |
1245 | ||
1246 | if (list_empty(&req->r_linger_item)) { | |
1247 | dout("%s %p tid %llu not registered\n", __func__, req, | |
1248 | req->r_tid); | |
1249 | return; | |
1250 | } | |
1251 | ||
1252 | dout("%s %p tid %llu\n", __func__, req, req->r_tid); | |
61c74035 | 1253 | list_del_init(&req->r_linger_item); |
af593064 | 1254 | |
a40c4f10 | 1255 | if (req->r_osd) { |
1d0326b1 | 1256 | list_del_init(&req->r_linger_osd_item); |
9dd2845c | 1257 | maybe_move_osd_to_lru(req->r_osd); |
fbdb9190 SW |
1258 | if (list_empty(&req->r_osd_item)) |
1259 | req->r_osd = NULL; | |
a40c4f10 | 1260 | } |
96e4dac6 | 1261 | ceph_osdc_put_request(req); |
a40c4f10 YS |
1262 | } |
1263 | ||
a40c4f10 YS |
1264 | void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc, |
1265 | struct ceph_osd_request *req) | |
1266 | { | |
1267 | if (!req->r_linger) { | |
1268 | dout("set_request_linger %p\n", req); | |
1269 | req->r_linger = 1; | |
a40c4f10 YS |
1270 | } |
1271 | } | |
1272 | EXPORT_SYMBOL(ceph_osdc_set_request_linger); | |
1273 | ||
63244fa1 ID |
1274 | static bool __pool_full(struct ceph_pg_pool_info *pi) |
1275 | { | |
1276 | return pi->flags & CEPH_POOL_FLAG_FULL; | |
1277 | } | |
1278 | ||
42c1b124 ID |
1279 | static bool have_pool_full(struct ceph_osd_client *osdc) |
1280 | { | |
1281 | struct rb_node *n; | |
1282 | ||
1283 | for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { | |
1284 | struct ceph_pg_pool_info *pi = | |
1285 | rb_entry(n, struct ceph_pg_pool_info, node); | |
1286 | ||
1287 | if (__pool_full(pi)) | |
1288 | return true; | |
1289 | } | |
1290 | ||
1291 | return false; | |
1292 | } | |
1293 | ||
d29adb34 JD |
1294 | /* |
1295 | * Returns whether a request should be blocked from being sent | |
1296 | * based on the current osdmap and osd_client settings. | |
1297 | * | |
1298 | * Caller should hold map_sem for read. | |
1299 | */ | |
63244fa1 ID |
1300 | static bool target_should_be_paused(struct ceph_osd_client *osdc, |
1301 | const struct ceph_osd_request_target *t, | |
1302 | struct ceph_pg_pool_info *pi) | |
1303 | { | |
1304 | bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD); | |
1305 | bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) || | |
1306 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) || | |
1307 | __pool_full(pi); | |
1308 | ||
1309 | WARN_ON(pi->id != t->base_oloc.pool); | |
1310 | return (t->flags & CEPH_OSD_FLAG_READ && pauserd) || | |
1311 | (t->flags & CEPH_OSD_FLAG_WRITE && pausewr); | |
1312 | } | |
1313 | ||
63244fa1 ID |
1314 | enum calc_target_result { |
1315 | CALC_TARGET_NO_ACTION = 0, | |
1316 | CALC_TARGET_NEED_RESEND, | |
1317 | CALC_TARGET_POOL_DNE, | |
1318 | }; | |
1319 | ||
1320 | static enum calc_target_result calc_target(struct ceph_osd_client *osdc, | |
1321 | struct ceph_osd_request_target *t, | |
1322 | u32 *last_force_resend, | |
1323 | bool any_change) | |
1324 | { | |
1325 | struct ceph_pg_pool_info *pi; | |
1326 | struct ceph_pg pgid, last_pgid; | |
1327 | struct ceph_osds up, acting; | |
1328 | bool force_resend = false; | |
1329 | bool need_check_tiering = false; | |
1330 | bool need_resend = false; | |
1331 | bool sort_bitwise = ceph_osdmap_flag(osdc->osdmap, | |
1332 | CEPH_OSDMAP_SORTBITWISE); | |
1333 | enum calc_target_result ct_res; | |
1334 | int ret; | |
1335 | ||
1336 | pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool); | |
1337 | if (!pi) { | |
1338 | t->osd = CEPH_HOMELESS_OSD; | |
1339 | ct_res = CALC_TARGET_POOL_DNE; | |
1340 | goto out; | |
1341 | } | |
1342 | ||
1343 | if (osdc->osdmap->epoch == pi->last_force_request_resend) { | |
1344 | if (last_force_resend && | |
1345 | *last_force_resend < pi->last_force_request_resend) { | |
1346 | *last_force_resend = pi->last_force_request_resend; | |
1347 | force_resend = true; | |
1348 | } else if (!last_force_resend) { | |
1349 | force_resend = true; | |
1350 | } | |
1351 | } | |
1352 | if (ceph_oid_empty(&t->target_oid) || force_resend) { | |
1353 | ceph_oid_copy(&t->target_oid, &t->base_oid); | |
1354 | need_check_tiering = true; | |
1355 | } | |
1356 | if (ceph_oloc_empty(&t->target_oloc) || force_resend) { | |
1357 | ceph_oloc_copy(&t->target_oloc, &t->base_oloc); | |
1358 | need_check_tiering = true; | |
1359 | } | |
1360 | ||
1361 | if (need_check_tiering && | |
1362 | (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) { | |
1363 | if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0) | |
1364 | t->target_oloc.pool = pi->read_tier; | |
1365 | if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0) | |
1366 | t->target_oloc.pool = pi->write_tier; | |
1367 | } | |
1368 | ||
1369 | ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid, | |
1370 | &t->target_oloc, &pgid); | |
1371 | if (ret) { | |
1372 | WARN_ON(ret != -ENOENT); | |
1373 | t->osd = CEPH_HOMELESS_OSD; | |
1374 | ct_res = CALC_TARGET_POOL_DNE; | |
1375 | goto out; | |
1376 | } | |
1377 | last_pgid.pool = pgid.pool; | |
1378 | last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask); | |
1379 | ||
1380 | ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting); | |
1381 | if (any_change && | |
1382 | ceph_is_new_interval(&t->acting, | |
1383 | &acting, | |
1384 | &t->up, | |
1385 | &up, | |
1386 | t->size, | |
1387 | pi->size, | |
1388 | t->min_size, | |
1389 | pi->min_size, | |
1390 | t->pg_num, | |
1391 | pi->pg_num, | |
1392 | t->sort_bitwise, | |
1393 | sort_bitwise, | |
1394 | &last_pgid)) | |
1395 | force_resend = true; | |
1396 | ||
1397 | if (t->paused && !target_should_be_paused(osdc, t, pi)) { | |
1398 | t->paused = false; | |
1399 | need_resend = true; | |
1400 | } | |
1401 | ||
1402 | if (ceph_pg_compare(&t->pgid, &pgid) || | |
1403 | ceph_osds_changed(&t->acting, &acting, any_change) || | |
1404 | force_resend) { | |
1405 | t->pgid = pgid; /* struct */ | |
1406 | ceph_osds_copy(&t->acting, &acting); | |
1407 | ceph_osds_copy(&t->up, &up); | |
1408 | t->size = pi->size; | |
1409 | t->min_size = pi->min_size; | |
1410 | t->pg_num = pi->pg_num; | |
1411 | t->pg_num_mask = pi->pg_num_mask; | |
1412 | t->sort_bitwise = sort_bitwise; | |
1413 | ||
1414 | t->osd = acting.primary; | |
1415 | need_resend = true; | |
1416 | } | |
1417 | ||
1418 | ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION; | |
1419 | out: | |
1420 | dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd); | |
1421 | return ct_res; | |
1422 | } | |
1423 | ||
f671b581 ID |
1424 | static void __enqueue_request(struct ceph_osd_request *req) |
1425 | { | |
1426 | struct ceph_osd_client *osdc = req->r_osdc; | |
1427 | ||
1428 | dout("%s %p tid %llu to osd%d\n", __func__, req, req->r_tid, | |
1429 | req->r_osd ? req->r_osd->o_osd : -1); | |
1430 | ||
1431 | if (req->r_osd) { | |
1432 | __remove_osd_from_lru(req->r_osd); | |
1433 | list_add_tail(&req->r_osd_item, &req->r_osd->o_requests); | |
1434 | list_move_tail(&req->r_req_lru_item, &osdc->req_unsent); | |
1435 | } else { | |
1436 | list_move_tail(&req->r_req_lru_item, &osdc->req_notarget); | |
1437 | } | |
1438 | } | |
1439 | ||
f24e9980 SW |
1440 | /* |
1441 | * Pick an osd (the first 'up' osd in the pg), allocate the osd struct | |
1442 | * (as needed), and set the request r_osd appropriately. If there is | |
25985edc | 1443 | * no up osd, set r_osd to NULL. Move the request to the appropriate list |
6f6c7006 | 1444 | * (unsent, homeless) or leave on in-flight lru. |
f24e9980 SW |
1445 | * |
1446 | * Return 0 if unchanged, 1 if changed, or negative on error. | |
1447 | * | |
1448 | * Caller should hold map_sem for read and request_mutex. | |
1449 | */ | |
6f6c7006 | 1450 | static int __map_request(struct ceph_osd_client *osdc, |
38d6453c | 1451 | struct ceph_osd_request *req, int force_resend) |
f24e9980 | 1452 | { |
a66dd383 | 1453 | enum calc_target_result ct_res; |
f24e9980 | 1454 | int err; |
f24e9980 | 1455 | |
6f6c7006 | 1456 | dout("map_request %p tid %lld\n", req, req->r_tid); |
17a13e40 | 1457 | |
a66dd383 ID |
1458 | ct_res = calc_target(osdc, &req->r_t, NULL, force_resend); |
1459 | switch (ct_res) { | |
1460 | case CALC_TARGET_POOL_DNE: | |
6f6c7006 | 1461 | list_move(&req->r_req_lru_item, &osdc->req_notarget); |
a66dd383 ID |
1462 | return -EIO; |
1463 | case CALC_TARGET_NO_ACTION: | |
f24e9980 | 1464 | return 0; /* no change */ |
a66dd383 ID |
1465 | default: |
1466 | BUG_ON(ct_res != CALC_TARGET_NEED_RESEND); | |
1467 | } | |
f24e9980 | 1468 | |
5b191d99 | 1469 | dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n", |
a66dd383 | 1470 | req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, req->r_t.osd, |
f24e9980 SW |
1471 | req->r_osd ? req->r_osd->o_osd : -1); |
1472 | ||
1473 | if (req->r_osd) { | |
1474 | __cancel_request(req); | |
1475 | list_del_init(&req->r_osd_item); | |
a390de02 | 1476 | list_del_init(&req->r_linger_osd_item); |
f24e9980 SW |
1477 | req->r_osd = NULL; |
1478 | } | |
1479 | ||
a66dd383 ID |
1480 | req->r_osd = lookup_osd(&osdc->osds, req->r_t.osd); |
1481 | if (!req->r_osd && req->r_t.osd >= 0) { | |
c99eb1c7 | 1482 | err = -ENOMEM; |
a66dd383 | 1483 | req->r_osd = create_osd(osdc, req->r_t.osd); |
6f6c7006 SW |
1484 | if (!req->r_osd) { |
1485 | list_move(&req->r_req_lru_item, &osdc->req_notarget); | |
c99eb1c7 | 1486 | goto out; |
6f6c7006 | 1487 | } |
f24e9980 | 1488 | |
6f3bfd45 | 1489 | dout("map_request osd %p is osd%d\n", req->r_osd, |
a66dd383 | 1490 | req->r_osd->o_osd); |
fcd00b68 | 1491 | insert_osd(&osdc->osds, req->r_osd); |
f24e9980 | 1492 | |
b7a9e5dd | 1493 | ceph_con_open(&req->r_osd->o_con, |
a66dd383 ID |
1494 | CEPH_ENTITY_TYPE_OSD, req->r_osd->o_osd, |
1495 | &osdc->osdmap->osd_addr[req->r_osd->o_osd]); | |
f24e9980 SW |
1496 | } |
1497 | ||
f671b581 | 1498 | __enqueue_request(req); |
d85b7056 | 1499 | err = 1; /* osd or pg changed */ |
f24e9980 SW |
1500 | |
1501 | out: | |
f24e9980 SW |
1502 | return err; |
1503 | } | |
1504 | ||
bb873b53 ID |
1505 | static void setup_request_data(struct ceph_osd_request *req, |
1506 | struct ceph_msg *msg) | |
f24e9980 | 1507 | { |
bb873b53 ID |
1508 | u32 data_len = 0; |
1509 | int i; | |
1510 | ||
1511 | if (!list_empty(&msg->data)) | |
1512 | return; | |
f24e9980 | 1513 | |
bb873b53 ID |
1514 | WARN_ON(msg->data_length); |
1515 | for (i = 0; i < req->r_num_ops; i++) { | |
1516 | struct ceph_osd_req_op *op = &req->r_ops[i]; | |
1517 | ||
1518 | switch (op->op) { | |
1519 | /* request */ | |
1520 | case CEPH_OSD_OP_WRITE: | |
1521 | case CEPH_OSD_OP_WRITEFULL: | |
1522 | WARN_ON(op->indata_len != op->extent.length); | |
1523 | ceph_osdc_msg_data_add(msg, &op->extent.osd_data); | |
1524 | break; | |
1525 | case CEPH_OSD_OP_SETXATTR: | |
1526 | case CEPH_OSD_OP_CMPXATTR: | |
1527 | WARN_ON(op->indata_len != op->xattr.name_len + | |
1528 | op->xattr.value_len); | |
1529 | ceph_osdc_msg_data_add(msg, &op->xattr.osd_data); | |
1530 | break; | |
1531 | ||
1532 | /* reply */ | |
1533 | case CEPH_OSD_OP_STAT: | |
1534 | ceph_osdc_msg_data_add(req->r_reply, | |
1535 | &op->raw_data_in); | |
1536 | break; | |
1537 | case CEPH_OSD_OP_READ: | |
1538 | ceph_osdc_msg_data_add(req->r_reply, | |
1539 | &op->extent.osd_data); | |
1540 | break; | |
1541 | ||
1542 | /* both */ | |
1543 | case CEPH_OSD_OP_CALL: | |
1544 | WARN_ON(op->indata_len != op->cls.class_len + | |
1545 | op->cls.method_len + | |
1546 | op->cls.indata_len); | |
1547 | ceph_osdc_msg_data_add(msg, &op->cls.request_info); | |
1548 | /* optional, can be NONE */ | |
1549 | ceph_osdc_msg_data_add(msg, &op->cls.request_data); | |
1550 | /* optional, can be NONE */ | |
1551 | ceph_osdc_msg_data_add(req->r_reply, | |
1552 | &op->cls.response_data); | |
1553 | break; | |
1554 | } | |
1555 | ||
1556 | data_len += op->indata_len; | |
1557 | } | |
1b83bef2 | 1558 | |
bb873b53 ID |
1559 | WARN_ON(data_len != msg->data_length); |
1560 | } | |
1561 | ||
1562 | static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg) | |
1563 | { | |
1564 | void *p = msg->front.iov_base; | |
1565 | void *const end = p + msg->front_alloc_len; | |
1566 | u32 data_len = 0; | |
1567 | int i; | |
1568 | ||
1569 | if (req->r_flags & CEPH_OSD_FLAG_WRITE) { | |
1570 | /* snapshots aren't writeable */ | |
1571 | WARN_ON(req->r_snapid != CEPH_NOSNAP); | |
1572 | } else { | |
1573 | WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec || | |
1574 | req->r_data_offset || req->r_snapc); | |
1575 | } | |
1576 | ||
1577 | setup_request_data(req, msg); | |
1578 | ||
1579 | ceph_encode_32(&p, 1); /* client_inc, always 1 */ | |
1580 | ceph_encode_32(&p, req->r_osdc->osdmap->epoch); | |
1581 | ceph_encode_32(&p, req->r_flags); | |
1582 | ceph_encode_timespec(p, &req->r_mtime); | |
1583 | p += sizeof(struct ceph_timespec); | |
1584 | /* aka reassert_version */ | |
1585 | memcpy(p, &req->r_replay_version, sizeof(req->r_replay_version)); | |
1586 | p += sizeof(req->r_replay_version); | |
1587 | ||
1588 | /* oloc */ | |
1589 | ceph_encode_8(&p, 4); | |
1590 | ceph_encode_8(&p, 4); | |
1591 | ceph_encode_32(&p, 8 + 4 + 4); | |
1592 | ceph_encode_64(&p, req->r_t.target_oloc.pool); | |
1593 | ceph_encode_32(&p, -1); /* preferred */ | |
1594 | ceph_encode_32(&p, 0); /* key len */ | |
1595 | ||
1596 | /* pgid */ | |
1597 | ceph_encode_8(&p, 1); | |
a66dd383 ID |
1598 | ceph_encode_64(&p, req->r_t.pgid.pool); |
1599 | ceph_encode_32(&p, req->r_t.pgid.seed); | |
bb873b53 | 1600 | ceph_encode_32(&p, -1); /* preferred */ |
2169aea6 | 1601 | |
bb873b53 ID |
1602 | /* oid */ |
1603 | ceph_encode_32(&p, req->r_t.target_oid.name_len); | |
1604 | memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len); | |
1605 | p += req->r_t.target_oid.name_len; | |
f24e9980 | 1606 | |
bb873b53 ID |
1607 | /* ops, can imply data */ |
1608 | ceph_encode_16(&p, req->r_num_ops); | |
1609 | for (i = 0; i < req->r_num_ops; i++) { | |
1610 | data_len += osd_req_encode_op(p, &req->r_ops[i]); | |
1611 | p += sizeof(struct ceph_osd_op); | |
1612 | } | |
26be8808 | 1613 | |
bb873b53 ID |
1614 | ceph_encode_64(&p, req->r_snapid); /* snapid */ |
1615 | if (req->r_snapc) { | |
1616 | ceph_encode_64(&p, req->r_snapc->seq); | |
1617 | ceph_encode_32(&p, req->r_snapc->num_snaps); | |
1618 | for (i = 0; i < req->r_snapc->num_snaps; i++) | |
1619 | ceph_encode_64(&p, req->r_snapc->snaps[i]); | |
1620 | } else { | |
1621 | ceph_encode_64(&p, 0); /* snap_seq */ | |
1622 | ceph_encode_32(&p, 0); /* snaps len */ | |
1623 | } | |
1624 | ||
1625 | ceph_encode_32(&p, req->r_attempts); /* retry_attempt */ | |
1626 | ||
1627 | BUG_ON(p > end); | |
1628 | msg->front.iov_len = p - msg->front.iov_base; | |
1629 | msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */ | |
1630 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
1631 | msg->hdr.data_len = cpu_to_le32(data_len); | |
1632 | /* | |
1633 | * The header "data_off" is a hint to the receiver allowing it | |
1634 | * to align received data into its buffers such that there's no | |
1635 | * need to re-copy it before writing it to disk (direct I/O). | |
1636 | */ | |
1637 | msg->hdr.data_off = cpu_to_le16(req->r_data_offset); | |
26be8808 | 1638 | |
bb873b53 ID |
1639 | dout("%s req %p oid %*pE oid_len %d front %zu data %u\n", __func__, |
1640 | req, req->r_t.target_oid.name_len, req->r_t.target_oid.name, | |
1641 | req->r_t.target_oid.name_len, msg->front.iov_len, data_len); | |
1642 | } | |
1643 | ||
1644 | /* | |
1645 | * @req has to be assigned a tid and registered. | |
1646 | */ | |
1647 | static void send_request(struct ceph_osd_request *req) | |
1648 | { | |
1649 | struct ceph_osd *osd = req->r_osd; | |
1650 | ||
1651 | WARN_ON(osd->o_osd != req->r_t.osd); | |
1652 | ||
1653 | req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR; | |
1654 | if (req->r_attempts) | |
1655 | req->r_flags |= CEPH_OSD_FLAG_RETRY; | |
1656 | else | |
1657 | WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY); | |
1658 | ||
1659 | encode_request(req, req->r_request); | |
1660 | ||
1661 | dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n", | |
1662 | __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, | |
1663 | req->r_t.osd, req->r_flags, req->r_attempts); | |
1664 | ||
1665 | req->r_t.paused = false; | |
1666 | req->r_stamp = jiffies; | |
1667 | req->r_attempts++; | |
1668 | ||
1669 | req->r_sent = osd->o_incarnation; | |
1670 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); | |
1671 | ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request)); | |
f24e9980 SW |
1672 | } |
1673 | ||
6f6c7006 SW |
1674 | /* |
1675 | * Send any requests in the queue (req_unsent). | |
1676 | */ | |
f9d25199 | 1677 | static void __send_queued(struct ceph_osd_client *osdc) |
6f6c7006 SW |
1678 | { |
1679 | struct ceph_osd_request *req, *tmp; | |
1680 | ||
f9d25199 | 1681 | dout("__send_queued\n"); |
bb873b53 ID |
1682 | list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item) { |
1683 | list_move_tail(&req->r_req_lru_item, &osdc->req_lru); | |
1684 | send_request(req); | |
1685 | } | |
6f6c7006 SW |
1686 | } |
1687 | ||
42c1b124 ID |
1688 | static void maybe_request_map(struct ceph_osd_client *osdc) |
1689 | { | |
1690 | bool continuous = false; | |
1691 | ||
1692 | WARN_ON(!osdc->osdmap->epoch); | |
1693 | ||
1694 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) || | |
1695 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) || | |
1696 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR)) { | |
1697 | dout("%s osdc %p continuous\n", __func__, osdc); | |
1698 | continuous = true; | |
1699 | } else { | |
1700 | dout("%s osdc %p onetime\n", __func__, osdc); | |
1701 | } | |
1702 | ||
1703 | if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP, | |
1704 | osdc->osdmap->epoch + 1, continuous)) | |
1705 | ceph_monc_renew_subs(&osdc->client->monc); | |
1706 | } | |
1707 | ||
0bbfdfe8 ID |
1708 | /* |
1709 | * Caller should hold map_sem for read and request_mutex. | |
1710 | */ | |
1711 | static int __ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
1712 | struct ceph_osd_request *req, | |
1713 | bool nofail) | |
1714 | { | |
1715 | int rc; | |
1716 | ||
1717 | __register_request(osdc, req); | |
1718 | req->r_sent = 0; | |
1719 | req->r_got_reply = 0; | |
1720 | rc = __map_request(osdc, req, 0); | |
1721 | if (rc < 0) { | |
1722 | if (nofail) { | |
1723 | dout("osdc_start_request failed map, " | |
1724 | " will retry %lld\n", req->r_tid); | |
1725 | rc = 0; | |
1726 | } else { | |
1727 | __unregister_request(osdc, req); | |
1728 | } | |
1729 | return rc; | |
1730 | } | |
1731 | ||
1732 | if (req->r_osd == NULL) { | |
1733 | dout("send_request %p no up osds in pg\n", req); | |
1734 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1735 | } else { | |
1736 | __send_queued(osdc); | |
1737 | } | |
1738 | ||
1739 | return 0; | |
1740 | } | |
1741 | ||
fe5da05e ID |
1742 | static void __complete_request(struct ceph_osd_request *req) |
1743 | { | |
1744 | if (req->r_callback) | |
1745 | req->r_callback(req); | |
1746 | else | |
1747 | complete_all(&req->r_completion); | |
1748 | } | |
1749 | ||
f24e9980 | 1750 | /* |
fbca9635 ID |
1751 | * Timeout callback, called every N seconds. When 1 or more OSD |
1752 | * requests has been active for more than N seconds, we send a keepalive | |
1753 | * (tag + timestamp) to its OSD to ensure any communications channel | |
1754 | * reset is detected. | |
f24e9980 SW |
1755 | */ |
1756 | static void handle_timeout(struct work_struct *work) | |
1757 | { | |
1758 | struct ceph_osd_client *osdc = | |
1759 | container_of(work, struct ceph_osd_client, timeout_work.work); | |
a319bf56 | 1760 | struct ceph_options *opts = osdc->client->options; |
83aff95e | 1761 | struct ceph_osd_request *req; |
f24e9980 | 1762 | struct ceph_osd *osd; |
422d2cb8 | 1763 | struct list_head slow_osds; |
f24e9980 SW |
1764 | dout("timeout\n"); |
1765 | down_read(&osdc->map_sem); | |
1766 | ||
1767 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
1768 | ||
1769 | mutex_lock(&osdc->request_mutex); | |
f24e9980 | 1770 | |
422d2cb8 YS |
1771 | /* |
1772 | * ping osds that are a bit slow. this ensures that if there | |
1773 | * is a break in the TCP connection we will notice, and reopen | |
1774 | * a connection with that osd (from the fault callback). | |
1775 | */ | |
1776 | INIT_LIST_HEAD(&slow_osds); | |
1777 | list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) { | |
a319bf56 ID |
1778 | if (time_before(jiffies, |
1779 | req->r_stamp + opts->osd_keepalive_timeout)) | |
422d2cb8 YS |
1780 | break; |
1781 | ||
1782 | osd = req->r_osd; | |
1783 | BUG_ON(!osd); | |
1784 | dout(" tid %llu is slow, will send keepalive on osd%d\n", | |
f24e9980 | 1785 | req->r_tid, osd->o_osd); |
422d2cb8 YS |
1786 | list_move_tail(&osd->o_keepalive_item, &slow_osds); |
1787 | } | |
1788 | while (!list_empty(&slow_osds)) { | |
1789 | osd = list_entry(slow_osds.next, struct ceph_osd, | |
1790 | o_keepalive_item); | |
1791 | list_del_init(&osd->o_keepalive_item); | |
f24e9980 SW |
1792 | ceph_con_keepalive(&osd->o_con); |
1793 | } | |
1794 | ||
f9d25199 | 1795 | __send_queued(osdc); |
f24e9980 | 1796 | mutex_unlock(&osdc->request_mutex); |
f24e9980 | 1797 | up_read(&osdc->map_sem); |
fbca9635 ID |
1798 | |
1799 | schedule_delayed_work(&osdc->timeout_work, | |
1800 | osdc->client->options->osd_keepalive_timeout); | |
f24e9980 SW |
1801 | } |
1802 | ||
f5a2041b YS |
1803 | static void handle_osds_timeout(struct work_struct *work) |
1804 | { | |
1805 | struct ceph_osd_client *osdc = | |
1806 | container_of(work, struct ceph_osd_client, | |
1807 | osds_timeout_work.work); | |
a319bf56 | 1808 | unsigned long delay = osdc->client->options->osd_idle_ttl / 4; |
42a2c09f | 1809 | struct ceph_osd *osd, *nosd; |
f5a2041b | 1810 | |
42a2c09f | 1811 | dout("%s osdc %p\n", __func__, osdc); |
f5a2041b | 1812 | down_read(&osdc->map_sem); |
42a2c09f ID |
1813 | mutex_lock(&osdc->request_mutex); |
1814 | ||
1815 | list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { | |
1816 | if (time_before(jiffies, osd->lru_ttl)) | |
1817 | break; | |
1818 | ||
1819 | remove_osd(osdc, osd); | |
1820 | } | |
f5a2041b | 1821 | |
42a2c09f ID |
1822 | mutex_unlock(&osdc->request_mutex); |
1823 | up_read(&osdc->map_sem); | |
f5a2041b YS |
1824 | schedule_delayed_work(&osdc->osds_timeout_work, |
1825 | round_jiffies_relative(delay)); | |
1826 | } | |
1827 | ||
205ee118 ID |
1828 | static int ceph_oloc_decode(void **p, void *end, |
1829 | struct ceph_object_locator *oloc) | |
1830 | { | |
1831 | u8 struct_v, struct_cv; | |
1832 | u32 len; | |
1833 | void *struct_end; | |
1834 | int ret = 0; | |
1835 | ||
1836 | ceph_decode_need(p, end, 1 + 1 + 4, e_inval); | |
1837 | struct_v = ceph_decode_8(p); | |
1838 | struct_cv = ceph_decode_8(p); | |
1839 | if (struct_v < 3) { | |
1840 | pr_warn("got v %d < 3 cv %d of ceph_object_locator\n", | |
1841 | struct_v, struct_cv); | |
1842 | goto e_inval; | |
1843 | } | |
1844 | if (struct_cv > 6) { | |
1845 | pr_warn("got v %d cv %d > 6 of ceph_object_locator\n", | |
1846 | struct_v, struct_cv); | |
1847 | goto e_inval; | |
1848 | } | |
1849 | len = ceph_decode_32(p); | |
1850 | ceph_decode_need(p, end, len, e_inval); | |
1851 | struct_end = *p + len; | |
1852 | ||
1853 | oloc->pool = ceph_decode_64(p); | |
1854 | *p += 4; /* skip preferred */ | |
1855 | ||
1856 | len = ceph_decode_32(p); | |
1857 | if (len > 0) { | |
1858 | pr_warn("ceph_object_locator::key is set\n"); | |
1859 | goto e_inval; | |
1860 | } | |
1861 | ||
1862 | if (struct_v >= 5) { | |
1863 | len = ceph_decode_32(p); | |
1864 | if (len > 0) { | |
1865 | pr_warn("ceph_object_locator::nspace is set\n"); | |
1866 | goto e_inval; | |
1867 | } | |
1868 | } | |
1869 | ||
1870 | if (struct_v >= 6) { | |
1871 | s64 hash = ceph_decode_64(p); | |
1872 | if (hash != -1) { | |
1873 | pr_warn("ceph_object_locator::hash is set\n"); | |
1874 | goto e_inval; | |
1875 | } | |
1876 | } | |
1877 | ||
1878 | /* skip the rest */ | |
1879 | *p = struct_end; | |
1880 | out: | |
1881 | return ret; | |
1882 | ||
1883 | e_inval: | |
1884 | ret = -EINVAL; | |
1885 | goto out; | |
1886 | } | |
1887 | ||
1888 | static int ceph_redirect_decode(void **p, void *end, | |
1889 | struct ceph_request_redirect *redir) | |
1890 | { | |
1891 | u8 struct_v, struct_cv; | |
1892 | u32 len; | |
1893 | void *struct_end; | |
1894 | int ret; | |
1895 | ||
1896 | ceph_decode_need(p, end, 1 + 1 + 4, e_inval); | |
1897 | struct_v = ceph_decode_8(p); | |
1898 | struct_cv = ceph_decode_8(p); | |
1899 | if (struct_cv > 1) { | |
1900 | pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n", | |
1901 | struct_v, struct_cv); | |
1902 | goto e_inval; | |
1903 | } | |
1904 | len = ceph_decode_32(p); | |
1905 | ceph_decode_need(p, end, len, e_inval); | |
1906 | struct_end = *p + len; | |
1907 | ||
1908 | ret = ceph_oloc_decode(p, end, &redir->oloc); | |
1909 | if (ret) | |
1910 | goto out; | |
1911 | ||
1912 | len = ceph_decode_32(p); | |
1913 | if (len > 0) { | |
1914 | pr_warn("ceph_request_redirect::object_name is set\n"); | |
1915 | goto e_inval; | |
1916 | } | |
1917 | ||
1918 | len = ceph_decode_32(p); | |
1919 | *p += len; /* skip osd_instructions */ | |
1920 | ||
1921 | /* skip the rest */ | |
1922 | *p = struct_end; | |
1923 | out: | |
1924 | return ret; | |
1925 | ||
1926 | e_inval: | |
1927 | ret = -EINVAL; | |
1928 | goto out; | |
1929 | } | |
1930 | ||
fe5da05e ID |
1931 | struct MOSDOpReply { |
1932 | struct ceph_pg pgid; | |
1933 | u64 flags; | |
1934 | int result; | |
1935 | u32 epoch; | |
1936 | int num_ops; | |
1937 | u32 outdata_len[CEPH_OSD_MAX_OPS]; | |
1938 | s32 rval[CEPH_OSD_MAX_OPS]; | |
1939 | int retry_attempt; | |
1940 | struct ceph_eversion replay_version; | |
1941 | u64 user_version; | |
1942 | struct ceph_request_redirect redirect; | |
1943 | }; | |
25845472 | 1944 | |
fe5da05e | 1945 | static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m) |
f24e9980 | 1946 | { |
fe5da05e ID |
1947 | void *p = msg->front.iov_base; |
1948 | void *const end = p + msg->front.iov_len; | |
1949 | u16 version = le16_to_cpu(msg->hdr.version); | |
1950 | struct ceph_eversion bad_replay_version; | |
b0b31a8f | 1951 | u8 decode_redir; |
fe5da05e ID |
1952 | u32 len; |
1953 | int ret; | |
1954 | int i; | |
1b83bef2 | 1955 | |
fe5da05e ID |
1956 | ceph_decode_32_safe(&p, end, len, e_inval); |
1957 | ceph_decode_need(&p, end, len, e_inval); | |
1958 | p += len; /* skip oid */ | |
1b83bef2 | 1959 | |
fe5da05e ID |
1960 | ret = ceph_decode_pgid(&p, end, &m->pgid); |
1961 | if (ret) | |
1962 | return ret; | |
1b83bef2 | 1963 | |
fe5da05e ID |
1964 | ceph_decode_64_safe(&p, end, m->flags, e_inval); |
1965 | ceph_decode_32_safe(&p, end, m->result, e_inval); | |
1966 | ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval); | |
1967 | memcpy(&bad_replay_version, p, sizeof(bad_replay_version)); | |
1968 | p += sizeof(bad_replay_version); | |
1969 | ceph_decode_32_safe(&p, end, m->epoch, e_inval); | |
1b83bef2 | 1970 | |
fe5da05e ID |
1971 | ceph_decode_32_safe(&p, end, m->num_ops, e_inval); |
1972 | if (m->num_ops > ARRAY_SIZE(m->outdata_len)) | |
1973 | goto e_inval; | |
1b83bef2 | 1974 | |
fe5da05e ID |
1975 | ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op), |
1976 | e_inval); | |
1977 | for (i = 0; i < m->num_ops; i++) { | |
1b83bef2 | 1978 | struct ceph_osd_op *op = p; |
1b83bef2 | 1979 | |
fe5da05e | 1980 | m->outdata_len[i] = le32_to_cpu(op->payload_len); |
1b83bef2 SW |
1981 | p += sizeof(*op); |
1982 | } | |
1b83bef2 | 1983 | |
fe5da05e ID |
1984 | ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval); |
1985 | for (i = 0; i < m->num_ops; i++) | |
1986 | ceph_decode_32_safe(&p, end, m->rval[i], e_inval); | |
f24e9980 | 1987 | |
fe5da05e ID |
1988 | if (version >= 5) { |
1989 | ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval); | |
1990 | memcpy(&m->replay_version, p, sizeof(m->replay_version)); | |
1991 | p += sizeof(m->replay_version); | |
1992 | ceph_decode_64_safe(&p, end, m->user_version, e_inval); | |
1993 | } else { | |
1994 | m->replay_version = bad_replay_version; /* struct */ | |
1995 | m->user_version = le64_to_cpu(m->replay_version.version); | |
1996 | } | |
eb845ff1 | 1997 | |
fe5da05e ID |
1998 | if (version >= 6) { |
1999 | if (version >= 7) | |
2000 | ceph_decode_8_safe(&p, end, decode_redir, e_inval); | |
b0b31a8f ID |
2001 | else |
2002 | decode_redir = 1; | |
2003 | } else { | |
2004 | decode_redir = 0; | |
2005 | } | |
2006 | ||
2007 | if (decode_redir) { | |
fe5da05e ID |
2008 | ret = ceph_redirect_decode(&p, end, &m->redirect); |
2009 | if (ret) | |
2010 | return ret; | |
205ee118 | 2011 | } else { |
fe5da05e | 2012 | ceph_oloc_init(&m->redirect.oloc); |
205ee118 | 2013 | } |
f24e9980 | 2014 | |
fe5da05e ID |
2015 | return 0; |
2016 | ||
2017 | e_inval: | |
2018 | return -EINVAL; | |
2019 | } | |
2020 | ||
2021 | /* | |
2022 | * We are done with @req if | |
2023 | * - @m is a safe reply, or | |
2024 | * - @m is an unsafe reply and we didn't want a safe one | |
2025 | */ | |
2026 | static bool done_request(const struct ceph_osd_request *req, | |
2027 | const struct MOSDOpReply *m) | |
2028 | { | |
2029 | return (m->result < 0 || | |
2030 | (m->flags & CEPH_OSD_FLAG_ONDISK) || | |
2031 | !(req->r_flags & CEPH_OSD_FLAG_ONDISK)); | |
2032 | } | |
205ee118 | 2033 | |
fe5da05e ID |
2034 | /* |
2035 | * handle osd op reply. either call the callback if it is specified, | |
2036 | * or do the completion to wake up the waiting thread. | |
2037 | * | |
2038 | * ->r_unsafe_callback is set? yes no | |
2039 | * | |
2040 | * first reply is OK (needed r_cb/r_completion, r_cb/r_completion, | |
2041 | * any or needed/got safe) r_safe_completion r_safe_completion | |
2042 | * | |
2043 | * first reply is unsafe r_unsafe_cb(true) (nothing) | |
2044 | * | |
2045 | * when we get the safe reply r_unsafe_cb(false), r_cb/r_completion, | |
2046 | * r_safe_completion r_safe_completion | |
2047 | */ | |
2048 | static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
2049 | { | |
2050 | struct ceph_osd_request *req; | |
2051 | struct MOSDOpReply m; | |
2052 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
2053 | u32 data_len = 0; | |
2054 | bool already_acked; | |
2055 | int ret; | |
2056 | int i; | |
2057 | ||
2058 | dout("%s msg %p tid %llu\n", __func__, msg, tid); | |
2059 | ||
2060 | down_read(&osdc->map_sem); | |
2061 | mutex_lock(&osdc->request_mutex); | |
2062 | req = lookup_request(&osdc->requests, tid); | |
2063 | if (!req) { | |
2064 | dout("%s no tid %llu\n", __func__, tid); | |
2065 | goto out_unlock; | |
2066 | } | |
2067 | ceph_osdc_get_request(req); | |
2068 | ||
2069 | ret = decode_MOSDOpReply(msg, &m); | |
2070 | if (ret) { | |
2071 | pr_err("failed to decode MOSDOpReply for tid %llu: %d\n", | |
2072 | req->r_tid, ret); | |
2073 | ceph_msg_dump(msg); | |
2074 | goto fail_request; | |
2075 | } | |
2076 | dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n", | |
2077 | __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed, | |
2078 | m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch), | |
2079 | le64_to_cpu(m.replay_version.version), m.user_version); | |
2080 | ||
2081 | if (m.retry_attempt >= 0) { | |
2082 | if (m.retry_attempt != req->r_attempts - 1) { | |
2083 | dout("req %p tid %llu retry_attempt %d != %d, ignoring\n", | |
2084 | req, req->r_tid, m.retry_attempt, | |
2085 | req->r_attempts - 1); | |
2086 | goto out_put; | |
2087 | } | |
2088 | } else { | |
2089 | WARN_ON(1); /* MOSDOpReply v4 is assumed */ | |
2090 | } | |
2091 | ||
2092 | if (!ceph_oloc_empty(&m.redirect.oloc)) { | |
2093 | dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid, | |
2094 | m.redirect.oloc.pool); | |
205ee118 | 2095 | __unregister_request(osdc, req); |
205ee118 | 2096 | |
fe5da05e | 2097 | ceph_oloc_copy(&req->r_t.target_oloc, &m.redirect.oloc); |
205ee118 ID |
2098 | |
2099 | /* | |
2100 | * Start redirect requests with nofail=true. If | |
2101 | * mapping fails, request will end up on the notarget | |
2102 | * list, waiting for the new osdmap (which can take | |
2103 | * a while), even though the original request mapped | |
2104 | * successfully. In the future we might want to follow | |
2105 | * original request's nofail setting here. | |
2106 | */ | |
fe5da05e ID |
2107 | ret = __ceph_osdc_start_request(osdc, req, true); |
2108 | BUG_ON(ret); | |
205ee118 | 2109 | |
fe5da05e | 2110 | goto out_put; |
205ee118 ID |
2111 | } |
2112 | ||
fe5da05e ID |
2113 | if (m.num_ops != req->r_num_ops) { |
2114 | pr_err("num_ops %d != %d for tid %llu\n", m.num_ops, | |
2115 | req->r_num_ops, req->r_tid); | |
2116 | goto fail_request; | |
f24e9980 | 2117 | } |
fe5da05e ID |
2118 | for (i = 0; i < req->r_num_ops; i++) { |
2119 | dout(" req %p tid %llu op %d rval %d len %u\n", req, | |
2120 | req->r_tid, i, m.rval[i], m.outdata_len[i]); | |
2121 | req->r_ops[i].rval = m.rval[i]; | |
2122 | req->r_ops[i].outdata_len = m.outdata_len[i]; | |
2123 | data_len += m.outdata_len[i]; | |
2124 | } | |
2125 | if (data_len != le32_to_cpu(msg->hdr.data_len)) { | |
2126 | pr_err("sum of lens %u != %u for tid %llu\n", data_len, | |
2127 | le32_to_cpu(msg->hdr.data_len), req->r_tid); | |
2128 | goto fail_request; | |
2129 | } | |
2130 | dout("%s req %p tid %llu acked %d result %d data_len %u\n", __func__, | |
2131 | req, req->r_tid, req->r_got_reply, m.result, data_len); | |
2132 | ||
2133 | already_acked = req->r_got_reply; | |
2134 | if (!already_acked) { | |
2135 | req->r_result = m.result ?: data_len; | |
2136 | req->r_replay_version = m.replay_version; /* struct */ | |
2137 | req->r_got_reply = true; | |
2138 | } else if (!(m.flags & CEPH_OSD_FLAG_ONDISK)) { | |
2139 | dout("req %p tid %llu dup ack\n", req, req->r_tid); | |
2140 | goto out_put; | |
2141 | } | |
2142 | ||
2143 | if (done_request(req, &m)) { | |
f24e9980 | 2144 | __unregister_request(osdc, req); |
fe5da05e ID |
2145 | if (req->r_linger) { |
2146 | WARN_ON(req->r_unsafe_callback); | |
2147 | __register_linger_request(osdc, req); | |
2148 | } | |
2149 | } | |
f24e9980 SW |
2150 | |
2151 | mutex_unlock(&osdc->request_mutex); | |
ff513ace | 2152 | up_read(&osdc->map_sem); |
f24e9980 | 2153 | |
fe5da05e ID |
2154 | if (done_request(req, &m)) { |
2155 | if (already_acked && req->r_unsafe_callback) { | |
2156 | dout("req %p tid %llu safe-cb\n", req, req->r_tid); | |
61c5d6bf | 2157 | req->r_unsafe_callback(req, false); |
fe5da05e ID |
2158 | } else { |
2159 | dout("req %p tid %llu cb\n", req, req->r_tid); | |
2160 | __complete_request(req); | |
2161 | } | |
2162 | } else { | |
2163 | if (req->r_unsafe_callback) { | |
2164 | dout("req %p tid %llu unsafe-cb\n", req, req->r_tid); | |
2165 | req->r_unsafe_callback(req, true); | |
2166 | } else { | |
2167 | WARN_ON(1); | |
2168 | } | |
61c5d6bf | 2169 | } |
fe5da05e ID |
2170 | if (m.flags & CEPH_OSD_FLAG_ONDISK) |
2171 | complete_all(&req->r_safe_completion); | |
f24e9980 | 2172 | |
f24e9980 SW |
2173 | ceph_osdc_put_request(req); |
2174 | return; | |
2175 | ||
fe5da05e | 2176 | fail_request: |
37c89bde LW |
2177 | req->r_result = -EIO; |
2178 | __unregister_request(osdc, req); | |
fe5da05e ID |
2179 | __complete_request(req); |
2180 | complete_all(&req->r_safe_completion); | |
2181 | out_put: | |
1b83bef2 | 2182 | ceph_osdc_put_request(req); |
fe5da05e | 2183 | out_unlock: |
8058fd45 | 2184 | mutex_unlock(&osdc->request_mutex); |
ff513ace | 2185 | up_read(&osdc->map_sem); |
f24e9980 SW |
2186 | } |
2187 | ||
42c1b124 ID |
2188 | static void set_pool_was_full(struct ceph_osd_client *osdc) |
2189 | { | |
2190 | struct rb_node *n; | |
2191 | ||
2192 | for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { | |
2193 | struct ceph_pg_pool_info *pi = | |
2194 | rb_entry(n, struct ceph_pg_pool_info, node); | |
2195 | ||
2196 | pi->was_full = __pool_full(pi); | |
2197 | } | |
2198 | } | |
2199 | ||
6f6c7006 | 2200 | static void reset_changed_osds(struct ceph_osd_client *osdc) |
f24e9980 | 2201 | { |
f24e9980 | 2202 | struct rb_node *p, *n; |
f24e9980 | 2203 | |
7eb71e03 | 2204 | dout("%s %p\n", __func__, osdc); |
6f6c7006 SW |
2205 | for (p = rb_first(&osdc->osds); p; p = n) { |
2206 | struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node); | |
f24e9980 | 2207 | |
6f6c7006 SW |
2208 | n = rb_next(p); |
2209 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || | |
2210 | memcmp(&osd->o_con.peer_addr, | |
2211 | ceph_osd_addr(osdc->osdmap, | |
2212 | osd->o_osd), | |
2213 | sizeof(struct ceph_entity_addr)) != 0) | |
2214 | __reset_osd(osdc, osd); | |
f24e9980 | 2215 | } |
422d2cb8 YS |
2216 | } |
2217 | ||
2218 | /* | |
6f6c7006 SW |
2219 | * Requeue requests whose mapping to an OSD has changed. If requests map to |
2220 | * no osd, request a new map. | |
422d2cb8 | 2221 | * |
e6d50f67 | 2222 | * Caller should hold map_sem for read. |
422d2cb8 | 2223 | */ |
9a1ea2db JD |
2224 | static void kick_requests(struct ceph_osd_client *osdc, bool force_resend, |
2225 | bool force_resend_writes) | |
422d2cb8 | 2226 | { |
a40c4f10 | 2227 | struct ceph_osd_request *req, *nreq; |
6f6c7006 SW |
2228 | struct rb_node *p; |
2229 | int needmap = 0; | |
2230 | int err; | |
9a1ea2db | 2231 | bool force_resend_req; |
422d2cb8 | 2232 | |
9a1ea2db JD |
2233 | dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "", |
2234 | force_resend_writes ? " (force resend writes)" : ""); | |
422d2cb8 | 2235 | mutex_lock(&osdc->request_mutex); |
6194ea89 | 2236 | for (p = rb_first(&osdc->requests); p; ) { |
6f6c7006 | 2237 | req = rb_entry(p, struct ceph_osd_request, r_node); |
6194ea89 | 2238 | p = rb_next(p); |
ab60b16d AE |
2239 | |
2240 | /* | |
2241 | * For linger requests that have not yet been | |
2242 | * registered, move them to the linger list; they'll | |
2243 | * be sent to the osd in the loop below. Unregister | |
2244 | * the request before re-registering it as a linger | |
2245 | * request to ensure the __map_request() below | |
2246 | * will decide it needs to be sent. | |
2247 | */ | |
2248 | if (req->r_linger && list_empty(&req->r_linger_item)) { | |
2249 | dout("%p tid %llu restart on osd%d\n", | |
2250 | req, req->r_tid, | |
2251 | req->r_osd ? req->r_osd->o_osd : -1); | |
96e4dac6 | 2252 | ceph_osdc_get_request(req); |
ab60b16d AE |
2253 | __unregister_request(osdc, req); |
2254 | __register_linger_request(osdc, req); | |
96e4dac6 | 2255 | ceph_osdc_put_request(req); |
ab60b16d AE |
2256 | continue; |
2257 | } | |
2258 | ||
9a1ea2db JD |
2259 | force_resend_req = force_resend || |
2260 | (force_resend_writes && | |
2261 | req->r_flags & CEPH_OSD_FLAG_WRITE); | |
2262 | err = __map_request(osdc, req, force_resend_req); | |
6f6c7006 SW |
2263 | if (err < 0) |
2264 | continue; /* error */ | |
2265 | if (req->r_osd == NULL) { | |
2266 | dout("%p tid %llu maps to no osd\n", req, req->r_tid); | |
2267 | needmap++; /* request a newer map */ | |
2268 | } else if (err > 0) { | |
6194ea89 SW |
2269 | if (!req->r_linger) { |
2270 | dout("%p tid %llu requeued on osd%d\n", req, | |
2271 | req->r_tid, | |
2272 | req->r_osd ? req->r_osd->o_osd : -1); | |
a40c4f10 | 2273 | req->r_flags |= CEPH_OSD_FLAG_RETRY; |
6194ea89 SW |
2274 | } |
2275 | } | |
a40c4f10 YS |
2276 | } |
2277 | ||
2278 | list_for_each_entry_safe(req, nreq, &osdc->req_linger, | |
2279 | r_linger_item) { | |
2280 | dout("linger req=%p req->r_osd=%p\n", req, req->r_osd); | |
2281 | ||
9a1ea2db JD |
2282 | err = __map_request(osdc, req, |
2283 | force_resend || force_resend_writes); | |
ab60b16d | 2284 | dout("__map_request returned %d\n", err); |
a40c4f10 YS |
2285 | if (err < 0) |
2286 | continue; /* hrm! */ | |
b0494532 ID |
2287 | if (req->r_osd == NULL || err > 0) { |
2288 | if (req->r_osd == NULL) { | |
2289 | dout("lingering %p tid %llu maps to no osd\n", | |
2290 | req, req->r_tid); | |
2291 | /* | |
2292 | * A homeless lingering request makes | |
2293 | * no sense, as it's job is to keep | |
2294 | * a particular OSD connection open. | |
2295 | * Request a newer map and kick the | |
2296 | * request, knowing that it won't be | |
2297 | * resent until we actually get a map | |
2298 | * that can tell us where to send it. | |
2299 | */ | |
2300 | needmap++; | |
2301 | } | |
a40c4f10 | 2302 | |
b0494532 ID |
2303 | dout("kicking lingering %p tid %llu osd%d\n", req, |
2304 | req->r_tid, req->r_osd ? req->r_osd->o_osd : -1); | |
2305 | __register_request(osdc, req); | |
2306 | __unregister_linger_request(osdc, req); | |
2307 | } | |
6f6c7006 | 2308 | } |
14d2f38d | 2309 | reset_changed_osds(osdc); |
f24e9980 SW |
2310 | mutex_unlock(&osdc->request_mutex); |
2311 | ||
2312 | if (needmap) { | |
2313 | dout("%d requests for down osds, need new map\n", needmap); | |
2314 | ceph_monc_request_next_osdmap(&osdc->client->monc); | |
2315 | } | |
422d2cb8 | 2316 | } |
6f6c7006 | 2317 | |
42c1b124 ID |
2318 | static int handle_one_map(struct ceph_osd_client *osdc, |
2319 | void *p, void *end, bool incremental) | |
2320 | { | |
2321 | struct ceph_osdmap *newmap; | |
2322 | struct rb_node *n; | |
2323 | bool skipped_map = false; | |
2324 | bool was_full; | |
2325 | ||
2326 | was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL); | |
2327 | set_pool_was_full(osdc); | |
2328 | ||
2329 | if (incremental) | |
2330 | newmap = osdmap_apply_incremental(&p, end, osdc->osdmap); | |
2331 | else | |
2332 | newmap = ceph_osdmap_decode(&p, end); | |
2333 | if (IS_ERR(newmap)) | |
2334 | return PTR_ERR(newmap); | |
2335 | ||
2336 | if (newmap != osdc->osdmap) { | |
2337 | /* | |
2338 | * Preserve ->was_full before destroying the old map. | |
2339 | * For pools that weren't in the old map, ->was_full | |
2340 | * should be false. | |
2341 | */ | |
2342 | for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) { | |
2343 | struct ceph_pg_pool_info *pi = | |
2344 | rb_entry(n, struct ceph_pg_pool_info, node); | |
2345 | struct ceph_pg_pool_info *old_pi; | |
2346 | ||
2347 | old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id); | |
2348 | if (old_pi) | |
2349 | pi->was_full = old_pi->was_full; | |
2350 | else | |
2351 | WARN_ON(pi->was_full); | |
2352 | } | |
2353 | ||
2354 | if (osdc->osdmap->epoch && | |
2355 | osdc->osdmap->epoch + 1 < newmap->epoch) { | |
2356 | WARN_ON(incremental); | |
2357 | skipped_map = true; | |
2358 | } | |
2359 | ||
2360 | ceph_osdmap_destroy(osdc->osdmap); | |
2361 | osdc->osdmap = newmap; | |
2362 | } | |
2363 | ||
2364 | was_full &= !ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL); | |
2365 | kick_requests(osdc, skipped_map, was_full); | |
2366 | ||
2367 | return 0; | |
2368 | } | |
6f6c7006 | 2369 | |
f24e9980 SW |
2370 | /* |
2371 | * Process updated osd map. | |
2372 | * | |
2373 | * The message contains any number of incremental and full maps, normally | |
2374 | * indicating some sort of topology change in the cluster. Kick requests | |
2375 | * off to different OSDs as needed. | |
2376 | */ | |
2377 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
2378 | { | |
42c1b124 ID |
2379 | void *p = msg->front.iov_base; |
2380 | void *const end = p + msg->front.iov_len; | |
f24e9980 SW |
2381 | u32 nr_maps, maplen; |
2382 | u32 epoch; | |
f24e9980 | 2383 | struct ceph_fsid fsid; |
42c1b124 ID |
2384 | bool handled_incremental = false; |
2385 | bool was_pauserd, was_pausewr; | |
2386 | bool pauserd, pausewr; | |
2387 | int err; | |
f24e9980 | 2388 | |
42c1b124 ID |
2389 | dout("%s have %u\n", __func__, osdc->osdmap->epoch); |
2390 | down_write(&osdc->map_sem); | |
f24e9980 SW |
2391 | |
2392 | /* verify fsid */ | |
2393 | ceph_decode_need(&p, end, sizeof(fsid), bad); | |
2394 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
0743304d | 2395 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
42c1b124 | 2396 | goto bad; |
f24e9980 | 2397 | |
42c1b124 ID |
2398 | was_pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD); |
2399 | was_pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) || | |
2400 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) || | |
2401 | have_pool_full(osdc); | |
9a1ea2db | 2402 | |
f24e9980 SW |
2403 | /* incremental maps */ |
2404 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
2405 | dout(" %d inc maps\n", nr_maps); | |
2406 | while (nr_maps > 0) { | |
2407 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
2408 | epoch = ceph_decode_32(&p); |
2409 | maplen = ceph_decode_32(&p); | |
f24e9980 | 2410 | ceph_decode_need(&p, end, maplen, bad); |
42c1b124 ID |
2411 | if (osdc->osdmap->epoch && |
2412 | osdc->osdmap->epoch + 1 == epoch) { | |
f24e9980 SW |
2413 | dout("applying incremental map %u len %d\n", |
2414 | epoch, maplen); | |
42c1b124 ID |
2415 | err = handle_one_map(osdc, p, p + maplen, true); |
2416 | if (err) | |
f24e9980 | 2417 | goto bad; |
42c1b124 | 2418 | handled_incremental = true; |
f24e9980 SW |
2419 | } else { |
2420 | dout("ignoring incremental map %u len %d\n", | |
2421 | epoch, maplen); | |
2422 | } | |
42c1b124 | 2423 | p += maplen; |
f24e9980 SW |
2424 | nr_maps--; |
2425 | } | |
42c1b124 | 2426 | if (handled_incremental) |
f24e9980 SW |
2427 | goto done; |
2428 | ||
2429 | /* full maps */ | |
2430 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
2431 | dout(" %d full maps\n", nr_maps); | |
2432 | while (nr_maps) { | |
2433 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
2434 | epoch = ceph_decode_32(&p); |
2435 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
2436 | ceph_decode_need(&p, end, maplen, bad); |
2437 | if (nr_maps > 1) { | |
2438 | dout("skipping non-latest full map %u len %d\n", | |
2439 | epoch, maplen); | |
e5253a7b | 2440 | } else if (osdc->osdmap->epoch >= epoch) { |
f24e9980 SW |
2441 | dout("skipping full map %u len %d, " |
2442 | "older than our %u\n", epoch, maplen, | |
2443 | osdc->osdmap->epoch); | |
2444 | } else { | |
2445 | dout("taking full map %u len %d\n", epoch, maplen); | |
42c1b124 ID |
2446 | err = handle_one_map(osdc, p, p + maplen, false); |
2447 | if (err) | |
f24e9980 | 2448 | goto bad; |
f24e9980 SW |
2449 | } |
2450 | p += maplen; | |
2451 | nr_maps--; | |
2452 | } | |
2453 | ||
2454 | done: | |
cd634fb6 SW |
2455 | /* |
2456 | * subscribe to subsequent osdmap updates if full to ensure | |
2457 | * we find out when we are no longer full and stop returning | |
2458 | * ENOSPC. | |
2459 | */ | |
42c1b124 ID |
2460 | pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD); |
2461 | pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) || | |
2462 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) || | |
2463 | have_pool_full(osdc); | |
2464 | if (was_pauserd || was_pausewr || pauserd || pausewr) | |
2465 | maybe_request_map(osdc); | |
cd634fb6 | 2466 | |
f9d25199 AE |
2467 | mutex_lock(&osdc->request_mutex); |
2468 | __send_queued(osdc); | |
2469 | mutex_unlock(&osdc->request_mutex); | |
42c1b124 ID |
2470 | |
2471 | ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP, | |
2472 | osdc->osdmap->epoch); | |
2473 | up_write(&osdc->map_sem); | |
03066f23 | 2474 | wake_up_all(&osdc->client->auth_wq); |
f24e9980 SW |
2475 | return; |
2476 | ||
2477 | bad: | |
2478 | pr_err("osdc handle_map corrupt msg\n"); | |
9ec7cab1 | 2479 | ceph_msg_dump(msg); |
f24e9980 | 2480 | up_write(&osdc->map_sem); |
f24e9980 SW |
2481 | } |
2482 | ||
a40c4f10 YS |
2483 | /* |
2484 | * watch/notify callback event infrastructure | |
2485 | * | |
2486 | * These callbacks are used both for watch and notify operations. | |
2487 | */ | |
2488 | static void __release_event(struct kref *kref) | |
2489 | { | |
2490 | struct ceph_osd_event *event = | |
2491 | container_of(kref, struct ceph_osd_event, kref); | |
2492 | ||
2493 | dout("__release_event %p\n", event); | |
2494 | kfree(event); | |
2495 | } | |
2496 | ||
2497 | static void get_event(struct ceph_osd_event *event) | |
2498 | { | |
2499 | kref_get(&event->kref); | |
2500 | } | |
2501 | ||
2502 | void ceph_osdc_put_event(struct ceph_osd_event *event) | |
2503 | { | |
2504 | kref_put(&event->kref, __release_event); | |
2505 | } | |
2506 | EXPORT_SYMBOL(ceph_osdc_put_event); | |
2507 | ||
2508 | static void __insert_event(struct ceph_osd_client *osdc, | |
2509 | struct ceph_osd_event *new) | |
2510 | { | |
2511 | struct rb_node **p = &osdc->event_tree.rb_node; | |
2512 | struct rb_node *parent = NULL; | |
2513 | struct ceph_osd_event *event = NULL; | |
2514 | ||
2515 | while (*p) { | |
2516 | parent = *p; | |
2517 | event = rb_entry(parent, struct ceph_osd_event, node); | |
2518 | if (new->cookie < event->cookie) | |
2519 | p = &(*p)->rb_left; | |
2520 | else if (new->cookie > event->cookie) | |
2521 | p = &(*p)->rb_right; | |
2522 | else | |
2523 | BUG(); | |
2524 | } | |
2525 | ||
2526 | rb_link_node(&new->node, parent, p); | |
2527 | rb_insert_color(&new->node, &osdc->event_tree); | |
2528 | } | |
2529 | ||
2530 | static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc, | |
2531 | u64 cookie) | |
2532 | { | |
2533 | struct rb_node **p = &osdc->event_tree.rb_node; | |
2534 | struct rb_node *parent = NULL; | |
2535 | struct ceph_osd_event *event = NULL; | |
2536 | ||
2537 | while (*p) { | |
2538 | parent = *p; | |
2539 | event = rb_entry(parent, struct ceph_osd_event, node); | |
2540 | if (cookie < event->cookie) | |
2541 | p = &(*p)->rb_left; | |
2542 | else if (cookie > event->cookie) | |
2543 | p = &(*p)->rb_right; | |
2544 | else | |
2545 | return event; | |
2546 | } | |
2547 | return NULL; | |
2548 | } | |
2549 | ||
2550 | static void __remove_event(struct ceph_osd_event *event) | |
2551 | { | |
2552 | struct ceph_osd_client *osdc = event->osdc; | |
2553 | ||
2554 | if (!RB_EMPTY_NODE(&event->node)) { | |
2555 | dout("__remove_event removed %p\n", event); | |
2556 | rb_erase(&event->node, &osdc->event_tree); | |
2557 | ceph_osdc_put_event(event); | |
2558 | } else { | |
2559 | dout("__remove_event didn't remove %p\n", event); | |
2560 | } | |
2561 | } | |
2562 | ||
2563 | int ceph_osdc_create_event(struct ceph_osd_client *osdc, | |
2564 | void (*event_cb)(u64, u64, u8, void *), | |
3c663bbd | 2565 | void *data, struct ceph_osd_event **pevent) |
a40c4f10 YS |
2566 | { |
2567 | struct ceph_osd_event *event; | |
2568 | ||
2569 | event = kmalloc(sizeof(*event), GFP_NOIO); | |
2570 | if (!event) | |
2571 | return -ENOMEM; | |
2572 | ||
2573 | dout("create_event %p\n", event); | |
2574 | event->cb = event_cb; | |
3c663bbd | 2575 | event->one_shot = 0; |
a40c4f10 YS |
2576 | event->data = data; |
2577 | event->osdc = osdc; | |
2578 | INIT_LIST_HEAD(&event->osd_node); | |
3ee5234d | 2579 | RB_CLEAR_NODE(&event->node); |
a40c4f10 YS |
2580 | kref_init(&event->kref); /* one ref for us */ |
2581 | kref_get(&event->kref); /* one ref for the caller */ | |
a40c4f10 YS |
2582 | |
2583 | spin_lock(&osdc->event_lock); | |
2584 | event->cookie = ++osdc->event_count; | |
2585 | __insert_event(osdc, event); | |
2586 | spin_unlock(&osdc->event_lock); | |
2587 | ||
2588 | *pevent = event; | |
2589 | return 0; | |
2590 | } | |
2591 | EXPORT_SYMBOL(ceph_osdc_create_event); | |
2592 | ||
2593 | void ceph_osdc_cancel_event(struct ceph_osd_event *event) | |
2594 | { | |
2595 | struct ceph_osd_client *osdc = event->osdc; | |
2596 | ||
2597 | dout("cancel_event %p\n", event); | |
2598 | spin_lock(&osdc->event_lock); | |
2599 | __remove_event(event); | |
2600 | spin_unlock(&osdc->event_lock); | |
2601 | ceph_osdc_put_event(event); /* caller's */ | |
2602 | } | |
2603 | EXPORT_SYMBOL(ceph_osdc_cancel_event); | |
2604 | ||
2605 | ||
2606 | static void do_event_work(struct work_struct *work) | |
2607 | { | |
2608 | struct ceph_osd_event_work *event_work = | |
2609 | container_of(work, struct ceph_osd_event_work, work); | |
2610 | struct ceph_osd_event *event = event_work->event; | |
2611 | u64 ver = event_work->ver; | |
2612 | u64 notify_id = event_work->notify_id; | |
2613 | u8 opcode = event_work->opcode; | |
2614 | ||
2615 | dout("do_event_work completing %p\n", event); | |
2616 | event->cb(ver, notify_id, opcode, event->data); | |
a40c4f10 YS |
2617 | dout("do_event_work completed %p\n", event); |
2618 | ceph_osdc_put_event(event); | |
2619 | kfree(event_work); | |
2620 | } | |
2621 | ||
2622 | ||
2623 | /* | |
2624 | * Process osd watch notifications | |
2625 | */ | |
3c663bbd AE |
2626 | static void handle_watch_notify(struct ceph_osd_client *osdc, |
2627 | struct ceph_msg *msg) | |
a40c4f10 YS |
2628 | { |
2629 | void *p, *end; | |
2630 | u8 proto_ver; | |
2631 | u64 cookie, ver, notify_id; | |
2632 | u8 opcode; | |
2633 | struct ceph_osd_event *event; | |
2634 | struct ceph_osd_event_work *event_work; | |
2635 | ||
2636 | p = msg->front.iov_base; | |
2637 | end = p + msg->front.iov_len; | |
2638 | ||
2639 | ceph_decode_8_safe(&p, end, proto_ver, bad); | |
2640 | ceph_decode_8_safe(&p, end, opcode, bad); | |
2641 | ceph_decode_64_safe(&p, end, cookie, bad); | |
2642 | ceph_decode_64_safe(&p, end, ver, bad); | |
2643 | ceph_decode_64_safe(&p, end, notify_id, bad); | |
2644 | ||
2645 | spin_lock(&osdc->event_lock); | |
2646 | event = __find_event(osdc, cookie); | |
2647 | if (event) { | |
3c663bbd | 2648 | BUG_ON(event->one_shot); |
a40c4f10 | 2649 | get_event(event); |
a40c4f10 YS |
2650 | } |
2651 | spin_unlock(&osdc->event_lock); | |
2652 | dout("handle_watch_notify cookie %lld ver %lld event %p\n", | |
2653 | cookie, ver, event); | |
2654 | if (event) { | |
2655 | event_work = kmalloc(sizeof(*event_work), GFP_NOIO); | |
a40c4f10 | 2656 | if (!event_work) { |
91883cd2 ID |
2657 | pr_err("couldn't allocate event_work\n"); |
2658 | ceph_osdc_put_event(event); | |
2659 | return; | |
a40c4f10 | 2660 | } |
6b0ae409 | 2661 | INIT_WORK(&event_work->work, do_event_work); |
a40c4f10 YS |
2662 | event_work->event = event; |
2663 | event_work->ver = ver; | |
2664 | event_work->notify_id = notify_id; | |
2665 | event_work->opcode = opcode; | |
a40c4f10 | 2666 | |
91883cd2 ID |
2667 | queue_work(osdc->notify_wq, &event_work->work); |
2668 | } | |
a40c4f10 | 2669 | |
a40c4f10 YS |
2670 | return; |
2671 | ||
2672 | bad: | |
2673 | pr_err("osdc handle_watch_notify corrupt msg\n"); | |
a40c4f10 YS |
2674 | } |
2675 | ||
70636773 AE |
2676 | /* |
2677 | * Register request, send initial attempt. | |
2678 | */ | |
2679 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
2680 | struct ceph_osd_request *req, | |
2681 | bool nofail) | |
2682 | { | |
0bbfdfe8 | 2683 | int rc; |
70636773 | 2684 | |
f24e9980 SW |
2685 | down_read(&osdc->map_sem); |
2686 | mutex_lock(&osdc->request_mutex); | |
0bbfdfe8 ID |
2687 | |
2688 | rc = __ceph_osdc_start_request(osdc, req, nofail); | |
2689 | ||
f24e9980 SW |
2690 | mutex_unlock(&osdc->request_mutex); |
2691 | up_read(&osdc->map_sem); | |
0bbfdfe8 | 2692 | |
f24e9980 SW |
2693 | return rc; |
2694 | } | |
3d14c5d2 | 2695 | EXPORT_SYMBOL(ceph_osdc_start_request); |
f24e9980 | 2696 | |
c9f9b93d ID |
2697 | /* |
2698 | * Unregister a registered request. The request is not completed (i.e. | |
2699 | * no callbacks or wakeups) - higher layers are supposed to know what | |
2700 | * they are canceling. | |
2701 | */ | |
2702 | void ceph_osdc_cancel_request(struct ceph_osd_request *req) | |
2703 | { | |
2704 | struct ceph_osd_client *osdc = req->r_osdc; | |
2705 | ||
2706 | mutex_lock(&osdc->request_mutex); | |
2707 | if (req->r_linger) | |
2708 | __unregister_linger_request(osdc, req); | |
2709 | __unregister_request(osdc, req); | |
2710 | mutex_unlock(&osdc->request_mutex); | |
2711 | ||
2712 | dout("%s %p tid %llu canceled\n", __func__, req, req->r_tid); | |
2713 | } | |
2714 | EXPORT_SYMBOL(ceph_osdc_cancel_request); | |
2715 | ||
f24e9980 SW |
2716 | /* |
2717 | * wait for a request to complete | |
2718 | */ | |
2719 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, | |
2720 | struct ceph_osd_request *req) | |
2721 | { | |
2722 | int rc; | |
2723 | ||
c9f9b93d ID |
2724 | dout("%s %p tid %llu\n", __func__, req, req->r_tid); |
2725 | ||
f24e9980 SW |
2726 | rc = wait_for_completion_interruptible(&req->r_completion); |
2727 | if (rc < 0) { | |
c9f9b93d ID |
2728 | dout("%s %p tid %llu interrupted\n", __func__, req, req->r_tid); |
2729 | ceph_osdc_cancel_request(req); | |
fe5da05e ID |
2730 | |
2731 | /* kludge - need to to wake ceph_osdc_sync() */ | |
2732 | complete_all(&req->r_safe_completion); | |
f24e9980 SW |
2733 | return rc; |
2734 | } | |
2735 | ||
c9f9b93d ID |
2736 | dout("%s %p tid %llu result %d\n", __func__, req, req->r_tid, |
2737 | req->r_result); | |
f24e9980 SW |
2738 | return req->r_result; |
2739 | } | |
3d14c5d2 | 2740 | EXPORT_SYMBOL(ceph_osdc_wait_request); |
f24e9980 SW |
2741 | |
2742 | /* | |
2743 | * sync - wait for all in-flight requests to flush. avoid starvation. | |
2744 | */ | |
2745 | void ceph_osdc_sync(struct ceph_osd_client *osdc) | |
2746 | { | |
2747 | struct ceph_osd_request *req; | |
2748 | u64 last_tid, next_tid = 0; | |
2749 | ||
2750 | mutex_lock(&osdc->request_mutex); | |
2751 | last_tid = osdc->last_tid; | |
2752 | while (1) { | |
2753 | req = __lookup_request_ge(osdc, next_tid); | |
2754 | if (!req) | |
2755 | break; | |
2756 | if (req->r_tid > last_tid) | |
2757 | break; | |
2758 | ||
2759 | next_tid = req->r_tid + 1; | |
2760 | if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0) | |
2761 | continue; | |
2762 | ||
2763 | ceph_osdc_get_request(req); | |
2764 | mutex_unlock(&osdc->request_mutex); | |
2765 | dout("sync waiting on tid %llu (last is %llu)\n", | |
2766 | req->r_tid, last_tid); | |
2767 | wait_for_completion(&req->r_safe_completion); | |
2768 | mutex_lock(&osdc->request_mutex); | |
2769 | ceph_osdc_put_request(req); | |
2770 | } | |
2771 | mutex_unlock(&osdc->request_mutex); | |
2772 | dout("sync done (thru tid %llu)\n", last_tid); | |
2773 | } | |
3d14c5d2 | 2774 | EXPORT_SYMBOL(ceph_osdc_sync); |
f24e9980 | 2775 | |
dd935f44 JD |
2776 | /* |
2777 | * Call all pending notify callbacks - for use after a watch is | |
2778 | * unregistered, to make sure no more callbacks for it will be invoked | |
2779 | */ | |
f6479449 | 2780 | void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc) |
dd935f44 JD |
2781 | { |
2782 | flush_workqueue(osdc->notify_wq); | |
2783 | } | |
2784 | EXPORT_SYMBOL(ceph_osdc_flush_notifies); | |
2785 | ||
2786 | ||
f24e9980 SW |
2787 | /* |
2788 | * init, shutdown | |
2789 | */ | |
2790 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) | |
2791 | { | |
2792 | int err; | |
2793 | ||
2794 | dout("init\n"); | |
2795 | osdc->client = client; | |
f24e9980 | 2796 | init_rwsem(&osdc->map_sem); |
f24e9980 | 2797 | mutex_init(&osdc->request_mutex); |
f24e9980 SW |
2798 | osdc->last_tid = 0; |
2799 | osdc->osds = RB_ROOT; | |
f5a2041b | 2800 | INIT_LIST_HEAD(&osdc->osd_lru); |
9dd2845c | 2801 | spin_lock_init(&osdc->osd_lru_lock); |
f24e9980 | 2802 | osdc->requests = RB_ROOT; |
422d2cb8 | 2803 | INIT_LIST_HEAD(&osdc->req_lru); |
6f6c7006 SW |
2804 | INIT_LIST_HEAD(&osdc->req_unsent); |
2805 | INIT_LIST_HEAD(&osdc->req_notarget); | |
a40c4f10 | 2806 | INIT_LIST_HEAD(&osdc->req_linger); |
f24e9980 SW |
2807 | osdc->num_requests = 0; |
2808 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); | |
f5a2041b | 2809 | INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); |
a40c4f10 YS |
2810 | spin_lock_init(&osdc->event_lock); |
2811 | osdc->event_tree = RB_ROOT; | |
2812 | osdc->event_count = 0; | |
f5a2041b | 2813 | |
5f44f142 | 2814 | err = -ENOMEM; |
e5253a7b ID |
2815 | osdc->osdmap = ceph_osdmap_alloc(); |
2816 | if (!osdc->osdmap) | |
2817 | goto out; | |
2818 | ||
9e767adb ID |
2819 | osdc->req_mempool = mempool_create_slab_pool(10, |
2820 | ceph_osd_request_cache); | |
f24e9980 | 2821 | if (!osdc->req_mempool) |
e5253a7b | 2822 | goto out_map; |
f24e9980 | 2823 | |
d50b409f | 2824 | err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, |
711da55d | 2825 | PAGE_SIZE, 10, true, "osd_op"); |
f24e9980 | 2826 | if (err < 0) |
5f44f142 | 2827 | goto out_mempool; |
d50b409f | 2828 | err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, |
711da55d | 2829 | PAGE_SIZE, 10, true, "osd_op_reply"); |
c16e7869 SW |
2830 | if (err < 0) |
2831 | goto out_msgpool; | |
a40c4f10 | 2832 | |
dbcae088 | 2833 | err = -ENOMEM; |
a40c4f10 | 2834 | osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); |
dbcae088 | 2835 | if (!osdc->notify_wq) |
c172ec5c ID |
2836 | goto out_msgpool_reply; |
2837 | ||
fbca9635 ID |
2838 | schedule_delayed_work(&osdc->timeout_work, |
2839 | osdc->client->options->osd_keepalive_timeout); | |
b37ee1b9 ID |
2840 | schedule_delayed_work(&osdc->osds_timeout_work, |
2841 | round_jiffies_relative(osdc->client->options->osd_idle_ttl)); | |
2842 | ||
f24e9980 | 2843 | return 0; |
5f44f142 | 2844 | |
c172ec5c ID |
2845 | out_msgpool_reply: |
2846 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); | |
c16e7869 SW |
2847 | out_msgpool: |
2848 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
5f44f142 SW |
2849 | out_mempool: |
2850 | mempool_destroy(osdc->req_mempool); | |
e5253a7b ID |
2851 | out_map: |
2852 | ceph_osdmap_destroy(osdc->osdmap); | |
5f44f142 SW |
2853 | out: |
2854 | return err; | |
f24e9980 SW |
2855 | } |
2856 | ||
2857 | void ceph_osdc_stop(struct ceph_osd_client *osdc) | |
2858 | { | |
a40c4f10 YS |
2859 | flush_workqueue(osdc->notify_wq); |
2860 | destroy_workqueue(osdc->notify_wq); | |
f24e9980 | 2861 | cancel_delayed_work_sync(&osdc->timeout_work); |
f5a2041b | 2862 | cancel_delayed_work_sync(&osdc->osds_timeout_work); |
42a2c09f ID |
2863 | |
2864 | mutex_lock(&osdc->request_mutex); | |
2865 | while (!RB_EMPTY_ROOT(&osdc->osds)) { | |
2866 | struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), | |
2867 | struct ceph_osd, o_node); | |
2868 | remove_osd(osdc, osd); | |
2869 | } | |
2870 | mutex_unlock(&osdc->request_mutex); | |
2871 | ||
e5253a7b | 2872 | ceph_osdmap_destroy(osdc->osdmap); |
f24e9980 SW |
2873 | mempool_destroy(osdc->req_mempool); |
2874 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
c16e7869 | 2875 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); |
f24e9980 SW |
2876 | } |
2877 | ||
2878 | /* | |
2879 | * Read some contiguous pages. If we cross a stripe boundary, shorten | |
2880 | * *plen. Return number of bytes read, or error. | |
2881 | */ | |
2882 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, | |
2883 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
2884 | u64 off, u64 *plen, | |
2885 | u32 truncate_seq, u64 truncate_size, | |
b7495fc2 | 2886 | struct page **pages, int num_pages, int page_align) |
f24e9980 SW |
2887 | { |
2888 | struct ceph_osd_request *req; | |
2889 | int rc = 0; | |
2890 | ||
2891 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, | |
2892 | vino.snap, off, *plen); | |
715e4cd4 | 2893 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1, |
f24e9980 | 2894 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, |
acead002 | 2895 | NULL, truncate_seq, truncate_size, |
153e5167 | 2896 | false); |
6816282d SW |
2897 | if (IS_ERR(req)) |
2898 | return PTR_ERR(req); | |
f24e9980 SW |
2899 | |
2900 | /* it may be a short read due to an object boundary */ | |
406e2c9f | 2901 | osd_req_op_extent_osd_data_pages(req, 0, |
a4ce40a9 | 2902 | pages, *plen, page_align, false, false); |
f24e9980 | 2903 | |
e0c59487 | 2904 | dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n", |
43bfe5de | 2905 | off, *plen, *plen, page_align); |
f24e9980 SW |
2906 | |
2907 | rc = ceph_osdc_start_request(osdc, req, false); | |
2908 | if (!rc) | |
2909 | rc = ceph_osdc_wait_request(osdc, req); | |
2910 | ||
2911 | ceph_osdc_put_request(req); | |
2912 | dout("readpages result %d\n", rc); | |
2913 | return rc; | |
2914 | } | |
3d14c5d2 | 2915 | EXPORT_SYMBOL(ceph_osdc_readpages); |
f24e9980 SW |
2916 | |
2917 | /* | |
2918 | * do a synchronous write on N pages | |
2919 | */ | |
2920 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, | |
2921 | struct ceph_file_layout *layout, | |
2922 | struct ceph_snap_context *snapc, | |
2923 | u64 off, u64 len, | |
2924 | u32 truncate_seq, u64 truncate_size, | |
2925 | struct timespec *mtime, | |
24808826 | 2926 | struct page **pages, int num_pages) |
f24e9980 SW |
2927 | { |
2928 | struct ceph_osd_request *req; | |
2929 | int rc = 0; | |
b7495fc2 | 2930 | int page_align = off & ~PAGE_MASK; |
f24e9980 | 2931 | |
715e4cd4 | 2932 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1, |
f24e9980 | 2933 | CEPH_OSD_OP_WRITE, |
24808826 | 2934 | CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE, |
acead002 | 2935 | snapc, truncate_seq, truncate_size, |
153e5167 | 2936 | true); |
6816282d SW |
2937 | if (IS_ERR(req)) |
2938 | return PTR_ERR(req); | |
f24e9980 SW |
2939 | |
2940 | /* it may be a short write due to an object boundary */ | |
406e2c9f | 2941 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, |
43bfe5de AE |
2942 | false, false); |
2943 | dout("writepages %llu~%llu (%llu bytes)\n", off, len, len); | |
f24e9980 | 2944 | |
bb873b53 | 2945 | req->r_mtime = *mtime; |
87f979d3 | 2946 | rc = ceph_osdc_start_request(osdc, req, true); |
f24e9980 SW |
2947 | if (!rc) |
2948 | rc = ceph_osdc_wait_request(osdc, req); | |
2949 | ||
2950 | ceph_osdc_put_request(req); | |
2951 | if (rc == 0) | |
2952 | rc = len; | |
2953 | dout("writepages result %d\n", rc); | |
2954 | return rc; | |
2955 | } | |
3d14c5d2 | 2956 | EXPORT_SYMBOL(ceph_osdc_writepages); |
f24e9980 | 2957 | |
5522ae0b AE |
2958 | int ceph_osdc_setup(void) |
2959 | { | |
3f1af42a ID |
2960 | size_t size = sizeof(struct ceph_osd_request) + |
2961 | CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op); | |
2962 | ||
5522ae0b | 2963 | BUG_ON(ceph_osd_request_cache); |
3f1af42a ID |
2964 | ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size, |
2965 | 0, 0, NULL); | |
5522ae0b AE |
2966 | |
2967 | return ceph_osd_request_cache ? 0 : -ENOMEM; | |
2968 | } | |
2969 | EXPORT_SYMBOL(ceph_osdc_setup); | |
2970 | ||
2971 | void ceph_osdc_cleanup(void) | |
2972 | { | |
2973 | BUG_ON(!ceph_osd_request_cache); | |
2974 | kmem_cache_destroy(ceph_osd_request_cache); | |
2975 | ceph_osd_request_cache = NULL; | |
2976 | } | |
2977 | EXPORT_SYMBOL(ceph_osdc_cleanup); | |
2978 | ||
f24e9980 SW |
2979 | /* |
2980 | * handle incoming message | |
2981 | */ | |
2982 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
2983 | { | |
2984 | struct ceph_osd *osd = con->private; | |
32c895e7 | 2985 | struct ceph_osd_client *osdc; |
f24e9980 SW |
2986 | int type = le16_to_cpu(msg->hdr.type); |
2987 | ||
2988 | if (!osd) | |
4a32f93d | 2989 | goto out; |
32c895e7 | 2990 | osdc = osd->o_osdc; |
f24e9980 SW |
2991 | |
2992 | switch (type) { | |
2993 | case CEPH_MSG_OSD_MAP: | |
2994 | ceph_osdc_handle_map(osdc, msg); | |
2995 | break; | |
2996 | case CEPH_MSG_OSD_OPREPLY: | |
70cf052d | 2997 | handle_reply(osdc, msg); |
f24e9980 | 2998 | break; |
a40c4f10 YS |
2999 | case CEPH_MSG_WATCH_NOTIFY: |
3000 | handle_watch_notify(osdc, msg); | |
3001 | break; | |
f24e9980 SW |
3002 | |
3003 | default: | |
3004 | pr_err("received unknown message type %d %s\n", type, | |
3005 | ceph_msg_type_name(type)); | |
3006 | } | |
4a32f93d | 3007 | out: |
f24e9980 SW |
3008 | ceph_msg_put(msg); |
3009 | } | |
3010 | ||
5b3a4db3 | 3011 | /* |
d15f9d69 ID |
3012 | * Lookup and return message for incoming reply. Don't try to do |
3013 | * anything about a larger than preallocated data portion of the | |
3014 | * message at the moment - for now, just skip the message. | |
5b3a4db3 SW |
3015 | */ |
3016 | static struct ceph_msg *get_reply(struct ceph_connection *con, | |
2450418c YS |
3017 | struct ceph_msg_header *hdr, |
3018 | int *skip) | |
f24e9980 SW |
3019 | { |
3020 | struct ceph_osd *osd = con->private; | |
3021 | struct ceph_osd_client *osdc = osd->o_osdc; | |
2450418c | 3022 | struct ceph_msg *m; |
0547a9b3 | 3023 | struct ceph_osd_request *req; |
3f0a4ac5 | 3024 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 3025 | int data_len = le32_to_cpu(hdr->data_len); |
0547a9b3 | 3026 | u64 tid; |
f24e9980 | 3027 | |
0547a9b3 YS |
3028 | tid = le64_to_cpu(hdr->tid); |
3029 | mutex_lock(&osdc->request_mutex); | |
fcd00b68 | 3030 | req = lookup_request(&osdc->requests, tid); |
0547a9b3 | 3031 | if (!req) { |
cd8140c6 ID |
3032 | dout("%s osd%d tid %llu unknown, skipping\n", __func__, |
3033 | osd->o_osd, tid); | |
0547a9b3 | 3034 | m = NULL; |
d15f9d69 | 3035 | *skip = 1; |
0547a9b3 YS |
3036 | goto out; |
3037 | } | |
c16e7869 | 3038 | |
ace6d3a9 | 3039 | ceph_msg_revoke_incoming(req->r_reply); |
0547a9b3 | 3040 | |
f2be82b0 | 3041 | if (front_len > req->r_reply->front_alloc_len) { |
d15f9d69 ID |
3042 | pr_warn("%s osd%d tid %llu front %d > preallocated %d\n", |
3043 | __func__, osd->o_osd, req->r_tid, front_len, | |
3044 | req->r_reply->front_alloc_len); | |
3f0a4ac5 ID |
3045 | m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS, |
3046 | false); | |
a79832f2 | 3047 | if (!m) |
c16e7869 SW |
3048 | goto out; |
3049 | ceph_msg_put(req->r_reply); | |
3050 | req->r_reply = m; | |
3051 | } | |
0fff87ec | 3052 | |
d15f9d69 ID |
3053 | if (data_len > req->r_reply->data_length) { |
3054 | pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n", | |
3055 | __func__, osd->o_osd, req->r_tid, data_len, | |
3056 | req->r_reply->data_length); | |
3057 | m = NULL; | |
3058 | *skip = 1; | |
3059 | goto out; | |
0547a9b3 | 3060 | } |
d15f9d69 ID |
3061 | |
3062 | m = ceph_msg_get(req->r_reply); | |
c16e7869 | 3063 | dout("get_reply tid %lld %p\n", tid, m); |
0547a9b3 YS |
3064 | |
3065 | out: | |
3066 | mutex_unlock(&osdc->request_mutex); | |
2450418c | 3067 | return m; |
5b3a4db3 SW |
3068 | } |
3069 | ||
3070 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, | |
3071 | struct ceph_msg_header *hdr, | |
3072 | int *skip) | |
3073 | { | |
3074 | struct ceph_osd *osd = con->private; | |
3075 | int type = le16_to_cpu(hdr->type); | |
3076 | int front = le32_to_cpu(hdr->front_len); | |
3077 | ||
1c20f2d2 | 3078 | *skip = 0; |
5b3a4db3 SW |
3079 | switch (type) { |
3080 | case CEPH_MSG_OSD_MAP: | |
a40c4f10 | 3081 | case CEPH_MSG_WATCH_NOTIFY: |
b61c2763 | 3082 | return ceph_msg_new(type, front, GFP_NOFS, false); |
5b3a4db3 SW |
3083 | case CEPH_MSG_OSD_OPREPLY: |
3084 | return get_reply(con, hdr, skip); | |
3085 | default: | |
3086 | pr_info("alloc_msg unexpected msg type %d from osd%d\n", type, | |
3087 | osd->o_osd); | |
3088 | *skip = 1; | |
3089 | return NULL; | |
3090 | } | |
f24e9980 SW |
3091 | } |
3092 | ||
3093 | /* | |
3094 | * Wrappers to refcount containing ceph_osd struct | |
3095 | */ | |
3096 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) | |
3097 | { | |
3098 | struct ceph_osd *osd = con->private; | |
3099 | if (get_osd(osd)) | |
3100 | return con; | |
3101 | return NULL; | |
3102 | } | |
3103 | ||
3104 | static void put_osd_con(struct ceph_connection *con) | |
3105 | { | |
3106 | struct ceph_osd *osd = con->private; | |
3107 | put_osd(osd); | |
3108 | } | |
3109 | ||
4e7a5dcd SW |
3110 | /* |
3111 | * authentication | |
3112 | */ | |
a3530df3 AE |
3113 | /* |
3114 | * Note: returned pointer is the address of a structure that's | |
3115 | * managed separately. Caller must *not* attempt to free it. | |
3116 | */ | |
3117 | static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, | |
8f43fb53 | 3118 | int *proto, int force_new) |
4e7a5dcd SW |
3119 | { |
3120 | struct ceph_osd *o = con->private; | |
3121 | struct ceph_osd_client *osdc = o->o_osdc; | |
3122 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
74f1869f | 3123 | struct ceph_auth_handshake *auth = &o->o_auth; |
4e7a5dcd | 3124 | |
74f1869f | 3125 | if (force_new && auth->authorizer) { |
6c1ea260 | 3126 | ceph_auth_destroy_authorizer(auth->authorizer); |
74f1869f AE |
3127 | auth->authorizer = NULL; |
3128 | } | |
27859f97 SW |
3129 | if (!auth->authorizer) { |
3130 | int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
3131 | auth); | |
4e7a5dcd | 3132 | if (ret) |
a3530df3 | 3133 | return ERR_PTR(ret); |
27859f97 SW |
3134 | } else { |
3135 | int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
0bed9b5c SW |
3136 | auth); |
3137 | if (ret) | |
3138 | return ERR_PTR(ret); | |
4e7a5dcd | 3139 | } |
4e7a5dcd | 3140 | *proto = ac->protocol; |
74f1869f | 3141 | |
a3530df3 | 3142 | return auth; |
4e7a5dcd SW |
3143 | } |
3144 | ||
3145 | ||
3146 | static int verify_authorizer_reply(struct ceph_connection *con, int len) | |
3147 | { | |
3148 | struct ceph_osd *o = con->private; | |
3149 | struct ceph_osd_client *osdc = o->o_osdc; | |
3150 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
3151 | ||
27859f97 | 3152 | return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len); |
4e7a5dcd SW |
3153 | } |
3154 | ||
9bd2e6f8 SW |
3155 | static int invalidate_authorizer(struct ceph_connection *con) |
3156 | { | |
3157 | struct ceph_osd *o = con->private; | |
3158 | struct ceph_osd_client *osdc = o->o_osdc; | |
3159 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
3160 | ||
27859f97 | 3161 | ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); |
9bd2e6f8 SW |
3162 | return ceph_monc_validate_auth(&osdc->client->monc); |
3163 | } | |
4e7a5dcd | 3164 | |
79dbd1ba | 3165 | static int osd_sign_message(struct ceph_msg *msg) |
33d07337 | 3166 | { |
79dbd1ba | 3167 | struct ceph_osd *o = msg->con->private; |
33d07337 | 3168 | struct ceph_auth_handshake *auth = &o->o_auth; |
79dbd1ba | 3169 | |
33d07337 YZ |
3170 | return ceph_auth_sign_message(auth, msg); |
3171 | } | |
3172 | ||
79dbd1ba | 3173 | static int osd_check_message_signature(struct ceph_msg *msg) |
33d07337 | 3174 | { |
79dbd1ba | 3175 | struct ceph_osd *o = msg->con->private; |
33d07337 | 3176 | struct ceph_auth_handshake *auth = &o->o_auth; |
79dbd1ba | 3177 | |
33d07337 YZ |
3178 | return ceph_auth_check_message_signature(auth, msg); |
3179 | } | |
3180 | ||
9e32789f | 3181 | static const struct ceph_connection_operations osd_con_ops = { |
f24e9980 SW |
3182 | .get = get_osd_con, |
3183 | .put = put_osd_con, | |
3184 | .dispatch = dispatch, | |
4e7a5dcd SW |
3185 | .get_authorizer = get_authorizer, |
3186 | .verify_authorizer_reply = verify_authorizer_reply, | |
9bd2e6f8 | 3187 | .invalidate_authorizer = invalidate_authorizer, |
f24e9980 | 3188 | .alloc_msg = alloc_msg, |
79dbd1ba ID |
3189 | .sign_message = osd_sign_message, |
3190 | .check_message_signature = osd_check_message_signature, | |
81b024e7 | 3191 | .fault = osd_reset, |
f24e9980 | 3192 | }; |