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 | |
f24e9980 SW |
28 | /* |
29 | * Implement client access to distributed object storage cluster. | |
30 | * | |
31 | * All data objects are stored within a cluster/cloud of OSDs, or | |
32 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to | |
33 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply | |
34 | * remote daemons serving up and coordinating consistent and safe | |
35 | * access to storage. | |
36 | * | |
37 | * Cluster membership and the mapping of data objects onto storage devices | |
38 | * are described by the osd map. | |
39 | * | |
40 | * We keep track of pending OSD requests (read, write), resubmit | |
41 | * requests to different OSDs when the cluster topology/data layout | |
42 | * change, or retry the affected requests when the communications | |
43 | * channel with an OSD is reset. | |
44 | */ | |
45 | ||
5aea3dcd ID |
46 | static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req); |
47 | static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req); | |
922dab61 ID |
48 | static void link_linger(struct ceph_osd *osd, |
49 | struct ceph_osd_linger_request *lreq); | |
50 | static void unlink_linger(struct ceph_osd *osd, | |
51 | struct ceph_osd_linger_request *lreq); | |
5aea3dcd ID |
52 | |
53 | #if 1 | |
54 | static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem) | |
55 | { | |
56 | bool wrlocked = true; | |
57 | ||
58 | if (unlikely(down_read_trylock(sem))) { | |
59 | wrlocked = false; | |
60 | up_read(sem); | |
61 | } | |
62 | ||
63 | return wrlocked; | |
64 | } | |
65 | static inline void verify_osdc_locked(struct ceph_osd_client *osdc) | |
66 | { | |
67 | WARN_ON(!rwsem_is_locked(&osdc->lock)); | |
68 | } | |
69 | static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) | |
70 | { | |
71 | WARN_ON(!rwsem_is_wrlocked(&osdc->lock)); | |
72 | } | |
73 | static inline void verify_osd_locked(struct ceph_osd *osd) | |
74 | { | |
75 | struct ceph_osd_client *osdc = osd->o_osdc; | |
76 | ||
77 | WARN_ON(!(mutex_is_locked(&osd->lock) && | |
78 | rwsem_is_locked(&osdc->lock)) && | |
79 | !rwsem_is_wrlocked(&osdc->lock)); | |
80 | } | |
922dab61 ID |
81 | static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) |
82 | { | |
83 | WARN_ON(!mutex_is_locked(&lreq->lock)); | |
84 | } | |
5aea3dcd ID |
85 | #else |
86 | static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { } | |
87 | static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { } | |
88 | static inline void verify_osd_locked(struct ceph_osd *osd) { } | |
922dab61 | 89 | static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { } |
5aea3dcd ID |
90 | #endif |
91 | ||
f24e9980 SW |
92 | /* |
93 | * calculate the mapping of a file extent onto an object, and fill out the | |
94 | * request accordingly. shorten extent as necessary if it crosses an | |
95 | * object boundary. | |
96 | * | |
97 | * fill osd op in request message. | |
98 | */ | |
dbe0fc41 | 99 | static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen, |
a19dadfb | 100 | u64 *objnum, u64 *objoff, u64 *objlen) |
f24e9980 | 101 | { |
60e56f13 | 102 | u64 orig_len = *plen; |
d63b77f4 | 103 | int r; |
f24e9980 | 104 | |
60e56f13 | 105 | /* object extent? */ |
75d1c941 AE |
106 | r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum, |
107 | objoff, objlen); | |
d63b77f4 SW |
108 | if (r < 0) |
109 | return r; | |
75d1c941 AE |
110 | if (*objlen < orig_len) { |
111 | *plen = *objlen; | |
60e56f13 AE |
112 | dout(" skipping last %llu, final file extent %llu~%llu\n", |
113 | orig_len - *plen, off, *plen); | |
114 | } | |
115 | ||
75d1c941 | 116 | dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen); |
f24e9980 | 117 | |
3ff5f385 | 118 | return 0; |
f24e9980 SW |
119 | } |
120 | ||
c54d47bf AE |
121 | static void ceph_osd_data_init(struct ceph_osd_data *osd_data) |
122 | { | |
123 | memset(osd_data, 0, sizeof (*osd_data)); | |
124 | osd_data->type = CEPH_OSD_DATA_TYPE_NONE; | |
125 | } | |
126 | ||
a4ce40a9 | 127 | static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
128 | struct page **pages, u64 length, u32 alignment, |
129 | bool pages_from_pool, bool own_pages) | |
130 | { | |
131 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; | |
132 | osd_data->pages = pages; | |
133 | osd_data->length = length; | |
134 | osd_data->alignment = alignment; | |
135 | osd_data->pages_from_pool = pages_from_pool; | |
136 | osd_data->own_pages = own_pages; | |
137 | } | |
43bfe5de | 138 | |
a4ce40a9 | 139 | static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
140 | struct ceph_pagelist *pagelist) |
141 | { | |
142 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST; | |
143 | osd_data->pagelist = pagelist; | |
144 | } | |
43bfe5de AE |
145 | |
146 | #ifdef CONFIG_BLOCK | |
a4ce40a9 | 147 | static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
148 | struct bio *bio, size_t bio_length) |
149 | { | |
150 | osd_data->type = CEPH_OSD_DATA_TYPE_BIO; | |
151 | osd_data->bio = bio; | |
152 | osd_data->bio_length = bio_length; | |
153 | } | |
43bfe5de AE |
154 | #endif /* CONFIG_BLOCK */ |
155 | ||
8a703a38 IC |
156 | #define osd_req_op_data(oreq, whch, typ, fld) \ |
157 | ({ \ | |
158 | struct ceph_osd_request *__oreq = (oreq); \ | |
159 | unsigned int __whch = (whch); \ | |
160 | BUG_ON(__whch >= __oreq->r_num_ops); \ | |
161 | &__oreq->r_ops[__whch].typ.fld; \ | |
162 | }) | |
863c7eb5 | 163 | |
49719778 AE |
164 | static struct ceph_osd_data * |
165 | osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which) | |
166 | { | |
167 | BUG_ON(which >= osd_req->r_num_ops); | |
168 | ||
169 | return &osd_req->r_ops[which].raw_data_in; | |
170 | } | |
171 | ||
a4ce40a9 AE |
172 | struct ceph_osd_data * |
173 | osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req, | |
406e2c9f | 174 | unsigned int which) |
a4ce40a9 | 175 | { |
863c7eb5 | 176 | return osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 AE |
177 | } |
178 | EXPORT_SYMBOL(osd_req_op_extent_osd_data); | |
179 | ||
49719778 AE |
180 | void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req, |
181 | unsigned int which, struct page **pages, | |
182 | u64 length, u32 alignment, | |
183 | bool pages_from_pool, bool own_pages) | |
184 | { | |
185 | struct ceph_osd_data *osd_data; | |
186 | ||
187 | osd_data = osd_req_op_raw_data_in(osd_req, which); | |
188 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, | |
189 | pages_from_pool, own_pages); | |
190 | } | |
191 | EXPORT_SYMBOL(osd_req_op_raw_data_in_pages); | |
192 | ||
a4ce40a9 | 193 | void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req, |
406e2c9f AE |
194 | unsigned int which, struct page **pages, |
195 | u64 length, u32 alignment, | |
a4ce40a9 AE |
196 | bool pages_from_pool, bool own_pages) |
197 | { | |
198 | struct ceph_osd_data *osd_data; | |
199 | ||
863c7eb5 | 200 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 AE |
201 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, |
202 | pages_from_pool, own_pages); | |
a4ce40a9 AE |
203 | } |
204 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages); | |
205 | ||
206 | void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req, | |
406e2c9f | 207 | unsigned int which, struct ceph_pagelist *pagelist) |
a4ce40a9 AE |
208 | { |
209 | struct ceph_osd_data *osd_data; | |
210 | ||
863c7eb5 | 211 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 | 212 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
a4ce40a9 AE |
213 | } |
214 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist); | |
215 | ||
216 | #ifdef CONFIG_BLOCK | |
217 | void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req, | |
406e2c9f | 218 | unsigned int which, struct bio *bio, size_t bio_length) |
a4ce40a9 AE |
219 | { |
220 | struct ceph_osd_data *osd_data; | |
863c7eb5 AE |
221 | |
222 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); | |
a4ce40a9 | 223 | ceph_osd_data_bio_init(osd_data, bio, bio_length); |
a4ce40a9 AE |
224 | } |
225 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio); | |
226 | #endif /* CONFIG_BLOCK */ | |
227 | ||
228 | static void osd_req_op_cls_request_info_pagelist( | |
229 | struct ceph_osd_request *osd_req, | |
230 | unsigned int which, struct ceph_pagelist *pagelist) | |
231 | { | |
232 | struct ceph_osd_data *osd_data; | |
233 | ||
863c7eb5 | 234 | osd_data = osd_req_op_data(osd_req, which, cls, request_info); |
a4ce40a9 | 235 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
a4ce40a9 AE |
236 | } |
237 | ||
04017e29 AE |
238 | void osd_req_op_cls_request_data_pagelist( |
239 | struct ceph_osd_request *osd_req, | |
240 | unsigned int which, struct ceph_pagelist *pagelist) | |
241 | { | |
242 | struct ceph_osd_data *osd_data; | |
243 | ||
863c7eb5 | 244 | osd_data = osd_req_op_data(osd_req, which, cls, request_data); |
04017e29 | 245 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
bb873b53 ID |
246 | osd_req->r_ops[which].cls.indata_len += pagelist->length; |
247 | osd_req->r_ops[which].indata_len += pagelist->length; | |
04017e29 AE |
248 | } |
249 | EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist); | |
250 | ||
6c57b554 AE |
251 | void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req, |
252 | unsigned int which, struct page **pages, u64 length, | |
253 | u32 alignment, bool pages_from_pool, bool own_pages) | |
254 | { | |
255 | struct ceph_osd_data *osd_data; | |
256 | ||
257 | osd_data = osd_req_op_data(osd_req, which, cls, request_data); | |
258 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, | |
259 | pages_from_pool, own_pages); | |
bb873b53 ID |
260 | osd_req->r_ops[which].cls.indata_len += length; |
261 | osd_req->r_ops[which].indata_len += length; | |
6c57b554 AE |
262 | } |
263 | EXPORT_SYMBOL(osd_req_op_cls_request_data_pages); | |
264 | ||
a4ce40a9 AE |
265 | void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req, |
266 | unsigned int which, struct page **pages, u64 length, | |
267 | u32 alignment, bool pages_from_pool, bool own_pages) | |
268 | { | |
269 | struct ceph_osd_data *osd_data; | |
270 | ||
863c7eb5 | 271 | osd_data = osd_req_op_data(osd_req, which, cls, response_data); |
a4ce40a9 AE |
272 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, |
273 | pages_from_pool, own_pages); | |
a4ce40a9 AE |
274 | } |
275 | EXPORT_SYMBOL(osd_req_op_cls_response_data_pages); | |
276 | ||
23c08a9c AE |
277 | static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data) |
278 | { | |
279 | switch (osd_data->type) { | |
280 | case CEPH_OSD_DATA_TYPE_NONE: | |
281 | return 0; | |
282 | case CEPH_OSD_DATA_TYPE_PAGES: | |
283 | return osd_data->length; | |
284 | case CEPH_OSD_DATA_TYPE_PAGELIST: | |
285 | return (u64)osd_data->pagelist->length; | |
286 | #ifdef CONFIG_BLOCK | |
287 | case CEPH_OSD_DATA_TYPE_BIO: | |
288 | return (u64)osd_data->bio_length; | |
289 | #endif /* CONFIG_BLOCK */ | |
290 | default: | |
291 | WARN(true, "unrecognized data type %d\n", (int)osd_data->type); | |
292 | return 0; | |
293 | } | |
294 | } | |
295 | ||
c54d47bf AE |
296 | static void ceph_osd_data_release(struct ceph_osd_data *osd_data) |
297 | { | |
5476492f | 298 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) { |
c54d47bf AE |
299 | int num_pages; |
300 | ||
301 | num_pages = calc_pages_for((u64)osd_data->alignment, | |
302 | (u64)osd_data->length); | |
303 | ceph_release_page_vector(osd_data->pages, num_pages); | |
304 | } | |
5476492f AE |
305 | ceph_osd_data_init(osd_data); |
306 | } | |
307 | ||
308 | static void osd_req_op_data_release(struct ceph_osd_request *osd_req, | |
309 | unsigned int which) | |
310 | { | |
311 | struct ceph_osd_req_op *op; | |
312 | ||
313 | BUG_ON(which >= osd_req->r_num_ops); | |
314 | op = &osd_req->r_ops[which]; | |
315 | ||
316 | switch (op->op) { | |
317 | case CEPH_OSD_OP_READ: | |
318 | case CEPH_OSD_OP_WRITE: | |
e30b7577 | 319 | case CEPH_OSD_OP_WRITEFULL: |
5476492f AE |
320 | ceph_osd_data_release(&op->extent.osd_data); |
321 | break; | |
322 | case CEPH_OSD_OP_CALL: | |
323 | ceph_osd_data_release(&op->cls.request_info); | |
04017e29 | 324 | ceph_osd_data_release(&op->cls.request_data); |
5476492f AE |
325 | ceph_osd_data_release(&op->cls.response_data); |
326 | break; | |
d74b50be YZ |
327 | case CEPH_OSD_OP_SETXATTR: |
328 | case CEPH_OSD_OP_CMPXATTR: | |
329 | ceph_osd_data_release(&op->xattr.osd_data); | |
330 | break; | |
66ba609f YZ |
331 | case CEPH_OSD_OP_STAT: |
332 | ceph_osd_data_release(&op->raw_data_in); | |
333 | break; | |
922dab61 ID |
334 | case CEPH_OSD_OP_NOTIFY_ACK: |
335 | ceph_osd_data_release(&op->notify_ack.request_data); | |
336 | break; | |
19079203 ID |
337 | case CEPH_OSD_OP_NOTIFY: |
338 | ceph_osd_data_release(&op->notify.request_data); | |
339 | ceph_osd_data_release(&op->notify.response_data); | |
340 | break; | |
a4ed38d7 DF |
341 | case CEPH_OSD_OP_LIST_WATCHERS: |
342 | ceph_osd_data_release(&op->list_watchers.response_data); | |
343 | break; | |
5476492f AE |
344 | default: |
345 | break; | |
346 | } | |
c54d47bf AE |
347 | } |
348 | ||
63244fa1 ID |
349 | /* |
350 | * Assumes @t is zero-initialized. | |
351 | */ | |
352 | static void target_init(struct ceph_osd_request_target *t) | |
353 | { | |
354 | ceph_oid_init(&t->base_oid); | |
355 | ceph_oloc_init(&t->base_oloc); | |
356 | ceph_oid_init(&t->target_oid); | |
357 | ceph_oloc_init(&t->target_oloc); | |
358 | ||
359 | ceph_osds_init(&t->acting); | |
360 | ceph_osds_init(&t->up); | |
361 | t->size = -1; | |
362 | t->min_size = -1; | |
363 | ||
364 | t->osd = CEPH_HOMELESS_OSD; | |
365 | } | |
366 | ||
922dab61 ID |
367 | static void target_copy(struct ceph_osd_request_target *dest, |
368 | const struct ceph_osd_request_target *src) | |
369 | { | |
370 | ceph_oid_copy(&dest->base_oid, &src->base_oid); | |
371 | ceph_oloc_copy(&dest->base_oloc, &src->base_oloc); | |
372 | ceph_oid_copy(&dest->target_oid, &src->target_oid); | |
373 | ceph_oloc_copy(&dest->target_oloc, &src->target_oloc); | |
374 | ||
375 | dest->pgid = src->pgid; /* struct */ | |
376 | dest->pg_num = src->pg_num; | |
377 | dest->pg_num_mask = src->pg_num_mask; | |
378 | ceph_osds_copy(&dest->acting, &src->acting); | |
379 | ceph_osds_copy(&dest->up, &src->up); | |
380 | dest->size = src->size; | |
381 | dest->min_size = src->min_size; | |
382 | dest->sort_bitwise = src->sort_bitwise; | |
383 | ||
384 | dest->flags = src->flags; | |
385 | dest->paused = src->paused; | |
386 | ||
387 | dest->osd = src->osd; | |
388 | } | |
389 | ||
63244fa1 ID |
390 | static void target_destroy(struct ceph_osd_request_target *t) |
391 | { | |
392 | ceph_oid_destroy(&t->base_oid); | |
30c156d9 | 393 | ceph_oloc_destroy(&t->base_oloc); |
63244fa1 | 394 | ceph_oid_destroy(&t->target_oid); |
30c156d9 | 395 | ceph_oloc_destroy(&t->target_oloc); |
63244fa1 ID |
396 | } |
397 | ||
f24e9980 SW |
398 | /* |
399 | * requests | |
400 | */ | |
3540bfdb ID |
401 | static void request_release_checks(struct ceph_osd_request *req) |
402 | { | |
403 | WARN_ON(!RB_EMPTY_NODE(&req->r_node)); | |
4609245e | 404 | WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node)); |
3540bfdb ID |
405 | WARN_ON(!list_empty(&req->r_unsafe_item)); |
406 | WARN_ON(req->r_osd); | |
407 | } | |
408 | ||
9e94af20 | 409 | static void ceph_osdc_release_request(struct kref *kref) |
f24e9980 | 410 | { |
9e94af20 ID |
411 | struct ceph_osd_request *req = container_of(kref, |
412 | struct ceph_osd_request, r_kref); | |
5476492f | 413 | unsigned int which; |
415e49a9 | 414 | |
9e94af20 ID |
415 | dout("%s %p (r_request %p r_reply %p)\n", __func__, req, |
416 | req->r_request, req->r_reply); | |
3540bfdb | 417 | request_release_checks(req); |
9e94af20 | 418 | |
415e49a9 SW |
419 | if (req->r_request) |
420 | ceph_msg_put(req->r_request); | |
5aea3dcd | 421 | if (req->r_reply) |
ab8cb34a | 422 | ceph_msg_put(req->r_reply); |
0fff87ec | 423 | |
5476492f AE |
424 | for (which = 0; which < req->r_num_ops; which++) |
425 | osd_req_op_data_release(req, which); | |
0fff87ec | 426 | |
a66dd383 | 427 | target_destroy(&req->r_t); |
415e49a9 | 428 | ceph_put_snap_context(req->r_snapc); |
d30291b9 | 429 | |
415e49a9 SW |
430 | if (req->r_mempool) |
431 | mempool_free(req, req->r_osdc->req_mempool); | |
3f1af42a | 432 | else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS) |
5522ae0b | 433 | kmem_cache_free(ceph_osd_request_cache, req); |
3f1af42a ID |
434 | else |
435 | kfree(req); | |
f24e9980 | 436 | } |
9e94af20 ID |
437 | |
438 | void ceph_osdc_get_request(struct ceph_osd_request *req) | |
439 | { | |
440 | dout("%s %p (was %d)\n", __func__, req, | |
2c935bc5 | 441 | kref_read(&req->r_kref)); |
9e94af20 ID |
442 | kref_get(&req->r_kref); |
443 | } | |
444 | EXPORT_SYMBOL(ceph_osdc_get_request); | |
445 | ||
446 | void ceph_osdc_put_request(struct ceph_osd_request *req) | |
447 | { | |
3ed97d63 ID |
448 | if (req) { |
449 | dout("%s %p (was %d)\n", __func__, req, | |
2c935bc5 | 450 | kref_read(&req->r_kref)); |
3ed97d63 ID |
451 | kref_put(&req->r_kref, ceph_osdc_release_request); |
452 | } | |
9e94af20 ID |
453 | } |
454 | EXPORT_SYMBOL(ceph_osdc_put_request); | |
68b4476b | 455 | |
3540bfdb ID |
456 | static void request_init(struct ceph_osd_request *req) |
457 | { | |
458 | /* req only, each op is zeroed in _osd_req_op_init() */ | |
459 | memset(req, 0, sizeof(*req)); | |
460 | ||
461 | kref_init(&req->r_kref); | |
462 | init_completion(&req->r_completion); | |
3540bfdb | 463 | RB_CLEAR_NODE(&req->r_node); |
4609245e | 464 | RB_CLEAR_NODE(&req->r_mc_node); |
3540bfdb ID |
465 | INIT_LIST_HEAD(&req->r_unsafe_item); |
466 | ||
467 | target_init(&req->r_t); | |
468 | } | |
469 | ||
922dab61 ID |
470 | /* |
471 | * This is ugly, but it allows us to reuse linger registration and ping | |
472 | * requests, keeping the structure of the code around send_linger{_ping}() | |
473 | * reasonable. Setting up a min_nr=2 mempool for each linger request | |
474 | * and dealing with copying ops (this blasts req only, watch op remains | |
475 | * intact) isn't any better. | |
476 | */ | |
477 | static void request_reinit(struct ceph_osd_request *req) | |
478 | { | |
479 | struct ceph_osd_client *osdc = req->r_osdc; | |
480 | bool mempool = req->r_mempool; | |
481 | unsigned int num_ops = req->r_num_ops; | |
482 | u64 snapid = req->r_snapid; | |
483 | struct ceph_snap_context *snapc = req->r_snapc; | |
484 | bool linger = req->r_linger; | |
485 | struct ceph_msg *request_msg = req->r_request; | |
486 | struct ceph_msg *reply_msg = req->r_reply; | |
487 | ||
488 | dout("%s req %p\n", __func__, req); | |
2c935bc5 | 489 | WARN_ON(kref_read(&req->r_kref) != 1); |
922dab61 ID |
490 | request_release_checks(req); |
491 | ||
2c935bc5 PZ |
492 | WARN_ON(kref_read(&request_msg->kref) != 1); |
493 | WARN_ON(kref_read(&reply_msg->kref) != 1); | |
922dab61 ID |
494 | target_destroy(&req->r_t); |
495 | ||
496 | request_init(req); | |
497 | req->r_osdc = osdc; | |
498 | req->r_mempool = mempool; | |
499 | req->r_num_ops = num_ops; | |
500 | req->r_snapid = snapid; | |
501 | req->r_snapc = snapc; | |
502 | req->r_linger = linger; | |
503 | req->r_request = request_msg; | |
504 | req->r_reply = reply_msg; | |
505 | } | |
506 | ||
3499e8a5 | 507 | struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, |
f24e9980 | 508 | struct ceph_snap_context *snapc, |
1b83bef2 | 509 | unsigned int num_ops, |
3499e8a5 | 510 | bool use_mempool, |
54a54007 | 511 | gfp_t gfp_flags) |
f24e9980 SW |
512 | { |
513 | struct ceph_osd_request *req; | |
1b83bef2 | 514 | |
f24e9980 | 515 | if (use_mempool) { |
3f1af42a | 516 | BUG_ON(num_ops > CEPH_OSD_SLAB_OPS); |
3499e8a5 | 517 | req = mempool_alloc(osdc->req_mempool, gfp_flags); |
3f1af42a ID |
518 | } else if (num_ops <= CEPH_OSD_SLAB_OPS) { |
519 | req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags); | |
f24e9980 | 520 | } else { |
3f1af42a ID |
521 | BUG_ON(num_ops > CEPH_OSD_MAX_OPS); |
522 | req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]), | |
523 | gfp_flags); | |
f24e9980 | 524 | } |
3f1af42a | 525 | if (unlikely(!req)) |
a79832f2 | 526 | return NULL; |
f24e9980 | 527 | |
3540bfdb | 528 | request_init(req); |
f24e9980 SW |
529 | req->r_osdc = osdc; |
530 | req->r_mempool = use_mempool; | |
79528734 | 531 | req->r_num_ops = num_ops; |
84127282 ID |
532 | req->r_snapid = CEPH_NOSNAP; |
533 | req->r_snapc = ceph_get_snap_context(snapc); | |
68b4476b | 534 | |
13d1ad16 ID |
535 | dout("%s req %p\n", __func__, req); |
536 | return req; | |
537 | } | |
538 | EXPORT_SYMBOL(ceph_osdc_alloc_request); | |
3f1af42a | 539 | |
30c156d9 YZ |
540 | static int ceph_oloc_encoding_size(struct ceph_object_locator *oloc) |
541 | { | |
542 | return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0); | |
543 | } | |
544 | ||
13d1ad16 ID |
545 | int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp) |
546 | { | |
547 | struct ceph_osd_client *osdc = req->r_osdc; | |
548 | struct ceph_msg *msg; | |
549 | int msg_size; | |
c16e7869 | 550 | |
d30291b9 | 551 | WARN_ON(ceph_oid_empty(&req->r_base_oid)); |
30c156d9 | 552 | WARN_ON(ceph_oloc_empty(&req->r_base_oloc)); |
d30291b9 | 553 | |
13d1ad16 | 554 | /* create request message */ |
ae458f5a ID |
555 | msg_size = 4 + 4 + 4; /* client_inc, osdmap_epoch, flags */ |
556 | msg_size += 4 + 4 + 4 + 8; /* mtime, reassert_version */ | |
30c156d9 YZ |
557 | msg_size += CEPH_ENCODING_START_BLK_LEN + |
558 | ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */ | |
ae458f5a | 559 | msg_size += 1 + 8 + 4 + 4; /* pgid */ |
13d1ad16 ID |
560 | msg_size += 4 + req->r_base_oid.name_len; /* oid */ |
561 | msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op); | |
ae458f5a ID |
562 | msg_size += 8; /* snapid */ |
563 | msg_size += 8; /* snap_seq */ | |
13d1ad16 | 564 | msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0); |
ae458f5a ID |
565 | msg_size += 4; /* retry_attempt */ |
566 | ||
13d1ad16 | 567 | if (req->r_mempool) |
8f3bc053 | 568 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
f24e9980 | 569 | else |
13d1ad16 ID |
570 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true); |
571 | if (!msg) | |
572 | return -ENOMEM; | |
68b4476b | 573 | |
f24e9980 | 574 | memset(msg->front.iov_base, 0, msg->front.iov_len); |
3499e8a5 | 575 | req->r_request = msg; |
3499e8a5 | 576 | |
13d1ad16 ID |
577 | /* create reply message */ |
578 | msg_size = OSD_OPREPLY_FRONT_LEN; | |
711da55d ID |
579 | msg_size += req->r_base_oid.name_len; |
580 | msg_size += req->r_num_ops * sizeof(struct ceph_osd_op); | |
13d1ad16 ID |
581 | |
582 | if (req->r_mempool) | |
583 | msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); | |
584 | else | |
585 | msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true); | |
586 | if (!msg) | |
587 | return -ENOMEM; | |
588 | ||
589 | req->r_reply = msg; | |
590 | ||
591 | return 0; | |
3499e8a5 | 592 | } |
13d1ad16 | 593 | EXPORT_SYMBOL(ceph_osdc_alloc_messages); |
3499e8a5 | 594 | |
a8dd0a37 | 595 | static bool osd_req_opcode_valid(u16 opcode) |
68b4476b | 596 | { |
a8dd0a37 | 597 | switch (opcode) { |
70b5bfa3 ID |
598 | #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true; |
599 | __CEPH_FORALL_OSD_OPS(GENERATE_CASE) | |
600 | #undef GENERATE_CASE | |
a8dd0a37 AE |
601 | default: |
602 | return false; | |
603 | } | |
604 | } | |
605 | ||
33803f33 AE |
606 | /* |
607 | * This is an osd op init function for opcodes that have no data or | |
608 | * other information associated with them. It also serves as a | |
609 | * common init routine for all the other init functions, below. | |
610 | */ | |
c99d2d4a | 611 | static struct ceph_osd_req_op * |
49719778 | 612 | _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which, |
144cba14 | 613 | u16 opcode, u32 flags) |
33803f33 | 614 | { |
c99d2d4a AE |
615 | struct ceph_osd_req_op *op; |
616 | ||
617 | BUG_ON(which >= osd_req->r_num_ops); | |
33803f33 AE |
618 | BUG_ON(!osd_req_opcode_valid(opcode)); |
619 | ||
c99d2d4a | 620 | op = &osd_req->r_ops[which]; |
33803f33 | 621 | memset(op, 0, sizeof (*op)); |
33803f33 | 622 | op->op = opcode; |
144cba14 | 623 | op->flags = flags; |
c99d2d4a AE |
624 | |
625 | return op; | |
33803f33 AE |
626 | } |
627 | ||
49719778 | 628 | void osd_req_op_init(struct ceph_osd_request *osd_req, |
144cba14 | 629 | unsigned int which, u16 opcode, u32 flags) |
49719778 | 630 | { |
144cba14 | 631 | (void)_osd_req_op_init(osd_req, which, opcode, flags); |
49719778 AE |
632 | } |
633 | EXPORT_SYMBOL(osd_req_op_init); | |
634 | ||
c99d2d4a AE |
635 | void osd_req_op_extent_init(struct ceph_osd_request *osd_req, |
636 | unsigned int which, u16 opcode, | |
33803f33 AE |
637 | u64 offset, u64 length, |
638 | u64 truncate_size, u32 truncate_seq) | |
639 | { | |
144cba14 YZ |
640 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
641 | opcode, 0); | |
33803f33 AE |
642 | size_t payload_len = 0; |
643 | ||
ad7a60de | 644 | BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && |
e30b7577 ID |
645 | opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO && |
646 | opcode != CEPH_OSD_OP_TRUNCATE); | |
33803f33 | 647 | |
33803f33 AE |
648 | op->extent.offset = offset; |
649 | op->extent.length = length; | |
650 | op->extent.truncate_size = truncate_size; | |
651 | op->extent.truncate_seq = truncate_seq; | |
e30b7577 | 652 | if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL) |
33803f33 AE |
653 | payload_len += length; |
654 | ||
de2aa102 | 655 | op->indata_len = payload_len; |
33803f33 AE |
656 | } |
657 | EXPORT_SYMBOL(osd_req_op_extent_init); | |
658 | ||
c99d2d4a AE |
659 | void osd_req_op_extent_update(struct ceph_osd_request *osd_req, |
660 | unsigned int which, u64 length) | |
e5975c7c | 661 | { |
c99d2d4a AE |
662 | struct ceph_osd_req_op *op; |
663 | u64 previous; | |
664 | ||
665 | BUG_ON(which >= osd_req->r_num_ops); | |
666 | op = &osd_req->r_ops[which]; | |
667 | previous = op->extent.length; | |
e5975c7c AE |
668 | |
669 | if (length == previous) | |
670 | return; /* Nothing to do */ | |
671 | BUG_ON(length > previous); | |
672 | ||
673 | op->extent.length = length; | |
d641df81 YZ |
674 | if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) |
675 | op->indata_len -= previous - length; | |
e5975c7c AE |
676 | } |
677 | EXPORT_SYMBOL(osd_req_op_extent_update); | |
678 | ||
2c63f49a YZ |
679 | void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req, |
680 | unsigned int which, u64 offset_inc) | |
681 | { | |
682 | struct ceph_osd_req_op *op, *prev_op; | |
683 | ||
684 | BUG_ON(which + 1 >= osd_req->r_num_ops); | |
685 | ||
686 | prev_op = &osd_req->r_ops[which]; | |
687 | op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags); | |
688 | /* dup previous one */ | |
689 | op->indata_len = prev_op->indata_len; | |
690 | op->outdata_len = prev_op->outdata_len; | |
691 | op->extent = prev_op->extent; | |
692 | /* adjust offset */ | |
693 | op->extent.offset += offset_inc; | |
694 | op->extent.length -= offset_inc; | |
695 | ||
696 | if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) | |
697 | op->indata_len -= offset_inc; | |
698 | } | |
699 | EXPORT_SYMBOL(osd_req_op_extent_dup_last); | |
700 | ||
c99d2d4a | 701 | void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which, |
04017e29 | 702 | u16 opcode, const char *class, const char *method) |
33803f33 | 703 | { |
144cba14 YZ |
704 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
705 | opcode, 0); | |
5f562df5 | 706 | struct ceph_pagelist *pagelist; |
33803f33 AE |
707 | size_t payload_len = 0; |
708 | size_t size; | |
709 | ||
710 | BUG_ON(opcode != CEPH_OSD_OP_CALL); | |
711 | ||
5f562df5 AE |
712 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); |
713 | BUG_ON(!pagelist); | |
714 | ceph_pagelist_init(pagelist); | |
715 | ||
33803f33 AE |
716 | op->cls.class_name = class; |
717 | size = strlen(class); | |
718 | BUG_ON(size > (size_t) U8_MAX); | |
719 | op->cls.class_len = size; | |
5f562df5 | 720 | ceph_pagelist_append(pagelist, class, size); |
33803f33 AE |
721 | payload_len += size; |
722 | ||
723 | op->cls.method_name = method; | |
724 | size = strlen(method); | |
725 | BUG_ON(size > (size_t) U8_MAX); | |
726 | op->cls.method_len = size; | |
5f562df5 | 727 | ceph_pagelist_append(pagelist, method, size); |
33803f33 AE |
728 | payload_len += size; |
729 | ||
a4ce40a9 | 730 | osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist); |
5f562df5 | 731 | |
de2aa102 | 732 | op->indata_len = payload_len; |
33803f33 AE |
733 | } |
734 | EXPORT_SYMBOL(osd_req_op_cls_init); | |
8c042b0d | 735 | |
d74b50be YZ |
736 | int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which, |
737 | u16 opcode, const char *name, const void *value, | |
738 | size_t size, u8 cmp_op, u8 cmp_mode) | |
739 | { | |
144cba14 YZ |
740 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
741 | opcode, 0); | |
d74b50be YZ |
742 | struct ceph_pagelist *pagelist; |
743 | size_t payload_len; | |
744 | ||
745 | BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR); | |
746 | ||
747 | pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS); | |
748 | if (!pagelist) | |
749 | return -ENOMEM; | |
750 | ||
751 | ceph_pagelist_init(pagelist); | |
752 | ||
753 | payload_len = strlen(name); | |
754 | op->xattr.name_len = payload_len; | |
755 | ceph_pagelist_append(pagelist, name, payload_len); | |
756 | ||
757 | op->xattr.value_len = size; | |
758 | ceph_pagelist_append(pagelist, value, size); | |
759 | payload_len += size; | |
760 | ||
761 | op->xattr.cmp_op = cmp_op; | |
762 | op->xattr.cmp_mode = cmp_mode; | |
763 | ||
764 | ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist); | |
de2aa102 | 765 | op->indata_len = payload_len; |
d74b50be YZ |
766 | return 0; |
767 | } | |
768 | EXPORT_SYMBOL(osd_req_op_xattr_init); | |
769 | ||
922dab61 ID |
770 | /* |
771 | * @watch_opcode: CEPH_OSD_WATCH_OP_* | |
772 | */ | |
773 | static void osd_req_op_watch_init(struct ceph_osd_request *req, int which, | |
774 | u64 cookie, u8 watch_opcode) | |
33803f33 | 775 | { |
922dab61 | 776 | struct ceph_osd_req_op *op; |
33803f33 | 777 | |
922dab61 | 778 | op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0); |
33803f33 | 779 | op->watch.cookie = cookie; |
922dab61 ID |
780 | op->watch.op = watch_opcode; |
781 | op->watch.gen = 0; | |
33803f33 | 782 | } |
33803f33 | 783 | |
c647b8a8 ID |
784 | void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req, |
785 | unsigned int which, | |
786 | u64 expected_object_size, | |
787 | u64 expected_write_size) | |
788 | { | |
789 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, | |
144cba14 YZ |
790 | CEPH_OSD_OP_SETALLOCHINT, |
791 | 0); | |
c647b8a8 ID |
792 | |
793 | op->alloc_hint.expected_object_size = expected_object_size; | |
794 | op->alloc_hint.expected_write_size = expected_write_size; | |
795 | ||
796 | /* | |
797 | * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed | |
798 | * not worth a feature bit. Set FAILOK per-op flag to make | |
799 | * sure older osds don't trip over an unsupported opcode. | |
800 | */ | |
801 | op->flags |= CEPH_OSD_OP_FLAG_FAILOK; | |
802 | } | |
803 | EXPORT_SYMBOL(osd_req_op_alloc_hint_init); | |
804 | ||
90af3602 | 805 | static void ceph_osdc_msg_data_add(struct ceph_msg *msg, |
ec9123c5 AE |
806 | struct ceph_osd_data *osd_data) |
807 | { | |
808 | u64 length = ceph_osd_data_length(osd_data); | |
809 | ||
810 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { | |
811 | BUG_ON(length > (u64) SIZE_MAX); | |
812 | if (length) | |
90af3602 | 813 | ceph_msg_data_add_pages(msg, osd_data->pages, |
ec9123c5 AE |
814 | length, osd_data->alignment); |
815 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { | |
816 | BUG_ON(!length); | |
90af3602 | 817 | ceph_msg_data_add_pagelist(msg, osd_data->pagelist); |
ec9123c5 AE |
818 | #ifdef CONFIG_BLOCK |
819 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { | |
90af3602 | 820 | ceph_msg_data_add_bio(msg, osd_data->bio, length); |
ec9123c5 AE |
821 | #endif |
822 | } else { | |
823 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); | |
824 | } | |
825 | } | |
826 | ||
bb873b53 ID |
827 | static u32 osd_req_encode_op(struct ceph_osd_op *dst, |
828 | const struct ceph_osd_req_op *src) | |
a8dd0a37 | 829 | { |
a8dd0a37 AE |
830 | if (WARN_ON(!osd_req_opcode_valid(src->op))) { |
831 | pr_err("unrecognized osd opcode %d\n", src->op); | |
832 | ||
833 | return 0; | |
834 | } | |
835 | ||
836 | switch (src->op) { | |
837 | case CEPH_OSD_OP_STAT: | |
838 | break; | |
839 | case CEPH_OSD_OP_READ: | |
840 | case CEPH_OSD_OP_WRITE: | |
e30b7577 | 841 | case CEPH_OSD_OP_WRITEFULL: |
ad7a60de | 842 | case CEPH_OSD_OP_ZERO: |
ad7a60de | 843 | case CEPH_OSD_OP_TRUNCATE: |
a8dd0a37 AE |
844 | dst->extent.offset = cpu_to_le64(src->extent.offset); |
845 | dst->extent.length = cpu_to_le64(src->extent.length); | |
846 | dst->extent.truncate_size = | |
847 | cpu_to_le64(src->extent.truncate_size); | |
848 | dst->extent.truncate_seq = | |
849 | cpu_to_le32(src->extent.truncate_seq); | |
850 | break; | |
851 | case CEPH_OSD_OP_CALL: | |
a8dd0a37 AE |
852 | dst->cls.class_len = src->cls.class_len; |
853 | dst->cls.method_len = src->cls.method_len; | |
bb873b53 | 854 | dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); |
a8dd0a37 AE |
855 | break; |
856 | case CEPH_OSD_OP_STARTSYNC: | |
857 | break; | |
a8dd0a37 AE |
858 | case CEPH_OSD_OP_WATCH: |
859 | dst->watch.cookie = cpu_to_le64(src->watch.cookie); | |
922dab61 ID |
860 | dst->watch.ver = cpu_to_le64(0); |
861 | dst->watch.op = src->watch.op; | |
862 | dst->watch.gen = cpu_to_le32(src->watch.gen); | |
863 | break; | |
864 | case CEPH_OSD_OP_NOTIFY_ACK: | |
a8dd0a37 | 865 | break; |
19079203 ID |
866 | case CEPH_OSD_OP_NOTIFY: |
867 | dst->notify.cookie = cpu_to_le64(src->notify.cookie); | |
868 | break; | |
a4ed38d7 DF |
869 | case CEPH_OSD_OP_LIST_WATCHERS: |
870 | break; | |
c647b8a8 ID |
871 | case CEPH_OSD_OP_SETALLOCHINT: |
872 | dst->alloc_hint.expected_object_size = | |
873 | cpu_to_le64(src->alloc_hint.expected_object_size); | |
874 | dst->alloc_hint.expected_write_size = | |
875 | cpu_to_le64(src->alloc_hint.expected_write_size); | |
876 | break; | |
d74b50be YZ |
877 | case CEPH_OSD_OP_SETXATTR: |
878 | case CEPH_OSD_OP_CMPXATTR: | |
879 | dst->xattr.name_len = cpu_to_le32(src->xattr.name_len); | |
880 | dst->xattr.value_len = cpu_to_le32(src->xattr.value_len); | |
881 | dst->xattr.cmp_op = src->xattr.cmp_op; | |
882 | dst->xattr.cmp_mode = src->xattr.cmp_mode; | |
d74b50be | 883 | break; |
864e9197 YZ |
884 | case CEPH_OSD_OP_CREATE: |
885 | case CEPH_OSD_OP_DELETE: | |
886 | break; | |
a8dd0a37 | 887 | default: |
4c46459c | 888 | pr_err("unsupported osd opcode %s\n", |
8f63ca2d | 889 | ceph_osd_op_name(src->op)); |
4c46459c | 890 | WARN_ON(1); |
a8dd0a37 AE |
891 | |
892 | return 0; | |
68b4476b | 893 | } |
7b25bf5f | 894 | |
a8dd0a37 | 895 | dst->op = cpu_to_le16(src->op); |
7b25bf5f | 896 | dst->flags = cpu_to_le32(src->flags); |
de2aa102 | 897 | dst->payload_len = cpu_to_le32(src->indata_len); |
175face2 | 898 | |
bb873b53 | 899 | return src->indata_len; |
68b4476b YS |
900 | } |
901 | ||
3499e8a5 YS |
902 | /* |
903 | * build new request AND message, calculate layout, and adjust file | |
904 | * extent as needed. | |
905 | * | |
906 | * if the file was recently truncated, we include information about its | |
907 | * old and new size so that the object can be updated appropriately. (we | |
908 | * avoid synchronously deleting truncated objects because it's slow.) | |
909 | * | |
910 | * if @do_sync, include a 'startsync' command so that the osd will flush | |
911 | * data quickly. | |
912 | */ | |
913 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, | |
914 | struct ceph_file_layout *layout, | |
915 | struct ceph_vino vino, | |
715e4cd4 YZ |
916 | u64 off, u64 *plen, |
917 | unsigned int which, int num_ops, | |
3499e8a5 YS |
918 | int opcode, int flags, |
919 | struct ceph_snap_context *snapc, | |
3499e8a5 YS |
920 | u32 truncate_seq, |
921 | u64 truncate_size, | |
153e5167 | 922 | bool use_mempool) |
3499e8a5 | 923 | { |
68b4476b | 924 | struct ceph_osd_request *req; |
75d1c941 AE |
925 | u64 objnum = 0; |
926 | u64 objoff = 0; | |
927 | u64 objlen = 0; | |
6816282d | 928 | int r; |
68b4476b | 929 | |
ad7a60de | 930 | BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && |
864e9197 YZ |
931 | opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE && |
932 | opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE); | |
68b4476b | 933 | |
acead002 | 934 | req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool, |
ae7ca4a3 | 935 | GFP_NOFS); |
13d1ad16 ID |
936 | if (!req) { |
937 | r = -ENOMEM; | |
938 | goto fail; | |
939 | } | |
79528734 | 940 | |
3499e8a5 | 941 | /* calculate max write size */ |
a19dadfb | 942 | r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen); |
13d1ad16 ID |
943 | if (r) |
944 | goto fail; | |
a19dadfb | 945 | |
864e9197 | 946 | if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) { |
144cba14 | 947 | osd_req_op_init(req, which, opcode, 0); |
864e9197 | 948 | } else { |
7627151e | 949 | u32 object_size = layout->object_size; |
864e9197 YZ |
950 | u32 object_base = off - objoff; |
951 | if (!(truncate_seq == 1 && truncate_size == -1ULL)) { | |
952 | if (truncate_size <= object_base) { | |
953 | truncate_size = 0; | |
954 | } else { | |
955 | truncate_size -= object_base; | |
956 | if (truncate_size > object_size) | |
957 | truncate_size = object_size; | |
958 | } | |
ccca4e37 | 959 | } |
715e4cd4 | 960 | osd_req_op_extent_init(req, which, opcode, objoff, objlen, |
864e9197 YZ |
961 | truncate_size, truncate_seq); |
962 | } | |
d18d1e28 | 963 | |
a1f4020a | 964 | req->r_abort_on_full = true; |
bb873b53 | 965 | req->r_flags = flags; |
7627151e | 966 | req->r_base_oloc.pool = layout->pool_id; |
30c156d9 | 967 | req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns); |
d30291b9 | 968 | ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum); |
dbe0fc41 | 969 | |
bb873b53 ID |
970 | req->r_snapid = vino.snap; |
971 | if (flags & CEPH_OSD_FLAG_WRITE) | |
972 | req->r_data_offset = off; | |
973 | ||
13d1ad16 ID |
974 | r = ceph_osdc_alloc_messages(req, GFP_NOFS); |
975 | if (r) | |
976 | goto fail; | |
977 | ||
f24e9980 | 978 | return req; |
13d1ad16 ID |
979 | |
980 | fail: | |
981 | ceph_osdc_put_request(req); | |
982 | return ERR_PTR(r); | |
f24e9980 | 983 | } |
3d14c5d2 | 984 | EXPORT_SYMBOL(ceph_osdc_new_request); |
f24e9980 SW |
985 | |
986 | /* | |
987 | * We keep osd requests in an rbtree, sorted by ->r_tid. | |
988 | */ | |
fcd00b68 | 989 | DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node) |
4609245e | 990 | DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node) |
f24e9980 | 991 | |
0247a0cf ID |
992 | static bool osd_homeless(struct ceph_osd *osd) |
993 | { | |
994 | return osd->o_osd == CEPH_HOMELESS_OSD; | |
995 | } | |
996 | ||
5aea3dcd | 997 | static bool osd_registered(struct ceph_osd *osd) |
f24e9980 | 998 | { |
5aea3dcd | 999 | verify_osdc_locked(osd->o_osdc); |
f24e9980 | 1000 | |
5aea3dcd | 1001 | return !RB_EMPTY_NODE(&osd->o_node); |
f24e9980 SW |
1002 | } |
1003 | ||
0247a0cf ID |
1004 | /* |
1005 | * Assumes @osd is zero-initialized. | |
1006 | */ | |
1007 | static void osd_init(struct ceph_osd *osd) | |
1008 | { | |
02113a0f | 1009 | refcount_set(&osd->o_ref, 1); |
0247a0cf | 1010 | RB_CLEAR_NODE(&osd->o_node); |
5aea3dcd | 1011 | osd->o_requests = RB_ROOT; |
922dab61 | 1012 | osd->o_linger_requests = RB_ROOT; |
0247a0cf ID |
1013 | INIT_LIST_HEAD(&osd->o_osd_lru); |
1014 | INIT_LIST_HEAD(&osd->o_keepalive_item); | |
1015 | osd->o_incarnation = 1; | |
5aea3dcd | 1016 | mutex_init(&osd->lock); |
0247a0cf ID |
1017 | } |
1018 | ||
1019 | static void osd_cleanup(struct ceph_osd *osd) | |
1020 | { | |
1021 | WARN_ON(!RB_EMPTY_NODE(&osd->o_node)); | |
5aea3dcd | 1022 | WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); |
922dab61 | 1023 | WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); |
0247a0cf ID |
1024 | WARN_ON(!list_empty(&osd->o_osd_lru)); |
1025 | WARN_ON(!list_empty(&osd->o_keepalive_item)); | |
1026 | ||
1027 | if (osd->o_auth.authorizer) { | |
1028 | WARN_ON(osd_homeless(osd)); | |
1029 | ceph_auth_destroy_authorizer(osd->o_auth.authorizer); | |
1030 | } | |
1031 | } | |
1032 | ||
f24e9980 SW |
1033 | /* |
1034 | * Track open sessions with osds. | |
1035 | */ | |
e10006f8 | 1036 | static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) |
f24e9980 SW |
1037 | { |
1038 | struct ceph_osd *osd; | |
1039 | ||
0247a0cf ID |
1040 | WARN_ON(onum == CEPH_HOMELESS_OSD); |
1041 | ||
7a28f59b | 1042 | osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL); |
0247a0cf | 1043 | osd_init(osd); |
f24e9980 | 1044 | osd->o_osdc = osdc; |
e10006f8 | 1045 | osd->o_osd = onum; |
f24e9980 | 1046 | |
b7a9e5dd | 1047 | ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); |
4e7a5dcd | 1048 | |
f24e9980 SW |
1049 | return osd; |
1050 | } | |
1051 | ||
1052 | static struct ceph_osd *get_osd(struct ceph_osd *osd) | |
1053 | { | |
02113a0f ER |
1054 | if (refcount_inc_not_zero(&osd->o_ref)) { |
1055 | dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1, | |
1056 | refcount_read(&osd->o_ref)); | |
f24e9980 SW |
1057 | return osd; |
1058 | } else { | |
1059 | dout("get_osd %p FAIL\n", osd); | |
1060 | return NULL; | |
1061 | } | |
1062 | } | |
1063 | ||
1064 | static void put_osd(struct ceph_osd *osd) | |
1065 | { | |
02113a0f ER |
1066 | dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref), |
1067 | refcount_read(&osd->o_ref) - 1); | |
1068 | if (refcount_dec_and_test(&osd->o_ref)) { | |
0247a0cf | 1069 | osd_cleanup(osd); |
f24e9980 | 1070 | kfree(osd); |
79494d1b | 1071 | } |
f24e9980 SW |
1072 | } |
1073 | ||
fcd00b68 ID |
1074 | DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node) |
1075 | ||
9dd2845c | 1076 | static void __move_osd_to_lru(struct ceph_osd *osd) |
f5a2041b | 1077 | { |
9dd2845c ID |
1078 | struct ceph_osd_client *osdc = osd->o_osdc; |
1079 | ||
1080 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
f5a2041b | 1081 | BUG_ON(!list_empty(&osd->o_osd_lru)); |
bbf37ec3 | 1082 | |
9dd2845c | 1083 | spin_lock(&osdc->osd_lru_lock); |
f5a2041b | 1084 | list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); |
9dd2845c ID |
1085 | spin_unlock(&osdc->osd_lru_lock); |
1086 | ||
a319bf56 | 1087 | osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl; |
f5a2041b YS |
1088 | } |
1089 | ||
9dd2845c | 1090 | static void maybe_move_osd_to_lru(struct ceph_osd *osd) |
bbf37ec3 | 1091 | { |
5aea3dcd | 1092 | if (RB_EMPTY_ROOT(&osd->o_requests) && |
922dab61 | 1093 | RB_EMPTY_ROOT(&osd->o_linger_requests)) |
9dd2845c | 1094 | __move_osd_to_lru(osd); |
bbf37ec3 ID |
1095 | } |
1096 | ||
f5a2041b YS |
1097 | static void __remove_osd_from_lru(struct ceph_osd *osd) |
1098 | { | |
9dd2845c ID |
1099 | struct ceph_osd_client *osdc = osd->o_osdc; |
1100 | ||
1101 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
1102 | ||
1103 | spin_lock(&osdc->osd_lru_lock); | |
f5a2041b YS |
1104 | if (!list_empty(&osd->o_osd_lru)) |
1105 | list_del_init(&osd->o_osd_lru); | |
9dd2845c | 1106 | spin_unlock(&osdc->osd_lru_lock); |
f5a2041b YS |
1107 | } |
1108 | ||
5aea3dcd ID |
1109 | /* |
1110 | * Close the connection and assign any leftover requests to the | |
1111 | * homeless session. | |
1112 | */ | |
1113 | static void close_osd(struct ceph_osd *osd) | |
1114 | { | |
1115 | struct ceph_osd_client *osdc = osd->o_osdc; | |
1116 | struct rb_node *n; | |
1117 | ||
1118 | verify_osdc_wrlocked(osdc); | |
1119 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
1120 | ||
1121 | ceph_con_close(&osd->o_con); | |
1122 | ||
1123 | for (n = rb_first(&osd->o_requests); n; ) { | |
1124 | struct ceph_osd_request *req = | |
1125 | rb_entry(n, struct ceph_osd_request, r_node); | |
1126 | ||
1127 | n = rb_next(n); /* unlink_request() */ | |
1128 | ||
1129 | dout(" reassigning req %p tid %llu\n", req, req->r_tid); | |
1130 | unlink_request(osd, req); | |
1131 | link_request(&osdc->homeless_osd, req); | |
1132 | } | |
922dab61 ID |
1133 | for (n = rb_first(&osd->o_linger_requests); n; ) { |
1134 | struct ceph_osd_linger_request *lreq = | |
1135 | rb_entry(n, struct ceph_osd_linger_request, node); | |
1136 | ||
1137 | n = rb_next(n); /* unlink_linger() */ | |
1138 | ||
1139 | dout(" reassigning lreq %p linger_id %llu\n", lreq, | |
1140 | lreq->linger_id); | |
1141 | unlink_linger(osd, lreq); | |
1142 | link_linger(&osdc->homeless_osd, lreq); | |
1143 | } | |
5aea3dcd ID |
1144 | |
1145 | __remove_osd_from_lru(osd); | |
1146 | erase_osd(&osdc->osds, osd); | |
1147 | put_osd(osd); | |
1148 | } | |
1149 | ||
f24e9980 SW |
1150 | /* |
1151 | * reset osd connect | |
1152 | */ | |
5aea3dcd | 1153 | static int reopen_osd(struct ceph_osd *osd) |
f24e9980 | 1154 | { |
c3acb181 | 1155 | struct ceph_entity_addr *peer_addr; |
f24e9980 | 1156 | |
5aea3dcd ID |
1157 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); |
1158 | ||
1159 | if (RB_EMPTY_ROOT(&osd->o_requests) && | |
922dab61 | 1160 | RB_EMPTY_ROOT(&osd->o_linger_requests)) { |
5aea3dcd | 1161 | close_osd(osd); |
c3acb181 AE |
1162 | return -ENODEV; |
1163 | } | |
1164 | ||
5aea3dcd | 1165 | peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd]; |
c3acb181 AE |
1166 | if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && |
1167 | !ceph_con_opened(&osd->o_con)) { | |
5aea3dcd | 1168 | struct rb_node *n; |
c3acb181 | 1169 | |
0b4af2e8 ID |
1170 | dout("osd addr hasn't changed and connection never opened, " |
1171 | "letting msgr retry\n"); | |
87b315a5 | 1172 | /* touch each r_stamp for handle_timeout()'s benfit */ |
5aea3dcd ID |
1173 | for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) { |
1174 | struct ceph_osd_request *req = | |
1175 | rb_entry(n, struct ceph_osd_request, r_node); | |
87b315a5 | 1176 | req->r_stamp = jiffies; |
5aea3dcd | 1177 | } |
c3acb181 AE |
1178 | |
1179 | return -EAGAIN; | |
f24e9980 | 1180 | } |
c3acb181 AE |
1181 | |
1182 | ceph_con_close(&osd->o_con); | |
1183 | ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); | |
1184 | osd->o_incarnation++; | |
1185 | ||
1186 | return 0; | |
f24e9980 SW |
1187 | } |
1188 | ||
5aea3dcd ID |
1189 | static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o, |
1190 | bool wrlocked) | |
f24e9980 | 1191 | { |
5aea3dcd | 1192 | struct ceph_osd *osd; |
35f9f8a0 | 1193 | |
5aea3dcd ID |
1194 | if (wrlocked) |
1195 | verify_osdc_wrlocked(osdc); | |
1196 | else | |
1197 | verify_osdc_locked(osdc); | |
f24e9980 | 1198 | |
5aea3dcd ID |
1199 | if (o != CEPH_HOMELESS_OSD) |
1200 | osd = lookup_osd(&osdc->osds, o); | |
1201 | else | |
1202 | osd = &osdc->homeless_osd; | |
1203 | if (!osd) { | |
1204 | if (!wrlocked) | |
1205 | return ERR_PTR(-EAGAIN); | |
0ba6478d | 1206 | |
5aea3dcd ID |
1207 | osd = create_osd(osdc, o); |
1208 | insert_osd(&osdc->osds, osd); | |
1209 | ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, | |
1210 | &osdc->osdmap->osd_addr[osd->o_osd]); | |
0ba6478d | 1211 | } |
f24e9980 | 1212 | |
5aea3dcd ID |
1213 | dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd); |
1214 | return osd; | |
f24e9980 SW |
1215 | } |
1216 | ||
1217 | /* | |
5aea3dcd ID |
1218 | * Create request <-> OSD session relation. |
1219 | * | |
1220 | * @req has to be assigned a tid, @osd may be homeless. | |
f24e9980 | 1221 | */ |
5aea3dcd | 1222 | static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req) |
f24e9980 | 1223 | { |
5aea3dcd ID |
1224 | verify_osd_locked(osd); |
1225 | WARN_ON(!req->r_tid || req->r_osd); | |
1226 | dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, | |
1227 | req, req->r_tid); | |
1228 | ||
1229 | if (!osd_homeless(osd)) | |
1230 | __remove_osd_from_lru(osd); | |
1231 | else | |
1232 | atomic_inc(&osd->o_osdc->num_homeless); | |
1233 | ||
1234 | get_osd(osd); | |
1235 | insert_request(&osd->o_requests, req); | |
1236 | req->r_osd = osd; | |
f24e9980 SW |
1237 | } |
1238 | ||
5aea3dcd ID |
1239 | static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req) |
1240 | { | |
1241 | verify_osd_locked(osd); | |
1242 | WARN_ON(req->r_osd != osd); | |
1243 | dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, | |
1244 | req, req->r_tid); | |
1245 | ||
1246 | req->r_osd = NULL; | |
1247 | erase_request(&osd->o_requests, req); | |
1248 | put_osd(osd); | |
1249 | ||
1250 | if (!osd_homeless(osd)) | |
1251 | maybe_move_osd_to_lru(osd); | |
1252 | else | |
1253 | atomic_dec(&osd->o_osdc->num_homeless); | |
1254 | } | |
1255 | ||
63244fa1 ID |
1256 | static bool __pool_full(struct ceph_pg_pool_info *pi) |
1257 | { | |
1258 | return pi->flags & CEPH_POOL_FLAG_FULL; | |
1259 | } | |
1260 | ||
42c1b124 ID |
1261 | static bool have_pool_full(struct ceph_osd_client *osdc) |
1262 | { | |
1263 | struct rb_node *n; | |
1264 | ||
1265 | for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { | |
1266 | struct ceph_pg_pool_info *pi = | |
1267 | rb_entry(n, struct ceph_pg_pool_info, node); | |
1268 | ||
1269 | if (__pool_full(pi)) | |
1270 | return true; | |
1271 | } | |
1272 | ||
1273 | return false; | |
1274 | } | |
1275 | ||
5aea3dcd ID |
1276 | static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id) |
1277 | { | |
1278 | struct ceph_pg_pool_info *pi; | |
1279 | ||
1280 | pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); | |
1281 | if (!pi) | |
1282 | return false; | |
1283 | ||
1284 | return __pool_full(pi); | |
1285 | } | |
1286 | ||
d29adb34 JD |
1287 | /* |
1288 | * Returns whether a request should be blocked from being sent | |
1289 | * based on the current osdmap and osd_client settings. | |
d29adb34 | 1290 | */ |
63244fa1 ID |
1291 | static bool target_should_be_paused(struct ceph_osd_client *osdc, |
1292 | const struct ceph_osd_request_target *t, | |
1293 | struct ceph_pg_pool_info *pi) | |
1294 | { | |
b7ec35b3 ID |
1295 | bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); |
1296 | bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || | |
1297 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || | |
63244fa1 ID |
1298 | __pool_full(pi); |
1299 | ||
1300 | WARN_ON(pi->id != t->base_oloc.pool); | |
1301 | return (t->flags & CEPH_OSD_FLAG_READ && pauserd) || | |
1302 | (t->flags & CEPH_OSD_FLAG_WRITE && pausewr); | |
1303 | } | |
1304 | ||
63244fa1 ID |
1305 | enum calc_target_result { |
1306 | CALC_TARGET_NO_ACTION = 0, | |
1307 | CALC_TARGET_NEED_RESEND, | |
1308 | CALC_TARGET_POOL_DNE, | |
1309 | }; | |
1310 | ||
1311 | static enum calc_target_result calc_target(struct ceph_osd_client *osdc, | |
1312 | struct ceph_osd_request_target *t, | |
1313 | u32 *last_force_resend, | |
1314 | bool any_change) | |
1315 | { | |
1316 | struct ceph_pg_pool_info *pi; | |
1317 | struct ceph_pg pgid, last_pgid; | |
1318 | struct ceph_osds up, acting; | |
1319 | bool force_resend = false; | |
1320 | bool need_check_tiering = false; | |
1321 | bool need_resend = false; | |
b7ec35b3 | 1322 | bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE); |
63244fa1 ID |
1323 | enum calc_target_result ct_res; |
1324 | int ret; | |
1325 | ||
1326 | pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool); | |
1327 | if (!pi) { | |
1328 | t->osd = CEPH_HOMELESS_OSD; | |
1329 | ct_res = CALC_TARGET_POOL_DNE; | |
1330 | goto out; | |
1331 | } | |
1332 | ||
1333 | if (osdc->osdmap->epoch == pi->last_force_request_resend) { | |
1334 | if (last_force_resend && | |
1335 | *last_force_resend < pi->last_force_request_resend) { | |
1336 | *last_force_resend = pi->last_force_request_resend; | |
1337 | force_resend = true; | |
1338 | } else if (!last_force_resend) { | |
1339 | force_resend = true; | |
1340 | } | |
1341 | } | |
1342 | if (ceph_oid_empty(&t->target_oid) || force_resend) { | |
1343 | ceph_oid_copy(&t->target_oid, &t->base_oid); | |
1344 | need_check_tiering = true; | |
1345 | } | |
1346 | if (ceph_oloc_empty(&t->target_oloc) || force_resend) { | |
1347 | ceph_oloc_copy(&t->target_oloc, &t->base_oloc); | |
1348 | need_check_tiering = true; | |
1349 | } | |
1350 | ||
1351 | if (need_check_tiering && | |
1352 | (t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) { | |
1353 | if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0) | |
1354 | t->target_oloc.pool = pi->read_tier; | |
1355 | if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0) | |
1356 | t->target_oloc.pool = pi->write_tier; | |
1357 | } | |
1358 | ||
1359 | ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid, | |
1360 | &t->target_oloc, &pgid); | |
1361 | if (ret) { | |
1362 | WARN_ON(ret != -ENOENT); | |
1363 | t->osd = CEPH_HOMELESS_OSD; | |
1364 | ct_res = CALC_TARGET_POOL_DNE; | |
1365 | goto out; | |
1366 | } | |
1367 | last_pgid.pool = pgid.pool; | |
1368 | last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask); | |
1369 | ||
1370 | ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting); | |
1371 | if (any_change && | |
1372 | ceph_is_new_interval(&t->acting, | |
1373 | &acting, | |
1374 | &t->up, | |
1375 | &up, | |
1376 | t->size, | |
1377 | pi->size, | |
1378 | t->min_size, | |
1379 | pi->min_size, | |
1380 | t->pg_num, | |
1381 | pi->pg_num, | |
1382 | t->sort_bitwise, | |
1383 | sort_bitwise, | |
1384 | &last_pgid)) | |
1385 | force_resend = true; | |
1386 | ||
1387 | if (t->paused && !target_should_be_paused(osdc, t, pi)) { | |
1388 | t->paused = false; | |
1389 | need_resend = true; | |
1390 | } | |
1391 | ||
1392 | if (ceph_pg_compare(&t->pgid, &pgid) || | |
1393 | ceph_osds_changed(&t->acting, &acting, any_change) || | |
1394 | force_resend) { | |
1395 | t->pgid = pgid; /* struct */ | |
1396 | ceph_osds_copy(&t->acting, &acting); | |
1397 | ceph_osds_copy(&t->up, &up); | |
1398 | t->size = pi->size; | |
1399 | t->min_size = pi->min_size; | |
1400 | t->pg_num = pi->pg_num; | |
1401 | t->pg_num_mask = pi->pg_num_mask; | |
1402 | t->sort_bitwise = sort_bitwise; | |
1403 | ||
1404 | t->osd = acting.primary; | |
1405 | need_resend = true; | |
1406 | } | |
1407 | ||
1408 | ct_res = need_resend ? CALC_TARGET_NEED_RESEND : CALC_TARGET_NO_ACTION; | |
1409 | out: | |
1410 | dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd); | |
1411 | return ct_res; | |
1412 | } | |
1413 | ||
bb873b53 ID |
1414 | static void setup_request_data(struct ceph_osd_request *req, |
1415 | struct ceph_msg *msg) | |
f24e9980 | 1416 | { |
bb873b53 ID |
1417 | u32 data_len = 0; |
1418 | int i; | |
1419 | ||
1420 | if (!list_empty(&msg->data)) | |
1421 | return; | |
f24e9980 | 1422 | |
bb873b53 ID |
1423 | WARN_ON(msg->data_length); |
1424 | for (i = 0; i < req->r_num_ops; i++) { | |
1425 | struct ceph_osd_req_op *op = &req->r_ops[i]; | |
1426 | ||
1427 | switch (op->op) { | |
1428 | /* request */ | |
1429 | case CEPH_OSD_OP_WRITE: | |
1430 | case CEPH_OSD_OP_WRITEFULL: | |
1431 | WARN_ON(op->indata_len != op->extent.length); | |
1432 | ceph_osdc_msg_data_add(msg, &op->extent.osd_data); | |
1433 | break; | |
1434 | case CEPH_OSD_OP_SETXATTR: | |
1435 | case CEPH_OSD_OP_CMPXATTR: | |
1436 | WARN_ON(op->indata_len != op->xattr.name_len + | |
1437 | op->xattr.value_len); | |
1438 | ceph_osdc_msg_data_add(msg, &op->xattr.osd_data); | |
1439 | break; | |
922dab61 ID |
1440 | case CEPH_OSD_OP_NOTIFY_ACK: |
1441 | ceph_osdc_msg_data_add(msg, | |
1442 | &op->notify_ack.request_data); | |
1443 | break; | |
bb873b53 ID |
1444 | |
1445 | /* reply */ | |
1446 | case CEPH_OSD_OP_STAT: | |
1447 | ceph_osdc_msg_data_add(req->r_reply, | |
1448 | &op->raw_data_in); | |
1449 | break; | |
1450 | case CEPH_OSD_OP_READ: | |
1451 | ceph_osdc_msg_data_add(req->r_reply, | |
1452 | &op->extent.osd_data); | |
1453 | break; | |
a4ed38d7 DF |
1454 | case CEPH_OSD_OP_LIST_WATCHERS: |
1455 | ceph_osdc_msg_data_add(req->r_reply, | |
1456 | &op->list_watchers.response_data); | |
1457 | break; | |
bb873b53 ID |
1458 | |
1459 | /* both */ | |
1460 | case CEPH_OSD_OP_CALL: | |
1461 | WARN_ON(op->indata_len != op->cls.class_len + | |
1462 | op->cls.method_len + | |
1463 | op->cls.indata_len); | |
1464 | ceph_osdc_msg_data_add(msg, &op->cls.request_info); | |
1465 | /* optional, can be NONE */ | |
1466 | ceph_osdc_msg_data_add(msg, &op->cls.request_data); | |
1467 | /* optional, can be NONE */ | |
1468 | ceph_osdc_msg_data_add(req->r_reply, | |
1469 | &op->cls.response_data); | |
1470 | break; | |
19079203 ID |
1471 | case CEPH_OSD_OP_NOTIFY: |
1472 | ceph_osdc_msg_data_add(msg, | |
1473 | &op->notify.request_data); | |
1474 | ceph_osdc_msg_data_add(req->r_reply, | |
1475 | &op->notify.response_data); | |
1476 | break; | |
bb873b53 ID |
1477 | } |
1478 | ||
1479 | data_len += op->indata_len; | |
1480 | } | |
1b83bef2 | 1481 | |
bb873b53 ID |
1482 | WARN_ON(data_len != msg->data_length); |
1483 | } | |
1484 | ||
1485 | static void encode_request(struct ceph_osd_request *req, struct ceph_msg *msg) | |
1486 | { | |
1487 | void *p = msg->front.iov_base; | |
1488 | void *const end = p + msg->front_alloc_len; | |
1489 | u32 data_len = 0; | |
1490 | int i; | |
1491 | ||
1492 | if (req->r_flags & CEPH_OSD_FLAG_WRITE) { | |
1493 | /* snapshots aren't writeable */ | |
1494 | WARN_ON(req->r_snapid != CEPH_NOSNAP); | |
1495 | } else { | |
1496 | WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec || | |
1497 | req->r_data_offset || req->r_snapc); | |
1498 | } | |
1499 | ||
1500 | setup_request_data(req, msg); | |
1501 | ||
1502 | ceph_encode_32(&p, 1); /* client_inc, always 1 */ | |
1503 | ceph_encode_32(&p, req->r_osdc->osdmap->epoch); | |
1504 | ceph_encode_32(&p, req->r_flags); | |
1505 | ceph_encode_timespec(p, &req->r_mtime); | |
1506 | p += sizeof(struct ceph_timespec); | |
aa26d662 JL |
1507 | |
1508 | /* reassert_version */ | |
1509 | memset(p, 0, sizeof(struct ceph_eversion)); | |
1510 | p += sizeof(struct ceph_eversion); | |
bb873b53 ID |
1511 | |
1512 | /* oloc */ | |
30c156d9 YZ |
1513 | ceph_start_encoding(&p, 5, 4, |
1514 | ceph_oloc_encoding_size(&req->r_t.target_oloc)); | |
bb873b53 ID |
1515 | ceph_encode_64(&p, req->r_t.target_oloc.pool); |
1516 | ceph_encode_32(&p, -1); /* preferred */ | |
1517 | ceph_encode_32(&p, 0); /* key len */ | |
30c156d9 YZ |
1518 | if (req->r_t.target_oloc.pool_ns) |
1519 | ceph_encode_string(&p, end, req->r_t.target_oloc.pool_ns->str, | |
1520 | req->r_t.target_oloc.pool_ns->len); | |
1521 | else | |
1522 | ceph_encode_32(&p, 0); | |
bb873b53 ID |
1523 | |
1524 | /* pgid */ | |
1525 | ceph_encode_8(&p, 1); | |
a66dd383 ID |
1526 | ceph_encode_64(&p, req->r_t.pgid.pool); |
1527 | ceph_encode_32(&p, req->r_t.pgid.seed); | |
bb873b53 | 1528 | ceph_encode_32(&p, -1); /* preferred */ |
2169aea6 | 1529 | |
bb873b53 ID |
1530 | /* oid */ |
1531 | ceph_encode_32(&p, req->r_t.target_oid.name_len); | |
1532 | memcpy(p, req->r_t.target_oid.name, req->r_t.target_oid.name_len); | |
1533 | p += req->r_t.target_oid.name_len; | |
f24e9980 | 1534 | |
bb873b53 ID |
1535 | /* ops, can imply data */ |
1536 | ceph_encode_16(&p, req->r_num_ops); | |
1537 | for (i = 0; i < req->r_num_ops; i++) { | |
1538 | data_len += osd_req_encode_op(p, &req->r_ops[i]); | |
1539 | p += sizeof(struct ceph_osd_op); | |
1540 | } | |
26be8808 | 1541 | |
bb873b53 ID |
1542 | ceph_encode_64(&p, req->r_snapid); /* snapid */ |
1543 | if (req->r_snapc) { | |
1544 | ceph_encode_64(&p, req->r_snapc->seq); | |
1545 | ceph_encode_32(&p, req->r_snapc->num_snaps); | |
1546 | for (i = 0; i < req->r_snapc->num_snaps; i++) | |
1547 | ceph_encode_64(&p, req->r_snapc->snaps[i]); | |
1548 | } else { | |
1549 | ceph_encode_64(&p, 0); /* snap_seq */ | |
1550 | ceph_encode_32(&p, 0); /* snaps len */ | |
1551 | } | |
1552 | ||
1553 | ceph_encode_32(&p, req->r_attempts); /* retry_attempt */ | |
1554 | ||
1555 | BUG_ON(p > end); | |
1556 | msg->front.iov_len = p - msg->front.iov_base; | |
1557 | msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */ | |
1558 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
1559 | msg->hdr.data_len = cpu_to_le32(data_len); | |
1560 | /* | |
1561 | * The header "data_off" is a hint to the receiver allowing it | |
1562 | * to align received data into its buffers such that there's no | |
1563 | * need to re-copy it before writing it to disk (direct I/O). | |
1564 | */ | |
1565 | msg->hdr.data_off = cpu_to_le16(req->r_data_offset); | |
26be8808 | 1566 | |
4a3262b1 ID |
1567 | dout("%s req %p oid %s oid_len %d front %zu data %u\n", __func__, |
1568 | req, req->r_t.target_oid.name, req->r_t.target_oid.name_len, | |
1569 | msg->front.iov_len, data_len); | |
bb873b53 ID |
1570 | } |
1571 | ||
1572 | /* | |
1573 | * @req has to be assigned a tid and registered. | |
1574 | */ | |
1575 | static void send_request(struct ceph_osd_request *req) | |
1576 | { | |
1577 | struct ceph_osd *osd = req->r_osd; | |
1578 | ||
5aea3dcd | 1579 | verify_osd_locked(osd); |
bb873b53 ID |
1580 | WARN_ON(osd->o_osd != req->r_t.osd); |
1581 | ||
5aea3dcd ID |
1582 | /* |
1583 | * We may have a previously queued request message hanging | |
1584 | * around. Cancel it to avoid corrupting the msgr. | |
1585 | */ | |
1586 | if (req->r_sent) | |
1587 | ceph_msg_revoke(req->r_request); | |
1588 | ||
bb873b53 ID |
1589 | req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR; |
1590 | if (req->r_attempts) | |
1591 | req->r_flags |= CEPH_OSD_FLAG_RETRY; | |
1592 | else | |
1593 | WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY); | |
1594 | ||
1595 | encode_request(req, req->r_request); | |
1596 | ||
1597 | dout("%s req %p tid %llu to pg %llu.%x osd%d flags 0x%x attempt %d\n", | |
1598 | __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, | |
1599 | req->r_t.osd, req->r_flags, req->r_attempts); | |
1600 | ||
1601 | req->r_t.paused = false; | |
1602 | req->r_stamp = jiffies; | |
1603 | req->r_attempts++; | |
1604 | ||
1605 | req->r_sent = osd->o_incarnation; | |
1606 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); | |
1607 | ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request)); | |
f24e9980 SW |
1608 | } |
1609 | ||
42c1b124 ID |
1610 | static void maybe_request_map(struct ceph_osd_client *osdc) |
1611 | { | |
1612 | bool continuous = false; | |
1613 | ||
5aea3dcd | 1614 | verify_osdc_locked(osdc); |
42c1b124 ID |
1615 | WARN_ON(!osdc->osdmap->epoch); |
1616 | ||
b7ec35b3 ID |
1617 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || |
1618 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) || | |
1619 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { | |
42c1b124 ID |
1620 | dout("%s osdc %p continuous\n", __func__, osdc); |
1621 | continuous = true; | |
1622 | } else { | |
1623 | dout("%s osdc %p onetime\n", __func__, osdc); | |
1624 | } | |
1625 | ||
1626 | if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP, | |
1627 | osdc->osdmap->epoch + 1, continuous)) | |
1628 | ceph_monc_renew_subs(&osdc->client->monc); | |
1629 | } | |
1630 | ||
a1f4020a | 1631 | static void complete_request(struct ceph_osd_request *req, int err); |
4609245e ID |
1632 | static void send_map_check(struct ceph_osd_request *req); |
1633 | ||
5aea3dcd | 1634 | static void __submit_request(struct ceph_osd_request *req, bool wrlocked) |
0bbfdfe8 | 1635 | { |
5aea3dcd ID |
1636 | struct ceph_osd_client *osdc = req->r_osdc; |
1637 | struct ceph_osd *osd; | |
4609245e | 1638 | enum calc_target_result ct_res; |
5aea3dcd ID |
1639 | bool need_send = false; |
1640 | bool promoted = false; | |
a1f4020a | 1641 | bool need_abort = false; |
0bbfdfe8 | 1642 | |
b18b9550 | 1643 | WARN_ON(req->r_tid); |
5aea3dcd ID |
1644 | dout("%s req %p wrlocked %d\n", __func__, req, wrlocked); |
1645 | ||
1646 | again: | |
4609245e ID |
1647 | ct_res = calc_target(osdc, &req->r_t, &req->r_last_force_resend, false); |
1648 | if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked) | |
1649 | goto promote; | |
1650 | ||
5aea3dcd ID |
1651 | osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked); |
1652 | if (IS_ERR(osd)) { | |
1653 | WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked); | |
1654 | goto promote; | |
0bbfdfe8 ID |
1655 | } |
1656 | ||
5aea3dcd | 1657 | if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && |
b7ec35b3 | 1658 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { |
5aea3dcd ID |
1659 | dout("req %p pausewr\n", req); |
1660 | req->r_t.paused = true; | |
1661 | maybe_request_map(osdc); | |
1662 | } else if ((req->r_flags & CEPH_OSD_FLAG_READ) && | |
b7ec35b3 | 1663 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { |
5aea3dcd ID |
1664 | dout("req %p pauserd\n", req); |
1665 | req->r_t.paused = true; | |
1666 | maybe_request_map(osdc); | |
1667 | } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && | |
1668 | !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY | | |
1669 | CEPH_OSD_FLAG_FULL_FORCE)) && | |
b7ec35b3 | 1670 | (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || |
5aea3dcd ID |
1671 | pool_full(osdc, req->r_t.base_oloc.pool))) { |
1672 | dout("req %p full/pool_full\n", req); | |
1673 | pr_warn_ratelimited("FULL or reached pool quota\n"); | |
1674 | req->r_t.paused = true; | |
1675 | maybe_request_map(osdc); | |
a1f4020a JL |
1676 | if (req->r_abort_on_full) |
1677 | need_abort = true; | |
5aea3dcd ID |
1678 | } else if (!osd_homeless(osd)) { |
1679 | need_send = true; | |
0bbfdfe8 | 1680 | } else { |
5aea3dcd | 1681 | maybe_request_map(osdc); |
0bbfdfe8 ID |
1682 | } |
1683 | ||
5aea3dcd ID |
1684 | mutex_lock(&osd->lock); |
1685 | /* | |
1686 | * Assign the tid atomically with send_request() to protect | |
1687 | * multiple writes to the same object from racing with each | |
1688 | * other, resulting in out of order ops on the OSDs. | |
1689 | */ | |
1690 | req->r_tid = atomic64_inc_return(&osdc->last_tid); | |
1691 | link_request(osd, req); | |
1692 | if (need_send) | |
1693 | send_request(req); | |
a1f4020a JL |
1694 | else if (need_abort) |
1695 | complete_request(req, -ENOSPC); | |
5aea3dcd ID |
1696 | mutex_unlock(&osd->lock); |
1697 | ||
4609245e ID |
1698 | if (ct_res == CALC_TARGET_POOL_DNE) |
1699 | send_map_check(req); | |
1700 | ||
5aea3dcd ID |
1701 | if (promoted) |
1702 | downgrade_write(&osdc->lock); | |
1703 | return; | |
1704 | ||
1705 | promote: | |
1706 | up_read(&osdc->lock); | |
1707 | down_write(&osdc->lock); | |
1708 | wrlocked = true; | |
1709 | promoted = true; | |
1710 | goto again; | |
1711 | } | |
1712 | ||
1713 | static void account_request(struct ceph_osd_request *req) | |
1714 | { | |
54ea0046 | 1715 | WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK)); |
b18b9550 | 1716 | WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE))); |
5aea3dcd | 1717 | |
b18b9550 | 1718 | req->r_flags |= CEPH_OSD_FLAG_ONDISK; |
5aea3dcd | 1719 | atomic_inc(&req->r_osdc->num_requests); |
7cc5e38f ID |
1720 | |
1721 | req->r_start_stamp = jiffies; | |
5aea3dcd ID |
1722 | } |
1723 | ||
1724 | static void submit_request(struct ceph_osd_request *req, bool wrlocked) | |
1725 | { | |
1726 | ceph_osdc_get_request(req); | |
1727 | account_request(req); | |
1728 | __submit_request(req, wrlocked); | |
1729 | } | |
1730 | ||
45ee2c1d | 1731 | static void finish_request(struct ceph_osd_request *req) |
5aea3dcd ID |
1732 | { |
1733 | struct ceph_osd_client *osdc = req->r_osdc; | |
1734 | struct ceph_osd *osd = req->r_osd; | |
1735 | ||
1736 | verify_osd_locked(osd); | |
1737 | dout("%s req %p tid %llu\n", __func__, req, req->r_tid); | |
1738 | ||
4609245e | 1739 | WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid)); |
5aea3dcd ID |
1740 | unlink_request(osd, req); |
1741 | atomic_dec(&osdc->num_requests); | |
1742 | ||
1743 | /* | |
1744 | * If an OSD has failed or returned and a request has been sent | |
1745 | * twice, it's possible to get a reply and end up here while the | |
1746 | * request message is queued for delivery. We will ignore the | |
1747 | * reply, so not a big deal, but better to try and catch it. | |
1748 | */ | |
1749 | ceph_msg_revoke(req->r_request); | |
1750 | ceph_msg_revoke_incoming(req->r_reply); | |
1751 | } | |
1752 | ||
fe5da05e ID |
1753 | static void __complete_request(struct ceph_osd_request *req) |
1754 | { | |
b18b9550 ID |
1755 | if (req->r_callback) { |
1756 | dout("%s req %p tid %llu cb %pf result %d\n", __func__, req, | |
1757 | req->r_tid, req->r_callback, req->r_result); | |
fe5da05e | 1758 | req->r_callback(req); |
b18b9550 | 1759 | } |
fe5da05e ID |
1760 | } |
1761 | ||
4609245e | 1762 | /* |
b18b9550 | 1763 | * This is open-coded in handle_reply(). |
4609245e ID |
1764 | */ |
1765 | static void complete_request(struct ceph_osd_request *req, int err) | |
1766 | { | |
1767 | dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); | |
1768 | ||
1769 | req->r_result = err; | |
45ee2c1d | 1770 | finish_request(req); |
4609245e | 1771 | __complete_request(req); |
b18b9550 | 1772 | complete_all(&req->r_completion); |
4609245e ID |
1773 | ceph_osdc_put_request(req); |
1774 | } | |
1775 | ||
1776 | static void cancel_map_check(struct ceph_osd_request *req) | |
1777 | { | |
1778 | struct ceph_osd_client *osdc = req->r_osdc; | |
1779 | struct ceph_osd_request *lookup_req; | |
1780 | ||
1781 | verify_osdc_wrlocked(osdc); | |
1782 | ||
1783 | lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); | |
1784 | if (!lookup_req) | |
1785 | return; | |
1786 | ||
1787 | WARN_ON(lookup_req != req); | |
1788 | erase_request_mc(&osdc->map_checks, req); | |
1789 | ceph_osdc_put_request(req); | |
1790 | } | |
1791 | ||
5aea3dcd ID |
1792 | static void cancel_request(struct ceph_osd_request *req) |
1793 | { | |
1794 | dout("%s req %p tid %llu\n", __func__, req, req->r_tid); | |
1795 | ||
4609245e | 1796 | cancel_map_check(req); |
45ee2c1d | 1797 | finish_request(req); |
b18b9550 | 1798 | complete_all(&req->r_completion); |
c297eb42 | 1799 | ceph_osdc_put_request(req); |
5aea3dcd ID |
1800 | } |
1801 | ||
7cc5e38f ID |
1802 | static void abort_request(struct ceph_osd_request *req, int err) |
1803 | { | |
1804 | dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); | |
1805 | ||
1806 | cancel_map_check(req); | |
1807 | complete_request(req, err); | |
1808 | } | |
1809 | ||
fc36d0a4 JL |
1810 | /* |
1811 | * Drop all pending requests that are stalled waiting on a full condition to | |
1812 | * clear, and complete them with ENOSPC as the return code. | |
1813 | */ | |
1814 | static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc) | |
1815 | { | |
1816 | struct rb_node *n; | |
1817 | ||
1818 | dout("enter abort_on_full\n"); | |
1819 | ||
1820 | if (!ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) && !have_pool_full(osdc)) | |
1821 | goto out; | |
1822 | ||
1823 | for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { | |
1824 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
1825 | struct rb_node *m; | |
1826 | ||
1827 | m = rb_first(&osd->o_requests); | |
1828 | while (m) { | |
1829 | struct ceph_osd_request *req = rb_entry(m, | |
1830 | struct ceph_osd_request, r_node); | |
1831 | m = rb_next(m); | |
1832 | ||
1833 | if (req->r_abort_on_full && | |
1834 | (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || | |
1835 | pool_full(osdc, req->r_t.target_oloc.pool))) | |
1836 | abort_request(req, -ENOSPC); | |
1837 | } | |
1838 | } | |
1839 | out: | |
1840 | dout("return abort_on_full\n"); | |
1841 | } | |
1842 | ||
4609245e ID |
1843 | static void check_pool_dne(struct ceph_osd_request *req) |
1844 | { | |
1845 | struct ceph_osd_client *osdc = req->r_osdc; | |
1846 | struct ceph_osdmap *map = osdc->osdmap; | |
1847 | ||
1848 | verify_osdc_wrlocked(osdc); | |
1849 | WARN_ON(!map->epoch); | |
1850 | ||
1851 | if (req->r_attempts) { | |
1852 | /* | |
1853 | * We sent a request earlier, which means that | |
1854 | * previously the pool existed, and now it does not | |
1855 | * (i.e., it was deleted). | |
1856 | */ | |
1857 | req->r_map_dne_bound = map->epoch; | |
1858 | dout("%s req %p tid %llu pool disappeared\n", __func__, req, | |
1859 | req->r_tid); | |
1860 | } else { | |
1861 | dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__, | |
1862 | req, req->r_tid, req->r_map_dne_bound, map->epoch); | |
1863 | } | |
1864 | ||
1865 | if (req->r_map_dne_bound) { | |
1866 | if (map->epoch >= req->r_map_dne_bound) { | |
1867 | /* we had a new enough map */ | |
1868 | pr_info_ratelimited("tid %llu pool does not exist\n", | |
1869 | req->r_tid); | |
1870 | complete_request(req, -ENOENT); | |
1871 | } | |
1872 | } else { | |
1873 | send_map_check(req); | |
1874 | } | |
1875 | } | |
1876 | ||
1877 | static void map_check_cb(struct ceph_mon_generic_request *greq) | |
1878 | { | |
1879 | struct ceph_osd_client *osdc = &greq->monc->client->osdc; | |
1880 | struct ceph_osd_request *req; | |
1881 | u64 tid = greq->private_data; | |
1882 | ||
1883 | WARN_ON(greq->result || !greq->u.newest); | |
1884 | ||
1885 | down_write(&osdc->lock); | |
1886 | req = lookup_request_mc(&osdc->map_checks, tid); | |
1887 | if (!req) { | |
1888 | dout("%s tid %llu dne\n", __func__, tid); | |
1889 | goto out_unlock; | |
1890 | } | |
1891 | ||
1892 | dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__, | |
1893 | req, req->r_tid, req->r_map_dne_bound, greq->u.newest); | |
1894 | if (!req->r_map_dne_bound) | |
1895 | req->r_map_dne_bound = greq->u.newest; | |
1896 | erase_request_mc(&osdc->map_checks, req); | |
1897 | check_pool_dne(req); | |
1898 | ||
1899 | ceph_osdc_put_request(req); | |
1900 | out_unlock: | |
1901 | up_write(&osdc->lock); | |
1902 | } | |
1903 | ||
1904 | static void send_map_check(struct ceph_osd_request *req) | |
1905 | { | |
1906 | struct ceph_osd_client *osdc = req->r_osdc; | |
1907 | struct ceph_osd_request *lookup_req; | |
1908 | int ret; | |
1909 | ||
1910 | verify_osdc_wrlocked(osdc); | |
1911 | ||
1912 | lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); | |
1913 | if (lookup_req) { | |
1914 | WARN_ON(lookup_req != req); | |
1915 | return; | |
1916 | } | |
1917 | ||
1918 | ceph_osdc_get_request(req); | |
1919 | insert_request_mc(&osdc->map_checks, req); | |
1920 | ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", | |
1921 | map_check_cb, req->r_tid); | |
1922 | WARN_ON(ret); | |
1923 | } | |
1924 | ||
922dab61 ID |
1925 | /* |
1926 | * lingering requests, watch/notify v2 infrastructure | |
1927 | */ | |
1928 | static void linger_release(struct kref *kref) | |
1929 | { | |
1930 | struct ceph_osd_linger_request *lreq = | |
1931 | container_of(kref, struct ceph_osd_linger_request, kref); | |
1932 | ||
1933 | dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq, | |
1934 | lreq->reg_req, lreq->ping_req); | |
1935 | WARN_ON(!RB_EMPTY_NODE(&lreq->node)); | |
1936 | WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node)); | |
4609245e | 1937 | WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node)); |
922dab61 | 1938 | WARN_ON(!list_empty(&lreq->scan_item)); |
b07d3c4b | 1939 | WARN_ON(!list_empty(&lreq->pending_lworks)); |
922dab61 ID |
1940 | WARN_ON(lreq->osd); |
1941 | ||
1942 | if (lreq->reg_req) | |
1943 | ceph_osdc_put_request(lreq->reg_req); | |
1944 | if (lreq->ping_req) | |
1945 | ceph_osdc_put_request(lreq->ping_req); | |
1946 | target_destroy(&lreq->t); | |
1947 | kfree(lreq); | |
1948 | } | |
1949 | ||
1950 | static void linger_put(struct ceph_osd_linger_request *lreq) | |
1951 | { | |
1952 | if (lreq) | |
1953 | kref_put(&lreq->kref, linger_release); | |
1954 | } | |
1955 | ||
1956 | static struct ceph_osd_linger_request * | |
1957 | linger_get(struct ceph_osd_linger_request *lreq) | |
1958 | { | |
1959 | kref_get(&lreq->kref); | |
1960 | return lreq; | |
1961 | } | |
1962 | ||
1963 | static struct ceph_osd_linger_request * | |
1964 | linger_alloc(struct ceph_osd_client *osdc) | |
1965 | { | |
1966 | struct ceph_osd_linger_request *lreq; | |
1967 | ||
1968 | lreq = kzalloc(sizeof(*lreq), GFP_NOIO); | |
1969 | if (!lreq) | |
1970 | return NULL; | |
1971 | ||
1972 | kref_init(&lreq->kref); | |
1973 | mutex_init(&lreq->lock); | |
1974 | RB_CLEAR_NODE(&lreq->node); | |
1975 | RB_CLEAR_NODE(&lreq->osdc_node); | |
4609245e | 1976 | RB_CLEAR_NODE(&lreq->mc_node); |
922dab61 | 1977 | INIT_LIST_HEAD(&lreq->scan_item); |
b07d3c4b | 1978 | INIT_LIST_HEAD(&lreq->pending_lworks); |
922dab61 | 1979 | init_completion(&lreq->reg_commit_wait); |
19079203 | 1980 | init_completion(&lreq->notify_finish_wait); |
922dab61 ID |
1981 | |
1982 | lreq->osdc = osdc; | |
1983 | target_init(&lreq->t); | |
1984 | ||
1985 | dout("%s lreq %p\n", __func__, lreq); | |
1986 | return lreq; | |
1987 | } | |
1988 | ||
1989 | DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node) | |
1990 | DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node) | |
4609245e | 1991 | DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node) |
922dab61 ID |
1992 | |
1993 | /* | |
1994 | * Create linger request <-> OSD session relation. | |
1995 | * | |
1996 | * @lreq has to be registered, @osd may be homeless. | |
1997 | */ | |
1998 | static void link_linger(struct ceph_osd *osd, | |
1999 | struct ceph_osd_linger_request *lreq) | |
2000 | { | |
2001 | verify_osd_locked(osd); | |
2002 | WARN_ON(!lreq->linger_id || lreq->osd); | |
2003 | dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, | |
2004 | osd->o_osd, lreq, lreq->linger_id); | |
2005 | ||
2006 | if (!osd_homeless(osd)) | |
2007 | __remove_osd_from_lru(osd); | |
2008 | else | |
2009 | atomic_inc(&osd->o_osdc->num_homeless); | |
2010 | ||
2011 | get_osd(osd); | |
2012 | insert_linger(&osd->o_linger_requests, lreq); | |
2013 | lreq->osd = osd; | |
2014 | } | |
2015 | ||
2016 | static void unlink_linger(struct ceph_osd *osd, | |
2017 | struct ceph_osd_linger_request *lreq) | |
2018 | { | |
2019 | verify_osd_locked(osd); | |
2020 | WARN_ON(lreq->osd != osd); | |
2021 | dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, | |
2022 | osd->o_osd, lreq, lreq->linger_id); | |
2023 | ||
2024 | lreq->osd = NULL; | |
2025 | erase_linger(&osd->o_linger_requests, lreq); | |
2026 | put_osd(osd); | |
2027 | ||
2028 | if (!osd_homeless(osd)) | |
2029 | maybe_move_osd_to_lru(osd); | |
2030 | else | |
2031 | atomic_dec(&osd->o_osdc->num_homeless); | |
2032 | } | |
2033 | ||
2034 | static bool __linger_registered(struct ceph_osd_linger_request *lreq) | |
2035 | { | |
2036 | verify_osdc_locked(lreq->osdc); | |
2037 | ||
2038 | return !RB_EMPTY_NODE(&lreq->osdc_node); | |
2039 | } | |
2040 | ||
2041 | static bool linger_registered(struct ceph_osd_linger_request *lreq) | |
2042 | { | |
2043 | struct ceph_osd_client *osdc = lreq->osdc; | |
2044 | bool registered; | |
2045 | ||
2046 | down_read(&osdc->lock); | |
2047 | registered = __linger_registered(lreq); | |
2048 | up_read(&osdc->lock); | |
2049 | ||
2050 | return registered; | |
2051 | } | |
2052 | ||
2053 | static void linger_register(struct ceph_osd_linger_request *lreq) | |
2054 | { | |
2055 | struct ceph_osd_client *osdc = lreq->osdc; | |
2056 | ||
2057 | verify_osdc_wrlocked(osdc); | |
2058 | WARN_ON(lreq->linger_id); | |
2059 | ||
2060 | linger_get(lreq); | |
2061 | lreq->linger_id = ++osdc->last_linger_id; | |
2062 | insert_linger_osdc(&osdc->linger_requests, lreq); | |
2063 | } | |
2064 | ||
2065 | static void linger_unregister(struct ceph_osd_linger_request *lreq) | |
2066 | { | |
2067 | struct ceph_osd_client *osdc = lreq->osdc; | |
2068 | ||
2069 | verify_osdc_wrlocked(osdc); | |
2070 | ||
2071 | erase_linger_osdc(&osdc->linger_requests, lreq); | |
2072 | linger_put(lreq); | |
2073 | } | |
2074 | ||
2075 | static void cancel_linger_request(struct ceph_osd_request *req) | |
2076 | { | |
2077 | struct ceph_osd_linger_request *lreq = req->r_priv; | |
2078 | ||
2079 | WARN_ON(!req->r_linger); | |
2080 | cancel_request(req); | |
2081 | linger_put(lreq); | |
2082 | } | |
2083 | ||
2084 | struct linger_work { | |
2085 | struct work_struct work; | |
2086 | struct ceph_osd_linger_request *lreq; | |
b07d3c4b ID |
2087 | struct list_head pending_item; |
2088 | unsigned long queued_stamp; | |
922dab61 ID |
2089 | |
2090 | union { | |
2091 | struct { | |
2092 | u64 notify_id; | |
2093 | u64 notifier_id; | |
2094 | void *payload; /* points into @msg front */ | |
2095 | size_t payload_len; | |
2096 | ||
2097 | struct ceph_msg *msg; /* for ceph_msg_put() */ | |
2098 | } notify; | |
2099 | struct { | |
2100 | int err; | |
2101 | } error; | |
2102 | }; | |
2103 | }; | |
2104 | ||
2105 | static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq, | |
2106 | work_func_t workfn) | |
2107 | { | |
2108 | struct linger_work *lwork; | |
2109 | ||
2110 | lwork = kzalloc(sizeof(*lwork), GFP_NOIO); | |
2111 | if (!lwork) | |
2112 | return NULL; | |
2113 | ||
2114 | INIT_WORK(&lwork->work, workfn); | |
b07d3c4b | 2115 | INIT_LIST_HEAD(&lwork->pending_item); |
922dab61 ID |
2116 | lwork->lreq = linger_get(lreq); |
2117 | ||
2118 | return lwork; | |
2119 | } | |
2120 | ||
2121 | static void lwork_free(struct linger_work *lwork) | |
2122 | { | |
2123 | struct ceph_osd_linger_request *lreq = lwork->lreq; | |
2124 | ||
b07d3c4b ID |
2125 | mutex_lock(&lreq->lock); |
2126 | list_del(&lwork->pending_item); | |
2127 | mutex_unlock(&lreq->lock); | |
2128 | ||
922dab61 ID |
2129 | linger_put(lreq); |
2130 | kfree(lwork); | |
2131 | } | |
2132 | ||
2133 | static void lwork_queue(struct linger_work *lwork) | |
2134 | { | |
2135 | struct ceph_osd_linger_request *lreq = lwork->lreq; | |
2136 | struct ceph_osd_client *osdc = lreq->osdc; | |
2137 | ||
2138 | verify_lreq_locked(lreq); | |
b07d3c4b ID |
2139 | WARN_ON(!list_empty(&lwork->pending_item)); |
2140 | ||
2141 | lwork->queued_stamp = jiffies; | |
2142 | list_add_tail(&lwork->pending_item, &lreq->pending_lworks); | |
922dab61 ID |
2143 | queue_work(osdc->notify_wq, &lwork->work); |
2144 | } | |
2145 | ||
2146 | static void do_watch_notify(struct work_struct *w) | |
2147 | { | |
2148 | struct linger_work *lwork = container_of(w, struct linger_work, work); | |
2149 | struct ceph_osd_linger_request *lreq = lwork->lreq; | |
2150 | ||
2151 | if (!linger_registered(lreq)) { | |
2152 | dout("%s lreq %p not registered\n", __func__, lreq); | |
2153 | goto out; | |
2154 | } | |
2155 | ||
19079203 | 2156 | WARN_ON(!lreq->is_watch); |
922dab61 ID |
2157 | dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n", |
2158 | __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id, | |
2159 | lwork->notify.payload_len); | |
2160 | lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id, | |
2161 | lwork->notify.notifier_id, lwork->notify.payload, | |
2162 | lwork->notify.payload_len); | |
2163 | ||
2164 | out: | |
2165 | ceph_msg_put(lwork->notify.msg); | |
2166 | lwork_free(lwork); | |
2167 | } | |
2168 | ||
2169 | static void do_watch_error(struct work_struct *w) | |
2170 | { | |
2171 | struct linger_work *lwork = container_of(w, struct linger_work, work); | |
2172 | struct ceph_osd_linger_request *lreq = lwork->lreq; | |
2173 | ||
2174 | if (!linger_registered(lreq)) { | |
2175 | dout("%s lreq %p not registered\n", __func__, lreq); | |
2176 | goto out; | |
2177 | } | |
2178 | ||
2179 | dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err); | |
2180 | lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err); | |
2181 | ||
2182 | out: | |
2183 | lwork_free(lwork); | |
2184 | } | |
2185 | ||
2186 | static void queue_watch_error(struct ceph_osd_linger_request *lreq) | |
2187 | { | |
2188 | struct linger_work *lwork; | |
2189 | ||
2190 | lwork = lwork_alloc(lreq, do_watch_error); | |
2191 | if (!lwork) { | |
2192 | pr_err("failed to allocate error-lwork\n"); | |
2193 | return; | |
2194 | } | |
2195 | ||
2196 | lwork->error.err = lreq->last_error; | |
2197 | lwork_queue(lwork); | |
2198 | } | |
2199 | ||
2200 | static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq, | |
2201 | int result) | |
2202 | { | |
2203 | if (!completion_done(&lreq->reg_commit_wait)) { | |
2204 | lreq->reg_commit_error = (result <= 0 ? result : 0); | |
2205 | complete_all(&lreq->reg_commit_wait); | |
2206 | } | |
2207 | } | |
2208 | ||
2209 | static void linger_commit_cb(struct ceph_osd_request *req) | |
2210 | { | |
2211 | struct ceph_osd_linger_request *lreq = req->r_priv; | |
2212 | ||
2213 | mutex_lock(&lreq->lock); | |
2214 | dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq, | |
2215 | lreq->linger_id, req->r_result); | |
922dab61 ID |
2216 | linger_reg_commit_complete(lreq, req->r_result); |
2217 | lreq->committed = true; | |
2218 | ||
19079203 ID |
2219 | if (!lreq->is_watch) { |
2220 | struct ceph_osd_data *osd_data = | |
2221 | osd_req_op_data(req, 0, notify, response_data); | |
2222 | void *p = page_address(osd_data->pages[0]); | |
2223 | ||
2224 | WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY || | |
2225 | osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); | |
2226 | ||
2227 | /* make note of the notify_id */ | |
2228 | if (req->r_ops[0].outdata_len >= sizeof(u64)) { | |
2229 | lreq->notify_id = ceph_decode_64(&p); | |
2230 | dout("lreq %p notify_id %llu\n", lreq, | |
2231 | lreq->notify_id); | |
2232 | } else { | |
2233 | dout("lreq %p no notify_id\n", lreq); | |
2234 | } | |
2235 | } | |
2236 | ||
922dab61 ID |
2237 | mutex_unlock(&lreq->lock); |
2238 | linger_put(lreq); | |
2239 | } | |
2240 | ||
2241 | static int normalize_watch_error(int err) | |
2242 | { | |
2243 | /* | |
2244 | * Translate ENOENT -> ENOTCONN so that a delete->disconnection | |
2245 | * notification and a failure to reconnect because we raced with | |
2246 | * the delete appear the same to the user. | |
2247 | */ | |
2248 | if (err == -ENOENT) | |
2249 | err = -ENOTCONN; | |
2250 | ||
2251 | return err; | |
2252 | } | |
2253 | ||
2254 | static void linger_reconnect_cb(struct ceph_osd_request *req) | |
2255 | { | |
2256 | struct ceph_osd_linger_request *lreq = req->r_priv; | |
2257 | ||
2258 | mutex_lock(&lreq->lock); | |
2259 | dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__, | |
2260 | lreq, lreq->linger_id, req->r_result, lreq->last_error); | |
2261 | if (req->r_result < 0) { | |
2262 | if (!lreq->last_error) { | |
2263 | lreq->last_error = normalize_watch_error(req->r_result); | |
2264 | queue_watch_error(lreq); | |
2265 | } | |
2266 | } | |
2267 | ||
2268 | mutex_unlock(&lreq->lock); | |
2269 | linger_put(lreq); | |
2270 | } | |
2271 | ||
2272 | static void send_linger(struct ceph_osd_linger_request *lreq) | |
2273 | { | |
2274 | struct ceph_osd_request *req = lreq->reg_req; | |
2275 | struct ceph_osd_req_op *op = &req->r_ops[0]; | |
2276 | ||
2277 | verify_osdc_wrlocked(req->r_osdc); | |
2278 | dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); | |
2279 | ||
2280 | if (req->r_osd) | |
2281 | cancel_linger_request(req); | |
2282 | ||
2283 | request_reinit(req); | |
2284 | ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); | |
2285 | ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); | |
2286 | req->r_flags = lreq->t.flags; | |
2287 | req->r_mtime = lreq->mtime; | |
2288 | ||
2289 | mutex_lock(&lreq->lock); | |
19079203 | 2290 | if (lreq->is_watch && lreq->committed) { |
922dab61 ID |
2291 | WARN_ON(op->op != CEPH_OSD_OP_WATCH || |
2292 | op->watch.cookie != lreq->linger_id); | |
2293 | op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT; | |
2294 | op->watch.gen = ++lreq->register_gen; | |
2295 | dout("lreq %p reconnect register_gen %u\n", lreq, | |
2296 | op->watch.gen); | |
2297 | req->r_callback = linger_reconnect_cb; | |
2298 | } else { | |
19079203 ID |
2299 | if (!lreq->is_watch) |
2300 | lreq->notify_id = 0; | |
2301 | else | |
2302 | WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH); | |
922dab61 ID |
2303 | dout("lreq %p register\n", lreq); |
2304 | req->r_callback = linger_commit_cb; | |
2305 | } | |
2306 | mutex_unlock(&lreq->lock); | |
2307 | ||
2308 | req->r_priv = linger_get(lreq); | |
2309 | req->r_linger = true; | |
2310 | ||
2311 | submit_request(req, true); | |
2312 | } | |
2313 | ||
2314 | static void linger_ping_cb(struct ceph_osd_request *req) | |
2315 | { | |
2316 | struct ceph_osd_linger_request *lreq = req->r_priv; | |
2317 | ||
2318 | mutex_lock(&lreq->lock); | |
2319 | dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n", | |
2320 | __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent, | |
2321 | lreq->last_error); | |
2322 | if (lreq->register_gen == req->r_ops[0].watch.gen) { | |
b07d3c4b ID |
2323 | if (!req->r_result) { |
2324 | lreq->watch_valid_thru = lreq->ping_sent; | |
2325 | } else if (!lreq->last_error) { | |
922dab61 ID |
2326 | lreq->last_error = normalize_watch_error(req->r_result); |
2327 | queue_watch_error(lreq); | |
2328 | } | |
2329 | } else { | |
2330 | dout("lreq %p register_gen %u ignoring old pong %u\n", lreq, | |
2331 | lreq->register_gen, req->r_ops[0].watch.gen); | |
2332 | } | |
2333 | ||
2334 | mutex_unlock(&lreq->lock); | |
2335 | linger_put(lreq); | |
2336 | } | |
2337 | ||
2338 | static void send_linger_ping(struct ceph_osd_linger_request *lreq) | |
2339 | { | |
2340 | struct ceph_osd_client *osdc = lreq->osdc; | |
2341 | struct ceph_osd_request *req = lreq->ping_req; | |
2342 | struct ceph_osd_req_op *op = &req->r_ops[0]; | |
2343 | ||
b7ec35b3 | 2344 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { |
922dab61 ID |
2345 | dout("%s PAUSERD\n", __func__); |
2346 | return; | |
2347 | } | |
2348 | ||
2349 | lreq->ping_sent = jiffies; | |
2350 | dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n", | |
2351 | __func__, lreq, lreq->linger_id, lreq->ping_sent, | |
2352 | lreq->register_gen); | |
2353 | ||
2354 | if (req->r_osd) | |
2355 | cancel_linger_request(req); | |
2356 | ||
2357 | request_reinit(req); | |
2358 | target_copy(&req->r_t, &lreq->t); | |
2359 | ||
2360 | WARN_ON(op->op != CEPH_OSD_OP_WATCH || | |
2361 | op->watch.cookie != lreq->linger_id || | |
2362 | op->watch.op != CEPH_OSD_WATCH_OP_PING); | |
2363 | op->watch.gen = lreq->register_gen; | |
2364 | req->r_callback = linger_ping_cb; | |
2365 | req->r_priv = linger_get(lreq); | |
2366 | req->r_linger = true; | |
2367 | ||
2368 | ceph_osdc_get_request(req); | |
2369 | account_request(req); | |
2370 | req->r_tid = atomic64_inc_return(&osdc->last_tid); | |
2371 | link_request(lreq->osd, req); | |
2372 | send_request(req); | |
2373 | } | |
2374 | ||
2375 | static void linger_submit(struct ceph_osd_linger_request *lreq) | |
2376 | { | |
2377 | struct ceph_osd_client *osdc = lreq->osdc; | |
2378 | struct ceph_osd *osd; | |
2379 | ||
2380 | calc_target(osdc, &lreq->t, &lreq->last_force_resend, false); | |
2381 | osd = lookup_create_osd(osdc, lreq->t.osd, true); | |
2382 | link_linger(osd, lreq); | |
2383 | ||
2384 | send_linger(lreq); | |
2385 | } | |
2386 | ||
4609245e ID |
2387 | static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq) |
2388 | { | |
2389 | struct ceph_osd_client *osdc = lreq->osdc; | |
2390 | struct ceph_osd_linger_request *lookup_lreq; | |
2391 | ||
2392 | verify_osdc_wrlocked(osdc); | |
2393 | ||
2394 | lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, | |
2395 | lreq->linger_id); | |
2396 | if (!lookup_lreq) | |
2397 | return; | |
2398 | ||
2399 | WARN_ON(lookup_lreq != lreq); | |
2400 | erase_linger_mc(&osdc->linger_map_checks, lreq); | |
2401 | linger_put(lreq); | |
2402 | } | |
2403 | ||
922dab61 ID |
2404 | /* |
2405 | * @lreq has to be both registered and linked. | |
2406 | */ | |
2407 | static void __linger_cancel(struct ceph_osd_linger_request *lreq) | |
2408 | { | |
19079203 | 2409 | if (lreq->is_watch && lreq->ping_req->r_osd) |
922dab61 ID |
2410 | cancel_linger_request(lreq->ping_req); |
2411 | if (lreq->reg_req->r_osd) | |
2412 | cancel_linger_request(lreq->reg_req); | |
4609245e | 2413 | cancel_linger_map_check(lreq); |
922dab61 ID |
2414 | unlink_linger(lreq->osd, lreq); |
2415 | linger_unregister(lreq); | |
2416 | } | |
2417 | ||
2418 | static void linger_cancel(struct ceph_osd_linger_request *lreq) | |
2419 | { | |
2420 | struct ceph_osd_client *osdc = lreq->osdc; | |
2421 | ||
2422 | down_write(&osdc->lock); | |
2423 | if (__linger_registered(lreq)) | |
2424 | __linger_cancel(lreq); | |
2425 | up_write(&osdc->lock); | |
2426 | } | |
2427 | ||
4609245e ID |
2428 | static void send_linger_map_check(struct ceph_osd_linger_request *lreq); |
2429 | ||
2430 | static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq) | |
2431 | { | |
2432 | struct ceph_osd_client *osdc = lreq->osdc; | |
2433 | struct ceph_osdmap *map = osdc->osdmap; | |
2434 | ||
2435 | verify_osdc_wrlocked(osdc); | |
2436 | WARN_ON(!map->epoch); | |
2437 | ||
2438 | if (lreq->register_gen) { | |
2439 | lreq->map_dne_bound = map->epoch; | |
2440 | dout("%s lreq %p linger_id %llu pool disappeared\n", __func__, | |
2441 | lreq, lreq->linger_id); | |
2442 | } else { | |
2443 | dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n", | |
2444 | __func__, lreq, lreq->linger_id, lreq->map_dne_bound, | |
2445 | map->epoch); | |
2446 | } | |
2447 | ||
2448 | if (lreq->map_dne_bound) { | |
2449 | if (map->epoch >= lreq->map_dne_bound) { | |
2450 | /* we had a new enough map */ | |
2451 | pr_info("linger_id %llu pool does not exist\n", | |
2452 | lreq->linger_id); | |
2453 | linger_reg_commit_complete(lreq, -ENOENT); | |
2454 | __linger_cancel(lreq); | |
2455 | } | |
2456 | } else { | |
2457 | send_linger_map_check(lreq); | |
2458 | } | |
2459 | } | |
2460 | ||
2461 | static void linger_map_check_cb(struct ceph_mon_generic_request *greq) | |
2462 | { | |
2463 | struct ceph_osd_client *osdc = &greq->monc->client->osdc; | |
2464 | struct ceph_osd_linger_request *lreq; | |
2465 | u64 linger_id = greq->private_data; | |
2466 | ||
2467 | WARN_ON(greq->result || !greq->u.newest); | |
2468 | ||
2469 | down_write(&osdc->lock); | |
2470 | lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id); | |
2471 | if (!lreq) { | |
2472 | dout("%s linger_id %llu dne\n", __func__, linger_id); | |
2473 | goto out_unlock; | |
2474 | } | |
2475 | ||
2476 | dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n", | |
2477 | __func__, lreq, lreq->linger_id, lreq->map_dne_bound, | |
2478 | greq->u.newest); | |
2479 | if (!lreq->map_dne_bound) | |
2480 | lreq->map_dne_bound = greq->u.newest; | |
2481 | erase_linger_mc(&osdc->linger_map_checks, lreq); | |
2482 | check_linger_pool_dne(lreq); | |
2483 | ||
2484 | linger_put(lreq); | |
2485 | out_unlock: | |
2486 | up_write(&osdc->lock); | |
2487 | } | |
2488 | ||
2489 | static void send_linger_map_check(struct ceph_osd_linger_request *lreq) | |
2490 | { | |
2491 | struct ceph_osd_client *osdc = lreq->osdc; | |
2492 | struct ceph_osd_linger_request *lookup_lreq; | |
2493 | int ret; | |
2494 | ||
2495 | verify_osdc_wrlocked(osdc); | |
2496 | ||
2497 | lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, | |
2498 | lreq->linger_id); | |
2499 | if (lookup_lreq) { | |
2500 | WARN_ON(lookup_lreq != lreq); | |
2501 | return; | |
2502 | } | |
2503 | ||
2504 | linger_get(lreq); | |
2505 | insert_linger_mc(&osdc->linger_map_checks, lreq); | |
2506 | ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", | |
2507 | linger_map_check_cb, lreq->linger_id); | |
2508 | WARN_ON(ret); | |
2509 | } | |
2510 | ||
922dab61 ID |
2511 | static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq) |
2512 | { | |
2513 | int ret; | |
2514 | ||
2515 | dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); | |
2516 | ret = wait_for_completion_interruptible(&lreq->reg_commit_wait); | |
2517 | return ret ?: lreq->reg_commit_error; | |
2518 | } | |
2519 | ||
19079203 ID |
2520 | static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq) |
2521 | { | |
2522 | int ret; | |
2523 | ||
2524 | dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); | |
2525 | ret = wait_for_completion_interruptible(&lreq->notify_finish_wait); | |
2526 | return ret ?: lreq->notify_finish_error; | |
2527 | } | |
2528 | ||
f24e9980 | 2529 | /* |
fbca9635 ID |
2530 | * Timeout callback, called every N seconds. When 1 or more OSD |
2531 | * requests has been active for more than N seconds, we send a keepalive | |
2532 | * (tag + timestamp) to its OSD to ensure any communications channel | |
2533 | * reset is detected. | |
f24e9980 SW |
2534 | */ |
2535 | static void handle_timeout(struct work_struct *work) | |
2536 | { | |
2537 | struct ceph_osd_client *osdc = | |
2538 | container_of(work, struct ceph_osd_client, timeout_work.work); | |
a319bf56 | 2539 | struct ceph_options *opts = osdc->client->options; |
5aea3dcd | 2540 | unsigned long cutoff = jiffies - opts->osd_keepalive_timeout; |
7cc5e38f | 2541 | unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout; |
5aea3dcd ID |
2542 | LIST_HEAD(slow_osds); |
2543 | struct rb_node *n, *p; | |
f24e9980 | 2544 | |
5aea3dcd ID |
2545 | dout("%s osdc %p\n", __func__, osdc); |
2546 | down_write(&osdc->lock); | |
f24e9980 | 2547 | |
422d2cb8 YS |
2548 | /* |
2549 | * ping osds that are a bit slow. this ensures that if there | |
2550 | * is a break in the TCP connection we will notice, and reopen | |
2551 | * a connection with that osd (from the fault callback). | |
2552 | */ | |
5aea3dcd ID |
2553 | for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { |
2554 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
2555 | bool found = false; | |
2556 | ||
7cc5e38f | 2557 | for (p = rb_first(&osd->o_requests); p; ) { |
5aea3dcd ID |
2558 | struct ceph_osd_request *req = |
2559 | rb_entry(p, struct ceph_osd_request, r_node); | |
2560 | ||
7cc5e38f ID |
2561 | p = rb_next(p); /* abort_request() */ |
2562 | ||
5aea3dcd ID |
2563 | if (time_before(req->r_stamp, cutoff)) { |
2564 | dout(" req %p tid %llu on osd%d is laggy\n", | |
2565 | req, req->r_tid, osd->o_osd); | |
2566 | found = true; | |
2567 | } | |
7cc5e38f ID |
2568 | if (opts->osd_request_timeout && |
2569 | time_before(req->r_start_stamp, expiry_cutoff)) { | |
2570 | pr_err_ratelimited("tid %llu on osd%d timeout\n", | |
2571 | req->r_tid, osd->o_osd); | |
2572 | abort_request(req, -ETIMEDOUT); | |
2573 | } | |
5aea3dcd | 2574 | } |
922dab61 ID |
2575 | for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) { |
2576 | struct ceph_osd_linger_request *lreq = | |
2577 | rb_entry(p, struct ceph_osd_linger_request, node); | |
2578 | ||
2579 | dout(" lreq %p linger_id %llu is served by osd%d\n", | |
2580 | lreq, lreq->linger_id, osd->o_osd); | |
2581 | found = true; | |
2582 | ||
2583 | mutex_lock(&lreq->lock); | |
19079203 | 2584 | if (lreq->is_watch && lreq->committed && !lreq->last_error) |
922dab61 ID |
2585 | send_linger_ping(lreq); |
2586 | mutex_unlock(&lreq->lock); | |
2587 | } | |
422d2cb8 | 2588 | |
5aea3dcd ID |
2589 | if (found) |
2590 | list_move_tail(&osd->o_keepalive_item, &slow_osds); | |
422d2cb8 | 2591 | } |
5aea3dcd | 2592 | |
7cc5e38f ID |
2593 | if (opts->osd_request_timeout) { |
2594 | for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) { | |
2595 | struct ceph_osd_request *req = | |
2596 | rb_entry(p, struct ceph_osd_request, r_node); | |
2597 | ||
2598 | p = rb_next(p); /* abort_request() */ | |
2599 | ||
2600 | if (time_before(req->r_start_stamp, expiry_cutoff)) { | |
2601 | pr_err_ratelimited("tid %llu on osd%d timeout\n", | |
2602 | req->r_tid, osdc->homeless_osd.o_osd); | |
2603 | abort_request(req, -ETIMEDOUT); | |
2604 | } | |
2605 | } | |
2606 | } | |
2607 | ||
5aea3dcd ID |
2608 | if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds)) |
2609 | maybe_request_map(osdc); | |
2610 | ||
422d2cb8 | 2611 | while (!list_empty(&slow_osds)) { |
5aea3dcd ID |
2612 | struct ceph_osd *osd = list_first_entry(&slow_osds, |
2613 | struct ceph_osd, | |
2614 | o_keepalive_item); | |
422d2cb8 | 2615 | list_del_init(&osd->o_keepalive_item); |
f24e9980 SW |
2616 | ceph_con_keepalive(&osd->o_con); |
2617 | } | |
2618 | ||
5aea3dcd | 2619 | up_write(&osdc->lock); |
fbca9635 ID |
2620 | schedule_delayed_work(&osdc->timeout_work, |
2621 | osdc->client->options->osd_keepalive_timeout); | |
f24e9980 SW |
2622 | } |
2623 | ||
f5a2041b YS |
2624 | static void handle_osds_timeout(struct work_struct *work) |
2625 | { | |
2626 | struct ceph_osd_client *osdc = | |
2627 | container_of(work, struct ceph_osd_client, | |
2628 | osds_timeout_work.work); | |
a319bf56 | 2629 | unsigned long delay = osdc->client->options->osd_idle_ttl / 4; |
42a2c09f | 2630 | struct ceph_osd *osd, *nosd; |
f5a2041b | 2631 | |
42a2c09f | 2632 | dout("%s osdc %p\n", __func__, osdc); |
5aea3dcd | 2633 | down_write(&osdc->lock); |
42a2c09f ID |
2634 | list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { |
2635 | if (time_before(jiffies, osd->lru_ttl)) | |
2636 | break; | |
2637 | ||
5aea3dcd | 2638 | WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); |
922dab61 | 2639 | WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); |
5aea3dcd | 2640 | close_osd(osd); |
42a2c09f | 2641 | } |
f5a2041b | 2642 | |
5aea3dcd | 2643 | up_write(&osdc->lock); |
f5a2041b YS |
2644 | schedule_delayed_work(&osdc->osds_timeout_work, |
2645 | round_jiffies_relative(delay)); | |
2646 | } | |
2647 | ||
205ee118 ID |
2648 | static int ceph_oloc_decode(void **p, void *end, |
2649 | struct ceph_object_locator *oloc) | |
2650 | { | |
2651 | u8 struct_v, struct_cv; | |
2652 | u32 len; | |
2653 | void *struct_end; | |
2654 | int ret = 0; | |
2655 | ||
2656 | ceph_decode_need(p, end, 1 + 1 + 4, e_inval); | |
2657 | struct_v = ceph_decode_8(p); | |
2658 | struct_cv = ceph_decode_8(p); | |
2659 | if (struct_v < 3) { | |
2660 | pr_warn("got v %d < 3 cv %d of ceph_object_locator\n", | |
2661 | struct_v, struct_cv); | |
2662 | goto e_inval; | |
2663 | } | |
2664 | if (struct_cv > 6) { | |
2665 | pr_warn("got v %d cv %d > 6 of ceph_object_locator\n", | |
2666 | struct_v, struct_cv); | |
2667 | goto e_inval; | |
2668 | } | |
2669 | len = ceph_decode_32(p); | |
2670 | ceph_decode_need(p, end, len, e_inval); | |
2671 | struct_end = *p + len; | |
2672 | ||
2673 | oloc->pool = ceph_decode_64(p); | |
2674 | *p += 4; /* skip preferred */ | |
2675 | ||
2676 | len = ceph_decode_32(p); | |
2677 | if (len > 0) { | |
2678 | pr_warn("ceph_object_locator::key is set\n"); | |
2679 | goto e_inval; | |
2680 | } | |
2681 | ||
2682 | if (struct_v >= 5) { | |
cd08e0a2 YZ |
2683 | bool changed = false; |
2684 | ||
205ee118 ID |
2685 | len = ceph_decode_32(p); |
2686 | if (len > 0) { | |
30c156d9 | 2687 | ceph_decode_need(p, end, len, e_inval); |
cd08e0a2 YZ |
2688 | if (!oloc->pool_ns || |
2689 | ceph_compare_string(oloc->pool_ns, *p, len)) | |
2690 | changed = true; | |
30c156d9 | 2691 | *p += len; |
cd08e0a2 YZ |
2692 | } else { |
2693 | if (oloc->pool_ns) | |
2694 | changed = true; | |
2695 | } | |
2696 | if (changed) { | |
2697 | /* redirect changes namespace */ | |
2698 | pr_warn("ceph_object_locator::nspace is changed\n"); | |
2699 | goto e_inval; | |
205ee118 ID |
2700 | } |
2701 | } | |
2702 | ||
2703 | if (struct_v >= 6) { | |
2704 | s64 hash = ceph_decode_64(p); | |
2705 | if (hash != -1) { | |
2706 | pr_warn("ceph_object_locator::hash is set\n"); | |
2707 | goto e_inval; | |
2708 | } | |
2709 | } | |
2710 | ||
2711 | /* skip the rest */ | |
2712 | *p = struct_end; | |
2713 | out: | |
2714 | return ret; | |
2715 | ||
2716 | e_inval: | |
2717 | ret = -EINVAL; | |
2718 | goto out; | |
2719 | } | |
2720 | ||
2721 | static int ceph_redirect_decode(void **p, void *end, | |
2722 | struct ceph_request_redirect *redir) | |
2723 | { | |
2724 | u8 struct_v, struct_cv; | |
2725 | u32 len; | |
2726 | void *struct_end; | |
2727 | int ret; | |
2728 | ||
2729 | ceph_decode_need(p, end, 1 + 1 + 4, e_inval); | |
2730 | struct_v = ceph_decode_8(p); | |
2731 | struct_cv = ceph_decode_8(p); | |
2732 | if (struct_cv > 1) { | |
2733 | pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n", | |
2734 | struct_v, struct_cv); | |
2735 | goto e_inval; | |
2736 | } | |
2737 | len = ceph_decode_32(p); | |
2738 | ceph_decode_need(p, end, len, e_inval); | |
2739 | struct_end = *p + len; | |
2740 | ||
2741 | ret = ceph_oloc_decode(p, end, &redir->oloc); | |
2742 | if (ret) | |
2743 | goto out; | |
2744 | ||
2745 | len = ceph_decode_32(p); | |
2746 | if (len > 0) { | |
2747 | pr_warn("ceph_request_redirect::object_name is set\n"); | |
2748 | goto e_inval; | |
2749 | } | |
2750 | ||
2751 | len = ceph_decode_32(p); | |
2752 | *p += len; /* skip osd_instructions */ | |
2753 | ||
2754 | /* skip the rest */ | |
2755 | *p = struct_end; | |
2756 | out: | |
2757 | return ret; | |
2758 | ||
2759 | e_inval: | |
2760 | ret = -EINVAL; | |
2761 | goto out; | |
2762 | } | |
2763 | ||
fe5da05e ID |
2764 | struct MOSDOpReply { |
2765 | struct ceph_pg pgid; | |
2766 | u64 flags; | |
2767 | int result; | |
2768 | u32 epoch; | |
2769 | int num_ops; | |
2770 | u32 outdata_len[CEPH_OSD_MAX_OPS]; | |
2771 | s32 rval[CEPH_OSD_MAX_OPS]; | |
2772 | int retry_attempt; | |
2773 | struct ceph_eversion replay_version; | |
2774 | u64 user_version; | |
2775 | struct ceph_request_redirect redirect; | |
2776 | }; | |
25845472 | 2777 | |
fe5da05e | 2778 | static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m) |
f24e9980 | 2779 | { |
fe5da05e ID |
2780 | void *p = msg->front.iov_base; |
2781 | void *const end = p + msg->front.iov_len; | |
2782 | u16 version = le16_to_cpu(msg->hdr.version); | |
2783 | struct ceph_eversion bad_replay_version; | |
b0b31a8f | 2784 | u8 decode_redir; |
fe5da05e ID |
2785 | u32 len; |
2786 | int ret; | |
2787 | int i; | |
1b83bef2 | 2788 | |
fe5da05e ID |
2789 | ceph_decode_32_safe(&p, end, len, e_inval); |
2790 | ceph_decode_need(&p, end, len, e_inval); | |
2791 | p += len; /* skip oid */ | |
1b83bef2 | 2792 | |
fe5da05e ID |
2793 | ret = ceph_decode_pgid(&p, end, &m->pgid); |
2794 | if (ret) | |
2795 | return ret; | |
1b83bef2 | 2796 | |
fe5da05e ID |
2797 | ceph_decode_64_safe(&p, end, m->flags, e_inval); |
2798 | ceph_decode_32_safe(&p, end, m->result, e_inval); | |
2799 | ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval); | |
2800 | memcpy(&bad_replay_version, p, sizeof(bad_replay_version)); | |
2801 | p += sizeof(bad_replay_version); | |
2802 | ceph_decode_32_safe(&p, end, m->epoch, e_inval); | |
1b83bef2 | 2803 | |
fe5da05e ID |
2804 | ceph_decode_32_safe(&p, end, m->num_ops, e_inval); |
2805 | if (m->num_ops > ARRAY_SIZE(m->outdata_len)) | |
2806 | goto e_inval; | |
1b83bef2 | 2807 | |
fe5da05e ID |
2808 | ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op), |
2809 | e_inval); | |
2810 | for (i = 0; i < m->num_ops; i++) { | |
1b83bef2 | 2811 | struct ceph_osd_op *op = p; |
1b83bef2 | 2812 | |
fe5da05e | 2813 | m->outdata_len[i] = le32_to_cpu(op->payload_len); |
1b83bef2 SW |
2814 | p += sizeof(*op); |
2815 | } | |
1b83bef2 | 2816 | |
fe5da05e ID |
2817 | ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval); |
2818 | for (i = 0; i < m->num_ops; i++) | |
2819 | ceph_decode_32_safe(&p, end, m->rval[i], e_inval); | |
f24e9980 | 2820 | |
fe5da05e ID |
2821 | if (version >= 5) { |
2822 | ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval); | |
2823 | memcpy(&m->replay_version, p, sizeof(m->replay_version)); | |
2824 | p += sizeof(m->replay_version); | |
2825 | ceph_decode_64_safe(&p, end, m->user_version, e_inval); | |
2826 | } else { | |
2827 | m->replay_version = bad_replay_version; /* struct */ | |
2828 | m->user_version = le64_to_cpu(m->replay_version.version); | |
2829 | } | |
eb845ff1 | 2830 | |
fe5da05e ID |
2831 | if (version >= 6) { |
2832 | if (version >= 7) | |
2833 | ceph_decode_8_safe(&p, end, decode_redir, e_inval); | |
b0b31a8f ID |
2834 | else |
2835 | decode_redir = 1; | |
2836 | } else { | |
2837 | decode_redir = 0; | |
2838 | } | |
2839 | ||
2840 | if (decode_redir) { | |
fe5da05e ID |
2841 | ret = ceph_redirect_decode(&p, end, &m->redirect); |
2842 | if (ret) | |
2843 | return ret; | |
205ee118 | 2844 | } else { |
fe5da05e | 2845 | ceph_oloc_init(&m->redirect.oloc); |
205ee118 | 2846 | } |
f24e9980 | 2847 | |
fe5da05e ID |
2848 | return 0; |
2849 | ||
2850 | e_inval: | |
2851 | return -EINVAL; | |
2852 | } | |
2853 | ||
2854 | /* | |
b18b9550 ID |
2855 | * Handle MOSDOpReply. Set ->r_result and call the callback if it is |
2856 | * specified. | |
fe5da05e | 2857 | */ |
5aea3dcd | 2858 | static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg) |
fe5da05e | 2859 | { |
5aea3dcd | 2860 | struct ceph_osd_client *osdc = osd->o_osdc; |
fe5da05e ID |
2861 | struct ceph_osd_request *req; |
2862 | struct MOSDOpReply m; | |
2863 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
2864 | u32 data_len = 0; | |
fe5da05e ID |
2865 | int ret; |
2866 | int i; | |
2867 | ||
2868 | dout("%s msg %p tid %llu\n", __func__, msg, tid); | |
2869 | ||
5aea3dcd ID |
2870 | down_read(&osdc->lock); |
2871 | if (!osd_registered(osd)) { | |
2872 | dout("%s osd%d unknown\n", __func__, osd->o_osd); | |
2873 | goto out_unlock_osdc; | |
2874 | } | |
2875 | WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num)); | |
2876 | ||
2877 | mutex_lock(&osd->lock); | |
2878 | req = lookup_request(&osd->o_requests, tid); | |
fe5da05e | 2879 | if (!req) { |
5aea3dcd ID |
2880 | dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid); |
2881 | goto out_unlock_session; | |
fe5da05e | 2882 | } |
fe5da05e | 2883 | |
cd08e0a2 | 2884 | m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns; |
fe5da05e | 2885 | ret = decode_MOSDOpReply(msg, &m); |
cd08e0a2 | 2886 | m.redirect.oloc.pool_ns = NULL; |
fe5da05e ID |
2887 | if (ret) { |
2888 | pr_err("failed to decode MOSDOpReply for tid %llu: %d\n", | |
2889 | req->r_tid, ret); | |
2890 | ceph_msg_dump(msg); | |
2891 | goto fail_request; | |
2892 | } | |
2893 | dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n", | |
2894 | __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed, | |
2895 | m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch), | |
2896 | le64_to_cpu(m.replay_version.version), m.user_version); | |
2897 | ||
2898 | if (m.retry_attempt >= 0) { | |
2899 | if (m.retry_attempt != req->r_attempts - 1) { | |
2900 | dout("req %p tid %llu retry_attempt %d != %d, ignoring\n", | |
2901 | req, req->r_tid, m.retry_attempt, | |
2902 | req->r_attempts - 1); | |
5aea3dcd | 2903 | goto out_unlock_session; |
fe5da05e ID |
2904 | } |
2905 | } else { | |
2906 | WARN_ON(1); /* MOSDOpReply v4 is assumed */ | |
2907 | } | |
2908 | ||
2909 | if (!ceph_oloc_empty(&m.redirect.oloc)) { | |
2910 | dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid, | |
2911 | m.redirect.oloc.pool); | |
5aea3dcd ID |
2912 | unlink_request(osd, req); |
2913 | mutex_unlock(&osd->lock); | |
205ee118 | 2914 | |
30c156d9 YZ |
2915 | /* |
2916 | * Not ceph_oloc_copy() - changing pool_ns is not | |
2917 | * supported. | |
2918 | */ | |
2919 | req->r_t.target_oloc.pool = m.redirect.oloc.pool; | |
5aea3dcd ID |
2920 | req->r_flags |= CEPH_OSD_FLAG_REDIRECTED; |
2921 | req->r_tid = 0; | |
2922 | __submit_request(req, false); | |
2923 | goto out_unlock_osdc; | |
205ee118 ID |
2924 | } |
2925 | ||
fe5da05e ID |
2926 | if (m.num_ops != req->r_num_ops) { |
2927 | pr_err("num_ops %d != %d for tid %llu\n", m.num_ops, | |
2928 | req->r_num_ops, req->r_tid); | |
2929 | goto fail_request; | |
f24e9980 | 2930 | } |
fe5da05e ID |
2931 | for (i = 0; i < req->r_num_ops; i++) { |
2932 | dout(" req %p tid %llu op %d rval %d len %u\n", req, | |
2933 | req->r_tid, i, m.rval[i], m.outdata_len[i]); | |
2934 | req->r_ops[i].rval = m.rval[i]; | |
2935 | req->r_ops[i].outdata_len = m.outdata_len[i]; | |
2936 | data_len += m.outdata_len[i]; | |
2937 | } | |
2938 | if (data_len != le32_to_cpu(msg->hdr.data_len)) { | |
2939 | pr_err("sum of lens %u != %u for tid %llu\n", data_len, | |
2940 | le32_to_cpu(msg->hdr.data_len), req->r_tid); | |
2941 | goto fail_request; | |
2942 | } | |
b18b9550 ID |
2943 | dout("%s req %p tid %llu result %d data_len %u\n", __func__, |
2944 | req, req->r_tid, m.result, data_len); | |
f24e9980 | 2945 | |
b18b9550 ID |
2946 | /* |
2947 | * Since we only ever request ONDISK, we should only ever get | |
2948 | * one (type of) reply back. | |
2949 | */ | |
2950 | WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK)); | |
2951 | req->r_result = m.result ?: data_len; | |
2952 | finish_request(req); | |
5aea3dcd ID |
2953 | mutex_unlock(&osd->lock); |
2954 | up_read(&osdc->lock); | |
f24e9980 | 2955 | |
b18b9550 ID |
2956 | __complete_request(req); |
2957 | complete_all(&req->r_completion); | |
2958 | ceph_osdc_put_request(req); | |
f24e9980 SW |
2959 | return; |
2960 | ||
fe5da05e | 2961 | fail_request: |
4609245e | 2962 | complete_request(req, -EIO); |
5aea3dcd ID |
2963 | out_unlock_session: |
2964 | mutex_unlock(&osd->lock); | |
2965 | out_unlock_osdc: | |
2966 | up_read(&osdc->lock); | |
f24e9980 SW |
2967 | } |
2968 | ||
42c1b124 ID |
2969 | static void set_pool_was_full(struct ceph_osd_client *osdc) |
2970 | { | |
2971 | struct rb_node *n; | |
2972 | ||
2973 | for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { | |
2974 | struct ceph_pg_pool_info *pi = | |
2975 | rb_entry(n, struct ceph_pg_pool_info, node); | |
2976 | ||
2977 | pi->was_full = __pool_full(pi); | |
2978 | } | |
2979 | } | |
2980 | ||
5aea3dcd | 2981 | static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id) |
f24e9980 | 2982 | { |
5aea3dcd | 2983 | struct ceph_pg_pool_info *pi; |
f24e9980 | 2984 | |
5aea3dcd ID |
2985 | pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); |
2986 | if (!pi) | |
2987 | return false; | |
f24e9980 | 2988 | |
5aea3dcd | 2989 | return pi->was_full && !__pool_full(pi); |
422d2cb8 YS |
2990 | } |
2991 | ||
922dab61 ID |
2992 | static enum calc_target_result |
2993 | recalc_linger_target(struct ceph_osd_linger_request *lreq) | |
2994 | { | |
2995 | struct ceph_osd_client *osdc = lreq->osdc; | |
2996 | enum calc_target_result ct_res; | |
2997 | ||
2998 | ct_res = calc_target(osdc, &lreq->t, &lreq->last_force_resend, true); | |
2999 | if (ct_res == CALC_TARGET_NEED_RESEND) { | |
3000 | struct ceph_osd *osd; | |
3001 | ||
3002 | osd = lookup_create_osd(osdc, lreq->t.osd, true); | |
3003 | if (osd != lreq->osd) { | |
3004 | unlink_linger(lreq->osd, lreq); | |
3005 | link_linger(osd, lreq); | |
3006 | } | |
3007 | } | |
3008 | ||
3009 | return ct_res; | |
3010 | } | |
3011 | ||
422d2cb8 | 3012 | /* |
5aea3dcd | 3013 | * Requeue requests whose mapping to an OSD has changed. |
422d2cb8 | 3014 | */ |
5aea3dcd ID |
3015 | static void scan_requests(struct ceph_osd *osd, |
3016 | bool force_resend, | |
3017 | bool cleared_full, | |
3018 | bool check_pool_cleared_full, | |
3019 | struct rb_root *need_resend, | |
3020 | struct list_head *need_resend_linger) | |
422d2cb8 | 3021 | { |
5aea3dcd ID |
3022 | struct ceph_osd_client *osdc = osd->o_osdc; |
3023 | struct rb_node *n; | |
3024 | bool force_resend_writes; | |
3025 | ||
922dab61 ID |
3026 | for (n = rb_first(&osd->o_linger_requests); n; ) { |
3027 | struct ceph_osd_linger_request *lreq = | |
3028 | rb_entry(n, struct ceph_osd_linger_request, node); | |
3029 | enum calc_target_result ct_res; | |
3030 | ||
3031 | n = rb_next(n); /* recalc_linger_target() */ | |
3032 | ||
3033 | dout("%s lreq %p linger_id %llu\n", __func__, lreq, | |
3034 | lreq->linger_id); | |
3035 | ct_res = recalc_linger_target(lreq); | |
3036 | switch (ct_res) { | |
3037 | case CALC_TARGET_NO_ACTION: | |
3038 | force_resend_writes = cleared_full || | |
3039 | (check_pool_cleared_full && | |
3040 | pool_cleared_full(osdc, lreq->t.base_oloc.pool)); | |
3041 | if (!force_resend && !force_resend_writes) | |
3042 | break; | |
3043 | ||
3044 | /* fall through */ | |
3045 | case CALC_TARGET_NEED_RESEND: | |
4609245e | 3046 | cancel_linger_map_check(lreq); |
922dab61 ID |
3047 | /* |
3048 | * scan_requests() for the previous epoch(s) | |
3049 | * may have already added it to the list, since | |
3050 | * it's not unlinked here. | |
3051 | */ | |
3052 | if (list_empty(&lreq->scan_item)) | |
3053 | list_add_tail(&lreq->scan_item, need_resend_linger); | |
3054 | break; | |
3055 | case CALC_TARGET_POOL_DNE: | |
4609245e | 3056 | check_linger_pool_dne(lreq); |
922dab61 ID |
3057 | break; |
3058 | } | |
3059 | } | |
3060 | ||
5aea3dcd ID |
3061 | for (n = rb_first(&osd->o_requests); n; ) { |
3062 | struct ceph_osd_request *req = | |
3063 | rb_entry(n, struct ceph_osd_request, r_node); | |
3064 | enum calc_target_result ct_res; | |
3065 | ||
4609245e | 3066 | n = rb_next(n); /* unlink_request(), check_pool_dne() */ |
5aea3dcd ID |
3067 | |
3068 | dout("%s req %p tid %llu\n", __func__, req, req->r_tid); | |
3069 | ct_res = calc_target(osdc, &req->r_t, | |
3070 | &req->r_last_force_resend, false); | |
3071 | switch (ct_res) { | |
3072 | case CALC_TARGET_NO_ACTION: | |
3073 | force_resend_writes = cleared_full || | |
3074 | (check_pool_cleared_full && | |
3075 | pool_cleared_full(osdc, req->r_t.base_oloc.pool)); | |
3076 | if (!force_resend && | |
3077 | (!(req->r_flags & CEPH_OSD_FLAG_WRITE) || | |
3078 | !force_resend_writes)) | |
3079 | break; | |
3080 | ||
3081 | /* fall through */ | |
3082 | case CALC_TARGET_NEED_RESEND: | |
4609245e | 3083 | cancel_map_check(req); |
5aea3dcd ID |
3084 | unlink_request(osd, req); |
3085 | insert_request(need_resend, req); | |
3086 | break; | |
3087 | case CALC_TARGET_POOL_DNE: | |
4609245e | 3088 | check_pool_dne(req); |
5aea3dcd | 3089 | break; |
b0494532 | 3090 | } |
6f6c7006 | 3091 | } |
422d2cb8 | 3092 | } |
6f6c7006 | 3093 | |
42c1b124 | 3094 | static int handle_one_map(struct ceph_osd_client *osdc, |
5aea3dcd ID |
3095 | void *p, void *end, bool incremental, |
3096 | struct rb_root *need_resend, | |
3097 | struct list_head *need_resend_linger) | |
42c1b124 ID |
3098 | { |
3099 | struct ceph_osdmap *newmap; | |
3100 | struct rb_node *n; | |
3101 | bool skipped_map = false; | |
3102 | bool was_full; | |
3103 | ||
b7ec35b3 | 3104 | was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); |
42c1b124 ID |
3105 | set_pool_was_full(osdc); |
3106 | ||
3107 | if (incremental) | |
3108 | newmap = osdmap_apply_incremental(&p, end, osdc->osdmap); | |
3109 | else | |
3110 | newmap = ceph_osdmap_decode(&p, end); | |
3111 | if (IS_ERR(newmap)) | |
3112 | return PTR_ERR(newmap); | |
3113 | ||
3114 | if (newmap != osdc->osdmap) { | |
3115 | /* | |
3116 | * Preserve ->was_full before destroying the old map. | |
3117 | * For pools that weren't in the old map, ->was_full | |
3118 | * should be false. | |
3119 | */ | |
3120 | for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) { | |
3121 | struct ceph_pg_pool_info *pi = | |
3122 | rb_entry(n, struct ceph_pg_pool_info, node); | |
3123 | struct ceph_pg_pool_info *old_pi; | |
3124 | ||
3125 | old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id); | |
3126 | if (old_pi) | |
3127 | pi->was_full = old_pi->was_full; | |
3128 | else | |
3129 | WARN_ON(pi->was_full); | |
3130 | } | |
3131 | ||
3132 | if (osdc->osdmap->epoch && | |
3133 | osdc->osdmap->epoch + 1 < newmap->epoch) { | |
3134 | WARN_ON(incremental); | |
3135 | skipped_map = true; | |
3136 | } | |
3137 | ||
3138 | ceph_osdmap_destroy(osdc->osdmap); | |
3139 | osdc->osdmap = newmap; | |
3140 | } | |
3141 | ||
b7ec35b3 | 3142 | was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); |
5aea3dcd ID |
3143 | scan_requests(&osdc->homeless_osd, skipped_map, was_full, true, |
3144 | need_resend, need_resend_linger); | |
3145 | ||
3146 | for (n = rb_first(&osdc->osds); n; ) { | |
3147 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
3148 | ||
3149 | n = rb_next(n); /* close_osd() */ | |
3150 | ||
3151 | scan_requests(osd, skipped_map, was_full, true, need_resend, | |
3152 | need_resend_linger); | |
3153 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || | |
3154 | memcmp(&osd->o_con.peer_addr, | |
3155 | ceph_osd_addr(osdc->osdmap, osd->o_osd), | |
3156 | sizeof(struct ceph_entity_addr))) | |
3157 | close_osd(osd); | |
3158 | } | |
42c1b124 ID |
3159 | |
3160 | return 0; | |
3161 | } | |
6f6c7006 | 3162 | |
5aea3dcd ID |
3163 | static void kick_requests(struct ceph_osd_client *osdc, |
3164 | struct rb_root *need_resend, | |
3165 | struct list_head *need_resend_linger) | |
3166 | { | |
922dab61 | 3167 | struct ceph_osd_linger_request *lreq, *nlreq; |
5aea3dcd ID |
3168 | struct rb_node *n; |
3169 | ||
3170 | for (n = rb_first(need_resend); n; ) { | |
3171 | struct ceph_osd_request *req = | |
3172 | rb_entry(n, struct ceph_osd_request, r_node); | |
3173 | struct ceph_osd *osd; | |
3174 | ||
3175 | n = rb_next(n); | |
3176 | erase_request(need_resend, req); /* before link_request() */ | |
3177 | ||
3178 | WARN_ON(req->r_osd); | |
3179 | calc_target(osdc, &req->r_t, NULL, false); | |
3180 | osd = lookup_create_osd(osdc, req->r_t.osd, true); | |
3181 | link_request(osd, req); | |
3182 | if (!req->r_linger) { | |
3183 | if (!osd_homeless(osd) && !req->r_t.paused) | |
3184 | send_request(req); | |
922dab61 ID |
3185 | } else { |
3186 | cancel_linger_request(req); | |
5aea3dcd ID |
3187 | } |
3188 | } | |
922dab61 ID |
3189 | |
3190 | list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) { | |
3191 | if (!osd_homeless(lreq->osd)) | |
3192 | send_linger(lreq); | |
3193 | ||
3194 | list_del_init(&lreq->scan_item); | |
3195 | } | |
5aea3dcd ID |
3196 | } |
3197 | ||
f24e9980 SW |
3198 | /* |
3199 | * Process updated osd map. | |
3200 | * | |
3201 | * The message contains any number of incremental and full maps, normally | |
3202 | * indicating some sort of topology change in the cluster. Kick requests | |
3203 | * off to different OSDs as needed. | |
3204 | */ | |
3205 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
3206 | { | |
42c1b124 ID |
3207 | void *p = msg->front.iov_base; |
3208 | void *const end = p + msg->front.iov_len; | |
f24e9980 SW |
3209 | u32 nr_maps, maplen; |
3210 | u32 epoch; | |
f24e9980 | 3211 | struct ceph_fsid fsid; |
5aea3dcd ID |
3212 | struct rb_root need_resend = RB_ROOT; |
3213 | LIST_HEAD(need_resend_linger); | |
42c1b124 ID |
3214 | bool handled_incremental = false; |
3215 | bool was_pauserd, was_pausewr; | |
3216 | bool pauserd, pausewr; | |
3217 | int err; | |
f24e9980 | 3218 | |
42c1b124 | 3219 | dout("%s have %u\n", __func__, osdc->osdmap->epoch); |
5aea3dcd | 3220 | down_write(&osdc->lock); |
f24e9980 SW |
3221 | |
3222 | /* verify fsid */ | |
3223 | ceph_decode_need(&p, end, sizeof(fsid), bad); | |
3224 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
0743304d | 3225 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
42c1b124 | 3226 | goto bad; |
f24e9980 | 3227 | |
b7ec35b3 ID |
3228 | was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); |
3229 | was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || | |
3230 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || | |
42c1b124 | 3231 | have_pool_full(osdc); |
9a1ea2db | 3232 | |
f24e9980 SW |
3233 | /* incremental maps */ |
3234 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
3235 | dout(" %d inc maps\n", nr_maps); | |
3236 | while (nr_maps > 0) { | |
3237 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
3238 | epoch = ceph_decode_32(&p); |
3239 | maplen = ceph_decode_32(&p); | |
f24e9980 | 3240 | ceph_decode_need(&p, end, maplen, bad); |
42c1b124 ID |
3241 | if (osdc->osdmap->epoch && |
3242 | osdc->osdmap->epoch + 1 == epoch) { | |
f24e9980 SW |
3243 | dout("applying incremental map %u len %d\n", |
3244 | epoch, maplen); | |
5aea3dcd ID |
3245 | err = handle_one_map(osdc, p, p + maplen, true, |
3246 | &need_resend, &need_resend_linger); | |
42c1b124 | 3247 | if (err) |
f24e9980 | 3248 | goto bad; |
42c1b124 | 3249 | handled_incremental = true; |
f24e9980 SW |
3250 | } else { |
3251 | dout("ignoring incremental map %u len %d\n", | |
3252 | epoch, maplen); | |
3253 | } | |
42c1b124 | 3254 | p += maplen; |
f24e9980 SW |
3255 | nr_maps--; |
3256 | } | |
42c1b124 | 3257 | if (handled_incremental) |
f24e9980 SW |
3258 | goto done; |
3259 | ||
3260 | /* full maps */ | |
3261 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
3262 | dout(" %d full maps\n", nr_maps); | |
3263 | while (nr_maps) { | |
3264 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
3265 | epoch = ceph_decode_32(&p); |
3266 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
3267 | ceph_decode_need(&p, end, maplen, bad); |
3268 | if (nr_maps > 1) { | |
3269 | dout("skipping non-latest full map %u len %d\n", | |
3270 | epoch, maplen); | |
e5253a7b | 3271 | } else if (osdc->osdmap->epoch >= epoch) { |
f24e9980 SW |
3272 | dout("skipping full map %u len %d, " |
3273 | "older than our %u\n", epoch, maplen, | |
3274 | osdc->osdmap->epoch); | |
3275 | } else { | |
3276 | dout("taking full map %u len %d\n", epoch, maplen); | |
5aea3dcd ID |
3277 | err = handle_one_map(osdc, p, p + maplen, false, |
3278 | &need_resend, &need_resend_linger); | |
42c1b124 | 3279 | if (err) |
f24e9980 | 3280 | goto bad; |
f24e9980 SW |
3281 | } |
3282 | p += maplen; | |
3283 | nr_maps--; | |
3284 | } | |
3285 | ||
3286 | done: | |
cd634fb6 SW |
3287 | /* |
3288 | * subscribe to subsequent osdmap updates if full to ensure | |
3289 | * we find out when we are no longer full and stop returning | |
3290 | * ENOSPC. | |
3291 | */ | |
b7ec35b3 ID |
3292 | pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); |
3293 | pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || | |
3294 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || | |
42c1b124 ID |
3295 | have_pool_full(osdc); |
3296 | if (was_pauserd || was_pausewr || pauserd || pausewr) | |
3297 | maybe_request_map(osdc); | |
cd634fb6 | 3298 | |
5aea3dcd | 3299 | kick_requests(osdc, &need_resend, &need_resend_linger); |
42c1b124 | 3300 | |
fc36d0a4 | 3301 | ceph_osdc_abort_on_full(osdc); |
42c1b124 ID |
3302 | ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP, |
3303 | osdc->osdmap->epoch); | |
5aea3dcd | 3304 | up_write(&osdc->lock); |
03066f23 | 3305 | wake_up_all(&osdc->client->auth_wq); |
f24e9980 SW |
3306 | return; |
3307 | ||
3308 | bad: | |
3309 | pr_err("osdc handle_map corrupt msg\n"); | |
9ec7cab1 | 3310 | ceph_msg_dump(msg); |
5aea3dcd ID |
3311 | up_write(&osdc->lock); |
3312 | } | |
3313 | ||
3314 | /* | |
3315 | * Resubmit requests pending on the given osd. | |
3316 | */ | |
3317 | static void kick_osd_requests(struct ceph_osd *osd) | |
3318 | { | |
3319 | struct rb_node *n; | |
3320 | ||
922dab61 | 3321 | for (n = rb_first(&osd->o_requests); n; ) { |
5aea3dcd ID |
3322 | struct ceph_osd_request *req = |
3323 | rb_entry(n, struct ceph_osd_request, r_node); | |
3324 | ||
922dab61 ID |
3325 | n = rb_next(n); /* cancel_linger_request() */ |
3326 | ||
5aea3dcd ID |
3327 | if (!req->r_linger) { |
3328 | if (!req->r_t.paused) | |
3329 | send_request(req); | |
922dab61 ID |
3330 | } else { |
3331 | cancel_linger_request(req); | |
5aea3dcd ID |
3332 | } |
3333 | } | |
922dab61 ID |
3334 | for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) { |
3335 | struct ceph_osd_linger_request *lreq = | |
3336 | rb_entry(n, struct ceph_osd_linger_request, node); | |
3337 | ||
3338 | send_linger(lreq); | |
3339 | } | |
5aea3dcd ID |
3340 | } |
3341 | ||
3342 | /* | |
3343 | * If the osd connection drops, we need to resubmit all requests. | |
3344 | */ | |
3345 | static void osd_fault(struct ceph_connection *con) | |
3346 | { | |
3347 | struct ceph_osd *osd = con->private; | |
3348 | struct ceph_osd_client *osdc = osd->o_osdc; | |
3349 | ||
3350 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
3351 | ||
3352 | down_write(&osdc->lock); | |
3353 | if (!osd_registered(osd)) { | |
3354 | dout("%s osd%d unknown\n", __func__, osd->o_osd); | |
3355 | goto out_unlock; | |
3356 | } | |
3357 | ||
3358 | if (!reopen_osd(osd)) | |
3359 | kick_osd_requests(osd); | |
3360 | maybe_request_map(osdc); | |
3361 | ||
3362 | out_unlock: | |
3363 | up_write(&osdc->lock); | |
f24e9980 SW |
3364 | } |
3365 | ||
a40c4f10 YS |
3366 | /* |
3367 | * Process osd watch notifications | |
3368 | */ | |
3c663bbd AE |
3369 | static void handle_watch_notify(struct ceph_osd_client *osdc, |
3370 | struct ceph_msg *msg) | |
a40c4f10 | 3371 | { |
922dab61 ID |
3372 | void *p = msg->front.iov_base; |
3373 | void *const end = p + msg->front.iov_len; | |
3374 | struct ceph_osd_linger_request *lreq; | |
3375 | struct linger_work *lwork; | |
3376 | u8 proto_ver, opcode; | |
3377 | u64 cookie, notify_id; | |
3378 | u64 notifier_id = 0; | |
19079203 | 3379 | s32 return_code = 0; |
922dab61 ID |
3380 | void *payload = NULL; |
3381 | u32 payload_len = 0; | |
a40c4f10 YS |
3382 | |
3383 | ceph_decode_8_safe(&p, end, proto_ver, bad); | |
3384 | ceph_decode_8_safe(&p, end, opcode, bad); | |
3385 | ceph_decode_64_safe(&p, end, cookie, bad); | |
922dab61 | 3386 | p += 8; /* skip ver */ |
a40c4f10 YS |
3387 | ceph_decode_64_safe(&p, end, notify_id, bad); |
3388 | ||
922dab61 ID |
3389 | if (proto_ver >= 1) { |
3390 | ceph_decode_32_safe(&p, end, payload_len, bad); | |
3391 | ceph_decode_need(&p, end, payload_len, bad); | |
3392 | payload = p; | |
3393 | p += payload_len; | |
3394 | } | |
3395 | ||
3396 | if (le16_to_cpu(msg->hdr.version) >= 2) | |
19079203 | 3397 | ceph_decode_32_safe(&p, end, return_code, bad); |
922dab61 ID |
3398 | |
3399 | if (le16_to_cpu(msg->hdr.version) >= 3) | |
3400 | ceph_decode_64_safe(&p, end, notifier_id, bad); | |
3401 | ||
3402 | down_read(&osdc->lock); | |
3403 | lreq = lookup_linger_osdc(&osdc->linger_requests, cookie); | |
3404 | if (!lreq) { | |
3405 | dout("%s opcode %d cookie %llu dne\n", __func__, opcode, | |
3406 | cookie); | |
3407 | goto out_unlock_osdc; | |
3408 | } | |
3409 | ||
3410 | mutex_lock(&lreq->lock); | |
19079203 ID |
3411 | dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__, |
3412 | opcode, cookie, lreq, lreq->is_watch); | |
922dab61 ID |
3413 | if (opcode == CEPH_WATCH_EVENT_DISCONNECT) { |
3414 | if (!lreq->last_error) { | |
3415 | lreq->last_error = -ENOTCONN; | |
3416 | queue_watch_error(lreq); | |
3417 | } | |
19079203 ID |
3418 | } else if (!lreq->is_watch) { |
3419 | /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */ | |
3420 | if (lreq->notify_id && lreq->notify_id != notify_id) { | |
3421 | dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq, | |
3422 | lreq->notify_id, notify_id); | |
3423 | } else if (!completion_done(&lreq->notify_finish_wait)) { | |
3424 | struct ceph_msg_data *data = | |
3425 | list_first_entry_or_null(&msg->data, | |
3426 | struct ceph_msg_data, | |
3427 | links); | |
3428 | ||
3429 | if (data) { | |
3430 | if (lreq->preply_pages) { | |
3431 | WARN_ON(data->type != | |
3432 | CEPH_MSG_DATA_PAGES); | |
3433 | *lreq->preply_pages = data->pages; | |
3434 | *lreq->preply_len = data->length; | |
3435 | } else { | |
3436 | ceph_release_page_vector(data->pages, | |
3437 | calc_pages_for(0, data->length)); | |
3438 | } | |
3439 | } | |
3440 | lreq->notify_finish_error = return_code; | |
3441 | complete_all(&lreq->notify_finish_wait); | |
3442 | } | |
922dab61 ID |
3443 | } else { |
3444 | /* CEPH_WATCH_EVENT_NOTIFY */ | |
3445 | lwork = lwork_alloc(lreq, do_watch_notify); | |
3446 | if (!lwork) { | |
3447 | pr_err("failed to allocate notify-lwork\n"); | |
3448 | goto out_unlock_lreq; | |
a40c4f10 | 3449 | } |
a40c4f10 | 3450 | |
922dab61 ID |
3451 | lwork->notify.notify_id = notify_id; |
3452 | lwork->notify.notifier_id = notifier_id; | |
3453 | lwork->notify.payload = payload; | |
3454 | lwork->notify.payload_len = payload_len; | |
3455 | lwork->notify.msg = ceph_msg_get(msg); | |
3456 | lwork_queue(lwork); | |
91883cd2 | 3457 | } |
a40c4f10 | 3458 | |
922dab61 ID |
3459 | out_unlock_lreq: |
3460 | mutex_unlock(&lreq->lock); | |
3461 | out_unlock_osdc: | |
3462 | up_read(&osdc->lock); | |
a40c4f10 YS |
3463 | return; |
3464 | ||
3465 | bad: | |
3466 | pr_err("osdc handle_watch_notify corrupt msg\n"); | |
a40c4f10 YS |
3467 | } |
3468 | ||
70636773 AE |
3469 | /* |
3470 | * Register request, send initial attempt. | |
3471 | */ | |
3472 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
3473 | struct ceph_osd_request *req, | |
3474 | bool nofail) | |
3475 | { | |
5aea3dcd ID |
3476 | down_read(&osdc->lock); |
3477 | submit_request(req, false); | |
3478 | up_read(&osdc->lock); | |
0bbfdfe8 | 3479 | |
5aea3dcd | 3480 | return 0; |
f24e9980 | 3481 | } |
3d14c5d2 | 3482 | EXPORT_SYMBOL(ceph_osdc_start_request); |
f24e9980 | 3483 | |
c9f9b93d | 3484 | /* |
c297eb42 ID |
3485 | * Unregister a registered request. The request is not completed: |
3486 | * ->r_result isn't set and __complete_request() isn't called. | |
c9f9b93d ID |
3487 | */ |
3488 | void ceph_osdc_cancel_request(struct ceph_osd_request *req) | |
3489 | { | |
3490 | struct ceph_osd_client *osdc = req->r_osdc; | |
3491 | ||
5aea3dcd | 3492 | down_write(&osdc->lock); |
5aea3dcd ID |
3493 | if (req->r_osd) |
3494 | cancel_request(req); | |
3495 | up_write(&osdc->lock); | |
c9f9b93d ID |
3496 | } |
3497 | EXPORT_SYMBOL(ceph_osdc_cancel_request); | |
3498 | ||
f24e9980 | 3499 | /* |
42b06965 | 3500 | * @timeout: in jiffies, 0 means "wait forever" |
f24e9980 | 3501 | */ |
42b06965 ID |
3502 | static int wait_request_timeout(struct ceph_osd_request *req, |
3503 | unsigned long timeout) | |
f24e9980 | 3504 | { |
42b06965 | 3505 | long left; |
c9f9b93d | 3506 | |
42b06965 | 3507 | dout("%s req %p tid %llu\n", __func__, req, req->r_tid); |
0e76abf2 | 3508 | left = wait_for_completion_killable_timeout(&req->r_completion, |
42b06965 ID |
3509 | ceph_timeout_jiffies(timeout)); |
3510 | if (left <= 0) { | |
3511 | left = left ?: -ETIMEDOUT; | |
c9f9b93d | 3512 | ceph_osdc_cancel_request(req); |
42b06965 ID |
3513 | } else { |
3514 | left = req->r_result; /* completed */ | |
f24e9980 SW |
3515 | } |
3516 | ||
42b06965 ID |
3517 | return left; |
3518 | } | |
3519 | ||
3520 | /* | |
3521 | * wait for a request to complete | |
3522 | */ | |
3523 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, | |
3524 | struct ceph_osd_request *req) | |
3525 | { | |
3526 | return wait_request_timeout(req, 0); | |
f24e9980 | 3527 | } |
3d14c5d2 | 3528 | EXPORT_SYMBOL(ceph_osdc_wait_request); |
f24e9980 SW |
3529 | |
3530 | /* | |
3531 | * sync - wait for all in-flight requests to flush. avoid starvation. | |
3532 | */ | |
3533 | void ceph_osdc_sync(struct ceph_osd_client *osdc) | |
3534 | { | |
5aea3dcd ID |
3535 | struct rb_node *n, *p; |
3536 | u64 last_tid = atomic64_read(&osdc->last_tid); | |
f24e9980 | 3537 | |
5aea3dcd ID |
3538 | again: |
3539 | down_read(&osdc->lock); | |
3540 | for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { | |
3541 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
3542 | ||
3543 | mutex_lock(&osd->lock); | |
3544 | for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) { | |
3545 | struct ceph_osd_request *req = | |
3546 | rb_entry(p, struct ceph_osd_request, r_node); | |
3547 | ||
3548 | if (req->r_tid > last_tid) | |
3549 | break; | |
3550 | ||
3551 | if (!(req->r_flags & CEPH_OSD_FLAG_WRITE)) | |
3552 | continue; | |
f24e9980 | 3553 | |
5aea3dcd ID |
3554 | ceph_osdc_get_request(req); |
3555 | mutex_unlock(&osd->lock); | |
3556 | up_read(&osdc->lock); | |
3557 | dout("%s waiting on req %p tid %llu last_tid %llu\n", | |
3558 | __func__, req, req->r_tid, last_tid); | |
b18b9550 | 3559 | wait_for_completion(&req->r_completion); |
5aea3dcd ID |
3560 | ceph_osdc_put_request(req); |
3561 | goto again; | |
3562 | } | |
f24e9980 | 3563 | |
5aea3dcd | 3564 | mutex_unlock(&osd->lock); |
f24e9980 | 3565 | } |
5aea3dcd ID |
3566 | |
3567 | up_read(&osdc->lock); | |
3568 | dout("%s done last_tid %llu\n", __func__, last_tid); | |
f24e9980 | 3569 | } |
3d14c5d2 | 3570 | EXPORT_SYMBOL(ceph_osdc_sync); |
f24e9980 | 3571 | |
922dab61 ID |
3572 | static struct ceph_osd_request * |
3573 | alloc_linger_request(struct ceph_osd_linger_request *lreq) | |
3574 | { | |
3575 | struct ceph_osd_request *req; | |
3576 | ||
3577 | req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO); | |
3578 | if (!req) | |
3579 | return NULL; | |
3580 | ||
3581 | ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); | |
3582 | ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); | |
3583 | ||
3584 | if (ceph_osdc_alloc_messages(req, GFP_NOIO)) { | |
3585 | ceph_osdc_put_request(req); | |
3586 | return NULL; | |
3587 | } | |
3588 | ||
3589 | return req; | |
3590 | } | |
3591 | ||
3592 | /* | |
3593 | * Returns a handle, caller owns a ref. | |
3594 | */ | |
3595 | struct ceph_osd_linger_request * | |
3596 | ceph_osdc_watch(struct ceph_osd_client *osdc, | |
3597 | struct ceph_object_id *oid, | |
3598 | struct ceph_object_locator *oloc, | |
3599 | rados_watchcb2_t wcb, | |
3600 | rados_watcherrcb_t errcb, | |
3601 | void *data) | |
3602 | { | |
3603 | struct ceph_osd_linger_request *lreq; | |
3604 | int ret; | |
3605 | ||
3606 | lreq = linger_alloc(osdc); | |
3607 | if (!lreq) | |
3608 | return ERR_PTR(-ENOMEM); | |
3609 | ||
19079203 | 3610 | lreq->is_watch = true; |
922dab61 ID |
3611 | lreq->wcb = wcb; |
3612 | lreq->errcb = errcb; | |
3613 | lreq->data = data; | |
b07d3c4b | 3614 | lreq->watch_valid_thru = jiffies; |
922dab61 ID |
3615 | |
3616 | ceph_oid_copy(&lreq->t.base_oid, oid); | |
3617 | ceph_oloc_copy(&lreq->t.base_oloc, oloc); | |
54ea0046 | 3618 | lreq->t.flags = CEPH_OSD_FLAG_WRITE; |
922dab61 ID |
3619 | lreq->mtime = CURRENT_TIME; |
3620 | ||
3621 | lreq->reg_req = alloc_linger_request(lreq); | |
3622 | if (!lreq->reg_req) { | |
3623 | ret = -ENOMEM; | |
3624 | goto err_put_lreq; | |
3625 | } | |
3626 | ||
3627 | lreq->ping_req = alloc_linger_request(lreq); | |
3628 | if (!lreq->ping_req) { | |
3629 | ret = -ENOMEM; | |
3630 | goto err_put_lreq; | |
3631 | } | |
3632 | ||
3633 | down_write(&osdc->lock); | |
3634 | linger_register(lreq); /* before osd_req_op_* */ | |
3635 | osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id, | |
3636 | CEPH_OSD_WATCH_OP_WATCH); | |
3637 | osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id, | |
3638 | CEPH_OSD_WATCH_OP_PING); | |
3639 | linger_submit(lreq); | |
3640 | up_write(&osdc->lock); | |
3641 | ||
3642 | ret = linger_reg_commit_wait(lreq); | |
3643 | if (ret) { | |
3644 | linger_cancel(lreq); | |
3645 | goto err_put_lreq; | |
3646 | } | |
3647 | ||
3648 | return lreq; | |
3649 | ||
3650 | err_put_lreq: | |
3651 | linger_put(lreq); | |
3652 | return ERR_PTR(ret); | |
3653 | } | |
3654 | EXPORT_SYMBOL(ceph_osdc_watch); | |
3655 | ||
3656 | /* | |
3657 | * Releases a ref. | |
3658 | * | |
3659 | * Times out after mount_timeout to preserve rbd unmap behaviour | |
3660 | * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap | |
3661 | * with mount_timeout"). | |
3662 | */ | |
3663 | int ceph_osdc_unwatch(struct ceph_osd_client *osdc, | |
3664 | struct ceph_osd_linger_request *lreq) | |
3665 | { | |
3666 | struct ceph_options *opts = osdc->client->options; | |
3667 | struct ceph_osd_request *req; | |
3668 | int ret; | |
3669 | ||
3670 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); | |
3671 | if (!req) | |
3672 | return -ENOMEM; | |
3673 | ||
3674 | ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); | |
3675 | ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); | |
54ea0046 | 3676 | req->r_flags = CEPH_OSD_FLAG_WRITE; |
922dab61 ID |
3677 | req->r_mtime = CURRENT_TIME; |
3678 | osd_req_op_watch_init(req, 0, lreq->linger_id, | |
3679 | CEPH_OSD_WATCH_OP_UNWATCH); | |
3680 | ||
3681 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
3682 | if (ret) | |
3683 | goto out_put_req; | |
3684 | ||
3685 | ceph_osdc_start_request(osdc, req, false); | |
3686 | linger_cancel(lreq); | |
3687 | linger_put(lreq); | |
3688 | ret = wait_request_timeout(req, opts->mount_timeout); | |
3689 | ||
3690 | out_put_req: | |
3691 | ceph_osdc_put_request(req); | |
3692 | return ret; | |
3693 | } | |
3694 | EXPORT_SYMBOL(ceph_osdc_unwatch); | |
3695 | ||
3696 | static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which, | |
3697 | u64 notify_id, u64 cookie, void *payload, | |
3698 | size_t payload_len) | |
3699 | { | |
3700 | struct ceph_osd_req_op *op; | |
3701 | struct ceph_pagelist *pl; | |
3702 | int ret; | |
3703 | ||
3704 | op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0); | |
3705 | ||
3706 | pl = kmalloc(sizeof(*pl), GFP_NOIO); | |
3707 | if (!pl) | |
3708 | return -ENOMEM; | |
3709 | ||
3710 | ceph_pagelist_init(pl); | |
3711 | ret = ceph_pagelist_encode_64(pl, notify_id); | |
3712 | ret |= ceph_pagelist_encode_64(pl, cookie); | |
3713 | if (payload) { | |
3714 | ret |= ceph_pagelist_encode_32(pl, payload_len); | |
3715 | ret |= ceph_pagelist_append(pl, payload, payload_len); | |
3716 | } else { | |
3717 | ret |= ceph_pagelist_encode_32(pl, 0); | |
3718 | } | |
3719 | if (ret) { | |
3720 | ceph_pagelist_release(pl); | |
3721 | return -ENOMEM; | |
3722 | } | |
3723 | ||
3724 | ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl); | |
3725 | op->indata_len = pl->length; | |
3726 | return 0; | |
3727 | } | |
3728 | ||
3729 | int ceph_osdc_notify_ack(struct ceph_osd_client *osdc, | |
3730 | struct ceph_object_id *oid, | |
3731 | struct ceph_object_locator *oloc, | |
3732 | u64 notify_id, | |
3733 | u64 cookie, | |
3734 | void *payload, | |
3735 | size_t payload_len) | |
3736 | { | |
3737 | struct ceph_osd_request *req; | |
3738 | int ret; | |
3739 | ||
3740 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); | |
3741 | if (!req) | |
3742 | return -ENOMEM; | |
3743 | ||
3744 | ceph_oid_copy(&req->r_base_oid, oid); | |
3745 | ceph_oloc_copy(&req->r_base_oloc, oloc); | |
3746 | req->r_flags = CEPH_OSD_FLAG_READ; | |
3747 | ||
3748 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
3749 | if (ret) | |
3750 | goto out_put_req; | |
3751 | ||
3752 | ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload, | |
3753 | payload_len); | |
3754 | if (ret) | |
3755 | goto out_put_req; | |
3756 | ||
3757 | ceph_osdc_start_request(osdc, req, false); | |
3758 | ret = ceph_osdc_wait_request(osdc, req); | |
3759 | ||
3760 | out_put_req: | |
3761 | ceph_osdc_put_request(req); | |
3762 | return ret; | |
3763 | } | |
3764 | EXPORT_SYMBOL(ceph_osdc_notify_ack); | |
3765 | ||
19079203 ID |
3766 | static int osd_req_op_notify_init(struct ceph_osd_request *req, int which, |
3767 | u64 cookie, u32 prot_ver, u32 timeout, | |
3768 | void *payload, size_t payload_len) | |
3769 | { | |
3770 | struct ceph_osd_req_op *op; | |
3771 | struct ceph_pagelist *pl; | |
3772 | int ret; | |
3773 | ||
3774 | op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0); | |
3775 | op->notify.cookie = cookie; | |
3776 | ||
3777 | pl = kmalloc(sizeof(*pl), GFP_NOIO); | |
3778 | if (!pl) | |
3779 | return -ENOMEM; | |
3780 | ||
3781 | ceph_pagelist_init(pl); | |
3782 | ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */ | |
3783 | ret |= ceph_pagelist_encode_32(pl, timeout); | |
3784 | ret |= ceph_pagelist_encode_32(pl, payload_len); | |
3785 | ret |= ceph_pagelist_append(pl, payload, payload_len); | |
3786 | if (ret) { | |
3787 | ceph_pagelist_release(pl); | |
3788 | return -ENOMEM; | |
3789 | } | |
3790 | ||
3791 | ceph_osd_data_pagelist_init(&op->notify.request_data, pl); | |
3792 | op->indata_len = pl->length; | |
3793 | return 0; | |
3794 | } | |
3795 | ||
3796 | /* | |
3797 | * @timeout: in seconds | |
3798 | * | |
3799 | * @preply_{pages,len} are initialized both on success and error. | |
3800 | * The caller is responsible for: | |
3801 | * | |
3802 | * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)) | |
3803 | */ | |
3804 | int ceph_osdc_notify(struct ceph_osd_client *osdc, | |
3805 | struct ceph_object_id *oid, | |
3806 | struct ceph_object_locator *oloc, | |
3807 | void *payload, | |
3808 | size_t payload_len, | |
3809 | u32 timeout, | |
3810 | struct page ***preply_pages, | |
3811 | size_t *preply_len) | |
3812 | { | |
3813 | struct ceph_osd_linger_request *lreq; | |
3814 | struct page **pages; | |
3815 | int ret; | |
3816 | ||
3817 | WARN_ON(!timeout); | |
3818 | if (preply_pages) { | |
3819 | *preply_pages = NULL; | |
3820 | *preply_len = 0; | |
3821 | } | |
3822 | ||
3823 | lreq = linger_alloc(osdc); | |
3824 | if (!lreq) | |
3825 | return -ENOMEM; | |
3826 | ||
3827 | lreq->preply_pages = preply_pages; | |
3828 | lreq->preply_len = preply_len; | |
3829 | ||
3830 | ceph_oid_copy(&lreq->t.base_oid, oid); | |
3831 | ceph_oloc_copy(&lreq->t.base_oloc, oloc); | |
3832 | lreq->t.flags = CEPH_OSD_FLAG_READ; | |
3833 | ||
3834 | lreq->reg_req = alloc_linger_request(lreq); | |
3835 | if (!lreq->reg_req) { | |
3836 | ret = -ENOMEM; | |
3837 | goto out_put_lreq; | |
3838 | } | |
3839 | ||
3840 | /* for notify_id */ | |
3841 | pages = ceph_alloc_page_vector(1, GFP_NOIO); | |
3842 | if (IS_ERR(pages)) { | |
3843 | ret = PTR_ERR(pages); | |
3844 | goto out_put_lreq; | |
3845 | } | |
3846 | ||
3847 | down_write(&osdc->lock); | |
3848 | linger_register(lreq); /* before osd_req_op_* */ | |
3849 | ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1, | |
3850 | timeout, payload, payload_len); | |
3851 | if (ret) { | |
3852 | linger_unregister(lreq); | |
3853 | up_write(&osdc->lock); | |
3854 | ceph_release_page_vector(pages, 1); | |
3855 | goto out_put_lreq; | |
3856 | } | |
3857 | ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify, | |
3858 | response_data), | |
3859 | pages, PAGE_SIZE, 0, false, true); | |
3860 | linger_submit(lreq); | |
3861 | up_write(&osdc->lock); | |
3862 | ||
3863 | ret = linger_reg_commit_wait(lreq); | |
3864 | if (!ret) | |
3865 | ret = linger_notify_finish_wait(lreq); | |
3866 | else | |
3867 | dout("lreq %p failed to initiate notify %d\n", lreq, ret); | |
3868 | ||
3869 | linger_cancel(lreq); | |
3870 | out_put_lreq: | |
3871 | linger_put(lreq); | |
3872 | return ret; | |
3873 | } | |
3874 | EXPORT_SYMBOL(ceph_osdc_notify); | |
3875 | ||
b07d3c4b ID |
3876 | /* |
3877 | * Return the number of milliseconds since the watch was last | |
3878 | * confirmed, or an error. If there is an error, the watch is no | |
3879 | * longer valid, and should be destroyed with ceph_osdc_unwatch(). | |
3880 | */ | |
3881 | int ceph_osdc_watch_check(struct ceph_osd_client *osdc, | |
3882 | struct ceph_osd_linger_request *lreq) | |
3883 | { | |
3884 | unsigned long stamp, age; | |
3885 | int ret; | |
3886 | ||
3887 | down_read(&osdc->lock); | |
3888 | mutex_lock(&lreq->lock); | |
3889 | stamp = lreq->watch_valid_thru; | |
3890 | if (!list_empty(&lreq->pending_lworks)) { | |
3891 | struct linger_work *lwork = | |
3892 | list_first_entry(&lreq->pending_lworks, | |
3893 | struct linger_work, | |
3894 | pending_item); | |
3895 | ||
3896 | if (time_before(lwork->queued_stamp, stamp)) | |
3897 | stamp = lwork->queued_stamp; | |
3898 | } | |
3899 | age = jiffies - stamp; | |
3900 | dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__, | |
3901 | lreq, lreq->linger_id, age, lreq->last_error); | |
3902 | /* we are truncating to msecs, so return a safe upper bound */ | |
3903 | ret = lreq->last_error ?: 1 + jiffies_to_msecs(age); | |
3904 | ||
3905 | mutex_unlock(&lreq->lock); | |
3906 | up_read(&osdc->lock); | |
3907 | return ret; | |
3908 | } | |
3909 | ||
a4ed38d7 DF |
3910 | static int decode_watcher(void **p, void *end, struct ceph_watch_item *item) |
3911 | { | |
3912 | u8 struct_v; | |
3913 | u32 struct_len; | |
3914 | int ret; | |
3915 | ||
3916 | ret = ceph_start_decoding(p, end, 2, "watch_item_t", | |
3917 | &struct_v, &struct_len); | |
3918 | if (ret) | |
3919 | return ret; | |
3920 | ||
3921 | ceph_decode_copy(p, &item->name, sizeof(item->name)); | |
3922 | item->cookie = ceph_decode_64(p); | |
3923 | *p += 4; /* skip timeout_seconds */ | |
3924 | if (struct_v >= 2) { | |
3925 | ceph_decode_copy(p, &item->addr, sizeof(item->addr)); | |
3926 | ceph_decode_addr(&item->addr); | |
3927 | } | |
3928 | ||
3929 | dout("%s %s%llu cookie %llu addr %s\n", __func__, | |
3930 | ENTITY_NAME(item->name), item->cookie, | |
3931 | ceph_pr_addr(&item->addr.in_addr)); | |
3932 | return 0; | |
3933 | } | |
3934 | ||
3935 | static int decode_watchers(void **p, void *end, | |
3936 | struct ceph_watch_item **watchers, | |
3937 | u32 *num_watchers) | |
3938 | { | |
3939 | u8 struct_v; | |
3940 | u32 struct_len; | |
3941 | int i; | |
3942 | int ret; | |
3943 | ||
3944 | ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t", | |
3945 | &struct_v, &struct_len); | |
3946 | if (ret) | |
3947 | return ret; | |
3948 | ||
3949 | *num_watchers = ceph_decode_32(p); | |
3950 | *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO); | |
3951 | if (!*watchers) | |
3952 | return -ENOMEM; | |
3953 | ||
3954 | for (i = 0; i < *num_watchers; i++) { | |
3955 | ret = decode_watcher(p, end, *watchers + i); | |
3956 | if (ret) { | |
3957 | kfree(*watchers); | |
3958 | return ret; | |
3959 | } | |
3960 | } | |
3961 | ||
3962 | return 0; | |
3963 | } | |
3964 | ||
3965 | /* | |
3966 | * On success, the caller is responsible for: | |
3967 | * | |
3968 | * kfree(watchers); | |
3969 | */ | |
3970 | int ceph_osdc_list_watchers(struct ceph_osd_client *osdc, | |
3971 | struct ceph_object_id *oid, | |
3972 | struct ceph_object_locator *oloc, | |
3973 | struct ceph_watch_item **watchers, | |
3974 | u32 *num_watchers) | |
3975 | { | |
3976 | struct ceph_osd_request *req; | |
3977 | struct page **pages; | |
3978 | int ret; | |
3979 | ||
3980 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); | |
3981 | if (!req) | |
3982 | return -ENOMEM; | |
3983 | ||
3984 | ceph_oid_copy(&req->r_base_oid, oid); | |
3985 | ceph_oloc_copy(&req->r_base_oloc, oloc); | |
3986 | req->r_flags = CEPH_OSD_FLAG_READ; | |
3987 | ||
3988 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
3989 | if (ret) | |
3990 | goto out_put_req; | |
3991 | ||
3992 | pages = ceph_alloc_page_vector(1, GFP_NOIO); | |
3993 | if (IS_ERR(pages)) { | |
3994 | ret = PTR_ERR(pages); | |
3995 | goto out_put_req; | |
3996 | } | |
3997 | ||
3998 | osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0); | |
3999 | ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers, | |
4000 | response_data), | |
4001 | pages, PAGE_SIZE, 0, false, true); | |
4002 | ||
4003 | ceph_osdc_start_request(osdc, req, false); | |
4004 | ret = ceph_osdc_wait_request(osdc, req); | |
4005 | if (ret >= 0) { | |
4006 | void *p = page_address(pages[0]); | |
4007 | void *const end = p + req->r_ops[0].outdata_len; | |
4008 | ||
4009 | ret = decode_watchers(&p, end, watchers, num_watchers); | |
4010 | } | |
4011 | ||
4012 | out_put_req: | |
4013 | ceph_osdc_put_request(req); | |
4014 | return ret; | |
4015 | } | |
4016 | EXPORT_SYMBOL(ceph_osdc_list_watchers); | |
4017 | ||
dd935f44 JD |
4018 | /* |
4019 | * Call all pending notify callbacks - for use after a watch is | |
4020 | * unregistered, to make sure no more callbacks for it will be invoked | |
4021 | */ | |
f6479449 | 4022 | void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc) |
dd935f44 | 4023 | { |
99d16943 | 4024 | dout("%s osdc %p\n", __func__, osdc); |
dd935f44 JD |
4025 | flush_workqueue(osdc->notify_wq); |
4026 | } | |
4027 | EXPORT_SYMBOL(ceph_osdc_flush_notifies); | |
4028 | ||
7cca78c9 ID |
4029 | void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc) |
4030 | { | |
4031 | down_read(&osdc->lock); | |
4032 | maybe_request_map(osdc); | |
4033 | up_read(&osdc->lock); | |
4034 | } | |
4035 | EXPORT_SYMBOL(ceph_osdc_maybe_request_map); | |
dd935f44 | 4036 | |
428a7158 DF |
4037 | /* |
4038 | * Execute an OSD class method on an object. | |
4039 | * | |
4040 | * @flags: CEPH_OSD_FLAG_* | |
2544a020 | 4041 | * @resp_len: in/out param for reply length |
428a7158 DF |
4042 | */ |
4043 | int ceph_osdc_call(struct ceph_osd_client *osdc, | |
4044 | struct ceph_object_id *oid, | |
4045 | struct ceph_object_locator *oloc, | |
4046 | const char *class, const char *method, | |
4047 | unsigned int flags, | |
4048 | struct page *req_page, size_t req_len, | |
4049 | struct page *resp_page, size_t *resp_len) | |
4050 | { | |
4051 | struct ceph_osd_request *req; | |
4052 | int ret; | |
4053 | ||
2544a020 ID |
4054 | if (req_len > PAGE_SIZE || (resp_page && *resp_len > PAGE_SIZE)) |
4055 | return -E2BIG; | |
4056 | ||
428a7158 DF |
4057 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); |
4058 | if (!req) | |
4059 | return -ENOMEM; | |
4060 | ||
4061 | ceph_oid_copy(&req->r_base_oid, oid); | |
4062 | ceph_oloc_copy(&req->r_base_oloc, oloc); | |
4063 | req->r_flags = flags; | |
4064 | ||
4065 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
4066 | if (ret) | |
4067 | goto out_put_req; | |
4068 | ||
4069 | osd_req_op_cls_init(req, 0, CEPH_OSD_OP_CALL, class, method); | |
4070 | if (req_page) | |
4071 | osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len, | |
4072 | 0, false, false); | |
4073 | if (resp_page) | |
4074 | osd_req_op_cls_response_data_pages(req, 0, &resp_page, | |
2544a020 | 4075 | *resp_len, 0, false, false); |
428a7158 DF |
4076 | |
4077 | ceph_osdc_start_request(osdc, req, false); | |
4078 | ret = ceph_osdc_wait_request(osdc, req); | |
4079 | if (ret >= 0) { | |
4080 | ret = req->r_ops[0].rval; | |
4081 | if (resp_page) | |
4082 | *resp_len = req->r_ops[0].outdata_len; | |
4083 | } | |
4084 | ||
4085 | out_put_req: | |
4086 | ceph_osdc_put_request(req); | |
4087 | return ret; | |
4088 | } | |
4089 | EXPORT_SYMBOL(ceph_osdc_call); | |
4090 | ||
f24e9980 SW |
4091 | /* |
4092 | * init, shutdown | |
4093 | */ | |
4094 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) | |
4095 | { | |
4096 | int err; | |
4097 | ||
4098 | dout("init\n"); | |
4099 | osdc->client = client; | |
5aea3dcd | 4100 | init_rwsem(&osdc->lock); |
f24e9980 | 4101 | osdc->osds = RB_ROOT; |
f5a2041b | 4102 | INIT_LIST_HEAD(&osdc->osd_lru); |
9dd2845c | 4103 | spin_lock_init(&osdc->osd_lru_lock); |
5aea3dcd ID |
4104 | osd_init(&osdc->homeless_osd); |
4105 | osdc->homeless_osd.o_osdc = osdc; | |
4106 | osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD; | |
264048af | 4107 | osdc->last_linger_id = CEPH_LINGER_ID_START; |
922dab61 | 4108 | osdc->linger_requests = RB_ROOT; |
4609245e ID |
4109 | osdc->map_checks = RB_ROOT; |
4110 | osdc->linger_map_checks = RB_ROOT; | |
f24e9980 | 4111 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); |
f5a2041b YS |
4112 | INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); |
4113 | ||
5f44f142 | 4114 | err = -ENOMEM; |
e5253a7b ID |
4115 | osdc->osdmap = ceph_osdmap_alloc(); |
4116 | if (!osdc->osdmap) | |
4117 | goto out; | |
4118 | ||
9e767adb ID |
4119 | osdc->req_mempool = mempool_create_slab_pool(10, |
4120 | ceph_osd_request_cache); | |
f24e9980 | 4121 | if (!osdc->req_mempool) |
e5253a7b | 4122 | goto out_map; |
f24e9980 | 4123 | |
d50b409f | 4124 | err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, |
711da55d | 4125 | PAGE_SIZE, 10, true, "osd_op"); |
f24e9980 | 4126 | if (err < 0) |
5f44f142 | 4127 | goto out_mempool; |
d50b409f | 4128 | err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, |
711da55d | 4129 | PAGE_SIZE, 10, true, "osd_op_reply"); |
c16e7869 SW |
4130 | if (err < 0) |
4131 | goto out_msgpool; | |
a40c4f10 | 4132 | |
dbcae088 | 4133 | err = -ENOMEM; |
a40c4f10 | 4134 | osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); |
dbcae088 | 4135 | if (!osdc->notify_wq) |
c172ec5c ID |
4136 | goto out_msgpool_reply; |
4137 | ||
fbca9635 ID |
4138 | schedule_delayed_work(&osdc->timeout_work, |
4139 | osdc->client->options->osd_keepalive_timeout); | |
b37ee1b9 ID |
4140 | schedule_delayed_work(&osdc->osds_timeout_work, |
4141 | round_jiffies_relative(osdc->client->options->osd_idle_ttl)); | |
4142 | ||
f24e9980 | 4143 | return 0; |
5f44f142 | 4144 | |
c172ec5c ID |
4145 | out_msgpool_reply: |
4146 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); | |
c16e7869 SW |
4147 | out_msgpool: |
4148 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
5f44f142 SW |
4149 | out_mempool: |
4150 | mempool_destroy(osdc->req_mempool); | |
e5253a7b ID |
4151 | out_map: |
4152 | ceph_osdmap_destroy(osdc->osdmap); | |
5f44f142 SW |
4153 | out: |
4154 | return err; | |
f24e9980 SW |
4155 | } |
4156 | ||
4157 | void ceph_osdc_stop(struct ceph_osd_client *osdc) | |
4158 | { | |
a40c4f10 YS |
4159 | flush_workqueue(osdc->notify_wq); |
4160 | destroy_workqueue(osdc->notify_wq); | |
f24e9980 | 4161 | cancel_delayed_work_sync(&osdc->timeout_work); |
f5a2041b | 4162 | cancel_delayed_work_sync(&osdc->osds_timeout_work); |
42a2c09f | 4163 | |
5aea3dcd | 4164 | down_write(&osdc->lock); |
42a2c09f ID |
4165 | while (!RB_EMPTY_ROOT(&osdc->osds)) { |
4166 | struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), | |
4167 | struct ceph_osd, o_node); | |
5aea3dcd | 4168 | close_osd(osd); |
42a2c09f | 4169 | } |
5aea3dcd | 4170 | up_write(&osdc->lock); |
02113a0f | 4171 | WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1); |
5aea3dcd ID |
4172 | osd_cleanup(&osdc->homeless_osd); |
4173 | ||
4174 | WARN_ON(!list_empty(&osdc->osd_lru)); | |
922dab61 | 4175 | WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests)); |
4609245e ID |
4176 | WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks)); |
4177 | WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks)); | |
5aea3dcd ID |
4178 | WARN_ON(atomic_read(&osdc->num_requests)); |
4179 | WARN_ON(atomic_read(&osdc->num_homeless)); | |
42a2c09f | 4180 | |
e5253a7b | 4181 | ceph_osdmap_destroy(osdc->osdmap); |
f24e9980 SW |
4182 | mempool_destroy(osdc->req_mempool); |
4183 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
c16e7869 | 4184 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); |
f24e9980 SW |
4185 | } |
4186 | ||
4187 | /* | |
4188 | * Read some contiguous pages. If we cross a stripe boundary, shorten | |
4189 | * *plen. Return number of bytes read, or error. | |
4190 | */ | |
4191 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, | |
4192 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
4193 | u64 off, u64 *plen, | |
4194 | u32 truncate_seq, u64 truncate_size, | |
b7495fc2 | 4195 | struct page **pages, int num_pages, int page_align) |
f24e9980 SW |
4196 | { |
4197 | struct ceph_osd_request *req; | |
4198 | int rc = 0; | |
4199 | ||
4200 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, | |
4201 | vino.snap, off, *plen); | |
715e4cd4 | 4202 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1, |
f24e9980 | 4203 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, |
acead002 | 4204 | NULL, truncate_seq, truncate_size, |
153e5167 | 4205 | false); |
6816282d SW |
4206 | if (IS_ERR(req)) |
4207 | return PTR_ERR(req); | |
f24e9980 SW |
4208 | |
4209 | /* it may be a short read due to an object boundary */ | |
406e2c9f | 4210 | osd_req_op_extent_osd_data_pages(req, 0, |
a4ce40a9 | 4211 | pages, *plen, page_align, false, false); |
f24e9980 | 4212 | |
e0c59487 | 4213 | dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n", |
43bfe5de | 4214 | off, *plen, *plen, page_align); |
f24e9980 SW |
4215 | |
4216 | rc = ceph_osdc_start_request(osdc, req, false); | |
4217 | if (!rc) | |
4218 | rc = ceph_osdc_wait_request(osdc, req); | |
4219 | ||
4220 | ceph_osdc_put_request(req); | |
4221 | dout("readpages result %d\n", rc); | |
4222 | return rc; | |
4223 | } | |
3d14c5d2 | 4224 | EXPORT_SYMBOL(ceph_osdc_readpages); |
f24e9980 SW |
4225 | |
4226 | /* | |
4227 | * do a synchronous write on N pages | |
4228 | */ | |
4229 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, | |
4230 | struct ceph_file_layout *layout, | |
4231 | struct ceph_snap_context *snapc, | |
4232 | u64 off, u64 len, | |
4233 | u32 truncate_seq, u64 truncate_size, | |
4234 | struct timespec *mtime, | |
24808826 | 4235 | struct page **pages, int num_pages) |
f24e9980 SW |
4236 | { |
4237 | struct ceph_osd_request *req; | |
4238 | int rc = 0; | |
b7495fc2 | 4239 | int page_align = off & ~PAGE_MASK; |
f24e9980 | 4240 | |
715e4cd4 | 4241 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1, |
54ea0046 | 4242 | CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, |
acead002 | 4243 | snapc, truncate_seq, truncate_size, |
153e5167 | 4244 | true); |
6816282d SW |
4245 | if (IS_ERR(req)) |
4246 | return PTR_ERR(req); | |
f24e9980 SW |
4247 | |
4248 | /* it may be a short write due to an object boundary */ | |
406e2c9f | 4249 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, |
43bfe5de AE |
4250 | false, false); |
4251 | dout("writepages %llu~%llu (%llu bytes)\n", off, len, len); | |
f24e9980 | 4252 | |
bb873b53 | 4253 | req->r_mtime = *mtime; |
87f979d3 | 4254 | rc = ceph_osdc_start_request(osdc, req, true); |
f24e9980 SW |
4255 | if (!rc) |
4256 | rc = ceph_osdc_wait_request(osdc, req); | |
4257 | ||
4258 | ceph_osdc_put_request(req); | |
4259 | if (rc == 0) | |
4260 | rc = len; | |
4261 | dout("writepages result %d\n", rc); | |
4262 | return rc; | |
4263 | } | |
3d14c5d2 | 4264 | EXPORT_SYMBOL(ceph_osdc_writepages); |
f24e9980 | 4265 | |
5522ae0b AE |
4266 | int ceph_osdc_setup(void) |
4267 | { | |
3f1af42a ID |
4268 | size_t size = sizeof(struct ceph_osd_request) + |
4269 | CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op); | |
4270 | ||
5522ae0b | 4271 | BUG_ON(ceph_osd_request_cache); |
3f1af42a ID |
4272 | ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size, |
4273 | 0, 0, NULL); | |
5522ae0b AE |
4274 | |
4275 | return ceph_osd_request_cache ? 0 : -ENOMEM; | |
4276 | } | |
4277 | EXPORT_SYMBOL(ceph_osdc_setup); | |
4278 | ||
4279 | void ceph_osdc_cleanup(void) | |
4280 | { | |
4281 | BUG_ON(!ceph_osd_request_cache); | |
4282 | kmem_cache_destroy(ceph_osd_request_cache); | |
4283 | ceph_osd_request_cache = NULL; | |
4284 | } | |
4285 | EXPORT_SYMBOL(ceph_osdc_cleanup); | |
4286 | ||
f24e9980 SW |
4287 | /* |
4288 | * handle incoming message | |
4289 | */ | |
4290 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
4291 | { | |
4292 | struct ceph_osd *osd = con->private; | |
5aea3dcd | 4293 | struct ceph_osd_client *osdc = osd->o_osdc; |
f24e9980 SW |
4294 | int type = le16_to_cpu(msg->hdr.type); |
4295 | ||
f24e9980 SW |
4296 | switch (type) { |
4297 | case CEPH_MSG_OSD_MAP: | |
4298 | ceph_osdc_handle_map(osdc, msg); | |
4299 | break; | |
4300 | case CEPH_MSG_OSD_OPREPLY: | |
5aea3dcd | 4301 | handle_reply(osd, msg); |
f24e9980 | 4302 | break; |
a40c4f10 YS |
4303 | case CEPH_MSG_WATCH_NOTIFY: |
4304 | handle_watch_notify(osdc, msg); | |
4305 | break; | |
f24e9980 SW |
4306 | |
4307 | default: | |
4308 | pr_err("received unknown message type %d %s\n", type, | |
4309 | ceph_msg_type_name(type)); | |
4310 | } | |
5aea3dcd | 4311 | |
f24e9980 SW |
4312 | ceph_msg_put(msg); |
4313 | } | |
4314 | ||
5b3a4db3 | 4315 | /* |
d15f9d69 ID |
4316 | * Lookup and return message for incoming reply. Don't try to do |
4317 | * anything about a larger than preallocated data portion of the | |
4318 | * message at the moment - for now, just skip the message. | |
5b3a4db3 SW |
4319 | */ |
4320 | static struct ceph_msg *get_reply(struct ceph_connection *con, | |
2450418c YS |
4321 | struct ceph_msg_header *hdr, |
4322 | int *skip) | |
f24e9980 SW |
4323 | { |
4324 | struct ceph_osd *osd = con->private; | |
4325 | struct ceph_osd_client *osdc = osd->o_osdc; | |
5aea3dcd | 4326 | struct ceph_msg *m = NULL; |
0547a9b3 | 4327 | struct ceph_osd_request *req; |
3f0a4ac5 | 4328 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 4329 | int data_len = le32_to_cpu(hdr->data_len); |
5aea3dcd | 4330 | u64 tid = le64_to_cpu(hdr->tid); |
f24e9980 | 4331 | |
5aea3dcd ID |
4332 | down_read(&osdc->lock); |
4333 | if (!osd_registered(osd)) { | |
4334 | dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd); | |
4335 | *skip = 1; | |
4336 | goto out_unlock_osdc; | |
4337 | } | |
4338 | WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num)); | |
4339 | ||
4340 | mutex_lock(&osd->lock); | |
4341 | req = lookup_request(&osd->o_requests, tid); | |
0547a9b3 | 4342 | if (!req) { |
cd8140c6 ID |
4343 | dout("%s osd%d tid %llu unknown, skipping\n", __func__, |
4344 | osd->o_osd, tid); | |
d15f9d69 | 4345 | *skip = 1; |
5aea3dcd | 4346 | goto out_unlock_session; |
0547a9b3 | 4347 | } |
c16e7869 | 4348 | |
ace6d3a9 | 4349 | ceph_msg_revoke_incoming(req->r_reply); |
0547a9b3 | 4350 | |
f2be82b0 | 4351 | if (front_len > req->r_reply->front_alloc_len) { |
d15f9d69 ID |
4352 | pr_warn("%s osd%d tid %llu front %d > preallocated %d\n", |
4353 | __func__, osd->o_osd, req->r_tid, front_len, | |
4354 | req->r_reply->front_alloc_len); | |
3f0a4ac5 ID |
4355 | m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS, |
4356 | false); | |
a79832f2 | 4357 | if (!m) |
5aea3dcd | 4358 | goto out_unlock_session; |
c16e7869 SW |
4359 | ceph_msg_put(req->r_reply); |
4360 | req->r_reply = m; | |
4361 | } | |
0fff87ec | 4362 | |
d15f9d69 ID |
4363 | if (data_len > req->r_reply->data_length) { |
4364 | pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n", | |
4365 | __func__, osd->o_osd, req->r_tid, data_len, | |
4366 | req->r_reply->data_length); | |
4367 | m = NULL; | |
4368 | *skip = 1; | |
5aea3dcd | 4369 | goto out_unlock_session; |
0547a9b3 | 4370 | } |
d15f9d69 ID |
4371 | |
4372 | m = ceph_msg_get(req->r_reply); | |
c16e7869 | 4373 | dout("get_reply tid %lld %p\n", tid, m); |
0547a9b3 | 4374 | |
5aea3dcd ID |
4375 | out_unlock_session: |
4376 | mutex_unlock(&osd->lock); | |
4377 | out_unlock_osdc: | |
4378 | up_read(&osdc->lock); | |
2450418c | 4379 | return m; |
5b3a4db3 SW |
4380 | } |
4381 | ||
19079203 ID |
4382 | /* |
4383 | * TODO: switch to a msg-owned pagelist | |
4384 | */ | |
4385 | static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr) | |
4386 | { | |
4387 | struct ceph_msg *m; | |
4388 | int type = le16_to_cpu(hdr->type); | |
4389 | u32 front_len = le32_to_cpu(hdr->front_len); | |
4390 | u32 data_len = le32_to_cpu(hdr->data_len); | |
4391 | ||
4392 | m = ceph_msg_new(type, front_len, GFP_NOIO, false); | |
4393 | if (!m) | |
4394 | return NULL; | |
4395 | ||
4396 | if (data_len) { | |
4397 | struct page **pages; | |
4398 | struct ceph_osd_data osd_data; | |
4399 | ||
4400 | pages = ceph_alloc_page_vector(calc_pages_for(0, data_len), | |
4401 | GFP_NOIO); | |
c22e853a | 4402 | if (IS_ERR(pages)) { |
19079203 ID |
4403 | ceph_msg_put(m); |
4404 | return NULL; | |
4405 | } | |
4406 | ||
4407 | ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false, | |
4408 | false); | |
4409 | ceph_osdc_msg_data_add(m, &osd_data); | |
4410 | } | |
4411 | ||
4412 | return m; | |
4413 | } | |
4414 | ||
5b3a4db3 SW |
4415 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, |
4416 | struct ceph_msg_header *hdr, | |
4417 | int *skip) | |
4418 | { | |
4419 | struct ceph_osd *osd = con->private; | |
4420 | int type = le16_to_cpu(hdr->type); | |
5b3a4db3 | 4421 | |
1c20f2d2 | 4422 | *skip = 0; |
5b3a4db3 SW |
4423 | switch (type) { |
4424 | case CEPH_MSG_OSD_MAP: | |
a40c4f10 | 4425 | case CEPH_MSG_WATCH_NOTIFY: |
19079203 | 4426 | return alloc_msg_with_page_vector(hdr); |
5b3a4db3 SW |
4427 | case CEPH_MSG_OSD_OPREPLY: |
4428 | return get_reply(con, hdr, skip); | |
4429 | default: | |
5aea3dcd ID |
4430 | pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__, |
4431 | osd->o_osd, type); | |
5b3a4db3 SW |
4432 | *skip = 1; |
4433 | return NULL; | |
4434 | } | |
f24e9980 SW |
4435 | } |
4436 | ||
4437 | /* | |
4438 | * Wrappers to refcount containing ceph_osd struct | |
4439 | */ | |
4440 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) | |
4441 | { | |
4442 | struct ceph_osd *osd = con->private; | |
4443 | if (get_osd(osd)) | |
4444 | return con; | |
4445 | return NULL; | |
4446 | } | |
4447 | ||
4448 | static void put_osd_con(struct ceph_connection *con) | |
4449 | { | |
4450 | struct ceph_osd *osd = con->private; | |
4451 | put_osd(osd); | |
4452 | } | |
4453 | ||
4e7a5dcd SW |
4454 | /* |
4455 | * authentication | |
4456 | */ | |
a3530df3 AE |
4457 | /* |
4458 | * Note: returned pointer is the address of a structure that's | |
4459 | * managed separately. Caller must *not* attempt to free it. | |
4460 | */ | |
4461 | static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, | |
8f43fb53 | 4462 | int *proto, int force_new) |
4e7a5dcd SW |
4463 | { |
4464 | struct ceph_osd *o = con->private; | |
4465 | struct ceph_osd_client *osdc = o->o_osdc; | |
4466 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
74f1869f | 4467 | struct ceph_auth_handshake *auth = &o->o_auth; |
4e7a5dcd | 4468 | |
74f1869f | 4469 | if (force_new && auth->authorizer) { |
6c1ea260 | 4470 | ceph_auth_destroy_authorizer(auth->authorizer); |
74f1869f AE |
4471 | auth->authorizer = NULL; |
4472 | } | |
27859f97 SW |
4473 | if (!auth->authorizer) { |
4474 | int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
4475 | auth); | |
4e7a5dcd | 4476 | if (ret) |
a3530df3 | 4477 | return ERR_PTR(ret); |
27859f97 SW |
4478 | } else { |
4479 | int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
0bed9b5c SW |
4480 | auth); |
4481 | if (ret) | |
4482 | return ERR_PTR(ret); | |
4e7a5dcd | 4483 | } |
4e7a5dcd | 4484 | *proto = ac->protocol; |
74f1869f | 4485 | |
a3530df3 | 4486 | return auth; |
4e7a5dcd SW |
4487 | } |
4488 | ||
4489 | ||
0dde5848 | 4490 | static int verify_authorizer_reply(struct ceph_connection *con) |
4e7a5dcd SW |
4491 | { |
4492 | struct ceph_osd *o = con->private; | |
4493 | struct ceph_osd_client *osdc = o->o_osdc; | |
4494 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
4495 | ||
0dde5848 | 4496 | return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer); |
4e7a5dcd SW |
4497 | } |
4498 | ||
9bd2e6f8 SW |
4499 | static int invalidate_authorizer(struct ceph_connection *con) |
4500 | { | |
4501 | struct ceph_osd *o = con->private; | |
4502 | struct ceph_osd_client *osdc = o->o_osdc; | |
4503 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
4504 | ||
27859f97 | 4505 | ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); |
9bd2e6f8 SW |
4506 | return ceph_monc_validate_auth(&osdc->client->monc); |
4507 | } | |
4e7a5dcd | 4508 | |
79dbd1ba | 4509 | static int osd_sign_message(struct ceph_msg *msg) |
33d07337 | 4510 | { |
79dbd1ba | 4511 | struct ceph_osd *o = msg->con->private; |
33d07337 | 4512 | struct ceph_auth_handshake *auth = &o->o_auth; |
79dbd1ba | 4513 | |
33d07337 YZ |
4514 | return ceph_auth_sign_message(auth, msg); |
4515 | } | |
4516 | ||
79dbd1ba | 4517 | static int osd_check_message_signature(struct ceph_msg *msg) |
33d07337 | 4518 | { |
79dbd1ba | 4519 | struct ceph_osd *o = msg->con->private; |
33d07337 | 4520 | struct ceph_auth_handshake *auth = &o->o_auth; |
79dbd1ba | 4521 | |
33d07337 YZ |
4522 | return ceph_auth_check_message_signature(auth, msg); |
4523 | } | |
4524 | ||
9e32789f | 4525 | static const struct ceph_connection_operations osd_con_ops = { |
f24e9980 SW |
4526 | .get = get_osd_con, |
4527 | .put = put_osd_con, | |
4528 | .dispatch = dispatch, | |
4e7a5dcd SW |
4529 | .get_authorizer = get_authorizer, |
4530 | .verify_authorizer_reply = verify_authorizer_reply, | |
9bd2e6f8 | 4531 | .invalidate_authorizer = invalidate_authorizer, |
f24e9980 | 4532 | .alloc_msg = alloc_msg, |
79dbd1ba ID |
4533 | .sign_message = osd_sign_message, |
4534 | .check_message_signature = osd_check_message_signature, | |
5aea3dcd | 4535 | .fault = osd_fault, |
f24e9980 | 4536 | }; |