]>
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 | |
8cb441c0 | 15 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
16 | #include <linux/ceph/libceph.h> |
17 | #include <linux/ceph/osd_client.h> | |
18 | #include <linux/ceph/messenger.h> | |
19 | #include <linux/ceph/decode.h> | |
20 | #include <linux/ceph/auth.h> | |
21 | #include <linux/ceph/pagelist.h> | |
f24e9980 | 22 | |
c16e7869 | 23 | #define OSD_OPREPLY_FRONT_LEN 512 |
0d59ab81 | 24 | |
5522ae0b AE |
25 | static struct kmem_cache *ceph_osd_request_cache; |
26 | ||
9e32789f | 27 | static const struct ceph_connection_operations osd_con_ops; |
f24e9980 | 28 | |
f24e9980 SW |
29 | /* |
30 | * Implement client access to distributed object storage cluster. | |
31 | * | |
32 | * All data objects are stored within a cluster/cloud of OSDs, or | |
33 | * "object storage devices." (Note that Ceph OSDs have _nothing_ to | |
34 | * do with the T10 OSD extensions to SCSI.) Ceph OSDs are simply | |
35 | * remote daemons serving up and coordinating consistent and safe | |
36 | * access to storage. | |
37 | * | |
38 | * Cluster membership and the mapping of data objects onto storage devices | |
39 | * are described by the osd map. | |
40 | * | |
41 | * We keep track of pending OSD requests (read, write), resubmit | |
42 | * requests to different OSDs when the cluster topology/data layout | |
43 | * change, or retry the affected requests when the communications | |
44 | * channel with an OSD is reset. | |
45 | */ | |
46 | ||
5aea3dcd ID |
47 | static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req); |
48 | static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req); | |
922dab61 ID |
49 | static void link_linger(struct ceph_osd *osd, |
50 | struct ceph_osd_linger_request *lreq); | |
51 | static void unlink_linger(struct ceph_osd *osd, | |
52 | struct ceph_osd_linger_request *lreq); | |
5aea3dcd ID |
53 | |
54 | #if 1 | |
55 | static inline bool rwsem_is_wrlocked(struct rw_semaphore *sem) | |
56 | { | |
57 | bool wrlocked = true; | |
58 | ||
59 | if (unlikely(down_read_trylock(sem))) { | |
60 | wrlocked = false; | |
61 | up_read(sem); | |
62 | } | |
63 | ||
64 | return wrlocked; | |
65 | } | |
66 | static inline void verify_osdc_locked(struct ceph_osd_client *osdc) | |
67 | { | |
68 | WARN_ON(!rwsem_is_locked(&osdc->lock)); | |
69 | } | |
70 | static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) | |
71 | { | |
72 | WARN_ON(!rwsem_is_wrlocked(&osdc->lock)); | |
73 | } | |
74 | static inline void verify_osd_locked(struct ceph_osd *osd) | |
75 | { | |
76 | struct ceph_osd_client *osdc = osd->o_osdc; | |
77 | ||
78 | WARN_ON(!(mutex_is_locked(&osd->lock) && | |
79 | rwsem_is_locked(&osdc->lock)) && | |
80 | !rwsem_is_wrlocked(&osdc->lock)); | |
81 | } | |
922dab61 ID |
82 | static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) |
83 | { | |
84 | WARN_ON(!mutex_is_locked(&lreq->lock)); | |
85 | } | |
5aea3dcd ID |
86 | #else |
87 | static inline void verify_osdc_locked(struct ceph_osd_client *osdc) { } | |
88 | static inline void verify_osdc_wrlocked(struct ceph_osd_client *osdc) { } | |
89 | static inline void verify_osd_locked(struct ceph_osd *osd) { } | |
922dab61 | 90 | static inline void verify_lreq_locked(struct ceph_osd_linger_request *lreq) { } |
5aea3dcd ID |
91 | #endif |
92 | ||
f24e9980 SW |
93 | /* |
94 | * calculate the mapping of a file extent onto an object, and fill out the | |
95 | * request accordingly. shorten extent as necessary if it crosses an | |
96 | * object boundary. | |
97 | * | |
98 | * fill osd op in request message. | |
99 | */ | |
dbe0fc41 | 100 | static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen, |
a19dadfb | 101 | u64 *objnum, u64 *objoff, u64 *objlen) |
f24e9980 | 102 | { |
60e56f13 | 103 | u64 orig_len = *plen; |
d63b77f4 | 104 | int r; |
f24e9980 | 105 | |
60e56f13 | 106 | /* object extent? */ |
75d1c941 AE |
107 | r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum, |
108 | objoff, objlen); | |
d63b77f4 SW |
109 | if (r < 0) |
110 | return r; | |
75d1c941 AE |
111 | if (*objlen < orig_len) { |
112 | *plen = *objlen; | |
60e56f13 AE |
113 | dout(" skipping last %llu, final file extent %llu~%llu\n", |
114 | orig_len - *plen, off, *plen); | |
115 | } | |
116 | ||
75d1c941 | 117 | dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen); |
f24e9980 | 118 | |
3ff5f385 | 119 | return 0; |
f24e9980 SW |
120 | } |
121 | ||
c54d47bf AE |
122 | static void ceph_osd_data_init(struct ceph_osd_data *osd_data) |
123 | { | |
124 | memset(osd_data, 0, sizeof (*osd_data)); | |
125 | osd_data->type = CEPH_OSD_DATA_TYPE_NONE; | |
126 | } | |
127 | ||
a4ce40a9 | 128 | static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
129 | struct page **pages, u64 length, u32 alignment, |
130 | bool pages_from_pool, bool own_pages) | |
131 | { | |
132 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGES; | |
133 | osd_data->pages = pages; | |
134 | osd_data->length = length; | |
135 | osd_data->alignment = alignment; | |
136 | osd_data->pages_from_pool = pages_from_pool; | |
137 | osd_data->own_pages = own_pages; | |
138 | } | |
43bfe5de | 139 | |
a4ce40a9 | 140 | static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
141 | struct ceph_pagelist *pagelist) |
142 | { | |
143 | osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST; | |
144 | osd_data->pagelist = pagelist; | |
145 | } | |
43bfe5de AE |
146 | |
147 | #ifdef CONFIG_BLOCK | |
a4ce40a9 | 148 | static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data, |
43bfe5de AE |
149 | struct bio *bio, size_t bio_length) |
150 | { | |
151 | osd_data->type = CEPH_OSD_DATA_TYPE_BIO; | |
152 | osd_data->bio = bio; | |
153 | osd_data->bio_length = bio_length; | |
154 | } | |
43bfe5de AE |
155 | #endif /* CONFIG_BLOCK */ |
156 | ||
8a703a38 IC |
157 | #define osd_req_op_data(oreq, whch, typ, fld) \ |
158 | ({ \ | |
159 | struct ceph_osd_request *__oreq = (oreq); \ | |
160 | unsigned int __whch = (whch); \ | |
161 | BUG_ON(__whch >= __oreq->r_num_ops); \ | |
162 | &__oreq->r_ops[__whch].typ.fld; \ | |
163 | }) | |
863c7eb5 | 164 | |
49719778 AE |
165 | static struct ceph_osd_data * |
166 | osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which) | |
167 | { | |
168 | BUG_ON(which >= osd_req->r_num_ops); | |
169 | ||
170 | return &osd_req->r_ops[which].raw_data_in; | |
171 | } | |
172 | ||
a4ce40a9 AE |
173 | struct ceph_osd_data * |
174 | osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req, | |
406e2c9f | 175 | unsigned int which) |
a4ce40a9 | 176 | { |
863c7eb5 | 177 | return osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 AE |
178 | } |
179 | EXPORT_SYMBOL(osd_req_op_extent_osd_data); | |
180 | ||
49719778 AE |
181 | void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req, |
182 | unsigned int which, struct page **pages, | |
183 | u64 length, u32 alignment, | |
184 | bool pages_from_pool, bool own_pages) | |
185 | { | |
186 | struct ceph_osd_data *osd_data; | |
187 | ||
188 | osd_data = osd_req_op_raw_data_in(osd_req, which); | |
189 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, | |
190 | pages_from_pool, own_pages); | |
191 | } | |
192 | EXPORT_SYMBOL(osd_req_op_raw_data_in_pages); | |
193 | ||
a4ce40a9 | 194 | void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req, |
406e2c9f AE |
195 | unsigned int which, struct page **pages, |
196 | u64 length, u32 alignment, | |
a4ce40a9 AE |
197 | bool pages_from_pool, bool own_pages) |
198 | { | |
199 | struct ceph_osd_data *osd_data; | |
200 | ||
863c7eb5 | 201 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 AE |
202 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, |
203 | pages_from_pool, own_pages); | |
a4ce40a9 AE |
204 | } |
205 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages); | |
206 | ||
207 | void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req, | |
406e2c9f | 208 | unsigned int which, struct ceph_pagelist *pagelist) |
a4ce40a9 AE |
209 | { |
210 | struct ceph_osd_data *osd_data; | |
211 | ||
863c7eb5 | 212 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); |
a4ce40a9 | 213 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
a4ce40a9 AE |
214 | } |
215 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist); | |
216 | ||
217 | #ifdef CONFIG_BLOCK | |
218 | void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req, | |
406e2c9f | 219 | unsigned int which, struct bio *bio, size_t bio_length) |
a4ce40a9 AE |
220 | { |
221 | struct ceph_osd_data *osd_data; | |
863c7eb5 AE |
222 | |
223 | osd_data = osd_req_op_data(osd_req, which, extent, osd_data); | |
a4ce40a9 | 224 | ceph_osd_data_bio_init(osd_data, bio, bio_length); |
a4ce40a9 AE |
225 | } |
226 | EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio); | |
227 | #endif /* CONFIG_BLOCK */ | |
228 | ||
229 | static void osd_req_op_cls_request_info_pagelist( | |
230 | struct ceph_osd_request *osd_req, | |
231 | unsigned int which, struct ceph_pagelist *pagelist) | |
232 | { | |
233 | struct ceph_osd_data *osd_data; | |
234 | ||
863c7eb5 | 235 | osd_data = osd_req_op_data(osd_req, which, cls, request_info); |
a4ce40a9 | 236 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
a4ce40a9 AE |
237 | } |
238 | ||
04017e29 AE |
239 | void osd_req_op_cls_request_data_pagelist( |
240 | struct ceph_osd_request *osd_req, | |
241 | unsigned int which, struct ceph_pagelist *pagelist) | |
242 | { | |
243 | struct ceph_osd_data *osd_data; | |
244 | ||
863c7eb5 | 245 | osd_data = osd_req_op_data(osd_req, which, cls, request_data); |
04017e29 | 246 | ceph_osd_data_pagelist_init(osd_data, pagelist); |
bb873b53 ID |
247 | osd_req->r_ops[which].cls.indata_len += pagelist->length; |
248 | osd_req->r_ops[which].indata_len += pagelist->length; | |
04017e29 AE |
249 | } |
250 | EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist); | |
251 | ||
6c57b554 AE |
252 | void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req, |
253 | unsigned int which, struct page **pages, u64 length, | |
254 | u32 alignment, bool pages_from_pool, bool own_pages) | |
255 | { | |
256 | struct ceph_osd_data *osd_data; | |
257 | ||
258 | osd_data = osd_req_op_data(osd_req, which, cls, request_data); | |
259 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, | |
260 | pages_from_pool, own_pages); | |
bb873b53 ID |
261 | osd_req->r_ops[which].cls.indata_len += length; |
262 | osd_req->r_ops[which].indata_len += length; | |
6c57b554 AE |
263 | } |
264 | EXPORT_SYMBOL(osd_req_op_cls_request_data_pages); | |
265 | ||
a4ce40a9 AE |
266 | void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req, |
267 | unsigned int which, struct page **pages, u64 length, | |
268 | u32 alignment, bool pages_from_pool, bool own_pages) | |
269 | { | |
270 | struct ceph_osd_data *osd_data; | |
271 | ||
863c7eb5 | 272 | osd_data = osd_req_op_data(osd_req, which, cls, response_data); |
a4ce40a9 AE |
273 | ceph_osd_data_pages_init(osd_data, pages, length, alignment, |
274 | pages_from_pool, own_pages); | |
a4ce40a9 AE |
275 | } |
276 | EXPORT_SYMBOL(osd_req_op_cls_response_data_pages); | |
277 | ||
23c08a9c AE |
278 | static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data) |
279 | { | |
280 | switch (osd_data->type) { | |
281 | case CEPH_OSD_DATA_TYPE_NONE: | |
282 | return 0; | |
283 | case CEPH_OSD_DATA_TYPE_PAGES: | |
284 | return osd_data->length; | |
285 | case CEPH_OSD_DATA_TYPE_PAGELIST: | |
286 | return (u64)osd_data->pagelist->length; | |
287 | #ifdef CONFIG_BLOCK | |
288 | case CEPH_OSD_DATA_TYPE_BIO: | |
289 | return (u64)osd_data->bio_length; | |
290 | #endif /* CONFIG_BLOCK */ | |
291 | default: | |
292 | WARN(true, "unrecognized data type %d\n", (int)osd_data->type); | |
293 | return 0; | |
294 | } | |
295 | } | |
296 | ||
c54d47bf AE |
297 | static void ceph_osd_data_release(struct ceph_osd_data *osd_data) |
298 | { | |
5476492f | 299 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) { |
c54d47bf AE |
300 | int num_pages; |
301 | ||
302 | num_pages = calc_pages_for((u64)osd_data->alignment, | |
303 | (u64)osd_data->length); | |
304 | ceph_release_page_vector(osd_data->pages, num_pages); | |
305 | } | |
5476492f AE |
306 | ceph_osd_data_init(osd_data); |
307 | } | |
308 | ||
309 | static void osd_req_op_data_release(struct ceph_osd_request *osd_req, | |
310 | unsigned int which) | |
311 | { | |
312 | struct ceph_osd_req_op *op; | |
313 | ||
314 | BUG_ON(which >= osd_req->r_num_ops); | |
315 | op = &osd_req->r_ops[which]; | |
316 | ||
317 | switch (op->op) { | |
318 | case CEPH_OSD_OP_READ: | |
319 | case CEPH_OSD_OP_WRITE: | |
e30b7577 | 320 | case CEPH_OSD_OP_WRITEFULL: |
5476492f AE |
321 | ceph_osd_data_release(&op->extent.osd_data); |
322 | break; | |
323 | case CEPH_OSD_OP_CALL: | |
324 | ceph_osd_data_release(&op->cls.request_info); | |
04017e29 | 325 | ceph_osd_data_release(&op->cls.request_data); |
5476492f AE |
326 | ceph_osd_data_release(&op->cls.response_data); |
327 | break; | |
d74b50be YZ |
328 | case CEPH_OSD_OP_SETXATTR: |
329 | case CEPH_OSD_OP_CMPXATTR: | |
330 | ceph_osd_data_release(&op->xattr.osd_data); | |
331 | break; | |
66ba609f YZ |
332 | case CEPH_OSD_OP_STAT: |
333 | ceph_osd_data_release(&op->raw_data_in); | |
334 | break; | |
922dab61 ID |
335 | case CEPH_OSD_OP_NOTIFY_ACK: |
336 | ceph_osd_data_release(&op->notify_ack.request_data); | |
337 | break; | |
19079203 ID |
338 | case CEPH_OSD_OP_NOTIFY: |
339 | ceph_osd_data_release(&op->notify.request_data); | |
340 | ceph_osd_data_release(&op->notify.response_data); | |
341 | break; | |
a4ed38d7 DF |
342 | case CEPH_OSD_OP_LIST_WATCHERS: |
343 | ceph_osd_data_release(&op->list_watchers.response_data); | |
344 | break; | |
5476492f AE |
345 | default: |
346 | break; | |
347 | } | |
c54d47bf AE |
348 | } |
349 | ||
63244fa1 ID |
350 | /* |
351 | * Assumes @t is zero-initialized. | |
352 | */ | |
353 | static void target_init(struct ceph_osd_request_target *t) | |
354 | { | |
355 | ceph_oid_init(&t->base_oid); | |
356 | ceph_oloc_init(&t->base_oloc); | |
357 | ceph_oid_init(&t->target_oid); | |
358 | ceph_oloc_init(&t->target_oloc); | |
359 | ||
360 | ceph_osds_init(&t->acting); | |
361 | ceph_osds_init(&t->up); | |
362 | t->size = -1; | |
363 | t->min_size = -1; | |
364 | ||
365 | t->osd = CEPH_HOMELESS_OSD; | |
366 | } | |
367 | ||
922dab61 ID |
368 | static void target_copy(struct ceph_osd_request_target *dest, |
369 | const struct ceph_osd_request_target *src) | |
370 | { | |
371 | ceph_oid_copy(&dest->base_oid, &src->base_oid); | |
372 | ceph_oloc_copy(&dest->base_oloc, &src->base_oloc); | |
373 | ceph_oid_copy(&dest->target_oid, &src->target_oid); | |
374 | ceph_oloc_copy(&dest->target_oloc, &src->target_oloc); | |
375 | ||
376 | dest->pgid = src->pgid; /* struct */ | |
dc98ff72 | 377 | dest->spgid = src->spgid; /* struct */ |
922dab61 ID |
378 | dest->pg_num = src->pg_num; |
379 | dest->pg_num_mask = src->pg_num_mask; | |
380 | ceph_osds_copy(&dest->acting, &src->acting); | |
381 | ceph_osds_copy(&dest->up, &src->up); | |
382 | dest->size = src->size; | |
383 | dest->min_size = src->min_size; | |
384 | dest->sort_bitwise = src->sort_bitwise; | |
385 | ||
386 | dest->flags = src->flags; | |
387 | dest->paused = src->paused; | |
388 | ||
04c7d789 | 389 | dest->epoch = src->epoch; |
dc93e0e2 ID |
390 | dest->last_force_resend = src->last_force_resend; |
391 | ||
922dab61 ID |
392 | dest->osd = src->osd; |
393 | } | |
394 | ||
63244fa1 ID |
395 | static void target_destroy(struct ceph_osd_request_target *t) |
396 | { | |
397 | ceph_oid_destroy(&t->base_oid); | |
30c156d9 | 398 | ceph_oloc_destroy(&t->base_oloc); |
63244fa1 | 399 | ceph_oid_destroy(&t->target_oid); |
30c156d9 | 400 | ceph_oloc_destroy(&t->target_oloc); |
63244fa1 ID |
401 | } |
402 | ||
f24e9980 SW |
403 | /* |
404 | * requests | |
405 | */ | |
3540bfdb ID |
406 | static void request_release_checks(struct ceph_osd_request *req) |
407 | { | |
408 | WARN_ON(!RB_EMPTY_NODE(&req->r_node)); | |
4609245e | 409 | WARN_ON(!RB_EMPTY_NODE(&req->r_mc_node)); |
3540bfdb ID |
410 | WARN_ON(!list_empty(&req->r_unsafe_item)); |
411 | WARN_ON(req->r_osd); | |
412 | } | |
413 | ||
9e94af20 | 414 | static void ceph_osdc_release_request(struct kref *kref) |
f24e9980 | 415 | { |
9e94af20 ID |
416 | struct ceph_osd_request *req = container_of(kref, |
417 | struct ceph_osd_request, r_kref); | |
5476492f | 418 | unsigned int which; |
415e49a9 | 419 | |
9e94af20 ID |
420 | dout("%s %p (r_request %p r_reply %p)\n", __func__, req, |
421 | req->r_request, req->r_reply); | |
3540bfdb | 422 | request_release_checks(req); |
9e94af20 | 423 | |
415e49a9 SW |
424 | if (req->r_request) |
425 | ceph_msg_put(req->r_request); | |
5aea3dcd | 426 | if (req->r_reply) |
ab8cb34a | 427 | ceph_msg_put(req->r_reply); |
0fff87ec | 428 | |
5476492f AE |
429 | for (which = 0; which < req->r_num_ops; which++) |
430 | osd_req_op_data_release(req, which); | |
0fff87ec | 431 | |
a66dd383 | 432 | target_destroy(&req->r_t); |
415e49a9 | 433 | ceph_put_snap_context(req->r_snapc); |
d30291b9 | 434 | |
415e49a9 SW |
435 | if (req->r_mempool) |
436 | mempool_free(req, req->r_osdc->req_mempool); | |
3f1af42a | 437 | else if (req->r_num_ops <= CEPH_OSD_SLAB_OPS) |
5522ae0b | 438 | kmem_cache_free(ceph_osd_request_cache, req); |
3f1af42a ID |
439 | else |
440 | kfree(req); | |
f24e9980 | 441 | } |
9e94af20 ID |
442 | |
443 | void ceph_osdc_get_request(struct ceph_osd_request *req) | |
444 | { | |
445 | dout("%s %p (was %d)\n", __func__, req, | |
2c935bc5 | 446 | kref_read(&req->r_kref)); |
9e94af20 ID |
447 | kref_get(&req->r_kref); |
448 | } | |
449 | EXPORT_SYMBOL(ceph_osdc_get_request); | |
450 | ||
451 | void ceph_osdc_put_request(struct ceph_osd_request *req) | |
452 | { | |
3ed97d63 ID |
453 | if (req) { |
454 | dout("%s %p (was %d)\n", __func__, req, | |
2c935bc5 | 455 | kref_read(&req->r_kref)); |
3ed97d63 ID |
456 | kref_put(&req->r_kref, ceph_osdc_release_request); |
457 | } | |
9e94af20 ID |
458 | } |
459 | EXPORT_SYMBOL(ceph_osdc_put_request); | |
68b4476b | 460 | |
3540bfdb ID |
461 | static void request_init(struct ceph_osd_request *req) |
462 | { | |
463 | /* req only, each op is zeroed in _osd_req_op_init() */ | |
464 | memset(req, 0, sizeof(*req)); | |
465 | ||
466 | kref_init(&req->r_kref); | |
467 | init_completion(&req->r_completion); | |
3540bfdb | 468 | RB_CLEAR_NODE(&req->r_node); |
4609245e | 469 | RB_CLEAR_NODE(&req->r_mc_node); |
3540bfdb ID |
470 | INIT_LIST_HEAD(&req->r_unsafe_item); |
471 | ||
472 | target_init(&req->r_t); | |
473 | } | |
474 | ||
922dab61 ID |
475 | /* |
476 | * This is ugly, but it allows us to reuse linger registration and ping | |
477 | * requests, keeping the structure of the code around send_linger{_ping}() | |
478 | * reasonable. Setting up a min_nr=2 mempool for each linger request | |
479 | * and dealing with copying ops (this blasts req only, watch op remains | |
480 | * intact) isn't any better. | |
481 | */ | |
482 | static void request_reinit(struct ceph_osd_request *req) | |
483 | { | |
484 | struct ceph_osd_client *osdc = req->r_osdc; | |
485 | bool mempool = req->r_mempool; | |
486 | unsigned int num_ops = req->r_num_ops; | |
487 | u64 snapid = req->r_snapid; | |
488 | struct ceph_snap_context *snapc = req->r_snapc; | |
489 | bool linger = req->r_linger; | |
490 | struct ceph_msg *request_msg = req->r_request; | |
491 | struct ceph_msg *reply_msg = req->r_reply; | |
492 | ||
493 | dout("%s req %p\n", __func__, req); | |
2c935bc5 | 494 | WARN_ON(kref_read(&req->r_kref) != 1); |
922dab61 ID |
495 | request_release_checks(req); |
496 | ||
2c935bc5 PZ |
497 | WARN_ON(kref_read(&request_msg->kref) != 1); |
498 | WARN_ON(kref_read(&reply_msg->kref) != 1); | |
922dab61 ID |
499 | target_destroy(&req->r_t); |
500 | ||
501 | request_init(req); | |
502 | req->r_osdc = osdc; | |
503 | req->r_mempool = mempool; | |
504 | req->r_num_ops = num_ops; | |
505 | req->r_snapid = snapid; | |
506 | req->r_snapc = snapc; | |
507 | req->r_linger = linger; | |
508 | req->r_request = request_msg; | |
509 | req->r_reply = reply_msg; | |
510 | } | |
511 | ||
3499e8a5 | 512 | struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc, |
f24e9980 | 513 | struct ceph_snap_context *snapc, |
1b83bef2 | 514 | unsigned int num_ops, |
3499e8a5 | 515 | bool use_mempool, |
54a54007 | 516 | gfp_t gfp_flags) |
f24e9980 SW |
517 | { |
518 | struct ceph_osd_request *req; | |
1b83bef2 | 519 | |
f24e9980 | 520 | if (use_mempool) { |
3f1af42a | 521 | BUG_ON(num_ops > CEPH_OSD_SLAB_OPS); |
3499e8a5 | 522 | req = mempool_alloc(osdc->req_mempool, gfp_flags); |
3f1af42a ID |
523 | } else if (num_ops <= CEPH_OSD_SLAB_OPS) { |
524 | req = kmem_cache_alloc(ceph_osd_request_cache, gfp_flags); | |
f24e9980 | 525 | } else { |
3f1af42a ID |
526 | BUG_ON(num_ops > CEPH_OSD_MAX_OPS); |
527 | req = kmalloc(sizeof(*req) + num_ops * sizeof(req->r_ops[0]), | |
528 | gfp_flags); | |
f24e9980 | 529 | } |
3f1af42a | 530 | if (unlikely(!req)) |
a79832f2 | 531 | return NULL; |
f24e9980 | 532 | |
3540bfdb | 533 | request_init(req); |
f24e9980 SW |
534 | req->r_osdc = osdc; |
535 | req->r_mempool = use_mempool; | |
79528734 | 536 | req->r_num_ops = num_ops; |
84127282 ID |
537 | req->r_snapid = CEPH_NOSNAP; |
538 | req->r_snapc = ceph_get_snap_context(snapc); | |
68b4476b | 539 | |
13d1ad16 ID |
540 | dout("%s req %p\n", __func__, req); |
541 | return req; | |
542 | } | |
543 | EXPORT_SYMBOL(ceph_osdc_alloc_request); | |
3f1af42a | 544 | |
2e59ffd1 | 545 | static int ceph_oloc_encoding_size(const struct ceph_object_locator *oloc) |
30c156d9 YZ |
546 | { |
547 | return 8 + 4 + 4 + 4 + (oloc->pool_ns ? oloc->pool_ns->len : 0); | |
548 | } | |
549 | ||
13d1ad16 ID |
550 | int ceph_osdc_alloc_messages(struct ceph_osd_request *req, gfp_t gfp) |
551 | { | |
552 | struct ceph_osd_client *osdc = req->r_osdc; | |
553 | struct ceph_msg *msg; | |
554 | int msg_size; | |
c16e7869 | 555 | |
d30291b9 | 556 | WARN_ON(ceph_oid_empty(&req->r_base_oid)); |
30c156d9 | 557 | WARN_ON(ceph_oloc_empty(&req->r_base_oloc)); |
d30291b9 | 558 | |
13d1ad16 | 559 | /* create request message */ |
8cb441c0 ID |
560 | msg_size = CEPH_ENCODING_START_BLK_LEN + |
561 | CEPH_PGID_ENCODING_LEN + 1; /* spgid */ | |
562 | msg_size += 4 + 4 + 4; /* hash, osdmap_epoch, flags */ | |
563 | msg_size += CEPH_ENCODING_START_BLK_LEN + | |
564 | sizeof(struct ceph_osd_reqid); /* reqid */ | |
565 | msg_size += sizeof(struct ceph_blkin_trace_info); /* trace */ | |
566 | msg_size += 4 + sizeof(struct ceph_timespec); /* client_inc, mtime */ | |
30c156d9 YZ |
567 | msg_size += CEPH_ENCODING_START_BLK_LEN + |
568 | ceph_oloc_encoding_size(&req->r_base_oloc); /* oloc */ | |
13d1ad16 ID |
569 | msg_size += 4 + req->r_base_oid.name_len; /* oid */ |
570 | msg_size += 2 + req->r_num_ops * sizeof(struct ceph_osd_op); | |
ae458f5a ID |
571 | msg_size += 8; /* snapid */ |
572 | msg_size += 8; /* snap_seq */ | |
13d1ad16 | 573 | msg_size += 4 + 8 * (req->r_snapc ? req->r_snapc->num_snaps : 0); |
8cb441c0 | 574 | msg_size += 4 + 8; /* retry_attempt, features */ |
ae458f5a | 575 | |
13d1ad16 | 576 | if (req->r_mempool) |
8f3bc053 | 577 | msg = ceph_msgpool_get(&osdc->msgpool_op, 0); |
f24e9980 | 578 | else |
13d1ad16 ID |
579 | msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp, true); |
580 | if (!msg) | |
581 | return -ENOMEM; | |
68b4476b | 582 | |
f24e9980 | 583 | memset(msg->front.iov_base, 0, msg->front.iov_len); |
3499e8a5 | 584 | req->r_request = msg; |
3499e8a5 | 585 | |
13d1ad16 ID |
586 | /* create reply message */ |
587 | msg_size = OSD_OPREPLY_FRONT_LEN; | |
711da55d ID |
588 | msg_size += req->r_base_oid.name_len; |
589 | msg_size += req->r_num_ops * sizeof(struct ceph_osd_op); | |
13d1ad16 ID |
590 | |
591 | if (req->r_mempool) | |
592 | msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0); | |
593 | else | |
594 | msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, msg_size, gfp, true); | |
595 | if (!msg) | |
596 | return -ENOMEM; | |
597 | ||
598 | req->r_reply = msg; | |
599 | ||
600 | return 0; | |
3499e8a5 | 601 | } |
13d1ad16 | 602 | EXPORT_SYMBOL(ceph_osdc_alloc_messages); |
3499e8a5 | 603 | |
a8dd0a37 | 604 | static bool osd_req_opcode_valid(u16 opcode) |
68b4476b | 605 | { |
a8dd0a37 | 606 | switch (opcode) { |
70b5bfa3 ID |
607 | #define GENERATE_CASE(op, opcode, str) case CEPH_OSD_OP_##op: return true; |
608 | __CEPH_FORALL_OSD_OPS(GENERATE_CASE) | |
609 | #undef GENERATE_CASE | |
a8dd0a37 AE |
610 | default: |
611 | return false; | |
612 | } | |
613 | } | |
614 | ||
33803f33 AE |
615 | /* |
616 | * This is an osd op init function for opcodes that have no data or | |
617 | * other information associated with them. It also serves as a | |
618 | * common init routine for all the other init functions, below. | |
619 | */ | |
c99d2d4a | 620 | static struct ceph_osd_req_op * |
49719778 | 621 | _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which, |
144cba14 | 622 | u16 opcode, u32 flags) |
33803f33 | 623 | { |
c99d2d4a AE |
624 | struct ceph_osd_req_op *op; |
625 | ||
626 | BUG_ON(which >= osd_req->r_num_ops); | |
33803f33 AE |
627 | BUG_ON(!osd_req_opcode_valid(opcode)); |
628 | ||
c99d2d4a | 629 | op = &osd_req->r_ops[which]; |
33803f33 | 630 | memset(op, 0, sizeof (*op)); |
33803f33 | 631 | op->op = opcode; |
144cba14 | 632 | op->flags = flags; |
c99d2d4a AE |
633 | |
634 | return op; | |
33803f33 AE |
635 | } |
636 | ||
49719778 | 637 | void osd_req_op_init(struct ceph_osd_request *osd_req, |
144cba14 | 638 | unsigned int which, u16 opcode, u32 flags) |
49719778 | 639 | { |
144cba14 | 640 | (void)_osd_req_op_init(osd_req, which, opcode, flags); |
49719778 AE |
641 | } |
642 | EXPORT_SYMBOL(osd_req_op_init); | |
643 | ||
c99d2d4a AE |
644 | void osd_req_op_extent_init(struct ceph_osd_request *osd_req, |
645 | unsigned int which, u16 opcode, | |
33803f33 AE |
646 | u64 offset, u64 length, |
647 | u64 truncate_size, u32 truncate_seq) | |
648 | { | |
144cba14 YZ |
649 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
650 | opcode, 0); | |
33803f33 AE |
651 | size_t payload_len = 0; |
652 | ||
ad7a60de | 653 | BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && |
e30b7577 ID |
654 | opcode != CEPH_OSD_OP_WRITEFULL && opcode != CEPH_OSD_OP_ZERO && |
655 | opcode != CEPH_OSD_OP_TRUNCATE); | |
33803f33 | 656 | |
33803f33 AE |
657 | op->extent.offset = offset; |
658 | op->extent.length = length; | |
659 | op->extent.truncate_size = truncate_size; | |
660 | op->extent.truncate_seq = truncate_seq; | |
e30b7577 | 661 | if (opcode == CEPH_OSD_OP_WRITE || opcode == CEPH_OSD_OP_WRITEFULL) |
33803f33 AE |
662 | payload_len += length; |
663 | ||
de2aa102 | 664 | op->indata_len = payload_len; |
33803f33 AE |
665 | } |
666 | EXPORT_SYMBOL(osd_req_op_extent_init); | |
667 | ||
c99d2d4a AE |
668 | void osd_req_op_extent_update(struct ceph_osd_request *osd_req, |
669 | unsigned int which, u64 length) | |
e5975c7c | 670 | { |
c99d2d4a AE |
671 | struct ceph_osd_req_op *op; |
672 | u64 previous; | |
673 | ||
674 | BUG_ON(which >= osd_req->r_num_ops); | |
675 | op = &osd_req->r_ops[which]; | |
676 | previous = op->extent.length; | |
e5975c7c AE |
677 | |
678 | if (length == previous) | |
679 | return; /* Nothing to do */ | |
680 | BUG_ON(length > previous); | |
681 | ||
682 | op->extent.length = length; | |
d641df81 YZ |
683 | if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) |
684 | op->indata_len -= previous - length; | |
e5975c7c AE |
685 | } |
686 | EXPORT_SYMBOL(osd_req_op_extent_update); | |
687 | ||
2c63f49a YZ |
688 | void osd_req_op_extent_dup_last(struct ceph_osd_request *osd_req, |
689 | unsigned int which, u64 offset_inc) | |
690 | { | |
691 | struct ceph_osd_req_op *op, *prev_op; | |
692 | ||
693 | BUG_ON(which + 1 >= osd_req->r_num_ops); | |
694 | ||
695 | prev_op = &osd_req->r_ops[which]; | |
696 | op = _osd_req_op_init(osd_req, which + 1, prev_op->op, prev_op->flags); | |
697 | /* dup previous one */ | |
698 | op->indata_len = prev_op->indata_len; | |
699 | op->outdata_len = prev_op->outdata_len; | |
700 | op->extent = prev_op->extent; | |
701 | /* adjust offset */ | |
702 | op->extent.offset += offset_inc; | |
703 | op->extent.length -= offset_inc; | |
704 | ||
705 | if (op->op == CEPH_OSD_OP_WRITE || op->op == CEPH_OSD_OP_WRITEFULL) | |
706 | op->indata_len -= offset_inc; | |
707 | } | |
708 | EXPORT_SYMBOL(osd_req_op_extent_dup_last); | |
709 | ||
c99d2d4a | 710 | void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which, |
04017e29 | 711 | u16 opcode, const char *class, const char *method) |
33803f33 | 712 | { |
144cba14 YZ |
713 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
714 | opcode, 0); | |
5f562df5 | 715 | struct ceph_pagelist *pagelist; |
33803f33 AE |
716 | size_t payload_len = 0; |
717 | size_t size; | |
718 | ||
719 | BUG_ON(opcode != CEPH_OSD_OP_CALL); | |
720 | ||
5f562df5 AE |
721 | pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS); |
722 | BUG_ON(!pagelist); | |
723 | ceph_pagelist_init(pagelist); | |
724 | ||
33803f33 AE |
725 | op->cls.class_name = class; |
726 | size = strlen(class); | |
727 | BUG_ON(size > (size_t) U8_MAX); | |
728 | op->cls.class_len = size; | |
5f562df5 | 729 | ceph_pagelist_append(pagelist, class, size); |
33803f33 AE |
730 | payload_len += size; |
731 | ||
732 | op->cls.method_name = method; | |
733 | size = strlen(method); | |
734 | BUG_ON(size > (size_t) U8_MAX); | |
735 | op->cls.method_len = size; | |
5f562df5 | 736 | ceph_pagelist_append(pagelist, method, size); |
33803f33 AE |
737 | payload_len += size; |
738 | ||
a4ce40a9 | 739 | osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist); |
5f562df5 | 740 | |
de2aa102 | 741 | op->indata_len = payload_len; |
33803f33 AE |
742 | } |
743 | EXPORT_SYMBOL(osd_req_op_cls_init); | |
8c042b0d | 744 | |
d74b50be YZ |
745 | int osd_req_op_xattr_init(struct ceph_osd_request *osd_req, unsigned int which, |
746 | u16 opcode, const char *name, const void *value, | |
747 | size_t size, u8 cmp_op, u8 cmp_mode) | |
748 | { | |
144cba14 YZ |
749 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, |
750 | opcode, 0); | |
d74b50be YZ |
751 | struct ceph_pagelist *pagelist; |
752 | size_t payload_len; | |
753 | ||
754 | BUG_ON(opcode != CEPH_OSD_OP_SETXATTR && opcode != CEPH_OSD_OP_CMPXATTR); | |
755 | ||
756 | pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS); | |
757 | if (!pagelist) | |
758 | return -ENOMEM; | |
759 | ||
760 | ceph_pagelist_init(pagelist); | |
761 | ||
762 | payload_len = strlen(name); | |
763 | op->xattr.name_len = payload_len; | |
764 | ceph_pagelist_append(pagelist, name, payload_len); | |
765 | ||
766 | op->xattr.value_len = size; | |
767 | ceph_pagelist_append(pagelist, value, size); | |
768 | payload_len += size; | |
769 | ||
770 | op->xattr.cmp_op = cmp_op; | |
771 | op->xattr.cmp_mode = cmp_mode; | |
772 | ||
773 | ceph_osd_data_pagelist_init(&op->xattr.osd_data, pagelist); | |
de2aa102 | 774 | op->indata_len = payload_len; |
d74b50be YZ |
775 | return 0; |
776 | } | |
777 | EXPORT_SYMBOL(osd_req_op_xattr_init); | |
778 | ||
922dab61 ID |
779 | /* |
780 | * @watch_opcode: CEPH_OSD_WATCH_OP_* | |
781 | */ | |
782 | static void osd_req_op_watch_init(struct ceph_osd_request *req, int which, | |
783 | u64 cookie, u8 watch_opcode) | |
33803f33 | 784 | { |
922dab61 | 785 | struct ceph_osd_req_op *op; |
33803f33 | 786 | |
922dab61 | 787 | op = _osd_req_op_init(req, which, CEPH_OSD_OP_WATCH, 0); |
33803f33 | 788 | op->watch.cookie = cookie; |
922dab61 ID |
789 | op->watch.op = watch_opcode; |
790 | op->watch.gen = 0; | |
33803f33 | 791 | } |
33803f33 | 792 | |
c647b8a8 ID |
793 | void osd_req_op_alloc_hint_init(struct ceph_osd_request *osd_req, |
794 | unsigned int which, | |
795 | u64 expected_object_size, | |
796 | u64 expected_write_size) | |
797 | { | |
798 | struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, | |
144cba14 YZ |
799 | CEPH_OSD_OP_SETALLOCHINT, |
800 | 0); | |
c647b8a8 ID |
801 | |
802 | op->alloc_hint.expected_object_size = expected_object_size; | |
803 | op->alloc_hint.expected_write_size = expected_write_size; | |
804 | ||
805 | /* | |
806 | * CEPH_OSD_OP_SETALLOCHINT op is advisory and therefore deemed | |
807 | * not worth a feature bit. Set FAILOK per-op flag to make | |
808 | * sure older osds don't trip over an unsupported opcode. | |
809 | */ | |
810 | op->flags |= CEPH_OSD_OP_FLAG_FAILOK; | |
811 | } | |
812 | EXPORT_SYMBOL(osd_req_op_alloc_hint_init); | |
813 | ||
90af3602 | 814 | static void ceph_osdc_msg_data_add(struct ceph_msg *msg, |
ec9123c5 AE |
815 | struct ceph_osd_data *osd_data) |
816 | { | |
817 | u64 length = ceph_osd_data_length(osd_data); | |
818 | ||
819 | if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) { | |
820 | BUG_ON(length > (u64) SIZE_MAX); | |
821 | if (length) | |
90af3602 | 822 | ceph_msg_data_add_pages(msg, osd_data->pages, |
ec9123c5 AE |
823 | length, osd_data->alignment); |
824 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) { | |
825 | BUG_ON(!length); | |
90af3602 | 826 | ceph_msg_data_add_pagelist(msg, osd_data->pagelist); |
ec9123c5 AE |
827 | #ifdef CONFIG_BLOCK |
828 | } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) { | |
90af3602 | 829 | ceph_msg_data_add_bio(msg, osd_data->bio, length); |
ec9123c5 AE |
830 | #endif |
831 | } else { | |
832 | BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE); | |
833 | } | |
834 | } | |
835 | ||
bb873b53 ID |
836 | static u32 osd_req_encode_op(struct ceph_osd_op *dst, |
837 | const struct ceph_osd_req_op *src) | |
a8dd0a37 | 838 | { |
a8dd0a37 AE |
839 | if (WARN_ON(!osd_req_opcode_valid(src->op))) { |
840 | pr_err("unrecognized osd opcode %d\n", src->op); | |
841 | ||
842 | return 0; | |
843 | } | |
844 | ||
845 | switch (src->op) { | |
846 | case CEPH_OSD_OP_STAT: | |
847 | break; | |
848 | case CEPH_OSD_OP_READ: | |
849 | case CEPH_OSD_OP_WRITE: | |
e30b7577 | 850 | case CEPH_OSD_OP_WRITEFULL: |
ad7a60de | 851 | case CEPH_OSD_OP_ZERO: |
ad7a60de | 852 | case CEPH_OSD_OP_TRUNCATE: |
a8dd0a37 AE |
853 | dst->extent.offset = cpu_to_le64(src->extent.offset); |
854 | dst->extent.length = cpu_to_le64(src->extent.length); | |
855 | dst->extent.truncate_size = | |
856 | cpu_to_le64(src->extent.truncate_size); | |
857 | dst->extent.truncate_seq = | |
858 | cpu_to_le32(src->extent.truncate_seq); | |
859 | break; | |
860 | case CEPH_OSD_OP_CALL: | |
a8dd0a37 AE |
861 | dst->cls.class_len = src->cls.class_len; |
862 | dst->cls.method_len = src->cls.method_len; | |
bb873b53 | 863 | dst->cls.indata_len = cpu_to_le32(src->cls.indata_len); |
a8dd0a37 AE |
864 | break; |
865 | case CEPH_OSD_OP_STARTSYNC: | |
866 | break; | |
a8dd0a37 AE |
867 | case CEPH_OSD_OP_WATCH: |
868 | dst->watch.cookie = cpu_to_le64(src->watch.cookie); | |
922dab61 ID |
869 | dst->watch.ver = cpu_to_le64(0); |
870 | dst->watch.op = src->watch.op; | |
871 | dst->watch.gen = cpu_to_le32(src->watch.gen); | |
872 | break; | |
873 | case CEPH_OSD_OP_NOTIFY_ACK: | |
a8dd0a37 | 874 | break; |
19079203 ID |
875 | case CEPH_OSD_OP_NOTIFY: |
876 | dst->notify.cookie = cpu_to_le64(src->notify.cookie); | |
877 | break; | |
a4ed38d7 DF |
878 | case CEPH_OSD_OP_LIST_WATCHERS: |
879 | break; | |
c647b8a8 ID |
880 | case CEPH_OSD_OP_SETALLOCHINT: |
881 | dst->alloc_hint.expected_object_size = | |
882 | cpu_to_le64(src->alloc_hint.expected_object_size); | |
883 | dst->alloc_hint.expected_write_size = | |
884 | cpu_to_le64(src->alloc_hint.expected_write_size); | |
885 | break; | |
d74b50be YZ |
886 | case CEPH_OSD_OP_SETXATTR: |
887 | case CEPH_OSD_OP_CMPXATTR: | |
888 | dst->xattr.name_len = cpu_to_le32(src->xattr.name_len); | |
889 | dst->xattr.value_len = cpu_to_le32(src->xattr.value_len); | |
890 | dst->xattr.cmp_op = src->xattr.cmp_op; | |
891 | dst->xattr.cmp_mode = src->xattr.cmp_mode; | |
d74b50be | 892 | break; |
864e9197 YZ |
893 | case CEPH_OSD_OP_CREATE: |
894 | case CEPH_OSD_OP_DELETE: | |
895 | break; | |
a8dd0a37 | 896 | default: |
4c46459c | 897 | pr_err("unsupported osd opcode %s\n", |
8f63ca2d | 898 | ceph_osd_op_name(src->op)); |
4c46459c | 899 | WARN_ON(1); |
a8dd0a37 AE |
900 | |
901 | return 0; | |
68b4476b | 902 | } |
7b25bf5f | 903 | |
a8dd0a37 | 904 | dst->op = cpu_to_le16(src->op); |
7b25bf5f | 905 | dst->flags = cpu_to_le32(src->flags); |
de2aa102 | 906 | dst->payload_len = cpu_to_le32(src->indata_len); |
175face2 | 907 | |
bb873b53 | 908 | return src->indata_len; |
68b4476b YS |
909 | } |
910 | ||
3499e8a5 YS |
911 | /* |
912 | * build new request AND message, calculate layout, and adjust file | |
913 | * extent as needed. | |
914 | * | |
915 | * if the file was recently truncated, we include information about its | |
916 | * old and new size so that the object can be updated appropriately. (we | |
917 | * avoid synchronously deleting truncated objects because it's slow.) | |
918 | * | |
919 | * if @do_sync, include a 'startsync' command so that the osd will flush | |
920 | * data quickly. | |
921 | */ | |
922 | struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc, | |
923 | struct ceph_file_layout *layout, | |
924 | struct ceph_vino vino, | |
715e4cd4 YZ |
925 | u64 off, u64 *plen, |
926 | unsigned int which, int num_ops, | |
3499e8a5 YS |
927 | int opcode, int flags, |
928 | struct ceph_snap_context *snapc, | |
3499e8a5 YS |
929 | u32 truncate_seq, |
930 | u64 truncate_size, | |
153e5167 | 931 | bool use_mempool) |
3499e8a5 | 932 | { |
68b4476b | 933 | struct ceph_osd_request *req; |
75d1c941 AE |
934 | u64 objnum = 0; |
935 | u64 objoff = 0; | |
936 | u64 objlen = 0; | |
6816282d | 937 | int r; |
68b4476b | 938 | |
ad7a60de | 939 | BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE && |
864e9197 YZ |
940 | opcode != CEPH_OSD_OP_ZERO && opcode != CEPH_OSD_OP_TRUNCATE && |
941 | opcode != CEPH_OSD_OP_CREATE && opcode != CEPH_OSD_OP_DELETE); | |
68b4476b | 942 | |
acead002 | 943 | req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool, |
ae7ca4a3 | 944 | GFP_NOFS); |
13d1ad16 ID |
945 | if (!req) { |
946 | r = -ENOMEM; | |
947 | goto fail; | |
948 | } | |
79528734 | 949 | |
3499e8a5 | 950 | /* calculate max write size */ |
a19dadfb | 951 | r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen); |
13d1ad16 ID |
952 | if (r) |
953 | goto fail; | |
a19dadfb | 954 | |
864e9197 | 955 | if (opcode == CEPH_OSD_OP_CREATE || opcode == CEPH_OSD_OP_DELETE) { |
144cba14 | 956 | osd_req_op_init(req, which, opcode, 0); |
864e9197 | 957 | } else { |
7627151e | 958 | u32 object_size = layout->object_size; |
864e9197 YZ |
959 | u32 object_base = off - objoff; |
960 | if (!(truncate_seq == 1 && truncate_size == -1ULL)) { | |
961 | if (truncate_size <= object_base) { | |
962 | truncate_size = 0; | |
963 | } else { | |
964 | truncate_size -= object_base; | |
965 | if (truncate_size > object_size) | |
966 | truncate_size = object_size; | |
967 | } | |
ccca4e37 | 968 | } |
715e4cd4 | 969 | osd_req_op_extent_init(req, which, opcode, objoff, objlen, |
864e9197 YZ |
970 | truncate_size, truncate_seq); |
971 | } | |
d18d1e28 | 972 | |
a1f4020a | 973 | req->r_abort_on_full = true; |
bb873b53 | 974 | req->r_flags = flags; |
7627151e | 975 | req->r_base_oloc.pool = layout->pool_id; |
30c156d9 | 976 | req->r_base_oloc.pool_ns = ceph_try_get_string(layout->pool_ns); |
d30291b9 | 977 | ceph_oid_printf(&req->r_base_oid, "%llx.%08llx", vino.ino, objnum); |
dbe0fc41 | 978 | |
bb873b53 ID |
979 | req->r_snapid = vino.snap; |
980 | if (flags & CEPH_OSD_FLAG_WRITE) | |
981 | req->r_data_offset = off; | |
982 | ||
13d1ad16 ID |
983 | r = ceph_osdc_alloc_messages(req, GFP_NOFS); |
984 | if (r) | |
985 | goto fail; | |
986 | ||
f24e9980 | 987 | return req; |
13d1ad16 ID |
988 | |
989 | fail: | |
990 | ceph_osdc_put_request(req); | |
991 | return ERR_PTR(r); | |
f24e9980 | 992 | } |
3d14c5d2 | 993 | EXPORT_SYMBOL(ceph_osdc_new_request); |
f24e9980 SW |
994 | |
995 | /* | |
996 | * We keep osd requests in an rbtree, sorted by ->r_tid. | |
997 | */ | |
fcd00b68 | 998 | DEFINE_RB_FUNCS(request, struct ceph_osd_request, r_tid, r_node) |
4609245e | 999 | DEFINE_RB_FUNCS(request_mc, struct ceph_osd_request, r_tid, r_mc_node) |
f24e9980 | 1000 | |
0247a0cf ID |
1001 | static bool osd_homeless(struct ceph_osd *osd) |
1002 | { | |
1003 | return osd->o_osd == CEPH_HOMELESS_OSD; | |
1004 | } | |
1005 | ||
5aea3dcd | 1006 | static bool osd_registered(struct ceph_osd *osd) |
f24e9980 | 1007 | { |
5aea3dcd | 1008 | verify_osdc_locked(osd->o_osdc); |
f24e9980 | 1009 | |
5aea3dcd | 1010 | return !RB_EMPTY_NODE(&osd->o_node); |
f24e9980 SW |
1011 | } |
1012 | ||
0247a0cf ID |
1013 | /* |
1014 | * Assumes @osd is zero-initialized. | |
1015 | */ | |
1016 | static void osd_init(struct ceph_osd *osd) | |
1017 | { | |
02113a0f | 1018 | refcount_set(&osd->o_ref, 1); |
0247a0cf | 1019 | RB_CLEAR_NODE(&osd->o_node); |
5aea3dcd | 1020 | osd->o_requests = RB_ROOT; |
922dab61 | 1021 | osd->o_linger_requests = RB_ROOT; |
0247a0cf ID |
1022 | INIT_LIST_HEAD(&osd->o_osd_lru); |
1023 | INIT_LIST_HEAD(&osd->o_keepalive_item); | |
1024 | osd->o_incarnation = 1; | |
5aea3dcd | 1025 | mutex_init(&osd->lock); |
0247a0cf ID |
1026 | } |
1027 | ||
1028 | static void osd_cleanup(struct ceph_osd *osd) | |
1029 | { | |
1030 | WARN_ON(!RB_EMPTY_NODE(&osd->o_node)); | |
5aea3dcd | 1031 | WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); |
922dab61 | 1032 | WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); |
0247a0cf ID |
1033 | WARN_ON(!list_empty(&osd->o_osd_lru)); |
1034 | WARN_ON(!list_empty(&osd->o_keepalive_item)); | |
1035 | ||
1036 | if (osd->o_auth.authorizer) { | |
1037 | WARN_ON(osd_homeless(osd)); | |
1038 | ceph_auth_destroy_authorizer(osd->o_auth.authorizer); | |
1039 | } | |
1040 | } | |
1041 | ||
f24e9980 SW |
1042 | /* |
1043 | * Track open sessions with osds. | |
1044 | */ | |
e10006f8 | 1045 | static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) |
f24e9980 SW |
1046 | { |
1047 | struct ceph_osd *osd; | |
1048 | ||
0247a0cf ID |
1049 | WARN_ON(onum == CEPH_HOMELESS_OSD); |
1050 | ||
7a28f59b | 1051 | osd = kzalloc(sizeof(*osd), GFP_NOIO | __GFP_NOFAIL); |
0247a0cf | 1052 | osd_init(osd); |
f24e9980 | 1053 | osd->o_osdc = osdc; |
e10006f8 | 1054 | osd->o_osd = onum; |
f24e9980 | 1055 | |
b7a9e5dd | 1056 | ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); |
4e7a5dcd | 1057 | |
f24e9980 SW |
1058 | return osd; |
1059 | } | |
1060 | ||
1061 | static struct ceph_osd *get_osd(struct ceph_osd *osd) | |
1062 | { | |
02113a0f ER |
1063 | if (refcount_inc_not_zero(&osd->o_ref)) { |
1064 | dout("get_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref)-1, | |
1065 | refcount_read(&osd->o_ref)); | |
f24e9980 SW |
1066 | return osd; |
1067 | } else { | |
1068 | dout("get_osd %p FAIL\n", osd); | |
1069 | return NULL; | |
1070 | } | |
1071 | } | |
1072 | ||
1073 | static void put_osd(struct ceph_osd *osd) | |
1074 | { | |
02113a0f ER |
1075 | dout("put_osd %p %d -> %d\n", osd, refcount_read(&osd->o_ref), |
1076 | refcount_read(&osd->o_ref) - 1); | |
1077 | if (refcount_dec_and_test(&osd->o_ref)) { | |
0247a0cf | 1078 | osd_cleanup(osd); |
f24e9980 | 1079 | kfree(osd); |
79494d1b | 1080 | } |
f24e9980 SW |
1081 | } |
1082 | ||
fcd00b68 ID |
1083 | DEFINE_RB_FUNCS(osd, struct ceph_osd, o_osd, o_node) |
1084 | ||
9dd2845c | 1085 | static void __move_osd_to_lru(struct ceph_osd *osd) |
f5a2041b | 1086 | { |
9dd2845c ID |
1087 | struct ceph_osd_client *osdc = osd->o_osdc; |
1088 | ||
1089 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
f5a2041b | 1090 | BUG_ON(!list_empty(&osd->o_osd_lru)); |
bbf37ec3 | 1091 | |
9dd2845c | 1092 | spin_lock(&osdc->osd_lru_lock); |
f5a2041b | 1093 | list_add_tail(&osd->o_osd_lru, &osdc->osd_lru); |
9dd2845c ID |
1094 | spin_unlock(&osdc->osd_lru_lock); |
1095 | ||
a319bf56 | 1096 | osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl; |
f5a2041b YS |
1097 | } |
1098 | ||
9dd2845c | 1099 | static void maybe_move_osd_to_lru(struct ceph_osd *osd) |
bbf37ec3 | 1100 | { |
5aea3dcd | 1101 | if (RB_EMPTY_ROOT(&osd->o_requests) && |
922dab61 | 1102 | RB_EMPTY_ROOT(&osd->o_linger_requests)) |
9dd2845c | 1103 | __move_osd_to_lru(osd); |
bbf37ec3 ID |
1104 | } |
1105 | ||
f5a2041b YS |
1106 | static void __remove_osd_from_lru(struct ceph_osd *osd) |
1107 | { | |
9dd2845c ID |
1108 | struct ceph_osd_client *osdc = osd->o_osdc; |
1109 | ||
1110 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
1111 | ||
1112 | spin_lock(&osdc->osd_lru_lock); | |
f5a2041b YS |
1113 | if (!list_empty(&osd->o_osd_lru)) |
1114 | list_del_init(&osd->o_osd_lru); | |
9dd2845c | 1115 | spin_unlock(&osdc->osd_lru_lock); |
f5a2041b YS |
1116 | } |
1117 | ||
5aea3dcd ID |
1118 | /* |
1119 | * Close the connection and assign any leftover requests to the | |
1120 | * homeless session. | |
1121 | */ | |
1122 | static void close_osd(struct ceph_osd *osd) | |
1123 | { | |
1124 | struct ceph_osd_client *osdc = osd->o_osdc; | |
1125 | struct rb_node *n; | |
1126 | ||
1127 | verify_osdc_wrlocked(osdc); | |
1128 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
1129 | ||
1130 | ceph_con_close(&osd->o_con); | |
1131 | ||
1132 | for (n = rb_first(&osd->o_requests); n; ) { | |
1133 | struct ceph_osd_request *req = | |
1134 | rb_entry(n, struct ceph_osd_request, r_node); | |
1135 | ||
1136 | n = rb_next(n); /* unlink_request() */ | |
1137 | ||
1138 | dout(" reassigning req %p tid %llu\n", req, req->r_tid); | |
1139 | unlink_request(osd, req); | |
1140 | link_request(&osdc->homeless_osd, req); | |
1141 | } | |
922dab61 ID |
1142 | for (n = rb_first(&osd->o_linger_requests); n; ) { |
1143 | struct ceph_osd_linger_request *lreq = | |
1144 | rb_entry(n, struct ceph_osd_linger_request, node); | |
1145 | ||
1146 | n = rb_next(n); /* unlink_linger() */ | |
1147 | ||
1148 | dout(" reassigning lreq %p linger_id %llu\n", lreq, | |
1149 | lreq->linger_id); | |
1150 | unlink_linger(osd, lreq); | |
1151 | link_linger(&osdc->homeless_osd, lreq); | |
1152 | } | |
5aea3dcd ID |
1153 | |
1154 | __remove_osd_from_lru(osd); | |
1155 | erase_osd(&osdc->osds, osd); | |
1156 | put_osd(osd); | |
1157 | } | |
1158 | ||
f24e9980 SW |
1159 | /* |
1160 | * reset osd connect | |
1161 | */ | |
5aea3dcd | 1162 | static int reopen_osd(struct ceph_osd *osd) |
f24e9980 | 1163 | { |
c3acb181 | 1164 | struct ceph_entity_addr *peer_addr; |
f24e9980 | 1165 | |
5aea3dcd ID |
1166 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); |
1167 | ||
1168 | if (RB_EMPTY_ROOT(&osd->o_requests) && | |
922dab61 | 1169 | RB_EMPTY_ROOT(&osd->o_linger_requests)) { |
5aea3dcd | 1170 | close_osd(osd); |
c3acb181 AE |
1171 | return -ENODEV; |
1172 | } | |
1173 | ||
5aea3dcd | 1174 | peer_addr = &osd->o_osdc->osdmap->osd_addr[osd->o_osd]; |
c3acb181 AE |
1175 | if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) && |
1176 | !ceph_con_opened(&osd->o_con)) { | |
5aea3dcd | 1177 | struct rb_node *n; |
c3acb181 | 1178 | |
0b4af2e8 ID |
1179 | dout("osd addr hasn't changed and connection never opened, " |
1180 | "letting msgr retry\n"); | |
87b315a5 | 1181 | /* touch each r_stamp for handle_timeout()'s benfit */ |
5aea3dcd ID |
1182 | for (n = rb_first(&osd->o_requests); n; n = rb_next(n)) { |
1183 | struct ceph_osd_request *req = | |
1184 | rb_entry(n, struct ceph_osd_request, r_node); | |
87b315a5 | 1185 | req->r_stamp = jiffies; |
5aea3dcd | 1186 | } |
c3acb181 AE |
1187 | |
1188 | return -EAGAIN; | |
f24e9980 | 1189 | } |
c3acb181 AE |
1190 | |
1191 | ceph_con_close(&osd->o_con); | |
1192 | ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr); | |
1193 | osd->o_incarnation++; | |
1194 | ||
1195 | return 0; | |
f24e9980 SW |
1196 | } |
1197 | ||
5aea3dcd ID |
1198 | static struct ceph_osd *lookup_create_osd(struct ceph_osd_client *osdc, int o, |
1199 | bool wrlocked) | |
f24e9980 | 1200 | { |
5aea3dcd | 1201 | struct ceph_osd *osd; |
35f9f8a0 | 1202 | |
5aea3dcd ID |
1203 | if (wrlocked) |
1204 | verify_osdc_wrlocked(osdc); | |
1205 | else | |
1206 | verify_osdc_locked(osdc); | |
f24e9980 | 1207 | |
5aea3dcd ID |
1208 | if (o != CEPH_HOMELESS_OSD) |
1209 | osd = lookup_osd(&osdc->osds, o); | |
1210 | else | |
1211 | osd = &osdc->homeless_osd; | |
1212 | if (!osd) { | |
1213 | if (!wrlocked) | |
1214 | return ERR_PTR(-EAGAIN); | |
0ba6478d | 1215 | |
5aea3dcd ID |
1216 | osd = create_osd(osdc, o); |
1217 | insert_osd(&osdc->osds, osd); | |
1218 | ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, | |
1219 | &osdc->osdmap->osd_addr[osd->o_osd]); | |
0ba6478d | 1220 | } |
f24e9980 | 1221 | |
5aea3dcd ID |
1222 | dout("%s osdc %p osd%d -> osd %p\n", __func__, osdc, o, osd); |
1223 | return osd; | |
f24e9980 SW |
1224 | } |
1225 | ||
1226 | /* | |
5aea3dcd ID |
1227 | * Create request <-> OSD session relation. |
1228 | * | |
1229 | * @req has to be assigned a tid, @osd may be homeless. | |
f24e9980 | 1230 | */ |
5aea3dcd | 1231 | static void link_request(struct ceph_osd *osd, struct ceph_osd_request *req) |
f24e9980 | 1232 | { |
5aea3dcd ID |
1233 | verify_osd_locked(osd); |
1234 | WARN_ON(!req->r_tid || req->r_osd); | |
1235 | dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, | |
1236 | req, req->r_tid); | |
1237 | ||
1238 | if (!osd_homeless(osd)) | |
1239 | __remove_osd_from_lru(osd); | |
1240 | else | |
1241 | atomic_inc(&osd->o_osdc->num_homeless); | |
1242 | ||
1243 | get_osd(osd); | |
1244 | insert_request(&osd->o_requests, req); | |
1245 | req->r_osd = osd; | |
f24e9980 SW |
1246 | } |
1247 | ||
5aea3dcd ID |
1248 | static void unlink_request(struct ceph_osd *osd, struct ceph_osd_request *req) |
1249 | { | |
1250 | verify_osd_locked(osd); | |
1251 | WARN_ON(req->r_osd != osd); | |
1252 | dout("%s osd %p osd%d req %p tid %llu\n", __func__, osd, osd->o_osd, | |
1253 | req, req->r_tid); | |
1254 | ||
1255 | req->r_osd = NULL; | |
1256 | erase_request(&osd->o_requests, req); | |
1257 | put_osd(osd); | |
1258 | ||
1259 | if (!osd_homeless(osd)) | |
1260 | maybe_move_osd_to_lru(osd); | |
1261 | else | |
1262 | atomic_dec(&osd->o_osdc->num_homeless); | |
1263 | } | |
1264 | ||
63244fa1 ID |
1265 | static bool __pool_full(struct ceph_pg_pool_info *pi) |
1266 | { | |
1267 | return pi->flags & CEPH_POOL_FLAG_FULL; | |
1268 | } | |
1269 | ||
42c1b124 ID |
1270 | static bool have_pool_full(struct ceph_osd_client *osdc) |
1271 | { | |
1272 | struct rb_node *n; | |
1273 | ||
1274 | for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { | |
1275 | struct ceph_pg_pool_info *pi = | |
1276 | rb_entry(n, struct ceph_pg_pool_info, node); | |
1277 | ||
1278 | if (__pool_full(pi)) | |
1279 | return true; | |
1280 | } | |
1281 | ||
1282 | return false; | |
1283 | } | |
1284 | ||
5aea3dcd ID |
1285 | static bool pool_full(struct ceph_osd_client *osdc, s64 pool_id) |
1286 | { | |
1287 | struct ceph_pg_pool_info *pi; | |
1288 | ||
1289 | pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); | |
1290 | if (!pi) | |
1291 | return false; | |
1292 | ||
1293 | return __pool_full(pi); | |
1294 | } | |
1295 | ||
d29adb34 JD |
1296 | /* |
1297 | * Returns whether a request should be blocked from being sent | |
1298 | * based on the current osdmap and osd_client settings. | |
d29adb34 | 1299 | */ |
63244fa1 ID |
1300 | static bool target_should_be_paused(struct ceph_osd_client *osdc, |
1301 | const struct ceph_osd_request_target *t, | |
1302 | struct ceph_pg_pool_info *pi) | |
1303 | { | |
b7ec35b3 ID |
1304 | bool pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); |
1305 | bool pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || | |
1306 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || | |
63244fa1 ID |
1307 | __pool_full(pi); |
1308 | ||
6d637a54 | 1309 | WARN_ON(pi->id != t->target_oloc.pool); |
58eb7932 JL |
1310 | return ((t->flags & CEPH_OSD_FLAG_READ) && pauserd) || |
1311 | ((t->flags & CEPH_OSD_FLAG_WRITE) && pausewr) || | |
1312 | (osdc->osdmap->epoch < osdc->epoch_barrier); | |
63244fa1 ID |
1313 | } |
1314 | ||
63244fa1 ID |
1315 | enum calc_target_result { |
1316 | CALC_TARGET_NO_ACTION = 0, | |
1317 | CALC_TARGET_NEED_RESEND, | |
1318 | CALC_TARGET_POOL_DNE, | |
1319 | }; | |
1320 | ||
1321 | static enum calc_target_result calc_target(struct ceph_osd_client *osdc, | |
1322 | struct ceph_osd_request_target *t, | |
7de030d6 | 1323 | struct ceph_connection *con, |
63244fa1 ID |
1324 | bool any_change) |
1325 | { | |
1326 | struct ceph_pg_pool_info *pi; | |
1327 | struct ceph_pg pgid, last_pgid; | |
1328 | struct ceph_osds up, acting; | |
1329 | bool force_resend = false; | |
84ed45df ID |
1330 | bool unpaused = false; |
1331 | bool legacy_change; | |
7de030d6 | 1332 | bool split = false; |
b7ec35b3 | 1333 | bool sort_bitwise = ceph_osdmap_flag(osdc, CEPH_OSDMAP_SORTBITWISE); |
63244fa1 ID |
1334 | enum calc_target_result ct_res; |
1335 | int ret; | |
1336 | ||
04c7d789 | 1337 | t->epoch = osdc->osdmap->epoch; |
63244fa1 ID |
1338 | pi = ceph_pg_pool_by_id(osdc->osdmap, t->base_oloc.pool); |
1339 | if (!pi) { | |
1340 | t->osd = CEPH_HOMELESS_OSD; | |
1341 | ct_res = CALC_TARGET_POOL_DNE; | |
1342 | goto out; | |
1343 | } | |
1344 | ||
1345 | if (osdc->osdmap->epoch == pi->last_force_request_resend) { | |
dc93e0e2 ID |
1346 | if (t->last_force_resend < pi->last_force_request_resend) { |
1347 | t->last_force_resend = pi->last_force_request_resend; | |
63244fa1 | 1348 | force_resend = true; |
dc93e0e2 | 1349 | } else if (t->last_force_resend == 0) { |
63244fa1 ID |
1350 | force_resend = true; |
1351 | } | |
1352 | } | |
63244fa1 | 1353 | |
db098ec4 ID |
1354 | /* apply tiering */ |
1355 | ceph_oid_copy(&t->target_oid, &t->base_oid); | |
1356 | ceph_oloc_copy(&t->target_oloc, &t->base_oloc); | |
1357 | if ((t->flags & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0) { | |
63244fa1 ID |
1358 | if (t->flags & CEPH_OSD_FLAG_READ && pi->read_tier >= 0) |
1359 | t->target_oloc.pool = pi->read_tier; | |
1360 | if (t->flags & CEPH_OSD_FLAG_WRITE && pi->write_tier >= 0) | |
1361 | t->target_oloc.pool = pi->write_tier; | |
6d637a54 ID |
1362 | |
1363 | pi = ceph_pg_pool_by_id(osdc->osdmap, t->target_oloc.pool); | |
1364 | if (!pi) { | |
1365 | t->osd = CEPH_HOMELESS_OSD; | |
1366 | ct_res = CALC_TARGET_POOL_DNE; | |
1367 | goto out; | |
1368 | } | |
63244fa1 ID |
1369 | } |
1370 | ||
1371 | ret = ceph_object_locator_to_pg(osdc->osdmap, &t->target_oid, | |
1372 | &t->target_oloc, &pgid); | |
1373 | if (ret) { | |
1374 | WARN_ON(ret != -ENOENT); | |
1375 | t->osd = CEPH_HOMELESS_OSD; | |
1376 | ct_res = CALC_TARGET_POOL_DNE; | |
1377 | goto out; | |
1378 | } | |
1379 | last_pgid.pool = pgid.pool; | |
1380 | last_pgid.seed = ceph_stable_mod(pgid.seed, t->pg_num, t->pg_num_mask); | |
1381 | ||
1382 | ceph_pg_to_up_acting_osds(osdc->osdmap, &pgid, &up, &acting); | |
1383 | if (any_change && | |
1384 | ceph_is_new_interval(&t->acting, | |
1385 | &acting, | |
1386 | &t->up, | |
1387 | &up, | |
1388 | t->size, | |
1389 | pi->size, | |
1390 | t->min_size, | |
1391 | pi->min_size, | |
1392 | t->pg_num, | |
1393 | pi->pg_num, | |
1394 | t->sort_bitwise, | |
1395 | sort_bitwise, | |
1396 | &last_pgid)) | |
1397 | force_resend = true; | |
1398 | ||
1399 | if (t->paused && !target_should_be_paused(osdc, t, pi)) { | |
1400 | t->paused = false; | |
84ed45df | 1401 | unpaused = true; |
63244fa1 | 1402 | } |
84ed45df ID |
1403 | legacy_change = ceph_pg_compare(&t->pgid, &pgid) || |
1404 | ceph_osds_changed(&t->acting, &acting, any_change); | |
7de030d6 ID |
1405 | if (t->pg_num) |
1406 | split = ceph_pg_is_split(&last_pgid, t->pg_num, pi->pg_num); | |
63244fa1 | 1407 | |
7de030d6 | 1408 | if (legacy_change || force_resend || split) { |
63244fa1 | 1409 | t->pgid = pgid; /* struct */ |
dc98ff72 | 1410 | ceph_pg_to_primary_shard(osdc->osdmap, &pgid, &t->spgid); |
63244fa1 ID |
1411 | ceph_osds_copy(&t->acting, &acting); |
1412 | ceph_osds_copy(&t->up, &up); | |
1413 | t->size = pi->size; | |
1414 | t->min_size = pi->min_size; | |
1415 | t->pg_num = pi->pg_num; | |
1416 | t->pg_num_mask = pi->pg_num_mask; | |
1417 | t->sort_bitwise = sort_bitwise; | |
1418 | ||
1419 | t->osd = acting.primary; | |
63244fa1 ID |
1420 | } |
1421 | ||
7de030d6 ID |
1422 | if (unpaused || legacy_change || force_resend || |
1423 | (split && con && CEPH_HAVE_FEATURE(con->peer_features, | |
1424 | RESEND_ON_SPLIT))) | |
84ed45df ID |
1425 | ct_res = CALC_TARGET_NEED_RESEND; |
1426 | else | |
1427 | ct_res = CALC_TARGET_NO_ACTION; | |
1428 | ||
63244fa1 ID |
1429 | out: |
1430 | dout("%s t %p -> ct_res %d osd %d\n", __func__, t, ct_res, t->osd); | |
1431 | return ct_res; | |
1432 | } | |
1433 | ||
bb873b53 ID |
1434 | static void setup_request_data(struct ceph_osd_request *req, |
1435 | struct ceph_msg *msg) | |
f24e9980 | 1436 | { |
bb873b53 ID |
1437 | u32 data_len = 0; |
1438 | int i; | |
1439 | ||
1440 | if (!list_empty(&msg->data)) | |
1441 | return; | |
f24e9980 | 1442 | |
bb873b53 ID |
1443 | WARN_ON(msg->data_length); |
1444 | for (i = 0; i < req->r_num_ops; i++) { | |
1445 | struct ceph_osd_req_op *op = &req->r_ops[i]; | |
1446 | ||
1447 | switch (op->op) { | |
1448 | /* request */ | |
1449 | case CEPH_OSD_OP_WRITE: | |
1450 | case CEPH_OSD_OP_WRITEFULL: | |
1451 | WARN_ON(op->indata_len != op->extent.length); | |
1452 | ceph_osdc_msg_data_add(msg, &op->extent.osd_data); | |
1453 | break; | |
1454 | case CEPH_OSD_OP_SETXATTR: | |
1455 | case CEPH_OSD_OP_CMPXATTR: | |
1456 | WARN_ON(op->indata_len != op->xattr.name_len + | |
1457 | op->xattr.value_len); | |
1458 | ceph_osdc_msg_data_add(msg, &op->xattr.osd_data); | |
1459 | break; | |
922dab61 ID |
1460 | case CEPH_OSD_OP_NOTIFY_ACK: |
1461 | ceph_osdc_msg_data_add(msg, | |
1462 | &op->notify_ack.request_data); | |
1463 | break; | |
bb873b53 ID |
1464 | |
1465 | /* reply */ | |
1466 | case CEPH_OSD_OP_STAT: | |
1467 | ceph_osdc_msg_data_add(req->r_reply, | |
1468 | &op->raw_data_in); | |
1469 | break; | |
1470 | case CEPH_OSD_OP_READ: | |
1471 | ceph_osdc_msg_data_add(req->r_reply, | |
1472 | &op->extent.osd_data); | |
1473 | break; | |
a4ed38d7 DF |
1474 | case CEPH_OSD_OP_LIST_WATCHERS: |
1475 | ceph_osdc_msg_data_add(req->r_reply, | |
1476 | &op->list_watchers.response_data); | |
1477 | break; | |
bb873b53 ID |
1478 | |
1479 | /* both */ | |
1480 | case CEPH_OSD_OP_CALL: | |
1481 | WARN_ON(op->indata_len != op->cls.class_len + | |
1482 | op->cls.method_len + | |
1483 | op->cls.indata_len); | |
1484 | ceph_osdc_msg_data_add(msg, &op->cls.request_info); | |
1485 | /* optional, can be NONE */ | |
1486 | ceph_osdc_msg_data_add(msg, &op->cls.request_data); | |
1487 | /* optional, can be NONE */ | |
1488 | ceph_osdc_msg_data_add(req->r_reply, | |
1489 | &op->cls.response_data); | |
1490 | break; | |
19079203 ID |
1491 | case CEPH_OSD_OP_NOTIFY: |
1492 | ceph_osdc_msg_data_add(msg, | |
1493 | &op->notify.request_data); | |
1494 | ceph_osdc_msg_data_add(req->r_reply, | |
1495 | &op->notify.response_data); | |
1496 | break; | |
bb873b53 ID |
1497 | } |
1498 | ||
1499 | data_len += op->indata_len; | |
1500 | } | |
1b83bef2 | 1501 | |
bb873b53 ID |
1502 | WARN_ON(data_len != msg->data_length); |
1503 | } | |
1504 | ||
2e59ffd1 ID |
1505 | static void encode_pgid(void **p, const struct ceph_pg *pgid) |
1506 | { | |
1507 | ceph_encode_8(p, 1); | |
1508 | ceph_encode_64(p, pgid->pool); | |
1509 | ceph_encode_32(p, pgid->seed); | |
1510 | ceph_encode_32(p, -1); /* preferred */ | |
1511 | } | |
1512 | ||
8cb441c0 ID |
1513 | static void encode_spgid(void **p, const struct ceph_spg *spgid) |
1514 | { | |
1515 | ceph_start_encoding(p, 1, 1, CEPH_PGID_ENCODING_LEN + 1); | |
1516 | encode_pgid(p, &spgid->pgid); | |
1517 | ceph_encode_8(p, spgid->shard); | |
1518 | } | |
1519 | ||
2e59ffd1 ID |
1520 | static void encode_oloc(void **p, void *end, |
1521 | const struct ceph_object_locator *oloc) | |
1522 | { | |
1523 | ceph_start_encoding(p, 5, 4, ceph_oloc_encoding_size(oloc)); | |
1524 | ceph_encode_64(p, oloc->pool); | |
1525 | ceph_encode_32(p, -1); /* preferred */ | |
1526 | ceph_encode_32(p, 0); /* key len */ | |
1527 | if (oloc->pool_ns) | |
1528 | ceph_encode_string(p, end, oloc->pool_ns->str, | |
1529 | oloc->pool_ns->len); | |
1530 | else | |
1531 | ceph_encode_32(p, 0); | |
1532 | } | |
1533 | ||
8cb441c0 ID |
1534 | static void encode_request_partial(struct ceph_osd_request *req, |
1535 | struct ceph_msg *msg) | |
bb873b53 ID |
1536 | { |
1537 | void *p = msg->front.iov_base; | |
1538 | void *const end = p + msg->front_alloc_len; | |
1539 | u32 data_len = 0; | |
1540 | int i; | |
1541 | ||
1542 | if (req->r_flags & CEPH_OSD_FLAG_WRITE) { | |
1543 | /* snapshots aren't writeable */ | |
1544 | WARN_ON(req->r_snapid != CEPH_NOSNAP); | |
1545 | } else { | |
1546 | WARN_ON(req->r_mtime.tv_sec || req->r_mtime.tv_nsec || | |
1547 | req->r_data_offset || req->r_snapc); | |
1548 | } | |
1549 | ||
1550 | setup_request_data(req, msg); | |
1551 | ||
8cb441c0 ID |
1552 | encode_spgid(&p, &req->r_t.spgid); /* actual spg */ |
1553 | ceph_encode_32(&p, req->r_t.pgid.seed); /* raw hash */ | |
bb873b53 ID |
1554 | ceph_encode_32(&p, req->r_osdc->osdmap->epoch); |
1555 | ceph_encode_32(&p, req->r_flags); | |
8cb441c0 ID |
1556 | |
1557 | /* reqid */ | |
1558 | ceph_start_encoding(&p, 2, 2, sizeof(struct ceph_osd_reqid)); | |
1559 | memset(p, 0, sizeof(struct ceph_osd_reqid)); | |
1560 | p += sizeof(struct ceph_osd_reqid); | |
1561 | ||
1562 | /* trace */ | |
1563 | memset(p, 0, sizeof(struct ceph_blkin_trace_info)); | |
1564 | p += sizeof(struct ceph_blkin_trace_info); | |
1565 | ||
1566 | ceph_encode_32(&p, 0); /* client_inc, always 0 */ | |
bb873b53 ID |
1567 | ceph_encode_timespec(p, &req->r_mtime); |
1568 | p += sizeof(struct ceph_timespec); | |
aa26d662 | 1569 | |
2e59ffd1 | 1570 | encode_oloc(&p, end, &req->r_t.target_oloc); |
2e59ffd1 ID |
1571 | ceph_encode_string(&p, end, req->r_t.target_oid.name, |
1572 | req->r_t.target_oid.name_len); | |
f24e9980 | 1573 | |
bb873b53 ID |
1574 | /* ops, can imply data */ |
1575 | ceph_encode_16(&p, req->r_num_ops); | |
1576 | for (i = 0; i < req->r_num_ops; i++) { | |
1577 | data_len += osd_req_encode_op(p, &req->r_ops[i]); | |
1578 | p += sizeof(struct ceph_osd_op); | |
1579 | } | |
26be8808 | 1580 | |
bb873b53 ID |
1581 | ceph_encode_64(&p, req->r_snapid); /* snapid */ |
1582 | if (req->r_snapc) { | |
1583 | ceph_encode_64(&p, req->r_snapc->seq); | |
1584 | ceph_encode_32(&p, req->r_snapc->num_snaps); | |
1585 | for (i = 0; i < req->r_snapc->num_snaps; i++) | |
1586 | ceph_encode_64(&p, req->r_snapc->snaps[i]); | |
1587 | } else { | |
1588 | ceph_encode_64(&p, 0); /* snap_seq */ | |
1589 | ceph_encode_32(&p, 0); /* snaps len */ | |
1590 | } | |
1591 | ||
1592 | ceph_encode_32(&p, req->r_attempts); /* retry_attempt */ | |
8cb441c0 | 1593 | BUG_ON(p != end - 8); /* space for features */ |
bb873b53 | 1594 | |
8cb441c0 ID |
1595 | msg->hdr.version = cpu_to_le16(8); /* MOSDOp v8 */ |
1596 | /* front_len is finalized in encode_request_finish() */ | |
bb873b53 ID |
1597 | msg->hdr.data_len = cpu_to_le32(data_len); |
1598 | /* | |
1599 | * The header "data_off" is a hint to the receiver allowing it | |
1600 | * to align received data into its buffers such that there's no | |
1601 | * need to re-copy it before writing it to disk (direct I/O). | |
1602 | */ | |
1603 | msg->hdr.data_off = cpu_to_le16(req->r_data_offset); | |
26be8808 | 1604 | |
8cb441c0 ID |
1605 | dout("%s req %p msg %p oid %s oid_len %d\n", __func__, req, msg, |
1606 | req->r_t.target_oid.name, req->r_t.target_oid.name_len); | |
1607 | } | |
1608 | ||
1609 | static void encode_request_finish(struct ceph_msg *msg) | |
1610 | { | |
1611 | void *p = msg->front.iov_base; | |
1612 | void *const end = p + msg->front_alloc_len; | |
1613 | ||
1614 | if (CEPH_HAVE_FEATURE(msg->con->peer_features, RESEND_ON_SPLIT)) { | |
1615 | /* luminous OSD -- encode features and be done */ | |
1616 | p = end - 8; | |
1617 | ceph_encode_64(&p, msg->con->peer_features); | |
1618 | } else { | |
1619 | struct { | |
1620 | char spgid[CEPH_ENCODING_START_BLK_LEN + | |
1621 | CEPH_PGID_ENCODING_LEN + 1]; | |
1622 | __le32 hash; | |
1623 | __le32 epoch; | |
1624 | __le32 flags; | |
1625 | char reqid[CEPH_ENCODING_START_BLK_LEN + | |
1626 | sizeof(struct ceph_osd_reqid)]; | |
1627 | char trace[sizeof(struct ceph_blkin_trace_info)]; | |
1628 | __le32 client_inc; | |
1629 | struct ceph_timespec mtime; | |
1630 | } __packed head; | |
1631 | struct ceph_pg pgid; | |
1632 | void *oloc, *oid, *tail; | |
1633 | int oloc_len, oid_len, tail_len; | |
1634 | int len; | |
1635 | ||
1636 | /* | |
1637 | * Pre-luminous OSD -- reencode v8 into v4 using @head | |
1638 | * as a temporary buffer. Encode the raw PG; the rest | |
1639 | * is just a matter of moving oloc, oid and tail blobs | |
1640 | * around. | |
1641 | */ | |
1642 | memcpy(&head, p, sizeof(head)); | |
1643 | p += sizeof(head); | |
1644 | ||
1645 | oloc = p; | |
1646 | p += CEPH_ENCODING_START_BLK_LEN; | |
1647 | pgid.pool = ceph_decode_64(&p); | |
1648 | p += 4 + 4; /* preferred, key len */ | |
1649 | len = ceph_decode_32(&p); | |
1650 | p += len; /* nspace */ | |
1651 | oloc_len = p - oloc; | |
1652 | ||
1653 | oid = p; | |
1654 | len = ceph_decode_32(&p); | |
1655 | p += len; | |
1656 | oid_len = p - oid; | |
1657 | ||
1658 | tail = p; | |
1659 | tail_len = (end - p) - 8; | |
1660 | ||
1661 | p = msg->front.iov_base; | |
1662 | ceph_encode_copy(&p, &head.client_inc, sizeof(head.client_inc)); | |
1663 | ceph_encode_copy(&p, &head.epoch, sizeof(head.epoch)); | |
1664 | ceph_encode_copy(&p, &head.flags, sizeof(head.flags)); | |
1665 | ceph_encode_copy(&p, &head.mtime, sizeof(head.mtime)); | |
1666 | ||
1667 | /* reassert_version */ | |
1668 | memset(p, 0, sizeof(struct ceph_eversion)); | |
1669 | p += sizeof(struct ceph_eversion); | |
1670 | ||
1671 | BUG_ON(p >= oloc); | |
1672 | memmove(p, oloc, oloc_len); | |
1673 | p += oloc_len; | |
1674 | ||
1675 | pgid.seed = le32_to_cpu(head.hash); | |
1676 | encode_pgid(&p, &pgid); /* raw pg */ | |
1677 | ||
1678 | BUG_ON(p >= oid); | |
1679 | memmove(p, oid, oid_len); | |
1680 | p += oid_len; | |
1681 | ||
1682 | /* tail -- ops, snapid, snapc, retry_attempt */ | |
1683 | BUG_ON(p >= tail); | |
1684 | memmove(p, tail, tail_len); | |
1685 | p += tail_len; | |
1686 | ||
1687 | msg->hdr.version = cpu_to_le16(4); /* MOSDOp v4 */ | |
1688 | } | |
1689 | ||
1690 | BUG_ON(p > end); | |
1691 | msg->front.iov_len = p - msg->front.iov_base; | |
1692 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
1693 | ||
1694 | dout("%s msg %p tid %llu %u+%u+%u v%d\n", __func__, msg, | |
1695 | le64_to_cpu(msg->hdr.tid), le32_to_cpu(msg->hdr.front_len), | |
1696 | le32_to_cpu(msg->hdr.middle_len), le32_to_cpu(msg->hdr.data_len), | |
1697 | le16_to_cpu(msg->hdr.version)); | |
bb873b53 ID |
1698 | } |
1699 | ||
1700 | /* | |
1701 | * @req has to be assigned a tid and registered. | |
1702 | */ | |
1703 | static void send_request(struct ceph_osd_request *req) | |
1704 | { | |
1705 | struct ceph_osd *osd = req->r_osd; | |
1706 | ||
5aea3dcd | 1707 | verify_osd_locked(osd); |
bb873b53 ID |
1708 | WARN_ON(osd->o_osd != req->r_t.osd); |
1709 | ||
5aea3dcd ID |
1710 | /* |
1711 | * We may have a previously queued request message hanging | |
1712 | * around. Cancel it to avoid corrupting the msgr. | |
1713 | */ | |
1714 | if (req->r_sent) | |
1715 | ceph_msg_revoke(req->r_request); | |
1716 | ||
bb873b53 ID |
1717 | req->r_flags |= CEPH_OSD_FLAG_KNOWN_REDIR; |
1718 | if (req->r_attempts) | |
1719 | req->r_flags |= CEPH_OSD_FLAG_RETRY; | |
1720 | else | |
1721 | WARN_ON(req->r_flags & CEPH_OSD_FLAG_RETRY); | |
1722 | ||
8cb441c0 | 1723 | encode_request_partial(req, req->r_request); |
bb873b53 | 1724 | |
04c7d789 | 1725 | dout("%s req %p tid %llu to pgid %llu.%x spgid %llu.%xs%d osd%d e%u flags 0x%x attempt %d\n", |
bb873b53 | 1726 | __func__, req, req->r_tid, req->r_t.pgid.pool, req->r_t.pgid.seed, |
dc98ff72 | 1727 | req->r_t.spgid.pgid.pool, req->r_t.spgid.pgid.seed, |
04c7d789 ID |
1728 | req->r_t.spgid.shard, osd->o_osd, req->r_t.epoch, req->r_flags, |
1729 | req->r_attempts); | |
bb873b53 ID |
1730 | |
1731 | req->r_t.paused = false; | |
1732 | req->r_stamp = jiffies; | |
1733 | req->r_attempts++; | |
1734 | ||
1735 | req->r_sent = osd->o_incarnation; | |
1736 | req->r_request->hdr.tid = cpu_to_le64(req->r_tid); | |
1737 | ceph_con_send(&osd->o_con, ceph_msg_get(req->r_request)); | |
f24e9980 SW |
1738 | } |
1739 | ||
42c1b124 ID |
1740 | static void maybe_request_map(struct ceph_osd_client *osdc) |
1741 | { | |
1742 | bool continuous = false; | |
1743 | ||
5aea3dcd | 1744 | verify_osdc_locked(osdc); |
42c1b124 ID |
1745 | WARN_ON(!osdc->osdmap->epoch); |
1746 | ||
b7ec35b3 ID |
1747 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || |
1748 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD) || | |
1749 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { | |
42c1b124 ID |
1750 | dout("%s osdc %p continuous\n", __func__, osdc); |
1751 | continuous = true; | |
1752 | } else { | |
1753 | dout("%s osdc %p onetime\n", __func__, osdc); | |
1754 | } | |
1755 | ||
1756 | if (ceph_monc_want_map(&osdc->client->monc, CEPH_SUB_OSDMAP, | |
1757 | osdc->osdmap->epoch + 1, continuous)) | |
1758 | ceph_monc_renew_subs(&osdc->client->monc); | |
1759 | } | |
1760 | ||
a1f4020a | 1761 | static void complete_request(struct ceph_osd_request *req, int err); |
4609245e ID |
1762 | static void send_map_check(struct ceph_osd_request *req); |
1763 | ||
5aea3dcd | 1764 | static void __submit_request(struct ceph_osd_request *req, bool wrlocked) |
0bbfdfe8 | 1765 | { |
5aea3dcd ID |
1766 | struct ceph_osd_client *osdc = req->r_osdc; |
1767 | struct ceph_osd *osd; | |
4609245e | 1768 | enum calc_target_result ct_res; |
5aea3dcd ID |
1769 | bool need_send = false; |
1770 | bool promoted = false; | |
a1f4020a | 1771 | bool need_abort = false; |
0bbfdfe8 | 1772 | |
b18b9550 | 1773 | WARN_ON(req->r_tid); |
5aea3dcd ID |
1774 | dout("%s req %p wrlocked %d\n", __func__, req, wrlocked); |
1775 | ||
1776 | again: | |
7de030d6 | 1777 | ct_res = calc_target(osdc, &req->r_t, NULL, false); |
4609245e ID |
1778 | if (ct_res == CALC_TARGET_POOL_DNE && !wrlocked) |
1779 | goto promote; | |
1780 | ||
5aea3dcd ID |
1781 | osd = lookup_create_osd(osdc, req->r_t.osd, wrlocked); |
1782 | if (IS_ERR(osd)) { | |
1783 | WARN_ON(PTR_ERR(osd) != -EAGAIN || wrlocked); | |
1784 | goto promote; | |
0bbfdfe8 ID |
1785 | } |
1786 | ||
58eb7932 JL |
1787 | if (osdc->osdmap->epoch < osdc->epoch_barrier) { |
1788 | dout("req %p epoch %u barrier %u\n", req, osdc->osdmap->epoch, | |
1789 | osdc->epoch_barrier); | |
1790 | req->r_t.paused = true; | |
1791 | maybe_request_map(osdc); | |
1792 | } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && | |
1793 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR)) { | |
5aea3dcd ID |
1794 | dout("req %p pausewr\n", req); |
1795 | req->r_t.paused = true; | |
1796 | maybe_request_map(osdc); | |
1797 | } else if ((req->r_flags & CEPH_OSD_FLAG_READ) && | |
b7ec35b3 | 1798 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { |
5aea3dcd ID |
1799 | dout("req %p pauserd\n", req); |
1800 | req->r_t.paused = true; | |
1801 | maybe_request_map(osdc); | |
1802 | } else if ((req->r_flags & CEPH_OSD_FLAG_WRITE) && | |
1803 | !(req->r_flags & (CEPH_OSD_FLAG_FULL_TRY | | |
1804 | CEPH_OSD_FLAG_FULL_FORCE)) && | |
b7ec35b3 | 1805 | (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || |
5aea3dcd ID |
1806 | pool_full(osdc, req->r_t.base_oloc.pool))) { |
1807 | dout("req %p full/pool_full\n", req); | |
1808 | pr_warn_ratelimited("FULL or reached pool quota\n"); | |
1809 | req->r_t.paused = true; | |
1810 | maybe_request_map(osdc); | |
a1f4020a JL |
1811 | if (req->r_abort_on_full) |
1812 | need_abort = true; | |
5aea3dcd ID |
1813 | } else if (!osd_homeless(osd)) { |
1814 | need_send = true; | |
0bbfdfe8 | 1815 | } else { |
5aea3dcd | 1816 | maybe_request_map(osdc); |
0bbfdfe8 ID |
1817 | } |
1818 | ||
5aea3dcd ID |
1819 | mutex_lock(&osd->lock); |
1820 | /* | |
1821 | * Assign the tid atomically with send_request() to protect | |
1822 | * multiple writes to the same object from racing with each | |
1823 | * other, resulting in out of order ops on the OSDs. | |
1824 | */ | |
1825 | req->r_tid = atomic64_inc_return(&osdc->last_tid); | |
1826 | link_request(osd, req); | |
1827 | if (need_send) | |
1828 | send_request(req); | |
a1f4020a JL |
1829 | else if (need_abort) |
1830 | complete_request(req, -ENOSPC); | |
5aea3dcd ID |
1831 | mutex_unlock(&osd->lock); |
1832 | ||
4609245e ID |
1833 | if (ct_res == CALC_TARGET_POOL_DNE) |
1834 | send_map_check(req); | |
1835 | ||
5aea3dcd ID |
1836 | if (promoted) |
1837 | downgrade_write(&osdc->lock); | |
1838 | return; | |
1839 | ||
1840 | promote: | |
1841 | up_read(&osdc->lock); | |
1842 | down_write(&osdc->lock); | |
1843 | wrlocked = true; | |
1844 | promoted = true; | |
1845 | goto again; | |
1846 | } | |
1847 | ||
1848 | static void account_request(struct ceph_osd_request *req) | |
1849 | { | |
54ea0046 | 1850 | WARN_ON(req->r_flags & (CEPH_OSD_FLAG_ACK | CEPH_OSD_FLAG_ONDISK)); |
b18b9550 | 1851 | WARN_ON(!(req->r_flags & (CEPH_OSD_FLAG_READ | CEPH_OSD_FLAG_WRITE))); |
5aea3dcd | 1852 | |
b18b9550 | 1853 | req->r_flags |= CEPH_OSD_FLAG_ONDISK; |
5aea3dcd | 1854 | atomic_inc(&req->r_osdc->num_requests); |
7cc5e38f ID |
1855 | |
1856 | req->r_start_stamp = jiffies; | |
5aea3dcd ID |
1857 | } |
1858 | ||
1859 | static void submit_request(struct ceph_osd_request *req, bool wrlocked) | |
1860 | { | |
1861 | ceph_osdc_get_request(req); | |
1862 | account_request(req); | |
1863 | __submit_request(req, wrlocked); | |
1864 | } | |
1865 | ||
45ee2c1d | 1866 | static void finish_request(struct ceph_osd_request *req) |
5aea3dcd ID |
1867 | { |
1868 | struct ceph_osd_client *osdc = req->r_osdc; | |
5aea3dcd | 1869 | |
04c7d789 | 1870 | WARN_ON(lookup_request_mc(&osdc->map_checks, req->r_tid)); |
5aea3dcd ID |
1871 | dout("%s req %p tid %llu\n", __func__, req, req->r_tid); |
1872 | ||
04c7d789 ID |
1873 | if (req->r_osd) |
1874 | unlink_request(req->r_osd, req); | |
5aea3dcd ID |
1875 | atomic_dec(&osdc->num_requests); |
1876 | ||
1877 | /* | |
1878 | * If an OSD has failed or returned and a request has been sent | |
1879 | * twice, it's possible to get a reply and end up here while the | |
1880 | * request message is queued for delivery. We will ignore the | |
1881 | * reply, so not a big deal, but better to try and catch it. | |
1882 | */ | |
1883 | ceph_msg_revoke(req->r_request); | |
1884 | ceph_msg_revoke_incoming(req->r_reply); | |
1885 | } | |
1886 | ||
fe5da05e ID |
1887 | static void __complete_request(struct ceph_osd_request *req) |
1888 | { | |
b18b9550 ID |
1889 | if (req->r_callback) { |
1890 | dout("%s req %p tid %llu cb %pf result %d\n", __func__, req, | |
1891 | req->r_tid, req->r_callback, req->r_result); | |
fe5da05e | 1892 | req->r_callback(req); |
b18b9550 | 1893 | } |
fe5da05e ID |
1894 | } |
1895 | ||
4609245e | 1896 | /* |
b18b9550 | 1897 | * This is open-coded in handle_reply(). |
4609245e ID |
1898 | */ |
1899 | static void complete_request(struct ceph_osd_request *req, int err) | |
1900 | { | |
1901 | dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); | |
1902 | ||
1903 | req->r_result = err; | |
45ee2c1d | 1904 | finish_request(req); |
4609245e | 1905 | __complete_request(req); |
b18b9550 | 1906 | complete_all(&req->r_completion); |
4609245e ID |
1907 | ceph_osdc_put_request(req); |
1908 | } | |
1909 | ||
1910 | static void cancel_map_check(struct ceph_osd_request *req) | |
1911 | { | |
1912 | struct ceph_osd_client *osdc = req->r_osdc; | |
1913 | struct ceph_osd_request *lookup_req; | |
1914 | ||
1915 | verify_osdc_wrlocked(osdc); | |
1916 | ||
1917 | lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); | |
1918 | if (!lookup_req) | |
1919 | return; | |
1920 | ||
1921 | WARN_ON(lookup_req != req); | |
1922 | erase_request_mc(&osdc->map_checks, req); | |
1923 | ceph_osdc_put_request(req); | |
1924 | } | |
1925 | ||
5aea3dcd ID |
1926 | static void cancel_request(struct ceph_osd_request *req) |
1927 | { | |
1928 | dout("%s req %p tid %llu\n", __func__, req, req->r_tid); | |
1929 | ||
4609245e | 1930 | cancel_map_check(req); |
45ee2c1d | 1931 | finish_request(req); |
b18b9550 | 1932 | complete_all(&req->r_completion); |
c297eb42 | 1933 | ceph_osdc_put_request(req); |
5aea3dcd ID |
1934 | } |
1935 | ||
7cc5e38f ID |
1936 | static void abort_request(struct ceph_osd_request *req, int err) |
1937 | { | |
1938 | dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err); | |
1939 | ||
1940 | cancel_map_check(req); | |
1941 | complete_request(req, err); | |
1942 | } | |
1943 | ||
58eb7932 JL |
1944 | static void update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb) |
1945 | { | |
1946 | if (likely(eb > osdc->epoch_barrier)) { | |
1947 | dout("updating epoch_barrier from %u to %u\n", | |
1948 | osdc->epoch_barrier, eb); | |
1949 | osdc->epoch_barrier = eb; | |
1950 | /* Request map if we're not to the barrier yet */ | |
1951 | if (eb > osdc->osdmap->epoch) | |
1952 | maybe_request_map(osdc); | |
1953 | } | |
1954 | } | |
1955 | ||
1956 | void ceph_osdc_update_epoch_barrier(struct ceph_osd_client *osdc, u32 eb) | |
1957 | { | |
1958 | down_read(&osdc->lock); | |
1959 | if (unlikely(eb > osdc->epoch_barrier)) { | |
1960 | up_read(&osdc->lock); | |
1961 | down_write(&osdc->lock); | |
1962 | update_epoch_barrier(osdc, eb); | |
1963 | up_write(&osdc->lock); | |
1964 | } else { | |
1965 | up_read(&osdc->lock); | |
1966 | } | |
1967 | } | |
1968 | EXPORT_SYMBOL(ceph_osdc_update_epoch_barrier); | |
1969 | ||
fc36d0a4 JL |
1970 | /* |
1971 | * Drop all pending requests that are stalled waiting on a full condition to | |
58eb7932 JL |
1972 | * clear, and complete them with ENOSPC as the return code. Set the |
1973 | * osdc->epoch_barrier to the latest map epoch that we've seen if any were | |
1974 | * cancelled. | |
fc36d0a4 JL |
1975 | */ |
1976 | static void ceph_osdc_abort_on_full(struct ceph_osd_client *osdc) | |
1977 | { | |
1978 | struct rb_node *n; | |
58eb7932 | 1979 | bool victims = false; |
fc36d0a4 JL |
1980 | |
1981 | dout("enter abort_on_full\n"); | |
1982 | ||
1983 | if (!ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) && !have_pool_full(osdc)) | |
1984 | goto out; | |
1985 | ||
58eb7932 JL |
1986 | /* Scan list and see if there is anything to abort */ |
1987 | for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { | |
1988 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
1989 | struct rb_node *m; | |
1990 | ||
1991 | m = rb_first(&osd->o_requests); | |
1992 | while (m) { | |
1993 | struct ceph_osd_request *req = rb_entry(m, | |
1994 | struct ceph_osd_request, r_node); | |
1995 | m = rb_next(m); | |
1996 | ||
1997 | if (req->r_abort_on_full) { | |
1998 | victims = true; | |
1999 | break; | |
2000 | } | |
2001 | } | |
2002 | if (victims) | |
2003 | break; | |
2004 | } | |
2005 | ||
2006 | if (!victims) | |
2007 | goto out; | |
2008 | ||
2009 | /* | |
2010 | * Update the barrier to current epoch if it's behind that point, | |
2011 | * since we know we have some calls to be aborted in the tree. | |
2012 | */ | |
2013 | update_epoch_barrier(osdc, osdc->osdmap->epoch); | |
2014 | ||
fc36d0a4 JL |
2015 | for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { |
2016 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
2017 | struct rb_node *m; | |
2018 | ||
2019 | m = rb_first(&osd->o_requests); | |
2020 | while (m) { | |
2021 | struct ceph_osd_request *req = rb_entry(m, | |
2022 | struct ceph_osd_request, r_node); | |
2023 | m = rb_next(m); | |
2024 | ||
2025 | if (req->r_abort_on_full && | |
2026 | (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || | |
2027 | pool_full(osdc, req->r_t.target_oloc.pool))) | |
2028 | abort_request(req, -ENOSPC); | |
2029 | } | |
2030 | } | |
2031 | out: | |
58eb7932 | 2032 | dout("return abort_on_full barrier=%u\n", osdc->epoch_barrier); |
fc36d0a4 JL |
2033 | } |
2034 | ||
4609245e ID |
2035 | static void check_pool_dne(struct ceph_osd_request *req) |
2036 | { | |
2037 | struct ceph_osd_client *osdc = req->r_osdc; | |
2038 | struct ceph_osdmap *map = osdc->osdmap; | |
2039 | ||
2040 | verify_osdc_wrlocked(osdc); | |
2041 | WARN_ON(!map->epoch); | |
2042 | ||
2043 | if (req->r_attempts) { | |
2044 | /* | |
2045 | * We sent a request earlier, which means that | |
2046 | * previously the pool existed, and now it does not | |
2047 | * (i.e., it was deleted). | |
2048 | */ | |
2049 | req->r_map_dne_bound = map->epoch; | |
2050 | dout("%s req %p tid %llu pool disappeared\n", __func__, req, | |
2051 | req->r_tid); | |
2052 | } else { | |
2053 | dout("%s req %p tid %llu map_dne_bound %u have %u\n", __func__, | |
2054 | req, req->r_tid, req->r_map_dne_bound, map->epoch); | |
2055 | } | |
2056 | ||
2057 | if (req->r_map_dne_bound) { | |
2058 | if (map->epoch >= req->r_map_dne_bound) { | |
2059 | /* we had a new enough map */ | |
2060 | pr_info_ratelimited("tid %llu pool does not exist\n", | |
2061 | req->r_tid); | |
2062 | complete_request(req, -ENOENT); | |
2063 | } | |
2064 | } else { | |
2065 | send_map_check(req); | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | static void map_check_cb(struct ceph_mon_generic_request *greq) | |
2070 | { | |
2071 | struct ceph_osd_client *osdc = &greq->monc->client->osdc; | |
2072 | struct ceph_osd_request *req; | |
2073 | u64 tid = greq->private_data; | |
2074 | ||
2075 | WARN_ON(greq->result || !greq->u.newest); | |
2076 | ||
2077 | down_write(&osdc->lock); | |
2078 | req = lookup_request_mc(&osdc->map_checks, tid); | |
2079 | if (!req) { | |
2080 | dout("%s tid %llu dne\n", __func__, tid); | |
2081 | goto out_unlock; | |
2082 | } | |
2083 | ||
2084 | dout("%s req %p tid %llu map_dne_bound %u newest %llu\n", __func__, | |
2085 | req, req->r_tid, req->r_map_dne_bound, greq->u.newest); | |
2086 | if (!req->r_map_dne_bound) | |
2087 | req->r_map_dne_bound = greq->u.newest; | |
2088 | erase_request_mc(&osdc->map_checks, req); | |
2089 | check_pool_dne(req); | |
2090 | ||
2091 | ceph_osdc_put_request(req); | |
2092 | out_unlock: | |
2093 | up_write(&osdc->lock); | |
2094 | } | |
2095 | ||
2096 | static void send_map_check(struct ceph_osd_request *req) | |
2097 | { | |
2098 | struct ceph_osd_client *osdc = req->r_osdc; | |
2099 | struct ceph_osd_request *lookup_req; | |
2100 | int ret; | |
2101 | ||
2102 | verify_osdc_wrlocked(osdc); | |
2103 | ||
2104 | lookup_req = lookup_request_mc(&osdc->map_checks, req->r_tid); | |
2105 | if (lookup_req) { | |
2106 | WARN_ON(lookup_req != req); | |
2107 | return; | |
2108 | } | |
2109 | ||
2110 | ceph_osdc_get_request(req); | |
2111 | insert_request_mc(&osdc->map_checks, req); | |
2112 | ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", | |
2113 | map_check_cb, req->r_tid); | |
2114 | WARN_ON(ret); | |
2115 | } | |
2116 | ||
922dab61 ID |
2117 | /* |
2118 | * lingering requests, watch/notify v2 infrastructure | |
2119 | */ | |
2120 | static void linger_release(struct kref *kref) | |
2121 | { | |
2122 | struct ceph_osd_linger_request *lreq = | |
2123 | container_of(kref, struct ceph_osd_linger_request, kref); | |
2124 | ||
2125 | dout("%s lreq %p reg_req %p ping_req %p\n", __func__, lreq, | |
2126 | lreq->reg_req, lreq->ping_req); | |
2127 | WARN_ON(!RB_EMPTY_NODE(&lreq->node)); | |
2128 | WARN_ON(!RB_EMPTY_NODE(&lreq->osdc_node)); | |
4609245e | 2129 | WARN_ON(!RB_EMPTY_NODE(&lreq->mc_node)); |
922dab61 | 2130 | WARN_ON(!list_empty(&lreq->scan_item)); |
b07d3c4b | 2131 | WARN_ON(!list_empty(&lreq->pending_lworks)); |
922dab61 ID |
2132 | WARN_ON(lreq->osd); |
2133 | ||
2134 | if (lreq->reg_req) | |
2135 | ceph_osdc_put_request(lreq->reg_req); | |
2136 | if (lreq->ping_req) | |
2137 | ceph_osdc_put_request(lreq->ping_req); | |
2138 | target_destroy(&lreq->t); | |
2139 | kfree(lreq); | |
2140 | } | |
2141 | ||
2142 | static void linger_put(struct ceph_osd_linger_request *lreq) | |
2143 | { | |
2144 | if (lreq) | |
2145 | kref_put(&lreq->kref, linger_release); | |
2146 | } | |
2147 | ||
2148 | static struct ceph_osd_linger_request * | |
2149 | linger_get(struct ceph_osd_linger_request *lreq) | |
2150 | { | |
2151 | kref_get(&lreq->kref); | |
2152 | return lreq; | |
2153 | } | |
2154 | ||
2155 | static struct ceph_osd_linger_request * | |
2156 | linger_alloc(struct ceph_osd_client *osdc) | |
2157 | { | |
2158 | struct ceph_osd_linger_request *lreq; | |
2159 | ||
2160 | lreq = kzalloc(sizeof(*lreq), GFP_NOIO); | |
2161 | if (!lreq) | |
2162 | return NULL; | |
2163 | ||
2164 | kref_init(&lreq->kref); | |
2165 | mutex_init(&lreq->lock); | |
2166 | RB_CLEAR_NODE(&lreq->node); | |
2167 | RB_CLEAR_NODE(&lreq->osdc_node); | |
4609245e | 2168 | RB_CLEAR_NODE(&lreq->mc_node); |
922dab61 | 2169 | INIT_LIST_HEAD(&lreq->scan_item); |
b07d3c4b | 2170 | INIT_LIST_HEAD(&lreq->pending_lworks); |
922dab61 | 2171 | init_completion(&lreq->reg_commit_wait); |
19079203 | 2172 | init_completion(&lreq->notify_finish_wait); |
922dab61 ID |
2173 | |
2174 | lreq->osdc = osdc; | |
2175 | target_init(&lreq->t); | |
2176 | ||
2177 | dout("%s lreq %p\n", __func__, lreq); | |
2178 | return lreq; | |
2179 | } | |
2180 | ||
2181 | DEFINE_RB_INSDEL_FUNCS(linger, struct ceph_osd_linger_request, linger_id, node) | |
2182 | DEFINE_RB_FUNCS(linger_osdc, struct ceph_osd_linger_request, linger_id, osdc_node) | |
4609245e | 2183 | DEFINE_RB_FUNCS(linger_mc, struct ceph_osd_linger_request, linger_id, mc_node) |
922dab61 ID |
2184 | |
2185 | /* | |
2186 | * Create linger request <-> OSD session relation. | |
2187 | * | |
2188 | * @lreq has to be registered, @osd may be homeless. | |
2189 | */ | |
2190 | static void link_linger(struct ceph_osd *osd, | |
2191 | struct ceph_osd_linger_request *lreq) | |
2192 | { | |
2193 | verify_osd_locked(osd); | |
2194 | WARN_ON(!lreq->linger_id || lreq->osd); | |
2195 | dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, | |
2196 | osd->o_osd, lreq, lreq->linger_id); | |
2197 | ||
2198 | if (!osd_homeless(osd)) | |
2199 | __remove_osd_from_lru(osd); | |
2200 | else | |
2201 | atomic_inc(&osd->o_osdc->num_homeless); | |
2202 | ||
2203 | get_osd(osd); | |
2204 | insert_linger(&osd->o_linger_requests, lreq); | |
2205 | lreq->osd = osd; | |
2206 | } | |
2207 | ||
2208 | static void unlink_linger(struct ceph_osd *osd, | |
2209 | struct ceph_osd_linger_request *lreq) | |
2210 | { | |
2211 | verify_osd_locked(osd); | |
2212 | WARN_ON(lreq->osd != osd); | |
2213 | dout("%s osd %p osd%d lreq %p linger_id %llu\n", __func__, osd, | |
2214 | osd->o_osd, lreq, lreq->linger_id); | |
2215 | ||
2216 | lreq->osd = NULL; | |
2217 | erase_linger(&osd->o_linger_requests, lreq); | |
2218 | put_osd(osd); | |
2219 | ||
2220 | if (!osd_homeless(osd)) | |
2221 | maybe_move_osd_to_lru(osd); | |
2222 | else | |
2223 | atomic_dec(&osd->o_osdc->num_homeless); | |
2224 | } | |
2225 | ||
2226 | static bool __linger_registered(struct ceph_osd_linger_request *lreq) | |
2227 | { | |
2228 | verify_osdc_locked(lreq->osdc); | |
2229 | ||
2230 | return !RB_EMPTY_NODE(&lreq->osdc_node); | |
2231 | } | |
2232 | ||
2233 | static bool linger_registered(struct ceph_osd_linger_request *lreq) | |
2234 | { | |
2235 | struct ceph_osd_client *osdc = lreq->osdc; | |
2236 | bool registered; | |
2237 | ||
2238 | down_read(&osdc->lock); | |
2239 | registered = __linger_registered(lreq); | |
2240 | up_read(&osdc->lock); | |
2241 | ||
2242 | return registered; | |
2243 | } | |
2244 | ||
2245 | static void linger_register(struct ceph_osd_linger_request *lreq) | |
2246 | { | |
2247 | struct ceph_osd_client *osdc = lreq->osdc; | |
2248 | ||
2249 | verify_osdc_wrlocked(osdc); | |
2250 | WARN_ON(lreq->linger_id); | |
2251 | ||
2252 | linger_get(lreq); | |
2253 | lreq->linger_id = ++osdc->last_linger_id; | |
2254 | insert_linger_osdc(&osdc->linger_requests, lreq); | |
2255 | } | |
2256 | ||
2257 | static void linger_unregister(struct ceph_osd_linger_request *lreq) | |
2258 | { | |
2259 | struct ceph_osd_client *osdc = lreq->osdc; | |
2260 | ||
2261 | verify_osdc_wrlocked(osdc); | |
2262 | ||
2263 | erase_linger_osdc(&osdc->linger_requests, lreq); | |
2264 | linger_put(lreq); | |
2265 | } | |
2266 | ||
2267 | static void cancel_linger_request(struct ceph_osd_request *req) | |
2268 | { | |
2269 | struct ceph_osd_linger_request *lreq = req->r_priv; | |
2270 | ||
2271 | WARN_ON(!req->r_linger); | |
2272 | cancel_request(req); | |
2273 | linger_put(lreq); | |
2274 | } | |
2275 | ||
2276 | struct linger_work { | |
2277 | struct work_struct work; | |
2278 | struct ceph_osd_linger_request *lreq; | |
b07d3c4b ID |
2279 | struct list_head pending_item; |
2280 | unsigned long queued_stamp; | |
922dab61 ID |
2281 | |
2282 | union { | |
2283 | struct { | |
2284 | u64 notify_id; | |
2285 | u64 notifier_id; | |
2286 | void *payload; /* points into @msg front */ | |
2287 | size_t payload_len; | |
2288 | ||
2289 | struct ceph_msg *msg; /* for ceph_msg_put() */ | |
2290 | } notify; | |
2291 | struct { | |
2292 | int err; | |
2293 | } error; | |
2294 | }; | |
2295 | }; | |
2296 | ||
2297 | static struct linger_work *lwork_alloc(struct ceph_osd_linger_request *lreq, | |
2298 | work_func_t workfn) | |
2299 | { | |
2300 | struct linger_work *lwork; | |
2301 | ||
2302 | lwork = kzalloc(sizeof(*lwork), GFP_NOIO); | |
2303 | if (!lwork) | |
2304 | return NULL; | |
2305 | ||
2306 | INIT_WORK(&lwork->work, workfn); | |
b07d3c4b | 2307 | INIT_LIST_HEAD(&lwork->pending_item); |
922dab61 ID |
2308 | lwork->lreq = linger_get(lreq); |
2309 | ||
2310 | return lwork; | |
2311 | } | |
2312 | ||
2313 | static void lwork_free(struct linger_work *lwork) | |
2314 | { | |
2315 | struct ceph_osd_linger_request *lreq = lwork->lreq; | |
2316 | ||
b07d3c4b ID |
2317 | mutex_lock(&lreq->lock); |
2318 | list_del(&lwork->pending_item); | |
2319 | mutex_unlock(&lreq->lock); | |
2320 | ||
922dab61 ID |
2321 | linger_put(lreq); |
2322 | kfree(lwork); | |
2323 | } | |
2324 | ||
2325 | static void lwork_queue(struct linger_work *lwork) | |
2326 | { | |
2327 | struct ceph_osd_linger_request *lreq = lwork->lreq; | |
2328 | struct ceph_osd_client *osdc = lreq->osdc; | |
2329 | ||
2330 | verify_lreq_locked(lreq); | |
b07d3c4b ID |
2331 | WARN_ON(!list_empty(&lwork->pending_item)); |
2332 | ||
2333 | lwork->queued_stamp = jiffies; | |
2334 | list_add_tail(&lwork->pending_item, &lreq->pending_lworks); | |
922dab61 ID |
2335 | queue_work(osdc->notify_wq, &lwork->work); |
2336 | } | |
2337 | ||
2338 | static void do_watch_notify(struct work_struct *w) | |
2339 | { | |
2340 | struct linger_work *lwork = container_of(w, struct linger_work, work); | |
2341 | struct ceph_osd_linger_request *lreq = lwork->lreq; | |
2342 | ||
2343 | if (!linger_registered(lreq)) { | |
2344 | dout("%s lreq %p not registered\n", __func__, lreq); | |
2345 | goto out; | |
2346 | } | |
2347 | ||
19079203 | 2348 | WARN_ON(!lreq->is_watch); |
922dab61 ID |
2349 | dout("%s lreq %p notify_id %llu notifier_id %llu payload_len %zu\n", |
2350 | __func__, lreq, lwork->notify.notify_id, lwork->notify.notifier_id, | |
2351 | lwork->notify.payload_len); | |
2352 | lreq->wcb(lreq->data, lwork->notify.notify_id, lreq->linger_id, | |
2353 | lwork->notify.notifier_id, lwork->notify.payload, | |
2354 | lwork->notify.payload_len); | |
2355 | ||
2356 | out: | |
2357 | ceph_msg_put(lwork->notify.msg); | |
2358 | lwork_free(lwork); | |
2359 | } | |
2360 | ||
2361 | static void do_watch_error(struct work_struct *w) | |
2362 | { | |
2363 | struct linger_work *lwork = container_of(w, struct linger_work, work); | |
2364 | struct ceph_osd_linger_request *lreq = lwork->lreq; | |
2365 | ||
2366 | if (!linger_registered(lreq)) { | |
2367 | dout("%s lreq %p not registered\n", __func__, lreq); | |
2368 | goto out; | |
2369 | } | |
2370 | ||
2371 | dout("%s lreq %p err %d\n", __func__, lreq, lwork->error.err); | |
2372 | lreq->errcb(lreq->data, lreq->linger_id, lwork->error.err); | |
2373 | ||
2374 | out: | |
2375 | lwork_free(lwork); | |
2376 | } | |
2377 | ||
2378 | static void queue_watch_error(struct ceph_osd_linger_request *lreq) | |
2379 | { | |
2380 | struct linger_work *lwork; | |
2381 | ||
2382 | lwork = lwork_alloc(lreq, do_watch_error); | |
2383 | if (!lwork) { | |
2384 | pr_err("failed to allocate error-lwork\n"); | |
2385 | return; | |
2386 | } | |
2387 | ||
2388 | lwork->error.err = lreq->last_error; | |
2389 | lwork_queue(lwork); | |
2390 | } | |
2391 | ||
2392 | static void linger_reg_commit_complete(struct ceph_osd_linger_request *lreq, | |
2393 | int result) | |
2394 | { | |
2395 | if (!completion_done(&lreq->reg_commit_wait)) { | |
2396 | lreq->reg_commit_error = (result <= 0 ? result : 0); | |
2397 | complete_all(&lreq->reg_commit_wait); | |
2398 | } | |
2399 | } | |
2400 | ||
2401 | static void linger_commit_cb(struct ceph_osd_request *req) | |
2402 | { | |
2403 | struct ceph_osd_linger_request *lreq = req->r_priv; | |
2404 | ||
2405 | mutex_lock(&lreq->lock); | |
2406 | dout("%s lreq %p linger_id %llu result %d\n", __func__, lreq, | |
2407 | lreq->linger_id, req->r_result); | |
922dab61 ID |
2408 | linger_reg_commit_complete(lreq, req->r_result); |
2409 | lreq->committed = true; | |
2410 | ||
19079203 ID |
2411 | if (!lreq->is_watch) { |
2412 | struct ceph_osd_data *osd_data = | |
2413 | osd_req_op_data(req, 0, notify, response_data); | |
2414 | void *p = page_address(osd_data->pages[0]); | |
2415 | ||
2416 | WARN_ON(req->r_ops[0].op != CEPH_OSD_OP_NOTIFY || | |
2417 | osd_data->type != CEPH_OSD_DATA_TYPE_PAGES); | |
2418 | ||
2419 | /* make note of the notify_id */ | |
2420 | if (req->r_ops[0].outdata_len >= sizeof(u64)) { | |
2421 | lreq->notify_id = ceph_decode_64(&p); | |
2422 | dout("lreq %p notify_id %llu\n", lreq, | |
2423 | lreq->notify_id); | |
2424 | } else { | |
2425 | dout("lreq %p no notify_id\n", lreq); | |
2426 | } | |
2427 | } | |
2428 | ||
922dab61 ID |
2429 | mutex_unlock(&lreq->lock); |
2430 | linger_put(lreq); | |
2431 | } | |
2432 | ||
2433 | static int normalize_watch_error(int err) | |
2434 | { | |
2435 | /* | |
2436 | * Translate ENOENT -> ENOTCONN so that a delete->disconnection | |
2437 | * notification and a failure to reconnect because we raced with | |
2438 | * the delete appear the same to the user. | |
2439 | */ | |
2440 | if (err == -ENOENT) | |
2441 | err = -ENOTCONN; | |
2442 | ||
2443 | return err; | |
2444 | } | |
2445 | ||
2446 | static void linger_reconnect_cb(struct ceph_osd_request *req) | |
2447 | { | |
2448 | struct ceph_osd_linger_request *lreq = req->r_priv; | |
2449 | ||
2450 | mutex_lock(&lreq->lock); | |
2451 | dout("%s lreq %p linger_id %llu result %d last_error %d\n", __func__, | |
2452 | lreq, lreq->linger_id, req->r_result, lreq->last_error); | |
2453 | if (req->r_result < 0) { | |
2454 | if (!lreq->last_error) { | |
2455 | lreq->last_error = normalize_watch_error(req->r_result); | |
2456 | queue_watch_error(lreq); | |
2457 | } | |
2458 | } | |
2459 | ||
2460 | mutex_unlock(&lreq->lock); | |
2461 | linger_put(lreq); | |
2462 | } | |
2463 | ||
2464 | static void send_linger(struct ceph_osd_linger_request *lreq) | |
2465 | { | |
2466 | struct ceph_osd_request *req = lreq->reg_req; | |
2467 | struct ceph_osd_req_op *op = &req->r_ops[0]; | |
2468 | ||
2469 | verify_osdc_wrlocked(req->r_osdc); | |
2470 | dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); | |
2471 | ||
2472 | if (req->r_osd) | |
2473 | cancel_linger_request(req); | |
2474 | ||
2475 | request_reinit(req); | |
2476 | ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); | |
2477 | ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); | |
2478 | req->r_flags = lreq->t.flags; | |
2479 | req->r_mtime = lreq->mtime; | |
2480 | ||
2481 | mutex_lock(&lreq->lock); | |
19079203 | 2482 | if (lreq->is_watch && lreq->committed) { |
922dab61 ID |
2483 | WARN_ON(op->op != CEPH_OSD_OP_WATCH || |
2484 | op->watch.cookie != lreq->linger_id); | |
2485 | op->watch.op = CEPH_OSD_WATCH_OP_RECONNECT; | |
2486 | op->watch.gen = ++lreq->register_gen; | |
2487 | dout("lreq %p reconnect register_gen %u\n", lreq, | |
2488 | op->watch.gen); | |
2489 | req->r_callback = linger_reconnect_cb; | |
2490 | } else { | |
19079203 ID |
2491 | if (!lreq->is_watch) |
2492 | lreq->notify_id = 0; | |
2493 | else | |
2494 | WARN_ON(op->watch.op != CEPH_OSD_WATCH_OP_WATCH); | |
922dab61 ID |
2495 | dout("lreq %p register\n", lreq); |
2496 | req->r_callback = linger_commit_cb; | |
2497 | } | |
2498 | mutex_unlock(&lreq->lock); | |
2499 | ||
2500 | req->r_priv = linger_get(lreq); | |
2501 | req->r_linger = true; | |
2502 | ||
2503 | submit_request(req, true); | |
2504 | } | |
2505 | ||
2506 | static void linger_ping_cb(struct ceph_osd_request *req) | |
2507 | { | |
2508 | struct ceph_osd_linger_request *lreq = req->r_priv; | |
2509 | ||
2510 | mutex_lock(&lreq->lock); | |
2511 | dout("%s lreq %p linger_id %llu result %d ping_sent %lu last_error %d\n", | |
2512 | __func__, lreq, lreq->linger_id, req->r_result, lreq->ping_sent, | |
2513 | lreq->last_error); | |
2514 | if (lreq->register_gen == req->r_ops[0].watch.gen) { | |
b07d3c4b ID |
2515 | if (!req->r_result) { |
2516 | lreq->watch_valid_thru = lreq->ping_sent; | |
2517 | } else if (!lreq->last_error) { | |
922dab61 ID |
2518 | lreq->last_error = normalize_watch_error(req->r_result); |
2519 | queue_watch_error(lreq); | |
2520 | } | |
2521 | } else { | |
2522 | dout("lreq %p register_gen %u ignoring old pong %u\n", lreq, | |
2523 | lreq->register_gen, req->r_ops[0].watch.gen); | |
2524 | } | |
2525 | ||
2526 | mutex_unlock(&lreq->lock); | |
2527 | linger_put(lreq); | |
2528 | } | |
2529 | ||
2530 | static void send_linger_ping(struct ceph_osd_linger_request *lreq) | |
2531 | { | |
2532 | struct ceph_osd_client *osdc = lreq->osdc; | |
2533 | struct ceph_osd_request *req = lreq->ping_req; | |
2534 | struct ceph_osd_req_op *op = &req->r_ops[0]; | |
2535 | ||
b7ec35b3 | 2536 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD)) { |
922dab61 ID |
2537 | dout("%s PAUSERD\n", __func__); |
2538 | return; | |
2539 | } | |
2540 | ||
2541 | lreq->ping_sent = jiffies; | |
2542 | dout("%s lreq %p linger_id %llu ping_sent %lu register_gen %u\n", | |
2543 | __func__, lreq, lreq->linger_id, lreq->ping_sent, | |
2544 | lreq->register_gen); | |
2545 | ||
2546 | if (req->r_osd) | |
2547 | cancel_linger_request(req); | |
2548 | ||
2549 | request_reinit(req); | |
2550 | target_copy(&req->r_t, &lreq->t); | |
2551 | ||
2552 | WARN_ON(op->op != CEPH_OSD_OP_WATCH || | |
2553 | op->watch.cookie != lreq->linger_id || | |
2554 | op->watch.op != CEPH_OSD_WATCH_OP_PING); | |
2555 | op->watch.gen = lreq->register_gen; | |
2556 | req->r_callback = linger_ping_cb; | |
2557 | req->r_priv = linger_get(lreq); | |
2558 | req->r_linger = true; | |
2559 | ||
2560 | ceph_osdc_get_request(req); | |
2561 | account_request(req); | |
2562 | req->r_tid = atomic64_inc_return(&osdc->last_tid); | |
2563 | link_request(lreq->osd, req); | |
2564 | send_request(req); | |
2565 | } | |
2566 | ||
2567 | static void linger_submit(struct ceph_osd_linger_request *lreq) | |
2568 | { | |
2569 | struct ceph_osd_client *osdc = lreq->osdc; | |
2570 | struct ceph_osd *osd; | |
2571 | ||
7de030d6 | 2572 | calc_target(osdc, &lreq->t, NULL, false); |
922dab61 ID |
2573 | osd = lookup_create_osd(osdc, lreq->t.osd, true); |
2574 | link_linger(osd, lreq); | |
2575 | ||
2576 | send_linger(lreq); | |
2577 | } | |
2578 | ||
4609245e ID |
2579 | static void cancel_linger_map_check(struct ceph_osd_linger_request *lreq) |
2580 | { | |
2581 | struct ceph_osd_client *osdc = lreq->osdc; | |
2582 | struct ceph_osd_linger_request *lookup_lreq; | |
2583 | ||
2584 | verify_osdc_wrlocked(osdc); | |
2585 | ||
2586 | lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, | |
2587 | lreq->linger_id); | |
2588 | if (!lookup_lreq) | |
2589 | return; | |
2590 | ||
2591 | WARN_ON(lookup_lreq != lreq); | |
2592 | erase_linger_mc(&osdc->linger_map_checks, lreq); | |
2593 | linger_put(lreq); | |
2594 | } | |
2595 | ||
922dab61 ID |
2596 | /* |
2597 | * @lreq has to be both registered and linked. | |
2598 | */ | |
2599 | static void __linger_cancel(struct ceph_osd_linger_request *lreq) | |
2600 | { | |
19079203 | 2601 | if (lreq->is_watch && lreq->ping_req->r_osd) |
922dab61 ID |
2602 | cancel_linger_request(lreq->ping_req); |
2603 | if (lreq->reg_req->r_osd) | |
2604 | cancel_linger_request(lreq->reg_req); | |
4609245e | 2605 | cancel_linger_map_check(lreq); |
922dab61 ID |
2606 | unlink_linger(lreq->osd, lreq); |
2607 | linger_unregister(lreq); | |
2608 | } | |
2609 | ||
2610 | static void linger_cancel(struct ceph_osd_linger_request *lreq) | |
2611 | { | |
2612 | struct ceph_osd_client *osdc = lreq->osdc; | |
2613 | ||
2614 | down_write(&osdc->lock); | |
2615 | if (__linger_registered(lreq)) | |
2616 | __linger_cancel(lreq); | |
2617 | up_write(&osdc->lock); | |
2618 | } | |
2619 | ||
4609245e ID |
2620 | static void send_linger_map_check(struct ceph_osd_linger_request *lreq); |
2621 | ||
2622 | static void check_linger_pool_dne(struct ceph_osd_linger_request *lreq) | |
2623 | { | |
2624 | struct ceph_osd_client *osdc = lreq->osdc; | |
2625 | struct ceph_osdmap *map = osdc->osdmap; | |
2626 | ||
2627 | verify_osdc_wrlocked(osdc); | |
2628 | WARN_ON(!map->epoch); | |
2629 | ||
2630 | if (lreq->register_gen) { | |
2631 | lreq->map_dne_bound = map->epoch; | |
2632 | dout("%s lreq %p linger_id %llu pool disappeared\n", __func__, | |
2633 | lreq, lreq->linger_id); | |
2634 | } else { | |
2635 | dout("%s lreq %p linger_id %llu map_dne_bound %u have %u\n", | |
2636 | __func__, lreq, lreq->linger_id, lreq->map_dne_bound, | |
2637 | map->epoch); | |
2638 | } | |
2639 | ||
2640 | if (lreq->map_dne_bound) { | |
2641 | if (map->epoch >= lreq->map_dne_bound) { | |
2642 | /* we had a new enough map */ | |
2643 | pr_info("linger_id %llu pool does not exist\n", | |
2644 | lreq->linger_id); | |
2645 | linger_reg_commit_complete(lreq, -ENOENT); | |
2646 | __linger_cancel(lreq); | |
2647 | } | |
2648 | } else { | |
2649 | send_linger_map_check(lreq); | |
2650 | } | |
2651 | } | |
2652 | ||
2653 | static void linger_map_check_cb(struct ceph_mon_generic_request *greq) | |
2654 | { | |
2655 | struct ceph_osd_client *osdc = &greq->monc->client->osdc; | |
2656 | struct ceph_osd_linger_request *lreq; | |
2657 | u64 linger_id = greq->private_data; | |
2658 | ||
2659 | WARN_ON(greq->result || !greq->u.newest); | |
2660 | ||
2661 | down_write(&osdc->lock); | |
2662 | lreq = lookup_linger_mc(&osdc->linger_map_checks, linger_id); | |
2663 | if (!lreq) { | |
2664 | dout("%s linger_id %llu dne\n", __func__, linger_id); | |
2665 | goto out_unlock; | |
2666 | } | |
2667 | ||
2668 | dout("%s lreq %p linger_id %llu map_dne_bound %u newest %llu\n", | |
2669 | __func__, lreq, lreq->linger_id, lreq->map_dne_bound, | |
2670 | greq->u.newest); | |
2671 | if (!lreq->map_dne_bound) | |
2672 | lreq->map_dne_bound = greq->u.newest; | |
2673 | erase_linger_mc(&osdc->linger_map_checks, lreq); | |
2674 | check_linger_pool_dne(lreq); | |
2675 | ||
2676 | linger_put(lreq); | |
2677 | out_unlock: | |
2678 | up_write(&osdc->lock); | |
2679 | } | |
2680 | ||
2681 | static void send_linger_map_check(struct ceph_osd_linger_request *lreq) | |
2682 | { | |
2683 | struct ceph_osd_client *osdc = lreq->osdc; | |
2684 | struct ceph_osd_linger_request *lookup_lreq; | |
2685 | int ret; | |
2686 | ||
2687 | verify_osdc_wrlocked(osdc); | |
2688 | ||
2689 | lookup_lreq = lookup_linger_mc(&osdc->linger_map_checks, | |
2690 | lreq->linger_id); | |
2691 | if (lookup_lreq) { | |
2692 | WARN_ON(lookup_lreq != lreq); | |
2693 | return; | |
2694 | } | |
2695 | ||
2696 | linger_get(lreq); | |
2697 | insert_linger_mc(&osdc->linger_map_checks, lreq); | |
2698 | ret = ceph_monc_get_version_async(&osdc->client->monc, "osdmap", | |
2699 | linger_map_check_cb, lreq->linger_id); | |
2700 | WARN_ON(ret); | |
2701 | } | |
2702 | ||
922dab61 ID |
2703 | static int linger_reg_commit_wait(struct ceph_osd_linger_request *lreq) |
2704 | { | |
2705 | int ret; | |
2706 | ||
2707 | dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); | |
2708 | ret = wait_for_completion_interruptible(&lreq->reg_commit_wait); | |
2709 | return ret ?: lreq->reg_commit_error; | |
2710 | } | |
2711 | ||
19079203 ID |
2712 | static int linger_notify_finish_wait(struct ceph_osd_linger_request *lreq) |
2713 | { | |
2714 | int ret; | |
2715 | ||
2716 | dout("%s lreq %p linger_id %llu\n", __func__, lreq, lreq->linger_id); | |
2717 | ret = wait_for_completion_interruptible(&lreq->notify_finish_wait); | |
2718 | return ret ?: lreq->notify_finish_error; | |
2719 | } | |
2720 | ||
f24e9980 | 2721 | /* |
fbca9635 ID |
2722 | * Timeout callback, called every N seconds. When 1 or more OSD |
2723 | * requests has been active for more than N seconds, we send a keepalive | |
2724 | * (tag + timestamp) to its OSD to ensure any communications channel | |
2725 | * reset is detected. | |
f24e9980 SW |
2726 | */ |
2727 | static void handle_timeout(struct work_struct *work) | |
2728 | { | |
2729 | struct ceph_osd_client *osdc = | |
2730 | container_of(work, struct ceph_osd_client, timeout_work.work); | |
a319bf56 | 2731 | struct ceph_options *opts = osdc->client->options; |
5aea3dcd | 2732 | unsigned long cutoff = jiffies - opts->osd_keepalive_timeout; |
7cc5e38f | 2733 | unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout; |
5aea3dcd ID |
2734 | LIST_HEAD(slow_osds); |
2735 | struct rb_node *n, *p; | |
f24e9980 | 2736 | |
5aea3dcd ID |
2737 | dout("%s osdc %p\n", __func__, osdc); |
2738 | down_write(&osdc->lock); | |
f24e9980 | 2739 | |
422d2cb8 YS |
2740 | /* |
2741 | * ping osds that are a bit slow. this ensures that if there | |
2742 | * is a break in the TCP connection we will notice, and reopen | |
2743 | * a connection with that osd (from the fault callback). | |
2744 | */ | |
5aea3dcd ID |
2745 | for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { |
2746 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
2747 | bool found = false; | |
2748 | ||
7cc5e38f | 2749 | for (p = rb_first(&osd->o_requests); p; ) { |
5aea3dcd ID |
2750 | struct ceph_osd_request *req = |
2751 | rb_entry(p, struct ceph_osd_request, r_node); | |
2752 | ||
7cc5e38f ID |
2753 | p = rb_next(p); /* abort_request() */ |
2754 | ||
5aea3dcd ID |
2755 | if (time_before(req->r_stamp, cutoff)) { |
2756 | dout(" req %p tid %llu on osd%d is laggy\n", | |
2757 | req, req->r_tid, osd->o_osd); | |
2758 | found = true; | |
2759 | } | |
7cc5e38f ID |
2760 | if (opts->osd_request_timeout && |
2761 | time_before(req->r_start_stamp, expiry_cutoff)) { | |
2762 | pr_err_ratelimited("tid %llu on osd%d timeout\n", | |
2763 | req->r_tid, osd->o_osd); | |
2764 | abort_request(req, -ETIMEDOUT); | |
2765 | } | |
5aea3dcd | 2766 | } |
922dab61 ID |
2767 | for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) { |
2768 | struct ceph_osd_linger_request *lreq = | |
2769 | rb_entry(p, struct ceph_osd_linger_request, node); | |
2770 | ||
2771 | dout(" lreq %p linger_id %llu is served by osd%d\n", | |
2772 | lreq, lreq->linger_id, osd->o_osd); | |
2773 | found = true; | |
2774 | ||
2775 | mutex_lock(&lreq->lock); | |
19079203 | 2776 | if (lreq->is_watch && lreq->committed && !lreq->last_error) |
922dab61 ID |
2777 | send_linger_ping(lreq); |
2778 | mutex_unlock(&lreq->lock); | |
2779 | } | |
422d2cb8 | 2780 | |
5aea3dcd ID |
2781 | if (found) |
2782 | list_move_tail(&osd->o_keepalive_item, &slow_osds); | |
422d2cb8 | 2783 | } |
5aea3dcd | 2784 | |
7cc5e38f ID |
2785 | if (opts->osd_request_timeout) { |
2786 | for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) { | |
2787 | struct ceph_osd_request *req = | |
2788 | rb_entry(p, struct ceph_osd_request, r_node); | |
2789 | ||
2790 | p = rb_next(p); /* abort_request() */ | |
2791 | ||
2792 | if (time_before(req->r_start_stamp, expiry_cutoff)) { | |
2793 | pr_err_ratelimited("tid %llu on osd%d timeout\n", | |
2794 | req->r_tid, osdc->homeless_osd.o_osd); | |
2795 | abort_request(req, -ETIMEDOUT); | |
2796 | } | |
2797 | } | |
2798 | } | |
2799 | ||
5aea3dcd ID |
2800 | if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds)) |
2801 | maybe_request_map(osdc); | |
2802 | ||
422d2cb8 | 2803 | while (!list_empty(&slow_osds)) { |
5aea3dcd ID |
2804 | struct ceph_osd *osd = list_first_entry(&slow_osds, |
2805 | struct ceph_osd, | |
2806 | o_keepalive_item); | |
422d2cb8 | 2807 | list_del_init(&osd->o_keepalive_item); |
f24e9980 SW |
2808 | ceph_con_keepalive(&osd->o_con); |
2809 | } | |
2810 | ||
5aea3dcd | 2811 | up_write(&osdc->lock); |
fbca9635 ID |
2812 | schedule_delayed_work(&osdc->timeout_work, |
2813 | osdc->client->options->osd_keepalive_timeout); | |
f24e9980 SW |
2814 | } |
2815 | ||
f5a2041b YS |
2816 | static void handle_osds_timeout(struct work_struct *work) |
2817 | { | |
2818 | struct ceph_osd_client *osdc = | |
2819 | container_of(work, struct ceph_osd_client, | |
2820 | osds_timeout_work.work); | |
a319bf56 | 2821 | unsigned long delay = osdc->client->options->osd_idle_ttl / 4; |
42a2c09f | 2822 | struct ceph_osd *osd, *nosd; |
f5a2041b | 2823 | |
42a2c09f | 2824 | dout("%s osdc %p\n", __func__, osdc); |
5aea3dcd | 2825 | down_write(&osdc->lock); |
42a2c09f ID |
2826 | list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) { |
2827 | if (time_before(jiffies, osd->lru_ttl)) | |
2828 | break; | |
2829 | ||
5aea3dcd | 2830 | WARN_ON(!RB_EMPTY_ROOT(&osd->o_requests)); |
922dab61 | 2831 | WARN_ON(!RB_EMPTY_ROOT(&osd->o_linger_requests)); |
5aea3dcd | 2832 | close_osd(osd); |
42a2c09f | 2833 | } |
f5a2041b | 2834 | |
5aea3dcd | 2835 | up_write(&osdc->lock); |
f5a2041b YS |
2836 | schedule_delayed_work(&osdc->osds_timeout_work, |
2837 | round_jiffies_relative(delay)); | |
2838 | } | |
2839 | ||
205ee118 ID |
2840 | static int ceph_oloc_decode(void **p, void *end, |
2841 | struct ceph_object_locator *oloc) | |
2842 | { | |
2843 | u8 struct_v, struct_cv; | |
2844 | u32 len; | |
2845 | void *struct_end; | |
2846 | int ret = 0; | |
2847 | ||
2848 | ceph_decode_need(p, end, 1 + 1 + 4, e_inval); | |
2849 | struct_v = ceph_decode_8(p); | |
2850 | struct_cv = ceph_decode_8(p); | |
2851 | if (struct_v < 3) { | |
2852 | pr_warn("got v %d < 3 cv %d of ceph_object_locator\n", | |
2853 | struct_v, struct_cv); | |
2854 | goto e_inval; | |
2855 | } | |
2856 | if (struct_cv > 6) { | |
2857 | pr_warn("got v %d cv %d > 6 of ceph_object_locator\n", | |
2858 | struct_v, struct_cv); | |
2859 | goto e_inval; | |
2860 | } | |
2861 | len = ceph_decode_32(p); | |
2862 | ceph_decode_need(p, end, len, e_inval); | |
2863 | struct_end = *p + len; | |
2864 | ||
2865 | oloc->pool = ceph_decode_64(p); | |
2866 | *p += 4; /* skip preferred */ | |
2867 | ||
2868 | len = ceph_decode_32(p); | |
2869 | if (len > 0) { | |
2870 | pr_warn("ceph_object_locator::key is set\n"); | |
2871 | goto e_inval; | |
2872 | } | |
2873 | ||
2874 | if (struct_v >= 5) { | |
cd08e0a2 YZ |
2875 | bool changed = false; |
2876 | ||
205ee118 ID |
2877 | len = ceph_decode_32(p); |
2878 | if (len > 0) { | |
30c156d9 | 2879 | ceph_decode_need(p, end, len, e_inval); |
cd08e0a2 YZ |
2880 | if (!oloc->pool_ns || |
2881 | ceph_compare_string(oloc->pool_ns, *p, len)) | |
2882 | changed = true; | |
30c156d9 | 2883 | *p += len; |
cd08e0a2 YZ |
2884 | } else { |
2885 | if (oloc->pool_ns) | |
2886 | changed = true; | |
2887 | } | |
2888 | if (changed) { | |
2889 | /* redirect changes namespace */ | |
2890 | pr_warn("ceph_object_locator::nspace is changed\n"); | |
2891 | goto e_inval; | |
205ee118 ID |
2892 | } |
2893 | } | |
2894 | ||
2895 | if (struct_v >= 6) { | |
2896 | s64 hash = ceph_decode_64(p); | |
2897 | if (hash != -1) { | |
2898 | pr_warn("ceph_object_locator::hash is set\n"); | |
2899 | goto e_inval; | |
2900 | } | |
2901 | } | |
2902 | ||
2903 | /* skip the rest */ | |
2904 | *p = struct_end; | |
2905 | out: | |
2906 | return ret; | |
2907 | ||
2908 | e_inval: | |
2909 | ret = -EINVAL; | |
2910 | goto out; | |
2911 | } | |
2912 | ||
2913 | static int ceph_redirect_decode(void **p, void *end, | |
2914 | struct ceph_request_redirect *redir) | |
2915 | { | |
2916 | u8 struct_v, struct_cv; | |
2917 | u32 len; | |
2918 | void *struct_end; | |
2919 | int ret; | |
2920 | ||
2921 | ceph_decode_need(p, end, 1 + 1 + 4, e_inval); | |
2922 | struct_v = ceph_decode_8(p); | |
2923 | struct_cv = ceph_decode_8(p); | |
2924 | if (struct_cv > 1) { | |
2925 | pr_warn("got v %d cv %d > 1 of ceph_request_redirect\n", | |
2926 | struct_v, struct_cv); | |
2927 | goto e_inval; | |
2928 | } | |
2929 | len = ceph_decode_32(p); | |
2930 | ceph_decode_need(p, end, len, e_inval); | |
2931 | struct_end = *p + len; | |
2932 | ||
2933 | ret = ceph_oloc_decode(p, end, &redir->oloc); | |
2934 | if (ret) | |
2935 | goto out; | |
2936 | ||
2937 | len = ceph_decode_32(p); | |
2938 | if (len > 0) { | |
2939 | pr_warn("ceph_request_redirect::object_name is set\n"); | |
2940 | goto e_inval; | |
2941 | } | |
2942 | ||
2943 | len = ceph_decode_32(p); | |
2944 | *p += len; /* skip osd_instructions */ | |
2945 | ||
2946 | /* skip the rest */ | |
2947 | *p = struct_end; | |
2948 | out: | |
2949 | return ret; | |
2950 | ||
2951 | e_inval: | |
2952 | ret = -EINVAL; | |
2953 | goto out; | |
2954 | } | |
2955 | ||
fe5da05e ID |
2956 | struct MOSDOpReply { |
2957 | struct ceph_pg pgid; | |
2958 | u64 flags; | |
2959 | int result; | |
2960 | u32 epoch; | |
2961 | int num_ops; | |
2962 | u32 outdata_len[CEPH_OSD_MAX_OPS]; | |
2963 | s32 rval[CEPH_OSD_MAX_OPS]; | |
2964 | int retry_attempt; | |
2965 | struct ceph_eversion replay_version; | |
2966 | u64 user_version; | |
2967 | struct ceph_request_redirect redirect; | |
2968 | }; | |
25845472 | 2969 | |
fe5da05e | 2970 | static int decode_MOSDOpReply(const struct ceph_msg *msg, struct MOSDOpReply *m) |
f24e9980 | 2971 | { |
fe5da05e ID |
2972 | void *p = msg->front.iov_base; |
2973 | void *const end = p + msg->front.iov_len; | |
2974 | u16 version = le16_to_cpu(msg->hdr.version); | |
2975 | struct ceph_eversion bad_replay_version; | |
b0b31a8f | 2976 | u8 decode_redir; |
fe5da05e ID |
2977 | u32 len; |
2978 | int ret; | |
2979 | int i; | |
1b83bef2 | 2980 | |
fe5da05e ID |
2981 | ceph_decode_32_safe(&p, end, len, e_inval); |
2982 | ceph_decode_need(&p, end, len, e_inval); | |
2983 | p += len; /* skip oid */ | |
1b83bef2 | 2984 | |
fe5da05e ID |
2985 | ret = ceph_decode_pgid(&p, end, &m->pgid); |
2986 | if (ret) | |
2987 | return ret; | |
1b83bef2 | 2988 | |
fe5da05e ID |
2989 | ceph_decode_64_safe(&p, end, m->flags, e_inval); |
2990 | ceph_decode_32_safe(&p, end, m->result, e_inval); | |
2991 | ceph_decode_need(&p, end, sizeof(bad_replay_version), e_inval); | |
2992 | memcpy(&bad_replay_version, p, sizeof(bad_replay_version)); | |
2993 | p += sizeof(bad_replay_version); | |
2994 | ceph_decode_32_safe(&p, end, m->epoch, e_inval); | |
1b83bef2 | 2995 | |
fe5da05e ID |
2996 | ceph_decode_32_safe(&p, end, m->num_ops, e_inval); |
2997 | if (m->num_ops > ARRAY_SIZE(m->outdata_len)) | |
2998 | goto e_inval; | |
1b83bef2 | 2999 | |
fe5da05e ID |
3000 | ceph_decode_need(&p, end, m->num_ops * sizeof(struct ceph_osd_op), |
3001 | e_inval); | |
3002 | for (i = 0; i < m->num_ops; i++) { | |
1b83bef2 | 3003 | struct ceph_osd_op *op = p; |
1b83bef2 | 3004 | |
fe5da05e | 3005 | m->outdata_len[i] = le32_to_cpu(op->payload_len); |
1b83bef2 SW |
3006 | p += sizeof(*op); |
3007 | } | |
1b83bef2 | 3008 | |
fe5da05e ID |
3009 | ceph_decode_32_safe(&p, end, m->retry_attempt, e_inval); |
3010 | for (i = 0; i < m->num_ops; i++) | |
3011 | ceph_decode_32_safe(&p, end, m->rval[i], e_inval); | |
f24e9980 | 3012 | |
fe5da05e ID |
3013 | if (version >= 5) { |
3014 | ceph_decode_need(&p, end, sizeof(m->replay_version), e_inval); | |
3015 | memcpy(&m->replay_version, p, sizeof(m->replay_version)); | |
3016 | p += sizeof(m->replay_version); | |
3017 | ceph_decode_64_safe(&p, end, m->user_version, e_inval); | |
3018 | } else { | |
3019 | m->replay_version = bad_replay_version; /* struct */ | |
3020 | m->user_version = le64_to_cpu(m->replay_version.version); | |
3021 | } | |
eb845ff1 | 3022 | |
fe5da05e ID |
3023 | if (version >= 6) { |
3024 | if (version >= 7) | |
3025 | ceph_decode_8_safe(&p, end, decode_redir, e_inval); | |
b0b31a8f ID |
3026 | else |
3027 | decode_redir = 1; | |
3028 | } else { | |
3029 | decode_redir = 0; | |
3030 | } | |
3031 | ||
3032 | if (decode_redir) { | |
fe5da05e ID |
3033 | ret = ceph_redirect_decode(&p, end, &m->redirect); |
3034 | if (ret) | |
3035 | return ret; | |
205ee118 | 3036 | } else { |
fe5da05e | 3037 | ceph_oloc_init(&m->redirect.oloc); |
205ee118 | 3038 | } |
f24e9980 | 3039 | |
fe5da05e ID |
3040 | return 0; |
3041 | ||
3042 | e_inval: | |
3043 | return -EINVAL; | |
3044 | } | |
3045 | ||
3046 | /* | |
b18b9550 ID |
3047 | * Handle MOSDOpReply. Set ->r_result and call the callback if it is |
3048 | * specified. | |
fe5da05e | 3049 | */ |
5aea3dcd | 3050 | static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg) |
fe5da05e | 3051 | { |
5aea3dcd | 3052 | struct ceph_osd_client *osdc = osd->o_osdc; |
fe5da05e ID |
3053 | struct ceph_osd_request *req; |
3054 | struct MOSDOpReply m; | |
3055 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
3056 | u32 data_len = 0; | |
fe5da05e ID |
3057 | int ret; |
3058 | int i; | |
3059 | ||
3060 | dout("%s msg %p tid %llu\n", __func__, msg, tid); | |
3061 | ||
5aea3dcd ID |
3062 | down_read(&osdc->lock); |
3063 | if (!osd_registered(osd)) { | |
3064 | dout("%s osd%d unknown\n", __func__, osd->o_osd); | |
3065 | goto out_unlock_osdc; | |
3066 | } | |
3067 | WARN_ON(osd->o_osd != le64_to_cpu(msg->hdr.src.num)); | |
3068 | ||
3069 | mutex_lock(&osd->lock); | |
3070 | req = lookup_request(&osd->o_requests, tid); | |
fe5da05e | 3071 | if (!req) { |
5aea3dcd ID |
3072 | dout("%s osd%d tid %llu unknown\n", __func__, osd->o_osd, tid); |
3073 | goto out_unlock_session; | |
fe5da05e | 3074 | } |
fe5da05e | 3075 | |
cd08e0a2 | 3076 | m.redirect.oloc.pool_ns = req->r_t.target_oloc.pool_ns; |
fe5da05e | 3077 | ret = decode_MOSDOpReply(msg, &m); |
cd08e0a2 | 3078 | m.redirect.oloc.pool_ns = NULL; |
fe5da05e ID |
3079 | if (ret) { |
3080 | pr_err("failed to decode MOSDOpReply for tid %llu: %d\n", | |
3081 | req->r_tid, ret); | |
3082 | ceph_msg_dump(msg); | |
3083 | goto fail_request; | |
3084 | } | |
3085 | dout("%s req %p tid %llu flags 0x%llx pgid %llu.%x epoch %u attempt %d v %u'%llu uv %llu\n", | |
3086 | __func__, req, req->r_tid, m.flags, m.pgid.pool, m.pgid.seed, | |
3087 | m.epoch, m.retry_attempt, le32_to_cpu(m.replay_version.epoch), | |
3088 | le64_to_cpu(m.replay_version.version), m.user_version); | |
3089 | ||
3090 | if (m.retry_attempt >= 0) { | |
3091 | if (m.retry_attempt != req->r_attempts - 1) { | |
3092 | dout("req %p tid %llu retry_attempt %d != %d, ignoring\n", | |
3093 | req, req->r_tid, m.retry_attempt, | |
3094 | req->r_attempts - 1); | |
5aea3dcd | 3095 | goto out_unlock_session; |
fe5da05e ID |
3096 | } |
3097 | } else { | |
3098 | WARN_ON(1); /* MOSDOpReply v4 is assumed */ | |
3099 | } | |
3100 | ||
3101 | if (!ceph_oloc_empty(&m.redirect.oloc)) { | |
3102 | dout("req %p tid %llu redirect pool %lld\n", req, req->r_tid, | |
3103 | m.redirect.oloc.pool); | |
5aea3dcd ID |
3104 | unlink_request(osd, req); |
3105 | mutex_unlock(&osd->lock); | |
205ee118 | 3106 | |
30c156d9 YZ |
3107 | /* |
3108 | * Not ceph_oloc_copy() - changing pool_ns is not | |
3109 | * supported. | |
3110 | */ | |
3111 | req->r_t.target_oloc.pool = m.redirect.oloc.pool; | |
5aea3dcd ID |
3112 | req->r_flags |= CEPH_OSD_FLAG_REDIRECTED; |
3113 | req->r_tid = 0; | |
3114 | __submit_request(req, false); | |
3115 | goto out_unlock_osdc; | |
205ee118 ID |
3116 | } |
3117 | ||
fe5da05e ID |
3118 | if (m.num_ops != req->r_num_ops) { |
3119 | pr_err("num_ops %d != %d for tid %llu\n", m.num_ops, | |
3120 | req->r_num_ops, req->r_tid); | |
3121 | goto fail_request; | |
f24e9980 | 3122 | } |
fe5da05e ID |
3123 | for (i = 0; i < req->r_num_ops; i++) { |
3124 | dout(" req %p tid %llu op %d rval %d len %u\n", req, | |
3125 | req->r_tid, i, m.rval[i], m.outdata_len[i]); | |
3126 | req->r_ops[i].rval = m.rval[i]; | |
3127 | req->r_ops[i].outdata_len = m.outdata_len[i]; | |
3128 | data_len += m.outdata_len[i]; | |
3129 | } | |
3130 | if (data_len != le32_to_cpu(msg->hdr.data_len)) { | |
3131 | pr_err("sum of lens %u != %u for tid %llu\n", data_len, | |
3132 | le32_to_cpu(msg->hdr.data_len), req->r_tid); | |
3133 | goto fail_request; | |
3134 | } | |
b18b9550 ID |
3135 | dout("%s req %p tid %llu result %d data_len %u\n", __func__, |
3136 | req, req->r_tid, m.result, data_len); | |
f24e9980 | 3137 | |
b18b9550 ID |
3138 | /* |
3139 | * Since we only ever request ONDISK, we should only ever get | |
3140 | * one (type of) reply back. | |
3141 | */ | |
3142 | WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK)); | |
3143 | req->r_result = m.result ?: data_len; | |
3144 | finish_request(req); | |
5aea3dcd ID |
3145 | mutex_unlock(&osd->lock); |
3146 | up_read(&osdc->lock); | |
f24e9980 | 3147 | |
b18b9550 ID |
3148 | __complete_request(req); |
3149 | complete_all(&req->r_completion); | |
3150 | ceph_osdc_put_request(req); | |
f24e9980 SW |
3151 | return; |
3152 | ||
fe5da05e | 3153 | fail_request: |
4609245e | 3154 | complete_request(req, -EIO); |
5aea3dcd ID |
3155 | out_unlock_session: |
3156 | mutex_unlock(&osd->lock); | |
3157 | out_unlock_osdc: | |
3158 | up_read(&osdc->lock); | |
f24e9980 SW |
3159 | } |
3160 | ||
42c1b124 ID |
3161 | static void set_pool_was_full(struct ceph_osd_client *osdc) |
3162 | { | |
3163 | struct rb_node *n; | |
3164 | ||
3165 | for (n = rb_first(&osdc->osdmap->pg_pools); n; n = rb_next(n)) { | |
3166 | struct ceph_pg_pool_info *pi = | |
3167 | rb_entry(n, struct ceph_pg_pool_info, node); | |
3168 | ||
3169 | pi->was_full = __pool_full(pi); | |
3170 | } | |
3171 | } | |
3172 | ||
5aea3dcd | 3173 | static bool pool_cleared_full(struct ceph_osd_client *osdc, s64 pool_id) |
f24e9980 | 3174 | { |
5aea3dcd | 3175 | struct ceph_pg_pool_info *pi; |
f24e9980 | 3176 | |
5aea3dcd ID |
3177 | pi = ceph_pg_pool_by_id(osdc->osdmap, pool_id); |
3178 | if (!pi) | |
3179 | return false; | |
f24e9980 | 3180 | |
5aea3dcd | 3181 | return pi->was_full && !__pool_full(pi); |
422d2cb8 YS |
3182 | } |
3183 | ||
922dab61 ID |
3184 | static enum calc_target_result |
3185 | recalc_linger_target(struct ceph_osd_linger_request *lreq) | |
3186 | { | |
3187 | struct ceph_osd_client *osdc = lreq->osdc; | |
3188 | enum calc_target_result ct_res; | |
3189 | ||
7de030d6 | 3190 | ct_res = calc_target(osdc, &lreq->t, NULL, true); |
922dab61 ID |
3191 | if (ct_res == CALC_TARGET_NEED_RESEND) { |
3192 | struct ceph_osd *osd; | |
3193 | ||
3194 | osd = lookup_create_osd(osdc, lreq->t.osd, true); | |
3195 | if (osd != lreq->osd) { | |
3196 | unlink_linger(lreq->osd, lreq); | |
3197 | link_linger(osd, lreq); | |
3198 | } | |
3199 | } | |
3200 | ||
3201 | return ct_res; | |
3202 | } | |
3203 | ||
422d2cb8 | 3204 | /* |
5aea3dcd | 3205 | * Requeue requests whose mapping to an OSD has changed. |
422d2cb8 | 3206 | */ |
5aea3dcd ID |
3207 | static void scan_requests(struct ceph_osd *osd, |
3208 | bool force_resend, | |
3209 | bool cleared_full, | |
3210 | bool check_pool_cleared_full, | |
3211 | struct rb_root *need_resend, | |
3212 | struct list_head *need_resend_linger) | |
422d2cb8 | 3213 | { |
5aea3dcd ID |
3214 | struct ceph_osd_client *osdc = osd->o_osdc; |
3215 | struct rb_node *n; | |
3216 | bool force_resend_writes; | |
3217 | ||
922dab61 ID |
3218 | for (n = rb_first(&osd->o_linger_requests); n; ) { |
3219 | struct ceph_osd_linger_request *lreq = | |
3220 | rb_entry(n, struct ceph_osd_linger_request, node); | |
3221 | enum calc_target_result ct_res; | |
3222 | ||
3223 | n = rb_next(n); /* recalc_linger_target() */ | |
3224 | ||
3225 | dout("%s lreq %p linger_id %llu\n", __func__, lreq, | |
3226 | lreq->linger_id); | |
3227 | ct_res = recalc_linger_target(lreq); | |
3228 | switch (ct_res) { | |
3229 | case CALC_TARGET_NO_ACTION: | |
3230 | force_resend_writes = cleared_full || | |
3231 | (check_pool_cleared_full && | |
3232 | pool_cleared_full(osdc, lreq->t.base_oloc.pool)); | |
3233 | if (!force_resend && !force_resend_writes) | |
3234 | break; | |
3235 | ||
3236 | /* fall through */ | |
3237 | case CALC_TARGET_NEED_RESEND: | |
4609245e | 3238 | cancel_linger_map_check(lreq); |
922dab61 ID |
3239 | /* |
3240 | * scan_requests() for the previous epoch(s) | |
3241 | * may have already added it to the list, since | |
3242 | * it's not unlinked here. | |
3243 | */ | |
3244 | if (list_empty(&lreq->scan_item)) | |
3245 | list_add_tail(&lreq->scan_item, need_resend_linger); | |
3246 | break; | |
3247 | case CALC_TARGET_POOL_DNE: | |
a10bcb19 | 3248 | list_del_init(&lreq->scan_item); |
4609245e | 3249 | check_linger_pool_dne(lreq); |
922dab61 ID |
3250 | break; |
3251 | } | |
3252 | } | |
3253 | ||
5aea3dcd ID |
3254 | for (n = rb_first(&osd->o_requests); n; ) { |
3255 | struct ceph_osd_request *req = | |
3256 | rb_entry(n, struct ceph_osd_request, r_node); | |
3257 | enum calc_target_result ct_res; | |
3258 | ||
4609245e | 3259 | n = rb_next(n); /* unlink_request(), check_pool_dne() */ |
5aea3dcd ID |
3260 | |
3261 | dout("%s req %p tid %llu\n", __func__, req, req->r_tid); | |
7de030d6 ID |
3262 | ct_res = calc_target(osdc, &req->r_t, &req->r_osd->o_con, |
3263 | false); | |
5aea3dcd ID |
3264 | switch (ct_res) { |
3265 | case CALC_TARGET_NO_ACTION: | |
3266 | force_resend_writes = cleared_full || | |
3267 | (check_pool_cleared_full && | |
3268 | pool_cleared_full(osdc, req->r_t.base_oloc.pool)); | |
3269 | if (!force_resend && | |
3270 | (!(req->r_flags & CEPH_OSD_FLAG_WRITE) || | |
3271 | !force_resend_writes)) | |
3272 | break; | |
3273 | ||
3274 | /* fall through */ | |
3275 | case CALC_TARGET_NEED_RESEND: | |
4609245e | 3276 | cancel_map_check(req); |
5aea3dcd ID |
3277 | unlink_request(osd, req); |
3278 | insert_request(need_resend, req); | |
3279 | break; | |
3280 | case CALC_TARGET_POOL_DNE: | |
4609245e | 3281 | check_pool_dne(req); |
5aea3dcd | 3282 | break; |
b0494532 | 3283 | } |
6f6c7006 | 3284 | } |
422d2cb8 | 3285 | } |
6f6c7006 | 3286 | |
42c1b124 | 3287 | static int handle_one_map(struct ceph_osd_client *osdc, |
5aea3dcd ID |
3288 | void *p, void *end, bool incremental, |
3289 | struct rb_root *need_resend, | |
3290 | struct list_head *need_resend_linger) | |
42c1b124 ID |
3291 | { |
3292 | struct ceph_osdmap *newmap; | |
3293 | struct rb_node *n; | |
3294 | bool skipped_map = false; | |
3295 | bool was_full; | |
3296 | ||
b7ec35b3 | 3297 | was_full = ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); |
42c1b124 ID |
3298 | set_pool_was_full(osdc); |
3299 | ||
3300 | if (incremental) | |
3301 | newmap = osdmap_apply_incremental(&p, end, osdc->osdmap); | |
3302 | else | |
3303 | newmap = ceph_osdmap_decode(&p, end); | |
3304 | if (IS_ERR(newmap)) | |
3305 | return PTR_ERR(newmap); | |
3306 | ||
3307 | if (newmap != osdc->osdmap) { | |
3308 | /* | |
3309 | * Preserve ->was_full before destroying the old map. | |
3310 | * For pools that weren't in the old map, ->was_full | |
3311 | * should be false. | |
3312 | */ | |
3313 | for (n = rb_first(&newmap->pg_pools); n; n = rb_next(n)) { | |
3314 | struct ceph_pg_pool_info *pi = | |
3315 | rb_entry(n, struct ceph_pg_pool_info, node); | |
3316 | struct ceph_pg_pool_info *old_pi; | |
3317 | ||
3318 | old_pi = ceph_pg_pool_by_id(osdc->osdmap, pi->id); | |
3319 | if (old_pi) | |
3320 | pi->was_full = old_pi->was_full; | |
3321 | else | |
3322 | WARN_ON(pi->was_full); | |
3323 | } | |
3324 | ||
3325 | if (osdc->osdmap->epoch && | |
3326 | osdc->osdmap->epoch + 1 < newmap->epoch) { | |
3327 | WARN_ON(incremental); | |
3328 | skipped_map = true; | |
3329 | } | |
3330 | ||
3331 | ceph_osdmap_destroy(osdc->osdmap); | |
3332 | osdc->osdmap = newmap; | |
3333 | } | |
3334 | ||
b7ec35b3 | 3335 | was_full &= !ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL); |
5aea3dcd ID |
3336 | scan_requests(&osdc->homeless_osd, skipped_map, was_full, true, |
3337 | need_resend, need_resend_linger); | |
3338 | ||
3339 | for (n = rb_first(&osdc->osds); n; ) { | |
3340 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
3341 | ||
3342 | n = rb_next(n); /* close_osd() */ | |
3343 | ||
3344 | scan_requests(osd, skipped_map, was_full, true, need_resend, | |
3345 | need_resend_linger); | |
3346 | if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) || | |
3347 | memcmp(&osd->o_con.peer_addr, | |
3348 | ceph_osd_addr(osdc->osdmap, osd->o_osd), | |
3349 | sizeof(struct ceph_entity_addr))) | |
3350 | close_osd(osd); | |
3351 | } | |
42c1b124 ID |
3352 | |
3353 | return 0; | |
3354 | } | |
6f6c7006 | 3355 | |
5aea3dcd ID |
3356 | static void kick_requests(struct ceph_osd_client *osdc, |
3357 | struct rb_root *need_resend, | |
3358 | struct list_head *need_resend_linger) | |
3359 | { | |
922dab61 | 3360 | struct ceph_osd_linger_request *lreq, *nlreq; |
04c7d789 | 3361 | enum calc_target_result ct_res; |
5aea3dcd ID |
3362 | struct rb_node *n; |
3363 | ||
04c7d789 ID |
3364 | /* make sure need_resend targets reflect latest map */ |
3365 | for (n = rb_first(need_resend); n; ) { | |
3366 | struct ceph_osd_request *req = | |
3367 | rb_entry(n, struct ceph_osd_request, r_node); | |
3368 | ||
3369 | n = rb_next(n); | |
3370 | ||
3371 | if (req->r_t.epoch < osdc->osdmap->epoch) { | |
3372 | ct_res = calc_target(osdc, &req->r_t, NULL, false); | |
3373 | if (ct_res == CALC_TARGET_POOL_DNE) { | |
3374 | erase_request(need_resend, req); | |
3375 | check_pool_dne(req); | |
3376 | } | |
3377 | } | |
3378 | } | |
3379 | ||
5aea3dcd ID |
3380 | for (n = rb_first(need_resend); n; ) { |
3381 | struct ceph_osd_request *req = | |
3382 | rb_entry(n, struct ceph_osd_request, r_node); | |
3383 | struct ceph_osd *osd; | |
3384 | ||
3385 | n = rb_next(n); | |
3386 | erase_request(need_resend, req); /* before link_request() */ | |
3387 | ||
5aea3dcd ID |
3388 | osd = lookup_create_osd(osdc, req->r_t.osd, true); |
3389 | link_request(osd, req); | |
3390 | if (!req->r_linger) { | |
3391 | if (!osd_homeless(osd) && !req->r_t.paused) | |
3392 | send_request(req); | |
922dab61 ID |
3393 | } else { |
3394 | cancel_linger_request(req); | |
5aea3dcd ID |
3395 | } |
3396 | } | |
922dab61 ID |
3397 | |
3398 | list_for_each_entry_safe(lreq, nlreq, need_resend_linger, scan_item) { | |
3399 | if (!osd_homeless(lreq->osd)) | |
3400 | send_linger(lreq); | |
3401 | ||
3402 | list_del_init(&lreq->scan_item); | |
3403 | } | |
5aea3dcd ID |
3404 | } |
3405 | ||
f24e9980 SW |
3406 | /* |
3407 | * Process updated osd map. | |
3408 | * | |
3409 | * The message contains any number of incremental and full maps, normally | |
3410 | * indicating some sort of topology change in the cluster. Kick requests | |
3411 | * off to different OSDs as needed. | |
3412 | */ | |
3413 | void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg) | |
3414 | { | |
42c1b124 ID |
3415 | void *p = msg->front.iov_base; |
3416 | void *const end = p + msg->front.iov_len; | |
f24e9980 SW |
3417 | u32 nr_maps, maplen; |
3418 | u32 epoch; | |
f24e9980 | 3419 | struct ceph_fsid fsid; |
5aea3dcd ID |
3420 | struct rb_root need_resend = RB_ROOT; |
3421 | LIST_HEAD(need_resend_linger); | |
42c1b124 ID |
3422 | bool handled_incremental = false; |
3423 | bool was_pauserd, was_pausewr; | |
3424 | bool pauserd, pausewr; | |
3425 | int err; | |
f24e9980 | 3426 | |
42c1b124 | 3427 | dout("%s have %u\n", __func__, osdc->osdmap->epoch); |
5aea3dcd | 3428 | down_write(&osdc->lock); |
f24e9980 SW |
3429 | |
3430 | /* verify fsid */ | |
3431 | ceph_decode_need(&p, end, sizeof(fsid), bad); | |
3432 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
0743304d | 3433 | if (ceph_check_fsid(osdc->client, &fsid) < 0) |
42c1b124 | 3434 | goto bad; |
f24e9980 | 3435 | |
b7ec35b3 ID |
3436 | was_pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); |
3437 | was_pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || | |
3438 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || | |
42c1b124 | 3439 | have_pool_full(osdc); |
9a1ea2db | 3440 | |
f24e9980 SW |
3441 | /* incremental maps */ |
3442 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
3443 | dout(" %d inc maps\n", nr_maps); | |
3444 | while (nr_maps > 0) { | |
3445 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
3446 | epoch = ceph_decode_32(&p); |
3447 | maplen = ceph_decode_32(&p); | |
f24e9980 | 3448 | ceph_decode_need(&p, end, maplen, bad); |
42c1b124 ID |
3449 | if (osdc->osdmap->epoch && |
3450 | osdc->osdmap->epoch + 1 == epoch) { | |
f24e9980 SW |
3451 | dout("applying incremental map %u len %d\n", |
3452 | epoch, maplen); | |
5aea3dcd ID |
3453 | err = handle_one_map(osdc, p, p + maplen, true, |
3454 | &need_resend, &need_resend_linger); | |
42c1b124 | 3455 | if (err) |
f24e9980 | 3456 | goto bad; |
42c1b124 | 3457 | handled_incremental = true; |
f24e9980 SW |
3458 | } else { |
3459 | dout("ignoring incremental map %u len %d\n", | |
3460 | epoch, maplen); | |
3461 | } | |
42c1b124 | 3462 | p += maplen; |
f24e9980 SW |
3463 | nr_maps--; |
3464 | } | |
42c1b124 | 3465 | if (handled_incremental) |
f24e9980 SW |
3466 | goto done; |
3467 | ||
3468 | /* full maps */ | |
3469 | ceph_decode_32_safe(&p, end, nr_maps, bad); | |
3470 | dout(" %d full maps\n", nr_maps); | |
3471 | while (nr_maps) { | |
3472 | ceph_decode_need(&p, end, 2*sizeof(u32), bad); | |
c89136ea SW |
3473 | epoch = ceph_decode_32(&p); |
3474 | maplen = ceph_decode_32(&p); | |
f24e9980 SW |
3475 | ceph_decode_need(&p, end, maplen, bad); |
3476 | if (nr_maps > 1) { | |
3477 | dout("skipping non-latest full map %u len %d\n", | |
3478 | epoch, maplen); | |
e5253a7b | 3479 | } else if (osdc->osdmap->epoch >= epoch) { |
f24e9980 SW |
3480 | dout("skipping full map %u len %d, " |
3481 | "older than our %u\n", epoch, maplen, | |
3482 | osdc->osdmap->epoch); | |
3483 | } else { | |
3484 | dout("taking full map %u len %d\n", epoch, maplen); | |
5aea3dcd ID |
3485 | err = handle_one_map(osdc, p, p + maplen, false, |
3486 | &need_resend, &need_resend_linger); | |
42c1b124 | 3487 | if (err) |
f24e9980 | 3488 | goto bad; |
f24e9980 SW |
3489 | } |
3490 | p += maplen; | |
3491 | nr_maps--; | |
3492 | } | |
3493 | ||
3494 | done: | |
cd634fb6 SW |
3495 | /* |
3496 | * subscribe to subsequent osdmap updates if full to ensure | |
3497 | * we find out when we are no longer full and stop returning | |
3498 | * ENOSPC. | |
3499 | */ | |
b7ec35b3 ID |
3500 | pauserd = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSERD); |
3501 | pausewr = ceph_osdmap_flag(osdc, CEPH_OSDMAP_PAUSEWR) || | |
3502 | ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) || | |
42c1b124 | 3503 | have_pool_full(osdc); |
58eb7932 JL |
3504 | if (was_pauserd || was_pausewr || pauserd || pausewr || |
3505 | osdc->osdmap->epoch < osdc->epoch_barrier) | |
42c1b124 | 3506 | maybe_request_map(osdc); |
cd634fb6 | 3507 | |
5aea3dcd | 3508 | kick_requests(osdc, &need_resend, &need_resend_linger); |
42c1b124 | 3509 | |
fc36d0a4 | 3510 | ceph_osdc_abort_on_full(osdc); |
42c1b124 ID |
3511 | ceph_monc_got_map(&osdc->client->monc, CEPH_SUB_OSDMAP, |
3512 | osdc->osdmap->epoch); | |
5aea3dcd | 3513 | up_write(&osdc->lock); |
03066f23 | 3514 | wake_up_all(&osdc->client->auth_wq); |
f24e9980 SW |
3515 | return; |
3516 | ||
3517 | bad: | |
3518 | pr_err("osdc handle_map corrupt msg\n"); | |
9ec7cab1 | 3519 | ceph_msg_dump(msg); |
5aea3dcd ID |
3520 | up_write(&osdc->lock); |
3521 | } | |
3522 | ||
3523 | /* | |
3524 | * Resubmit requests pending on the given osd. | |
3525 | */ | |
3526 | static void kick_osd_requests(struct ceph_osd *osd) | |
3527 | { | |
3528 | struct rb_node *n; | |
3529 | ||
922dab61 | 3530 | for (n = rb_first(&osd->o_requests); n; ) { |
5aea3dcd ID |
3531 | struct ceph_osd_request *req = |
3532 | rb_entry(n, struct ceph_osd_request, r_node); | |
3533 | ||
922dab61 ID |
3534 | n = rb_next(n); /* cancel_linger_request() */ |
3535 | ||
5aea3dcd ID |
3536 | if (!req->r_linger) { |
3537 | if (!req->r_t.paused) | |
3538 | send_request(req); | |
922dab61 ID |
3539 | } else { |
3540 | cancel_linger_request(req); | |
5aea3dcd ID |
3541 | } |
3542 | } | |
922dab61 ID |
3543 | for (n = rb_first(&osd->o_linger_requests); n; n = rb_next(n)) { |
3544 | struct ceph_osd_linger_request *lreq = | |
3545 | rb_entry(n, struct ceph_osd_linger_request, node); | |
3546 | ||
3547 | send_linger(lreq); | |
3548 | } | |
5aea3dcd ID |
3549 | } |
3550 | ||
3551 | /* | |
3552 | * If the osd connection drops, we need to resubmit all requests. | |
3553 | */ | |
3554 | static void osd_fault(struct ceph_connection *con) | |
3555 | { | |
3556 | struct ceph_osd *osd = con->private; | |
3557 | struct ceph_osd_client *osdc = osd->o_osdc; | |
3558 | ||
3559 | dout("%s osd %p osd%d\n", __func__, osd, osd->o_osd); | |
3560 | ||
3561 | down_write(&osdc->lock); | |
3562 | if (!osd_registered(osd)) { | |
3563 | dout("%s osd%d unknown\n", __func__, osd->o_osd); | |
3564 | goto out_unlock; | |
3565 | } | |
3566 | ||
3567 | if (!reopen_osd(osd)) | |
3568 | kick_osd_requests(osd); | |
3569 | maybe_request_map(osdc); | |
3570 | ||
3571 | out_unlock: | |
3572 | up_write(&osdc->lock); | |
f24e9980 SW |
3573 | } |
3574 | ||
a40c4f10 YS |
3575 | /* |
3576 | * Process osd watch notifications | |
3577 | */ | |
3c663bbd AE |
3578 | static void handle_watch_notify(struct ceph_osd_client *osdc, |
3579 | struct ceph_msg *msg) | |
a40c4f10 | 3580 | { |
922dab61 ID |
3581 | void *p = msg->front.iov_base; |
3582 | void *const end = p + msg->front.iov_len; | |
3583 | struct ceph_osd_linger_request *lreq; | |
3584 | struct linger_work *lwork; | |
3585 | u8 proto_ver, opcode; | |
3586 | u64 cookie, notify_id; | |
3587 | u64 notifier_id = 0; | |
19079203 | 3588 | s32 return_code = 0; |
922dab61 ID |
3589 | void *payload = NULL; |
3590 | u32 payload_len = 0; | |
a40c4f10 YS |
3591 | |
3592 | ceph_decode_8_safe(&p, end, proto_ver, bad); | |
3593 | ceph_decode_8_safe(&p, end, opcode, bad); | |
3594 | ceph_decode_64_safe(&p, end, cookie, bad); | |
922dab61 | 3595 | p += 8; /* skip ver */ |
a40c4f10 YS |
3596 | ceph_decode_64_safe(&p, end, notify_id, bad); |
3597 | ||
922dab61 ID |
3598 | if (proto_ver >= 1) { |
3599 | ceph_decode_32_safe(&p, end, payload_len, bad); | |
3600 | ceph_decode_need(&p, end, payload_len, bad); | |
3601 | payload = p; | |
3602 | p += payload_len; | |
3603 | } | |
3604 | ||
3605 | if (le16_to_cpu(msg->hdr.version) >= 2) | |
19079203 | 3606 | ceph_decode_32_safe(&p, end, return_code, bad); |
922dab61 ID |
3607 | |
3608 | if (le16_to_cpu(msg->hdr.version) >= 3) | |
3609 | ceph_decode_64_safe(&p, end, notifier_id, bad); | |
3610 | ||
3611 | down_read(&osdc->lock); | |
3612 | lreq = lookup_linger_osdc(&osdc->linger_requests, cookie); | |
3613 | if (!lreq) { | |
3614 | dout("%s opcode %d cookie %llu dne\n", __func__, opcode, | |
3615 | cookie); | |
3616 | goto out_unlock_osdc; | |
3617 | } | |
3618 | ||
3619 | mutex_lock(&lreq->lock); | |
19079203 ID |
3620 | dout("%s opcode %d cookie %llu lreq %p is_watch %d\n", __func__, |
3621 | opcode, cookie, lreq, lreq->is_watch); | |
922dab61 ID |
3622 | if (opcode == CEPH_WATCH_EVENT_DISCONNECT) { |
3623 | if (!lreq->last_error) { | |
3624 | lreq->last_error = -ENOTCONN; | |
3625 | queue_watch_error(lreq); | |
3626 | } | |
19079203 ID |
3627 | } else if (!lreq->is_watch) { |
3628 | /* CEPH_WATCH_EVENT_NOTIFY_COMPLETE */ | |
3629 | if (lreq->notify_id && lreq->notify_id != notify_id) { | |
3630 | dout("lreq %p notify_id %llu != %llu, ignoring\n", lreq, | |
3631 | lreq->notify_id, notify_id); | |
3632 | } else if (!completion_done(&lreq->notify_finish_wait)) { | |
3633 | struct ceph_msg_data *data = | |
3634 | list_first_entry_or_null(&msg->data, | |
3635 | struct ceph_msg_data, | |
3636 | links); | |
3637 | ||
3638 | if (data) { | |
3639 | if (lreq->preply_pages) { | |
3640 | WARN_ON(data->type != | |
3641 | CEPH_MSG_DATA_PAGES); | |
3642 | *lreq->preply_pages = data->pages; | |
3643 | *lreq->preply_len = data->length; | |
3644 | } else { | |
3645 | ceph_release_page_vector(data->pages, | |
3646 | calc_pages_for(0, data->length)); | |
3647 | } | |
3648 | } | |
3649 | lreq->notify_finish_error = return_code; | |
3650 | complete_all(&lreq->notify_finish_wait); | |
3651 | } | |
922dab61 ID |
3652 | } else { |
3653 | /* CEPH_WATCH_EVENT_NOTIFY */ | |
3654 | lwork = lwork_alloc(lreq, do_watch_notify); | |
3655 | if (!lwork) { | |
3656 | pr_err("failed to allocate notify-lwork\n"); | |
3657 | goto out_unlock_lreq; | |
a40c4f10 | 3658 | } |
a40c4f10 | 3659 | |
922dab61 ID |
3660 | lwork->notify.notify_id = notify_id; |
3661 | lwork->notify.notifier_id = notifier_id; | |
3662 | lwork->notify.payload = payload; | |
3663 | lwork->notify.payload_len = payload_len; | |
3664 | lwork->notify.msg = ceph_msg_get(msg); | |
3665 | lwork_queue(lwork); | |
91883cd2 | 3666 | } |
a40c4f10 | 3667 | |
922dab61 ID |
3668 | out_unlock_lreq: |
3669 | mutex_unlock(&lreq->lock); | |
3670 | out_unlock_osdc: | |
3671 | up_read(&osdc->lock); | |
a40c4f10 YS |
3672 | return; |
3673 | ||
3674 | bad: | |
3675 | pr_err("osdc handle_watch_notify corrupt msg\n"); | |
a40c4f10 YS |
3676 | } |
3677 | ||
70636773 AE |
3678 | /* |
3679 | * Register request, send initial attempt. | |
3680 | */ | |
3681 | int ceph_osdc_start_request(struct ceph_osd_client *osdc, | |
3682 | struct ceph_osd_request *req, | |
3683 | bool nofail) | |
3684 | { | |
5aea3dcd ID |
3685 | down_read(&osdc->lock); |
3686 | submit_request(req, false); | |
3687 | up_read(&osdc->lock); | |
0bbfdfe8 | 3688 | |
5aea3dcd | 3689 | return 0; |
f24e9980 | 3690 | } |
3d14c5d2 | 3691 | EXPORT_SYMBOL(ceph_osdc_start_request); |
f24e9980 | 3692 | |
c9f9b93d | 3693 | /* |
c297eb42 ID |
3694 | * Unregister a registered request. The request is not completed: |
3695 | * ->r_result isn't set and __complete_request() isn't called. | |
c9f9b93d ID |
3696 | */ |
3697 | void ceph_osdc_cancel_request(struct ceph_osd_request *req) | |
3698 | { | |
3699 | struct ceph_osd_client *osdc = req->r_osdc; | |
3700 | ||
5aea3dcd | 3701 | down_write(&osdc->lock); |
5aea3dcd ID |
3702 | if (req->r_osd) |
3703 | cancel_request(req); | |
3704 | up_write(&osdc->lock); | |
c9f9b93d ID |
3705 | } |
3706 | EXPORT_SYMBOL(ceph_osdc_cancel_request); | |
3707 | ||
f24e9980 | 3708 | /* |
42b06965 | 3709 | * @timeout: in jiffies, 0 means "wait forever" |
f24e9980 | 3710 | */ |
42b06965 ID |
3711 | static int wait_request_timeout(struct ceph_osd_request *req, |
3712 | unsigned long timeout) | |
f24e9980 | 3713 | { |
42b06965 | 3714 | long left; |
c9f9b93d | 3715 | |
42b06965 | 3716 | dout("%s req %p tid %llu\n", __func__, req, req->r_tid); |
0e76abf2 | 3717 | left = wait_for_completion_killable_timeout(&req->r_completion, |
42b06965 ID |
3718 | ceph_timeout_jiffies(timeout)); |
3719 | if (left <= 0) { | |
3720 | left = left ?: -ETIMEDOUT; | |
c9f9b93d | 3721 | ceph_osdc_cancel_request(req); |
42b06965 ID |
3722 | } else { |
3723 | left = req->r_result; /* completed */ | |
f24e9980 SW |
3724 | } |
3725 | ||
42b06965 ID |
3726 | return left; |
3727 | } | |
3728 | ||
3729 | /* | |
3730 | * wait for a request to complete | |
3731 | */ | |
3732 | int ceph_osdc_wait_request(struct ceph_osd_client *osdc, | |
3733 | struct ceph_osd_request *req) | |
3734 | { | |
3735 | return wait_request_timeout(req, 0); | |
f24e9980 | 3736 | } |
3d14c5d2 | 3737 | EXPORT_SYMBOL(ceph_osdc_wait_request); |
f24e9980 SW |
3738 | |
3739 | /* | |
3740 | * sync - wait for all in-flight requests to flush. avoid starvation. | |
3741 | */ | |
3742 | void ceph_osdc_sync(struct ceph_osd_client *osdc) | |
3743 | { | |
5aea3dcd ID |
3744 | struct rb_node *n, *p; |
3745 | u64 last_tid = atomic64_read(&osdc->last_tid); | |
f24e9980 | 3746 | |
5aea3dcd ID |
3747 | again: |
3748 | down_read(&osdc->lock); | |
3749 | for (n = rb_first(&osdc->osds); n; n = rb_next(n)) { | |
3750 | struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node); | |
3751 | ||
3752 | mutex_lock(&osd->lock); | |
3753 | for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) { | |
3754 | struct ceph_osd_request *req = | |
3755 | rb_entry(p, struct ceph_osd_request, r_node); | |
3756 | ||
3757 | if (req->r_tid > last_tid) | |
3758 | break; | |
3759 | ||
3760 | if (!(req->r_flags & CEPH_OSD_FLAG_WRITE)) | |
3761 | continue; | |
f24e9980 | 3762 | |
5aea3dcd ID |
3763 | ceph_osdc_get_request(req); |
3764 | mutex_unlock(&osd->lock); | |
3765 | up_read(&osdc->lock); | |
3766 | dout("%s waiting on req %p tid %llu last_tid %llu\n", | |
3767 | __func__, req, req->r_tid, last_tid); | |
b18b9550 | 3768 | wait_for_completion(&req->r_completion); |
5aea3dcd ID |
3769 | ceph_osdc_put_request(req); |
3770 | goto again; | |
3771 | } | |
f24e9980 | 3772 | |
5aea3dcd | 3773 | mutex_unlock(&osd->lock); |
f24e9980 | 3774 | } |
5aea3dcd ID |
3775 | |
3776 | up_read(&osdc->lock); | |
3777 | dout("%s done last_tid %llu\n", __func__, last_tid); | |
f24e9980 | 3778 | } |
3d14c5d2 | 3779 | EXPORT_SYMBOL(ceph_osdc_sync); |
f24e9980 | 3780 | |
922dab61 ID |
3781 | static struct ceph_osd_request * |
3782 | alloc_linger_request(struct ceph_osd_linger_request *lreq) | |
3783 | { | |
3784 | struct ceph_osd_request *req; | |
3785 | ||
3786 | req = ceph_osdc_alloc_request(lreq->osdc, NULL, 1, false, GFP_NOIO); | |
3787 | if (!req) | |
3788 | return NULL; | |
3789 | ||
3790 | ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); | |
3791 | ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); | |
3792 | ||
3793 | if (ceph_osdc_alloc_messages(req, GFP_NOIO)) { | |
3794 | ceph_osdc_put_request(req); | |
3795 | return NULL; | |
3796 | } | |
3797 | ||
3798 | return req; | |
3799 | } | |
3800 | ||
3801 | /* | |
3802 | * Returns a handle, caller owns a ref. | |
3803 | */ | |
3804 | struct ceph_osd_linger_request * | |
3805 | ceph_osdc_watch(struct ceph_osd_client *osdc, | |
3806 | struct ceph_object_id *oid, | |
3807 | struct ceph_object_locator *oloc, | |
3808 | rados_watchcb2_t wcb, | |
3809 | rados_watcherrcb_t errcb, | |
3810 | void *data) | |
3811 | { | |
3812 | struct ceph_osd_linger_request *lreq; | |
3813 | int ret; | |
3814 | ||
3815 | lreq = linger_alloc(osdc); | |
3816 | if (!lreq) | |
3817 | return ERR_PTR(-ENOMEM); | |
3818 | ||
19079203 | 3819 | lreq->is_watch = true; |
922dab61 ID |
3820 | lreq->wcb = wcb; |
3821 | lreq->errcb = errcb; | |
3822 | lreq->data = data; | |
b07d3c4b | 3823 | lreq->watch_valid_thru = jiffies; |
922dab61 ID |
3824 | |
3825 | ceph_oid_copy(&lreq->t.base_oid, oid); | |
3826 | ceph_oloc_copy(&lreq->t.base_oloc, oloc); | |
54ea0046 | 3827 | lreq->t.flags = CEPH_OSD_FLAG_WRITE; |
1134e091 | 3828 | ktime_get_real_ts(&lreq->mtime); |
922dab61 ID |
3829 | |
3830 | lreq->reg_req = alloc_linger_request(lreq); | |
3831 | if (!lreq->reg_req) { | |
3832 | ret = -ENOMEM; | |
3833 | goto err_put_lreq; | |
3834 | } | |
3835 | ||
3836 | lreq->ping_req = alloc_linger_request(lreq); | |
3837 | if (!lreq->ping_req) { | |
3838 | ret = -ENOMEM; | |
3839 | goto err_put_lreq; | |
3840 | } | |
3841 | ||
3842 | down_write(&osdc->lock); | |
3843 | linger_register(lreq); /* before osd_req_op_* */ | |
3844 | osd_req_op_watch_init(lreq->reg_req, 0, lreq->linger_id, | |
3845 | CEPH_OSD_WATCH_OP_WATCH); | |
3846 | osd_req_op_watch_init(lreq->ping_req, 0, lreq->linger_id, | |
3847 | CEPH_OSD_WATCH_OP_PING); | |
3848 | linger_submit(lreq); | |
3849 | up_write(&osdc->lock); | |
3850 | ||
3851 | ret = linger_reg_commit_wait(lreq); | |
3852 | if (ret) { | |
3853 | linger_cancel(lreq); | |
3854 | goto err_put_lreq; | |
3855 | } | |
3856 | ||
3857 | return lreq; | |
3858 | ||
3859 | err_put_lreq: | |
3860 | linger_put(lreq); | |
3861 | return ERR_PTR(ret); | |
3862 | } | |
3863 | EXPORT_SYMBOL(ceph_osdc_watch); | |
3864 | ||
3865 | /* | |
3866 | * Releases a ref. | |
3867 | * | |
3868 | * Times out after mount_timeout to preserve rbd unmap behaviour | |
3869 | * introduced in 2894e1d76974 ("rbd: timeout watch teardown on unmap | |
3870 | * with mount_timeout"). | |
3871 | */ | |
3872 | int ceph_osdc_unwatch(struct ceph_osd_client *osdc, | |
3873 | struct ceph_osd_linger_request *lreq) | |
3874 | { | |
3875 | struct ceph_options *opts = osdc->client->options; | |
3876 | struct ceph_osd_request *req; | |
3877 | int ret; | |
3878 | ||
3879 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); | |
3880 | if (!req) | |
3881 | return -ENOMEM; | |
3882 | ||
3883 | ceph_oid_copy(&req->r_base_oid, &lreq->t.base_oid); | |
3884 | ceph_oloc_copy(&req->r_base_oloc, &lreq->t.base_oloc); | |
54ea0046 | 3885 | req->r_flags = CEPH_OSD_FLAG_WRITE; |
1134e091 | 3886 | ktime_get_real_ts(&req->r_mtime); |
922dab61 ID |
3887 | osd_req_op_watch_init(req, 0, lreq->linger_id, |
3888 | CEPH_OSD_WATCH_OP_UNWATCH); | |
3889 | ||
3890 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
3891 | if (ret) | |
3892 | goto out_put_req; | |
3893 | ||
3894 | ceph_osdc_start_request(osdc, req, false); | |
3895 | linger_cancel(lreq); | |
3896 | linger_put(lreq); | |
3897 | ret = wait_request_timeout(req, opts->mount_timeout); | |
3898 | ||
3899 | out_put_req: | |
3900 | ceph_osdc_put_request(req); | |
3901 | return ret; | |
3902 | } | |
3903 | EXPORT_SYMBOL(ceph_osdc_unwatch); | |
3904 | ||
3905 | static int osd_req_op_notify_ack_init(struct ceph_osd_request *req, int which, | |
3906 | u64 notify_id, u64 cookie, void *payload, | |
3907 | size_t payload_len) | |
3908 | { | |
3909 | struct ceph_osd_req_op *op; | |
3910 | struct ceph_pagelist *pl; | |
3911 | int ret; | |
3912 | ||
3913 | op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY_ACK, 0); | |
3914 | ||
3915 | pl = kmalloc(sizeof(*pl), GFP_NOIO); | |
3916 | if (!pl) | |
3917 | return -ENOMEM; | |
3918 | ||
3919 | ceph_pagelist_init(pl); | |
3920 | ret = ceph_pagelist_encode_64(pl, notify_id); | |
3921 | ret |= ceph_pagelist_encode_64(pl, cookie); | |
3922 | if (payload) { | |
3923 | ret |= ceph_pagelist_encode_32(pl, payload_len); | |
3924 | ret |= ceph_pagelist_append(pl, payload, payload_len); | |
3925 | } else { | |
3926 | ret |= ceph_pagelist_encode_32(pl, 0); | |
3927 | } | |
3928 | if (ret) { | |
3929 | ceph_pagelist_release(pl); | |
3930 | return -ENOMEM; | |
3931 | } | |
3932 | ||
3933 | ceph_osd_data_pagelist_init(&op->notify_ack.request_data, pl); | |
3934 | op->indata_len = pl->length; | |
3935 | return 0; | |
3936 | } | |
3937 | ||
3938 | int ceph_osdc_notify_ack(struct ceph_osd_client *osdc, | |
3939 | struct ceph_object_id *oid, | |
3940 | struct ceph_object_locator *oloc, | |
3941 | u64 notify_id, | |
3942 | u64 cookie, | |
3943 | void *payload, | |
3944 | size_t payload_len) | |
3945 | { | |
3946 | struct ceph_osd_request *req; | |
3947 | int ret; | |
3948 | ||
3949 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); | |
3950 | if (!req) | |
3951 | return -ENOMEM; | |
3952 | ||
3953 | ceph_oid_copy(&req->r_base_oid, oid); | |
3954 | ceph_oloc_copy(&req->r_base_oloc, oloc); | |
3955 | req->r_flags = CEPH_OSD_FLAG_READ; | |
3956 | ||
3957 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
3958 | if (ret) | |
3959 | goto out_put_req; | |
3960 | ||
3961 | ret = osd_req_op_notify_ack_init(req, 0, notify_id, cookie, payload, | |
3962 | payload_len); | |
3963 | if (ret) | |
3964 | goto out_put_req; | |
3965 | ||
3966 | ceph_osdc_start_request(osdc, req, false); | |
3967 | ret = ceph_osdc_wait_request(osdc, req); | |
3968 | ||
3969 | out_put_req: | |
3970 | ceph_osdc_put_request(req); | |
3971 | return ret; | |
3972 | } | |
3973 | EXPORT_SYMBOL(ceph_osdc_notify_ack); | |
3974 | ||
19079203 ID |
3975 | static int osd_req_op_notify_init(struct ceph_osd_request *req, int which, |
3976 | u64 cookie, u32 prot_ver, u32 timeout, | |
3977 | void *payload, size_t payload_len) | |
3978 | { | |
3979 | struct ceph_osd_req_op *op; | |
3980 | struct ceph_pagelist *pl; | |
3981 | int ret; | |
3982 | ||
3983 | op = _osd_req_op_init(req, which, CEPH_OSD_OP_NOTIFY, 0); | |
3984 | op->notify.cookie = cookie; | |
3985 | ||
3986 | pl = kmalloc(sizeof(*pl), GFP_NOIO); | |
3987 | if (!pl) | |
3988 | return -ENOMEM; | |
3989 | ||
3990 | ceph_pagelist_init(pl); | |
3991 | ret = ceph_pagelist_encode_32(pl, 1); /* prot_ver */ | |
3992 | ret |= ceph_pagelist_encode_32(pl, timeout); | |
3993 | ret |= ceph_pagelist_encode_32(pl, payload_len); | |
3994 | ret |= ceph_pagelist_append(pl, payload, payload_len); | |
3995 | if (ret) { | |
3996 | ceph_pagelist_release(pl); | |
3997 | return -ENOMEM; | |
3998 | } | |
3999 | ||
4000 | ceph_osd_data_pagelist_init(&op->notify.request_data, pl); | |
4001 | op->indata_len = pl->length; | |
4002 | return 0; | |
4003 | } | |
4004 | ||
4005 | /* | |
4006 | * @timeout: in seconds | |
4007 | * | |
4008 | * @preply_{pages,len} are initialized both on success and error. | |
4009 | * The caller is responsible for: | |
4010 | * | |
4011 | * ceph_release_page_vector(reply_pages, calc_pages_for(0, reply_len)) | |
4012 | */ | |
4013 | int ceph_osdc_notify(struct ceph_osd_client *osdc, | |
4014 | struct ceph_object_id *oid, | |
4015 | struct ceph_object_locator *oloc, | |
4016 | void *payload, | |
4017 | size_t payload_len, | |
4018 | u32 timeout, | |
4019 | struct page ***preply_pages, | |
4020 | size_t *preply_len) | |
4021 | { | |
4022 | struct ceph_osd_linger_request *lreq; | |
4023 | struct page **pages; | |
4024 | int ret; | |
4025 | ||
4026 | WARN_ON(!timeout); | |
4027 | if (preply_pages) { | |
4028 | *preply_pages = NULL; | |
4029 | *preply_len = 0; | |
4030 | } | |
4031 | ||
4032 | lreq = linger_alloc(osdc); | |
4033 | if (!lreq) | |
4034 | return -ENOMEM; | |
4035 | ||
4036 | lreq->preply_pages = preply_pages; | |
4037 | lreq->preply_len = preply_len; | |
4038 | ||
4039 | ceph_oid_copy(&lreq->t.base_oid, oid); | |
4040 | ceph_oloc_copy(&lreq->t.base_oloc, oloc); | |
4041 | lreq->t.flags = CEPH_OSD_FLAG_READ; | |
4042 | ||
4043 | lreq->reg_req = alloc_linger_request(lreq); | |
4044 | if (!lreq->reg_req) { | |
4045 | ret = -ENOMEM; | |
4046 | goto out_put_lreq; | |
4047 | } | |
4048 | ||
4049 | /* for notify_id */ | |
4050 | pages = ceph_alloc_page_vector(1, GFP_NOIO); | |
4051 | if (IS_ERR(pages)) { | |
4052 | ret = PTR_ERR(pages); | |
4053 | goto out_put_lreq; | |
4054 | } | |
4055 | ||
4056 | down_write(&osdc->lock); | |
4057 | linger_register(lreq); /* before osd_req_op_* */ | |
4058 | ret = osd_req_op_notify_init(lreq->reg_req, 0, lreq->linger_id, 1, | |
4059 | timeout, payload, payload_len); | |
4060 | if (ret) { | |
4061 | linger_unregister(lreq); | |
4062 | up_write(&osdc->lock); | |
4063 | ceph_release_page_vector(pages, 1); | |
4064 | goto out_put_lreq; | |
4065 | } | |
4066 | ceph_osd_data_pages_init(osd_req_op_data(lreq->reg_req, 0, notify, | |
4067 | response_data), | |
4068 | pages, PAGE_SIZE, 0, false, true); | |
4069 | linger_submit(lreq); | |
4070 | up_write(&osdc->lock); | |
4071 | ||
4072 | ret = linger_reg_commit_wait(lreq); | |
4073 | if (!ret) | |
4074 | ret = linger_notify_finish_wait(lreq); | |
4075 | else | |
4076 | dout("lreq %p failed to initiate notify %d\n", lreq, ret); | |
4077 | ||
4078 | linger_cancel(lreq); | |
4079 | out_put_lreq: | |
4080 | linger_put(lreq); | |
4081 | return ret; | |
4082 | } | |
4083 | EXPORT_SYMBOL(ceph_osdc_notify); | |
4084 | ||
b07d3c4b ID |
4085 | /* |
4086 | * Return the number of milliseconds since the watch was last | |
4087 | * confirmed, or an error. If there is an error, the watch is no | |
4088 | * longer valid, and should be destroyed with ceph_osdc_unwatch(). | |
4089 | */ | |
4090 | int ceph_osdc_watch_check(struct ceph_osd_client *osdc, | |
4091 | struct ceph_osd_linger_request *lreq) | |
4092 | { | |
4093 | unsigned long stamp, age; | |
4094 | int ret; | |
4095 | ||
4096 | down_read(&osdc->lock); | |
4097 | mutex_lock(&lreq->lock); | |
4098 | stamp = lreq->watch_valid_thru; | |
4099 | if (!list_empty(&lreq->pending_lworks)) { | |
4100 | struct linger_work *lwork = | |
4101 | list_first_entry(&lreq->pending_lworks, | |
4102 | struct linger_work, | |
4103 | pending_item); | |
4104 | ||
4105 | if (time_before(lwork->queued_stamp, stamp)) | |
4106 | stamp = lwork->queued_stamp; | |
4107 | } | |
4108 | age = jiffies - stamp; | |
4109 | dout("%s lreq %p linger_id %llu age %lu last_error %d\n", __func__, | |
4110 | lreq, lreq->linger_id, age, lreq->last_error); | |
4111 | /* we are truncating to msecs, so return a safe upper bound */ | |
4112 | ret = lreq->last_error ?: 1 + jiffies_to_msecs(age); | |
4113 | ||
4114 | mutex_unlock(&lreq->lock); | |
4115 | up_read(&osdc->lock); | |
4116 | return ret; | |
4117 | } | |
4118 | ||
a4ed38d7 DF |
4119 | static int decode_watcher(void **p, void *end, struct ceph_watch_item *item) |
4120 | { | |
4121 | u8 struct_v; | |
4122 | u32 struct_len; | |
4123 | int ret; | |
4124 | ||
4125 | ret = ceph_start_decoding(p, end, 2, "watch_item_t", | |
4126 | &struct_v, &struct_len); | |
4127 | if (ret) | |
4128 | return ret; | |
4129 | ||
4130 | ceph_decode_copy(p, &item->name, sizeof(item->name)); | |
4131 | item->cookie = ceph_decode_64(p); | |
4132 | *p += 4; /* skip timeout_seconds */ | |
4133 | if (struct_v >= 2) { | |
4134 | ceph_decode_copy(p, &item->addr, sizeof(item->addr)); | |
4135 | ceph_decode_addr(&item->addr); | |
4136 | } | |
4137 | ||
4138 | dout("%s %s%llu cookie %llu addr %s\n", __func__, | |
4139 | ENTITY_NAME(item->name), item->cookie, | |
4140 | ceph_pr_addr(&item->addr.in_addr)); | |
4141 | return 0; | |
4142 | } | |
4143 | ||
4144 | static int decode_watchers(void **p, void *end, | |
4145 | struct ceph_watch_item **watchers, | |
4146 | u32 *num_watchers) | |
4147 | { | |
4148 | u8 struct_v; | |
4149 | u32 struct_len; | |
4150 | int i; | |
4151 | int ret; | |
4152 | ||
4153 | ret = ceph_start_decoding(p, end, 1, "obj_list_watch_response_t", | |
4154 | &struct_v, &struct_len); | |
4155 | if (ret) | |
4156 | return ret; | |
4157 | ||
4158 | *num_watchers = ceph_decode_32(p); | |
4159 | *watchers = kcalloc(*num_watchers, sizeof(**watchers), GFP_NOIO); | |
4160 | if (!*watchers) | |
4161 | return -ENOMEM; | |
4162 | ||
4163 | for (i = 0; i < *num_watchers; i++) { | |
4164 | ret = decode_watcher(p, end, *watchers + i); | |
4165 | if (ret) { | |
4166 | kfree(*watchers); | |
4167 | return ret; | |
4168 | } | |
4169 | } | |
4170 | ||
4171 | return 0; | |
4172 | } | |
4173 | ||
4174 | /* | |
4175 | * On success, the caller is responsible for: | |
4176 | * | |
4177 | * kfree(watchers); | |
4178 | */ | |
4179 | int ceph_osdc_list_watchers(struct ceph_osd_client *osdc, | |
4180 | struct ceph_object_id *oid, | |
4181 | struct ceph_object_locator *oloc, | |
4182 | struct ceph_watch_item **watchers, | |
4183 | u32 *num_watchers) | |
4184 | { | |
4185 | struct ceph_osd_request *req; | |
4186 | struct page **pages; | |
4187 | int ret; | |
4188 | ||
4189 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); | |
4190 | if (!req) | |
4191 | return -ENOMEM; | |
4192 | ||
4193 | ceph_oid_copy(&req->r_base_oid, oid); | |
4194 | ceph_oloc_copy(&req->r_base_oloc, oloc); | |
4195 | req->r_flags = CEPH_OSD_FLAG_READ; | |
4196 | ||
4197 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
4198 | if (ret) | |
4199 | goto out_put_req; | |
4200 | ||
4201 | pages = ceph_alloc_page_vector(1, GFP_NOIO); | |
4202 | if (IS_ERR(pages)) { | |
4203 | ret = PTR_ERR(pages); | |
4204 | goto out_put_req; | |
4205 | } | |
4206 | ||
4207 | osd_req_op_init(req, 0, CEPH_OSD_OP_LIST_WATCHERS, 0); | |
4208 | ceph_osd_data_pages_init(osd_req_op_data(req, 0, list_watchers, | |
4209 | response_data), | |
4210 | pages, PAGE_SIZE, 0, false, true); | |
4211 | ||
4212 | ceph_osdc_start_request(osdc, req, false); | |
4213 | ret = ceph_osdc_wait_request(osdc, req); | |
4214 | if (ret >= 0) { | |
4215 | void *p = page_address(pages[0]); | |
4216 | void *const end = p + req->r_ops[0].outdata_len; | |
4217 | ||
4218 | ret = decode_watchers(&p, end, watchers, num_watchers); | |
4219 | } | |
4220 | ||
4221 | out_put_req: | |
4222 | ceph_osdc_put_request(req); | |
4223 | return ret; | |
4224 | } | |
4225 | EXPORT_SYMBOL(ceph_osdc_list_watchers); | |
4226 | ||
dd935f44 JD |
4227 | /* |
4228 | * Call all pending notify callbacks - for use after a watch is | |
4229 | * unregistered, to make sure no more callbacks for it will be invoked | |
4230 | */ | |
f6479449 | 4231 | void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc) |
dd935f44 | 4232 | { |
99d16943 | 4233 | dout("%s osdc %p\n", __func__, osdc); |
dd935f44 JD |
4234 | flush_workqueue(osdc->notify_wq); |
4235 | } | |
4236 | EXPORT_SYMBOL(ceph_osdc_flush_notifies); | |
4237 | ||
7cca78c9 ID |
4238 | void ceph_osdc_maybe_request_map(struct ceph_osd_client *osdc) |
4239 | { | |
4240 | down_read(&osdc->lock); | |
4241 | maybe_request_map(osdc); | |
4242 | up_read(&osdc->lock); | |
4243 | } | |
4244 | EXPORT_SYMBOL(ceph_osdc_maybe_request_map); | |
dd935f44 | 4245 | |
428a7158 DF |
4246 | /* |
4247 | * Execute an OSD class method on an object. | |
4248 | * | |
4249 | * @flags: CEPH_OSD_FLAG_* | |
2544a020 | 4250 | * @resp_len: in/out param for reply length |
428a7158 DF |
4251 | */ |
4252 | int ceph_osdc_call(struct ceph_osd_client *osdc, | |
4253 | struct ceph_object_id *oid, | |
4254 | struct ceph_object_locator *oloc, | |
4255 | const char *class, const char *method, | |
4256 | unsigned int flags, | |
4257 | struct page *req_page, size_t req_len, | |
4258 | struct page *resp_page, size_t *resp_len) | |
4259 | { | |
4260 | struct ceph_osd_request *req; | |
4261 | int ret; | |
4262 | ||
2544a020 ID |
4263 | if (req_len > PAGE_SIZE || (resp_page && *resp_len > PAGE_SIZE)) |
4264 | return -E2BIG; | |
4265 | ||
428a7158 DF |
4266 | req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_NOIO); |
4267 | if (!req) | |
4268 | return -ENOMEM; | |
4269 | ||
4270 | ceph_oid_copy(&req->r_base_oid, oid); | |
4271 | ceph_oloc_copy(&req->r_base_oloc, oloc); | |
4272 | req->r_flags = flags; | |
4273 | ||
4274 | ret = ceph_osdc_alloc_messages(req, GFP_NOIO); | |
4275 | if (ret) | |
4276 | goto out_put_req; | |
4277 | ||
4278 | osd_req_op_cls_init(req, 0, CEPH_OSD_OP_CALL, class, method); | |
4279 | if (req_page) | |
4280 | osd_req_op_cls_request_data_pages(req, 0, &req_page, req_len, | |
4281 | 0, false, false); | |
4282 | if (resp_page) | |
4283 | osd_req_op_cls_response_data_pages(req, 0, &resp_page, | |
2544a020 | 4284 | *resp_len, 0, false, false); |
428a7158 DF |
4285 | |
4286 | ceph_osdc_start_request(osdc, req, false); | |
4287 | ret = ceph_osdc_wait_request(osdc, req); | |
4288 | if (ret >= 0) { | |
4289 | ret = req->r_ops[0].rval; | |
4290 | if (resp_page) | |
4291 | *resp_len = req->r_ops[0].outdata_len; | |
4292 | } | |
4293 | ||
4294 | out_put_req: | |
4295 | ceph_osdc_put_request(req); | |
4296 | return ret; | |
4297 | } | |
4298 | EXPORT_SYMBOL(ceph_osdc_call); | |
4299 | ||
f24e9980 SW |
4300 | /* |
4301 | * init, shutdown | |
4302 | */ | |
4303 | int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client) | |
4304 | { | |
4305 | int err; | |
4306 | ||
4307 | dout("init\n"); | |
4308 | osdc->client = client; | |
5aea3dcd | 4309 | init_rwsem(&osdc->lock); |
f24e9980 | 4310 | osdc->osds = RB_ROOT; |
f5a2041b | 4311 | INIT_LIST_HEAD(&osdc->osd_lru); |
9dd2845c | 4312 | spin_lock_init(&osdc->osd_lru_lock); |
5aea3dcd ID |
4313 | osd_init(&osdc->homeless_osd); |
4314 | osdc->homeless_osd.o_osdc = osdc; | |
4315 | osdc->homeless_osd.o_osd = CEPH_HOMELESS_OSD; | |
264048af | 4316 | osdc->last_linger_id = CEPH_LINGER_ID_START; |
922dab61 | 4317 | osdc->linger_requests = RB_ROOT; |
4609245e ID |
4318 | osdc->map_checks = RB_ROOT; |
4319 | osdc->linger_map_checks = RB_ROOT; | |
f24e9980 | 4320 | INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout); |
f5a2041b YS |
4321 | INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout); |
4322 | ||
5f44f142 | 4323 | err = -ENOMEM; |
e5253a7b ID |
4324 | osdc->osdmap = ceph_osdmap_alloc(); |
4325 | if (!osdc->osdmap) | |
4326 | goto out; | |
4327 | ||
9e767adb ID |
4328 | osdc->req_mempool = mempool_create_slab_pool(10, |
4329 | ceph_osd_request_cache); | |
f24e9980 | 4330 | if (!osdc->req_mempool) |
e5253a7b | 4331 | goto out_map; |
f24e9980 | 4332 | |
d50b409f | 4333 | err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP, |
711da55d | 4334 | PAGE_SIZE, 10, true, "osd_op"); |
f24e9980 | 4335 | if (err < 0) |
5f44f142 | 4336 | goto out_mempool; |
d50b409f | 4337 | err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY, |
711da55d | 4338 | PAGE_SIZE, 10, true, "osd_op_reply"); |
c16e7869 SW |
4339 | if (err < 0) |
4340 | goto out_msgpool; | |
a40c4f10 | 4341 | |
dbcae088 | 4342 | err = -ENOMEM; |
a40c4f10 | 4343 | osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify"); |
dbcae088 | 4344 | if (!osdc->notify_wq) |
c172ec5c ID |
4345 | goto out_msgpool_reply; |
4346 | ||
fbca9635 ID |
4347 | schedule_delayed_work(&osdc->timeout_work, |
4348 | osdc->client->options->osd_keepalive_timeout); | |
b37ee1b9 ID |
4349 | schedule_delayed_work(&osdc->osds_timeout_work, |
4350 | round_jiffies_relative(osdc->client->options->osd_idle_ttl)); | |
4351 | ||
f24e9980 | 4352 | return 0; |
5f44f142 | 4353 | |
c172ec5c ID |
4354 | out_msgpool_reply: |
4355 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); | |
c16e7869 SW |
4356 | out_msgpool: |
4357 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
5f44f142 SW |
4358 | out_mempool: |
4359 | mempool_destroy(osdc->req_mempool); | |
e5253a7b ID |
4360 | out_map: |
4361 | ceph_osdmap_destroy(osdc->osdmap); | |
5f44f142 SW |
4362 | out: |
4363 | return err; | |
f24e9980 SW |
4364 | } |
4365 | ||
4366 | void ceph_osdc_stop(struct ceph_osd_client *osdc) | |
4367 | { | |
a40c4f10 YS |
4368 | flush_workqueue(osdc->notify_wq); |
4369 | destroy_workqueue(osdc->notify_wq); | |
f24e9980 | 4370 | cancel_delayed_work_sync(&osdc->timeout_work); |
f5a2041b | 4371 | cancel_delayed_work_sync(&osdc->osds_timeout_work); |
42a2c09f | 4372 | |
5aea3dcd | 4373 | down_write(&osdc->lock); |
42a2c09f ID |
4374 | while (!RB_EMPTY_ROOT(&osdc->osds)) { |
4375 | struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds), | |
4376 | struct ceph_osd, o_node); | |
5aea3dcd | 4377 | close_osd(osd); |
42a2c09f | 4378 | } |
5aea3dcd | 4379 | up_write(&osdc->lock); |
02113a0f | 4380 | WARN_ON(refcount_read(&osdc->homeless_osd.o_ref) != 1); |
5aea3dcd ID |
4381 | osd_cleanup(&osdc->homeless_osd); |
4382 | ||
4383 | WARN_ON(!list_empty(&osdc->osd_lru)); | |
922dab61 | 4384 | WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_requests)); |
4609245e ID |
4385 | WARN_ON(!RB_EMPTY_ROOT(&osdc->map_checks)); |
4386 | WARN_ON(!RB_EMPTY_ROOT(&osdc->linger_map_checks)); | |
5aea3dcd ID |
4387 | WARN_ON(atomic_read(&osdc->num_requests)); |
4388 | WARN_ON(atomic_read(&osdc->num_homeless)); | |
42a2c09f | 4389 | |
e5253a7b | 4390 | ceph_osdmap_destroy(osdc->osdmap); |
f24e9980 SW |
4391 | mempool_destroy(osdc->req_mempool); |
4392 | ceph_msgpool_destroy(&osdc->msgpool_op); | |
c16e7869 | 4393 | ceph_msgpool_destroy(&osdc->msgpool_op_reply); |
f24e9980 SW |
4394 | } |
4395 | ||
4396 | /* | |
4397 | * Read some contiguous pages. If we cross a stripe boundary, shorten | |
4398 | * *plen. Return number of bytes read, or error. | |
4399 | */ | |
4400 | int ceph_osdc_readpages(struct ceph_osd_client *osdc, | |
4401 | struct ceph_vino vino, struct ceph_file_layout *layout, | |
4402 | u64 off, u64 *plen, | |
4403 | u32 truncate_seq, u64 truncate_size, | |
b7495fc2 | 4404 | struct page **pages, int num_pages, int page_align) |
f24e9980 SW |
4405 | { |
4406 | struct ceph_osd_request *req; | |
4407 | int rc = 0; | |
4408 | ||
4409 | dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino, | |
4410 | vino.snap, off, *plen); | |
715e4cd4 | 4411 | req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 0, 1, |
f24e9980 | 4412 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, |
acead002 | 4413 | NULL, truncate_seq, truncate_size, |
153e5167 | 4414 | false); |
6816282d SW |
4415 | if (IS_ERR(req)) |
4416 | return PTR_ERR(req); | |
f24e9980 SW |
4417 | |
4418 | /* it may be a short read due to an object boundary */ | |
406e2c9f | 4419 | osd_req_op_extent_osd_data_pages(req, 0, |
a4ce40a9 | 4420 | pages, *plen, page_align, false, false); |
f24e9980 | 4421 | |
e0c59487 | 4422 | dout("readpages final extent is %llu~%llu (%llu bytes align %d)\n", |
43bfe5de | 4423 | off, *plen, *plen, page_align); |
f24e9980 SW |
4424 | |
4425 | rc = ceph_osdc_start_request(osdc, req, false); | |
4426 | if (!rc) | |
4427 | rc = ceph_osdc_wait_request(osdc, req); | |
4428 | ||
4429 | ceph_osdc_put_request(req); | |
4430 | dout("readpages result %d\n", rc); | |
4431 | return rc; | |
4432 | } | |
3d14c5d2 | 4433 | EXPORT_SYMBOL(ceph_osdc_readpages); |
f24e9980 SW |
4434 | |
4435 | /* | |
4436 | * do a synchronous write on N pages | |
4437 | */ | |
4438 | int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino, | |
4439 | struct ceph_file_layout *layout, | |
4440 | struct ceph_snap_context *snapc, | |
4441 | u64 off, u64 len, | |
4442 | u32 truncate_seq, u64 truncate_size, | |
4443 | struct timespec *mtime, | |
24808826 | 4444 | struct page **pages, int num_pages) |
f24e9980 SW |
4445 | { |
4446 | struct ceph_osd_request *req; | |
4447 | int rc = 0; | |
b7495fc2 | 4448 | int page_align = off & ~PAGE_MASK; |
f24e9980 | 4449 | |
715e4cd4 | 4450 | req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 0, 1, |
54ea0046 | 4451 | CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, |
acead002 | 4452 | snapc, truncate_seq, truncate_size, |
153e5167 | 4453 | true); |
6816282d SW |
4454 | if (IS_ERR(req)) |
4455 | return PTR_ERR(req); | |
f24e9980 SW |
4456 | |
4457 | /* it may be a short write due to an object boundary */ | |
406e2c9f | 4458 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align, |
43bfe5de AE |
4459 | false, false); |
4460 | dout("writepages %llu~%llu (%llu bytes)\n", off, len, len); | |
f24e9980 | 4461 | |
bb873b53 | 4462 | req->r_mtime = *mtime; |
87f979d3 | 4463 | rc = ceph_osdc_start_request(osdc, req, true); |
f24e9980 SW |
4464 | if (!rc) |
4465 | rc = ceph_osdc_wait_request(osdc, req); | |
4466 | ||
4467 | ceph_osdc_put_request(req); | |
4468 | if (rc == 0) | |
4469 | rc = len; | |
4470 | dout("writepages result %d\n", rc); | |
4471 | return rc; | |
4472 | } | |
3d14c5d2 | 4473 | EXPORT_SYMBOL(ceph_osdc_writepages); |
f24e9980 | 4474 | |
5522ae0b AE |
4475 | int ceph_osdc_setup(void) |
4476 | { | |
3f1af42a ID |
4477 | size_t size = sizeof(struct ceph_osd_request) + |
4478 | CEPH_OSD_SLAB_OPS * sizeof(struct ceph_osd_req_op); | |
4479 | ||
5522ae0b | 4480 | BUG_ON(ceph_osd_request_cache); |
3f1af42a ID |
4481 | ceph_osd_request_cache = kmem_cache_create("ceph_osd_request", size, |
4482 | 0, 0, NULL); | |
5522ae0b AE |
4483 | |
4484 | return ceph_osd_request_cache ? 0 : -ENOMEM; | |
4485 | } | |
4486 | EXPORT_SYMBOL(ceph_osdc_setup); | |
4487 | ||
4488 | void ceph_osdc_cleanup(void) | |
4489 | { | |
4490 | BUG_ON(!ceph_osd_request_cache); | |
4491 | kmem_cache_destroy(ceph_osd_request_cache); | |
4492 | ceph_osd_request_cache = NULL; | |
4493 | } | |
4494 | EXPORT_SYMBOL(ceph_osdc_cleanup); | |
4495 | ||
f24e9980 SW |
4496 | /* |
4497 | * handle incoming message | |
4498 | */ | |
4499 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
4500 | { | |
4501 | struct ceph_osd *osd = con->private; | |
5aea3dcd | 4502 | struct ceph_osd_client *osdc = osd->o_osdc; |
f24e9980 SW |
4503 | int type = le16_to_cpu(msg->hdr.type); |
4504 | ||
f24e9980 SW |
4505 | switch (type) { |
4506 | case CEPH_MSG_OSD_MAP: | |
4507 | ceph_osdc_handle_map(osdc, msg); | |
4508 | break; | |
4509 | case CEPH_MSG_OSD_OPREPLY: | |
5aea3dcd | 4510 | handle_reply(osd, msg); |
f24e9980 | 4511 | break; |
a40c4f10 YS |
4512 | case CEPH_MSG_WATCH_NOTIFY: |
4513 | handle_watch_notify(osdc, msg); | |
4514 | break; | |
f24e9980 SW |
4515 | |
4516 | default: | |
4517 | pr_err("received unknown message type %d %s\n", type, | |
4518 | ceph_msg_type_name(type)); | |
4519 | } | |
5aea3dcd | 4520 | |
f24e9980 SW |
4521 | ceph_msg_put(msg); |
4522 | } | |
4523 | ||
5b3a4db3 | 4524 | /* |
d15f9d69 ID |
4525 | * Lookup and return message for incoming reply. Don't try to do |
4526 | * anything about a larger than preallocated data portion of the | |
4527 | * message at the moment - for now, just skip the message. | |
5b3a4db3 SW |
4528 | */ |
4529 | static struct ceph_msg *get_reply(struct ceph_connection *con, | |
2450418c YS |
4530 | struct ceph_msg_header *hdr, |
4531 | int *skip) | |
f24e9980 SW |
4532 | { |
4533 | struct ceph_osd *osd = con->private; | |
4534 | struct ceph_osd_client *osdc = osd->o_osdc; | |
5aea3dcd | 4535 | struct ceph_msg *m = NULL; |
0547a9b3 | 4536 | struct ceph_osd_request *req; |
3f0a4ac5 | 4537 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 4538 | int data_len = le32_to_cpu(hdr->data_len); |
5aea3dcd | 4539 | u64 tid = le64_to_cpu(hdr->tid); |
f24e9980 | 4540 | |
5aea3dcd ID |
4541 | down_read(&osdc->lock); |
4542 | if (!osd_registered(osd)) { | |
4543 | dout("%s osd%d unknown, skipping\n", __func__, osd->o_osd); | |
4544 | *skip = 1; | |
4545 | goto out_unlock_osdc; | |
4546 | } | |
4547 | WARN_ON(osd->o_osd != le64_to_cpu(hdr->src.num)); | |
4548 | ||
4549 | mutex_lock(&osd->lock); | |
4550 | req = lookup_request(&osd->o_requests, tid); | |
0547a9b3 | 4551 | if (!req) { |
cd8140c6 ID |
4552 | dout("%s osd%d tid %llu unknown, skipping\n", __func__, |
4553 | osd->o_osd, tid); | |
d15f9d69 | 4554 | *skip = 1; |
5aea3dcd | 4555 | goto out_unlock_session; |
0547a9b3 | 4556 | } |
c16e7869 | 4557 | |
ace6d3a9 | 4558 | ceph_msg_revoke_incoming(req->r_reply); |
0547a9b3 | 4559 | |
f2be82b0 | 4560 | if (front_len > req->r_reply->front_alloc_len) { |
d15f9d69 ID |
4561 | pr_warn("%s osd%d tid %llu front %d > preallocated %d\n", |
4562 | __func__, osd->o_osd, req->r_tid, front_len, | |
4563 | req->r_reply->front_alloc_len); | |
3f0a4ac5 ID |
4564 | m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS, |
4565 | false); | |
a79832f2 | 4566 | if (!m) |
5aea3dcd | 4567 | goto out_unlock_session; |
c16e7869 SW |
4568 | ceph_msg_put(req->r_reply); |
4569 | req->r_reply = m; | |
4570 | } | |
0fff87ec | 4571 | |
d15f9d69 ID |
4572 | if (data_len > req->r_reply->data_length) { |
4573 | pr_warn("%s osd%d tid %llu data %d > preallocated %zu, skipping\n", | |
4574 | __func__, osd->o_osd, req->r_tid, data_len, | |
4575 | req->r_reply->data_length); | |
4576 | m = NULL; | |
4577 | *skip = 1; | |
5aea3dcd | 4578 | goto out_unlock_session; |
0547a9b3 | 4579 | } |
d15f9d69 ID |
4580 | |
4581 | m = ceph_msg_get(req->r_reply); | |
c16e7869 | 4582 | dout("get_reply tid %lld %p\n", tid, m); |
0547a9b3 | 4583 | |
5aea3dcd ID |
4584 | out_unlock_session: |
4585 | mutex_unlock(&osd->lock); | |
4586 | out_unlock_osdc: | |
4587 | up_read(&osdc->lock); | |
2450418c | 4588 | return m; |
5b3a4db3 SW |
4589 | } |
4590 | ||
19079203 ID |
4591 | /* |
4592 | * TODO: switch to a msg-owned pagelist | |
4593 | */ | |
4594 | static struct ceph_msg *alloc_msg_with_page_vector(struct ceph_msg_header *hdr) | |
4595 | { | |
4596 | struct ceph_msg *m; | |
4597 | int type = le16_to_cpu(hdr->type); | |
4598 | u32 front_len = le32_to_cpu(hdr->front_len); | |
4599 | u32 data_len = le32_to_cpu(hdr->data_len); | |
4600 | ||
4601 | m = ceph_msg_new(type, front_len, GFP_NOIO, false); | |
4602 | if (!m) | |
4603 | return NULL; | |
4604 | ||
4605 | if (data_len) { | |
4606 | struct page **pages; | |
4607 | struct ceph_osd_data osd_data; | |
4608 | ||
4609 | pages = ceph_alloc_page_vector(calc_pages_for(0, data_len), | |
4610 | GFP_NOIO); | |
c22e853a | 4611 | if (IS_ERR(pages)) { |
19079203 ID |
4612 | ceph_msg_put(m); |
4613 | return NULL; | |
4614 | } | |
4615 | ||
4616 | ceph_osd_data_pages_init(&osd_data, pages, data_len, 0, false, | |
4617 | false); | |
4618 | ceph_osdc_msg_data_add(m, &osd_data); | |
4619 | } | |
4620 | ||
4621 | return m; | |
4622 | } | |
4623 | ||
5b3a4db3 SW |
4624 | static struct ceph_msg *alloc_msg(struct ceph_connection *con, |
4625 | struct ceph_msg_header *hdr, | |
4626 | int *skip) | |
4627 | { | |
4628 | struct ceph_osd *osd = con->private; | |
4629 | int type = le16_to_cpu(hdr->type); | |
5b3a4db3 | 4630 | |
1c20f2d2 | 4631 | *skip = 0; |
5b3a4db3 SW |
4632 | switch (type) { |
4633 | case CEPH_MSG_OSD_MAP: | |
a40c4f10 | 4634 | case CEPH_MSG_WATCH_NOTIFY: |
19079203 | 4635 | return alloc_msg_with_page_vector(hdr); |
5b3a4db3 SW |
4636 | case CEPH_MSG_OSD_OPREPLY: |
4637 | return get_reply(con, hdr, skip); | |
4638 | default: | |
5aea3dcd ID |
4639 | pr_warn("%s osd%d unknown msg type %d, skipping\n", __func__, |
4640 | osd->o_osd, type); | |
5b3a4db3 SW |
4641 | *skip = 1; |
4642 | return NULL; | |
4643 | } | |
f24e9980 SW |
4644 | } |
4645 | ||
4646 | /* | |
4647 | * Wrappers to refcount containing ceph_osd struct | |
4648 | */ | |
4649 | static struct ceph_connection *get_osd_con(struct ceph_connection *con) | |
4650 | { | |
4651 | struct ceph_osd *osd = con->private; | |
4652 | if (get_osd(osd)) | |
4653 | return con; | |
4654 | return NULL; | |
4655 | } | |
4656 | ||
4657 | static void put_osd_con(struct ceph_connection *con) | |
4658 | { | |
4659 | struct ceph_osd *osd = con->private; | |
4660 | put_osd(osd); | |
4661 | } | |
4662 | ||
4e7a5dcd SW |
4663 | /* |
4664 | * authentication | |
4665 | */ | |
a3530df3 AE |
4666 | /* |
4667 | * Note: returned pointer is the address of a structure that's | |
4668 | * managed separately. Caller must *not* attempt to free it. | |
4669 | */ | |
4670 | static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con, | |
8f43fb53 | 4671 | int *proto, int force_new) |
4e7a5dcd SW |
4672 | { |
4673 | struct ceph_osd *o = con->private; | |
4674 | struct ceph_osd_client *osdc = o->o_osdc; | |
4675 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
74f1869f | 4676 | struct ceph_auth_handshake *auth = &o->o_auth; |
4e7a5dcd | 4677 | |
74f1869f | 4678 | if (force_new && auth->authorizer) { |
6c1ea260 | 4679 | ceph_auth_destroy_authorizer(auth->authorizer); |
74f1869f AE |
4680 | auth->authorizer = NULL; |
4681 | } | |
27859f97 SW |
4682 | if (!auth->authorizer) { |
4683 | int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
4684 | auth); | |
4e7a5dcd | 4685 | if (ret) |
a3530df3 | 4686 | return ERR_PTR(ret); |
27859f97 SW |
4687 | } else { |
4688 | int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD, | |
0bed9b5c SW |
4689 | auth); |
4690 | if (ret) | |
4691 | return ERR_PTR(ret); | |
4e7a5dcd | 4692 | } |
4e7a5dcd | 4693 | *proto = ac->protocol; |
74f1869f | 4694 | |
a3530df3 | 4695 | return auth; |
4e7a5dcd SW |
4696 | } |
4697 | ||
4698 | ||
0dde5848 | 4699 | static int verify_authorizer_reply(struct ceph_connection *con) |
4e7a5dcd SW |
4700 | { |
4701 | struct ceph_osd *o = con->private; | |
4702 | struct ceph_osd_client *osdc = o->o_osdc; | |
4703 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
4704 | ||
0dde5848 | 4705 | return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer); |
4e7a5dcd SW |
4706 | } |
4707 | ||
9bd2e6f8 SW |
4708 | static int invalidate_authorizer(struct ceph_connection *con) |
4709 | { | |
4710 | struct ceph_osd *o = con->private; | |
4711 | struct ceph_osd_client *osdc = o->o_osdc; | |
4712 | struct ceph_auth_client *ac = osdc->client->monc.auth; | |
4713 | ||
27859f97 | 4714 | ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD); |
9bd2e6f8 SW |
4715 | return ceph_monc_validate_auth(&osdc->client->monc); |
4716 | } | |
4e7a5dcd | 4717 | |
8cb441c0 ID |
4718 | static void osd_reencode_message(struct ceph_msg *msg) |
4719 | { | |
4720 | encode_request_finish(msg); | |
4721 | } | |
4722 | ||
79dbd1ba | 4723 | static int osd_sign_message(struct ceph_msg *msg) |
33d07337 | 4724 | { |
79dbd1ba | 4725 | struct ceph_osd *o = msg->con->private; |
33d07337 | 4726 | struct ceph_auth_handshake *auth = &o->o_auth; |
79dbd1ba | 4727 | |
33d07337 YZ |
4728 | return ceph_auth_sign_message(auth, msg); |
4729 | } | |
4730 | ||
79dbd1ba | 4731 | static int osd_check_message_signature(struct ceph_msg *msg) |
33d07337 | 4732 | { |
79dbd1ba | 4733 | struct ceph_osd *o = msg->con->private; |
33d07337 | 4734 | struct ceph_auth_handshake *auth = &o->o_auth; |
79dbd1ba | 4735 | |
33d07337 YZ |
4736 | return ceph_auth_check_message_signature(auth, msg); |
4737 | } | |
4738 | ||
9e32789f | 4739 | static const struct ceph_connection_operations osd_con_ops = { |
f24e9980 SW |
4740 | .get = get_osd_con, |
4741 | .put = put_osd_con, | |
4742 | .dispatch = dispatch, | |
4e7a5dcd SW |
4743 | .get_authorizer = get_authorizer, |
4744 | .verify_authorizer_reply = verify_authorizer_reply, | |
9bd2e6f8 | 4745 | .invalidate_authorizer = invalidate_authorizer, |
f24e9980 | 4746 | .alloc_msg = alloc_msg, |
8cb441c0 | 4747 | .reencode_message = osd_reencode_message, |
79dbd1ba ID |
4748 | .sign_message = osd_sign_message, |
4749 | .check_message_signature = osd_check_message_signature, | |
5aea3dcd | 4750 | .fault = osd_fault, |
f24e9980 | 4751 | }; |