Commit | Line | Data |
---|---|---|
d7e09d03 PT |
1 | /* |
2 | * GPL HEADER START | |
3 | * | |
4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * General Public License version 2 for more details (a copy is included | |
14 | * in the LICENSE file that accompanied this code). | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * version 2 along with this program; If not, see | |
18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf | |
19 | * | |
20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
21 | * CA 95054 USA or visit www.sun.com if you need additional information or | |
22 | * have any questions. | |
23 | * | |
24 | * GPL HEADER END | |
25 | */ | |
26 | /* | |
27 | * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. | |
28 | * Use is subject to license terms. | |
29 | * | |
1dc563a6 | 30 | * Copyright (c) 2011, 2015, Intel Corporation. |
d7e09d03 PT |
31 | */ |
32 | /* | |
33 | * This file is part of Lustre, http://www.lustre.org/ | |
34 | * Lustre is a trademark of Sun Microsystems, Inc. | |
35 | */ | |
36 | ||
37 | #define DEBUG_SUBSYSTEM S_OSC | |
38 | ||
9fdaf8c0 | 39 | #include "../../include/linux/libcfs/libcfs.h" |
d7e09d03 | 40 | |
3ee30015 GKH |
41 | #include "../include/lustre_dlm.h" |
42 | #include "../include/lustre_net.h" | |
43 | #include "../include/lustre/lustre_user.h" | |
44 | #include "../include/obd_cksum.h" | |
d7e09d03 | 45 | |
3ee30015 GKH |
46 | #include "../include/lustre_ha.h" |
47 | #include "../include/lprocfs_status.h" | |
3ee30015 GKH |
48 | #include "../include/lustre_debug.h" |
49 | #include "../include/lustre_param.h" | |
50 | #include "../include/lustre_fid.h" | |
dd45f477 | 51 | #include "../include/obd_class.h" |
aefd9d71 | 52 | #include "../include/obd.h" |
d7e09d03 PT |
53 | #include "osc_internal.h" |
54 | #include "osc_cl_internal.h" | |
55 | ||
aefd9d71 LX |
56 | atomic_t osc_pool_req_count; |
57 | unsigned int osc_reqpool_maxreqcount; | |
58 | struct ptlrpc_request_pool *osc_rq_pool; | |
59 | ||
60 | /* max memory used for request pool, unit is MB */ | |
61 | static unsigned int osc_reqpool_mem_max = 5; | |
62 | module_param(osc_reqpool_mem_max, uint, 0444); | |
63 | ||
f024bad4 JH |
64 | struct osc_brw_async_args { |
65 | struct obdo *aa_oa; | |
66 | int aa_requested_nob; | |
67 | int aa_nio_count; | |
68 | u32 aa_page_count; | |
69 | int aa_resends; | |
70 | struct brw_page **aa_ppga; | |
71 | struct client_obd *aa_cli; | |
72 | struct list_head aa_oaps; | |
73 | struct list_head aa_exts; | |
f024bad4 JH |
74 | struct cl_req *aa_clerq; |
75 | }; | |
76 | ||
77 | struct osc_async_args { | |
78 | struct obd_info *aa_oi; | |
79 | }; | |
80 | ||
81 | struct osc_setattr_args { | |
82 | struct obdo *sa_oa; | |
83 | obd_enqueue_update_f sa_upcall; | |
84 | void *sa_cookie; | |
85 | }; | |
86 | ||
87 | struct osc_fsync_args { | |
88 | struct obd_info *fa_oi; | |
89 | obd_enqueue_update_f fa_upcall; | |
90 | void *fa_cookie; | |
91 | }; | |
92 | ||
93 | struct osc_enqueue_args { | |
94 | struct obd_export *oa_exp; | |
95 | __u64 *oa_flags; | |
96 | obd_enqueue_update_f oa_upcall; | |
97 | void *oa_cookie; | |
98 | struct ost_lvb *oa_lvb; | |
99 | struct lustre_handle *oa_lockh; | |
100 | struct ldlm_enqueue_info *oa_ei; | |
101 | unsigned int oa_agl:1; | |
102 | }; | |
103 | ||
21aef7d9 | 104 | static void osc_release_ppga(struct brw_page **ppga, u32 count); |
d7e09d03 PT |
105 | static int brw_interpret(const struct lu_env *env, |
106 | struct ptlrpc_request *req, void *data, int rc); | |
d7e09d03 PT |
107 | |
108 | /* Pack OSC object metadata for disk storage (LE byte order). */ | |
109 | static int osc_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, | |
110 | struct lov_stripe_md *lsm) | |
111 | { | |
112 | int lmm_size; | |
d7e09d03 PT |
113 | |
114 | lmm_size = sizeof(**lmmp); | |
7f1ae4c0 | 115 | if (!lmmp) |
0a3bdb00 | 116 | return lmm_size; |
d7e09d03 | 117 | |
7f1ae4c0 | 118 | if (*lmmp && !lsm) { |
7795178d | 119 | kfree(*lmmp); |
d7e09d03 | 120 | *lmmp = NULL; |
0a3bdb00 | 121 | return 0; |
7f1ae4c0 | 122 | } else if (unlikely(lsm && ostid_id(&lsm->lsm_oi) == 0)) { |
0a3bdb00 | 123 | return -EBADF; |
d7e09d03 PT |
124 | } |
125 | ||
7f1ae4c0 | 126 | if (!*lmmp) { |
7795178d | 127 | *lmmp = kzalloc(lmm_size, GFP_NOFS); |
3408e9ae | 128 | if (!*lmmp) |
0a3bdb00 | 129 | return -ENOMEM; |
d7e09d03 PT |
130 | } |
131 | ||
132 | if (lsm) | |
133 | ostid_cpu_to_le(&lsm->lsm_oi, &(*lmmp)->lmm_oi); | |
134 | ||
0a3bdb00 | 135 | return lmm_size; |
d7e09d03 PT |
136 | } |
137 | ||
138 | /* Unpack OSC object metadata from disk storage (LE byte order). */ | |
139 | static int osc_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp, | |
140 | struct lov_mds_md *lmm, int lmm_bytes) | |
141 | { | |
142 | int lsm_size; | |
143 | struct obd_import *imp = class_exp2cliimp(exp); | |
d7e09d03 | 144 | |
7f1ae4c0 | 145 | if (lmm) { |
d7e09d03 PT |
146 | if (lmm_bytes < sizeof(*lmm)) { |
147 | CERROR("%s: lov_mds_md too small: %d, need %d\n", | |
148 | exp->exp_obd->obd_name, lmm_bytes, | |
149 | (int)sizeof(*lmm)); | |
0a3bdb00 | 150 | return -EINVAL; |
d7e09d03 PT |
151 | } |
152 | /* XXX LOV_MAGIC etc check? */ | |
153 | ||
154 | if (unlikely(ostid_id(&lmm->lmm_oi) == 0)) { | |
155 | CERROR("%s: zero lmm_object_id: rc = %d\n", | |
156 | exp->exp_obd->obd_name, -EINVAL); | |
0a3bdb00 | 157 | return -EINVAL; |
d7e09d03 PT |
158 | } |
159 | } | |
160 | ||
161 | lsm_size = lov_stripe_md_size(1); | |
7f1ae4c0 | 162 | if (!lsmp) |
0a3bdb00 | 163 | return lsm_size; |
d7e09d03 | 164 | |
7f1ae4c0 | 165 | if (*lsmp && !lmm) { |
7795178d JL |
166 | kfree((*lsmp)->lsm_oinfo[0]); |
167 | kfree(*lsmp); | |
d7e09d03 | 168 | *lsmp = NULL; |
0a3bdb00 | 169 | return 0; |
d7e09d03 PT |
170 | } |
171 | ||
7f1ae4c0 | 172 | if (!*lsmp) { |
7795178d | 173 | *lsmp = kzalloc(lsm_size, GFP_NOFS); |
7f1ae4c0 | 174 | if (unlikely(!*lsmp)) |
0a3bdb00 | 175 | return -ENOMEM; |
7795178d JL |
176 | (*lsmp)->lsm_oinfo[0] = kzalloc(sizeof(struct lov_oinfo), |
177 | GFP_NOFS); | |
7f1ae4c0 | 178 | if (unlikely(!(*lsmp)->lsm_oinfo[0])) { |
7795178d | 179 | kfree(*lsmp); |
0a3bdb00 | 180 | return -ENOMEM; |
d7e09d03 PT |
181 | } |
182 | loi_init((*lsmp)->lsm_oinfo[0]); | |
183 | } else if (unlikely(ostid_id(&(*lsmp)->lsm_oi) == 0)) { | |
0a3bdb00 | 184 | return -EBADF; |
d7e09d03 PT |
185 | } |
186 | ||
7f1ae4c0 | 187 | if (lmm) |
d7e09d03 PT |
188 | /* XXX zero *lsmp? */ |
189 | ostid_le_to_cpu(&lmm->lmm_oi, &(*lsmp)->lsm_oi); | |
190 | ||
7f1ae4c0 | 191 | if (imp && |
d7e09d03 PT |
192 | (imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES)) |
193 | (*lsmp)->lsm_maxbytes = imp->imp_connect_data.ocd_maxbytes; | |
194 | else | |
195 | (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES; | |
196 | ||
0a3bdb00 | 197 | return lsm_size; |
d7e09d03 PT |
198 | } |
199 | ||
d7e09d03 PT |
200 | static inline void osc_pack_req_body(struct ptlrpc_request *req, |
201 | struct obd_info *oinfo) | |
202 | { | |
203 | struct ost_body *body; | |
204 | ||
205 | body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); | |
206 | LASSERT(body); | |
207 | ||
3b2f75fd | 208 | lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, |
209 | oinfo->oi_oa); | |
d7e09d03 PT |
210 | } |
211 | ||
212 | static int osc_getattr_interpret(const struct lu_env *env, | |
213 | struct ptlrpc_request *req, | |
214 | struct osc_async_args *aa, int rc) | |
215 | { | |
216 | struct ost_body *body; | |
d7e09d03 PT |
217 | |
218 | if (rc != 0) | |
26c4ea46 | 219 | goto out; |
d7e09d03 PT |
220 | |
221 | body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); | |
222 | if (body) { | |
223 | CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode); | |
3b2f75fd | 224 | lustre_get_wire_obdo(&req->rq_import->imp_connect_data, |
225 | aa->aa_oi->oi_oa, &body->oa); | |
d7e09d03 PT |
226 | |
227 | /* This should really be sent by the OST */ | |
228 | aa->aa_oi->oi_oa->o_blksize = DT_MAX_BRW_SIZE; | |
229 | aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ; | |
230 | } else { | |
231 | CDEBUG(D_INFO, "can't unpack ost_body\n"); | |
232 | rc = -EPROTO; | |
233 | aa->aa_oi->oi_oa->o_valid = 0; | |
234 | } | |
235 | out: | |
236 | rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc); | |
0a3bdb00 | 237 | return rc; |
d7e09d03 PT |
238 | } |
239 | ||
240 | static int osc_getattr_async(struct obd_export *exp, struct obd_info *oinfo, | |
241 | struct ptlrpc_request_set *set) | |
242 | { | |
243 | struct ptlrpc_request *req; | |
244 | struct osc_async_args *aa; | |
29ac6840 | 245 | int rc; |
d7e09d03 PT |
246 | |
247 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR); | |
7f1ae4c0 | 248 | if (!req) |
0a3bdb00 | 249 | return -ENOMEM; |
d7e09d03 | 250 | |
d7e09d03 PT |
251 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR); |
252 | if (rc) { | |
253 | ptlrpc_request_free(req); | |
0a3bdb00 | 254 | return rc; |
d7e09d03 PT |
255 | } |
256 | ||
257 | osc_pack_req_body(req, oinfo); | |
258 | ||
259 | ptlrpc_request_set_replen(req); | |
260 | req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_getattr_interpret; | |
261 | ||
262 | CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args)); | |
263 | aa = ptlrpc_req_async_args(req); | |
264 | aa->aa_oi = oinfo; | |
265 | ||
266 | ptlrpc_set_add_req(set, req); | |
0a3bdb00 | 267 | return 0; |
d7e09d03 PT |
268 | } |
269 | ||
270 | static int osc_getattr(const struct lu_env *env, struct obd_export *exp, | |
271 | struct obd_info *oinfo) | |
272 | { | |
273 | struct ptlrpc_request *req; | |
29ac6840 CH |
274 | struct ost_body *body; |
275 | int rc; | |
d7e09d03 PT |
276 | |
277 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_GETATTR); | |
7f1ae4c0 | 278 | if (!req) |
0a3bdb00 | 279 | return -ENOMEM; |
d7e09d03 | 280 | |
d7e09d03 PT |
281 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GETATTR); |
282 | if (rc) { | |
283 | ptlrpc_request_free(req); | |
0a3bdb00 | 284 | return rc; |
d7e09d03 PT |
285 | } |
286 | ||
287 | osc_pack_req_body(req, oinfo); | |
288 | ||
289 | ptlrpc_request_set_replen(req); | |
290 | ||
291 | rc = ptlrpc_queue_wait(req); | |
292 | if (rc) | |
26c4ea46 | 293 | goto out; |
d7e09d03 PT |
294 | |
295 | body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); | |
7f1ae4c0 | 296 | if (!body) { |
26c4ea46 TJ |
297 | rc = -EPROTO; |
298 | goto out; | |
299 | } | |
d7e09d03 PT |
300 | |
301 | CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode); | |
3b2f75fd | 302 | lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa, |
303 | &body->oa); | |
d7e09d03 PT |
304 | |
305 | oinfo->oi_oa->o_blksize = cli_brw_size(exp->exp_obd); | |
306 | oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ; | |
307 | ||
d7e09d03 PT |
308 | out: |
309 | ptlrpc_req_finished(req); | |
310 | return rc; | |
311 | } | |
312 | ||
313 | static int osc_setattr(const struct lu_env *env, struct obd_export *exp, | |
314 | struct obd_info *oinfo, struct obd_trans_info *oti) | |
315 | { | |
316 | struct ptlrpc_request *req; | |
29ac6840 CH |
317 | struct ost_body *body; |
318 | int rc; | |
d7e09d03 PT |
319 | |
320 | LASSERT(oinfo->oi_oa->o_valid & OBD_MD_FLGROUP); | |
321 | ||
322 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR); | |
7f1ae4c0 | 323 | if (!req) |
0a3bdb00 | 324 | return -ENOMEM; |
d7e09d03 | 325 | |
d7e09d03 PT |
326 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR); |
327 | if (rc) { | |
328 | ptlrpc_request_free(req); | |
0a3bdb00 | 329 | return rc; |
d7e09d03 PT |
330 | } |
331 | ||
332 | osc_pack_req_body(req, oinfo); | |
333 | ||
334 | ptlrpc_request_set_replen(req); | |
335 | ||
336 | rc = ptlrpc_queue_wait(req); | |
337 | if (rc) | |
26c4ea46 | 338 | goto out; |
d7e09d03 PT |
339 | |
340 | body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); | |
7f1ae4c0 | 341 | if (!body) { |
26c4ea46 TJ |
342 | rc = -EPROTO; |
343 | goto out; | |
344 | } | |
d7e09d03 | 345 | |
3b2f75fd | 346 | lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa, |
347 | &body->oa); | |
d7e09d03 | 348 | |
d7e09d03 PT |
349 | out: |
350 | ptlrpc_req_finished(req); | |
0a3bdb00 | 351 | return rc; |
d7e09d03 PT |
352 | } |
353 | ||
354 | static int osc_setattr_interpret(const struct lu_env *env, | |
355 | struct ptlrpc_request *req, | |
356 | struct osc_setattr_args *sa, int rc) | |
357 | { | |
358 | struct ost_body *body; | |
d7e09d03 PT |
359 | |
360 | if (rc != 0) | |
26c4ea46 | 361 | goto out; |
d7e09d03 PT |
362 | |
363 | body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); | |
7f1ae4c0 | 364 | if (!body) { |
26c4ea46 TJ |
365 | rc = -EPROTO; |
366 | goto out; | |
367 | } | |
d7e09d03 | 368 | |
3b2f75fd | 369 | lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa, |
370 | &body->oa); | |
d7e09d03 PT |
371 | out: |
372 | rc = sa->sa_upcall(sa->sa_cookie, rc); | |
0a3bdb00 | 373 | return rc; |
d7e09d03 PT |
374 | } |
375 | ||
376 | int osc_setattr_async_base(struct obd_export *exp, struct obd_info *oinfo, | |
377 | struct obd_trans_info *oti, | |
378 | obd_enqueue_update_f upcall, void *cookie, | |
379 | struct ptlrpc_request_set *rqset) | |
380 | { | |
29ac6840 | 381 | struct ptlrpc_request *req; |
d7e09d03 | 382 | struct osc_setattr_args *sa; |
29ac6840 | 383 | int rc; |
d7e09d03 PT |
384 | |
385 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SETATTR); | |
7f1ae4c0 | 386 | if (!req) |
0a3bdb00 | 387 | return -ENOMEM; |
d7e09d03 | 388 | |
d7e09d03 PT |
389 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SETATTR); |
390 | if (rc) { | |
391 | ptlrpc_request_free(req); | |
0a3bdb00 | 392 | return rc; |
d7e09d03 PT |
393 | } |
394 | ||
395 | if (oti && oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) | |
396 | oinfo->oi_oa->o_lcookie = *oti->oti_logcookies; | |
397 | ||
398 | osc_pack_req_body(req, oinfo); | |
399 | ||
400 | ptlrpc_request_set_replen(req); | |
401 | ||
402 | /* do mds to ost setattr asynchronously */ | |
403 | if (!rqset) { | |
404 | /* Do not wait for response. */ | |
c5c4c6fa | 405 | ptlrpcd_add_req(req); |
d7e09d03 PT |
406 | } else { |
407 | req->rq_interpret_reply = | |
408 | (ptlrpc_interpterer_t)osc_setattr_interpret; | |
409 | ||
e72f36e2 | 410 | CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args)); |
d7e09d03 PT |
411 | sa = ptlrpc_req_async_args(req); |
412 | sa->sa_oa = oinfo->oi_oa; | |
413 | sa->sa_upcall = upcall; | |
414 | sa->sa_cookie = cookie; | |
415 | ||
416 | if (rqset == PTLRPCD_SET) | |
c5c4c6fa | 417 | ptlrpcd_add_req(req); |
d7e09d03 PT |
418 | else |
419 | ptlrpc_set_add_req(rqset, req); | |
420 | } | |
421 | ||
0a3bdb00 | 422 | return 0; |
d7e09d03 PT |
423 | } |
424 | ||
425 | static int osc_setattr_async(struct obd_export *exp, struct obd_info *oinfo, | |
426 | struct obd_trans_info *oti, | |
427 | struct ptlrpc_request_set *rqset) | |
428 | { | |
429 | return osc_setattr_async_base(exp, oinfo, oti, | |
430 | oinfo->oi_cb_up, oinfo, rqset); | |
431 | } | |
432 | ||
74d4ec11 SB |
433 | static int osc_real_create(struct obd_export *exp, struct obdo *oa, |
434 | struct lov_stripe_md **ea, | |
435 | struct obd_trans_info *oti) | |
d7e09d03 PT |
436 | { |
437 | struct ptlrpc_request *req; | |
29ac6840 CH |
438 | struct ost_body *body; |
439 | struct lov_stripe_md *lsm; | |
440 | int rc; | |
d7e09d03 PT |
441 | |
442 | LASSERT(oa); | |
443 | LASSERT(ea); | |
444 | ||
445 | lsm = *ea; | |
446 | if (!lsm) { | |
447 | rc = obd_alloc_memmd(exp, &lsm); | |
448 | if (rc < 0) | |
0a3bdb00 | 449 | return rc; |
d7e09d03 PT |
450 | } |
451 | ||
452 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE); | |
7f1ae4c0 | 453 | if (!req) { |
26c4ea46 TJ |
454 | rc = -ENOMEM; |
455 | goto out; | |
456 | } | |
d7e09d03 PT |
457 | |
458 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_CREATE); | |
459 | if (rc) { | |
460 | ptlrpc_request_free(req); | |
26c4ea46 | 461 | goto out; |
d7e09d03 PT |
462 | } |
463 | ||
464 | body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); | |
465 | LASSERT(body); | |
3b2f75fd | 466 | |
467 | lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa); | |
d7e09d03 PT |
468 | |
469 | ptlrpc_request_set_replen(req); | |
470 | ||
471 | if ((oa->o_valid & OBD_MD_FLFLAGS) && | |
472 | oa->o_flags == OBD_FL_DELORPHAN) { | |
473 | DEBUG_REQ(D_HA, req, | |
474 | "delorphan from OST integration"); | |
475 | /* Don't resend the delorphan req */ | |
476 | req->rq_no_resend = req->rq_no_delay = 1; | |
477 | } | |
478 | ||
479 | rc = ptlrpc_queue_wait(req); | |
480 | if (rc) | |
26c4ea46 | 481 | goto out_req; |
d7e09d03 PT |
482 | |
483 | body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); | |
7f1ae4c0 | 484 | if (!body) { |
26c4ea46 TJ |
485 | rc = -EPROTO; |
486 | goto out_req; | |
487 | } | |
d7e09d03 | 488 | |
3b2f75fd | 489 | CDEBUG(D_INFO, "oa flags %x\n", oa->o_flags); |
490 | lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa); | |
d7e09d03 PT |
491 | |
492 | oa->o_blksize = cli_brw_size(exp->exp_obd); | |
493 | oa->o_valid |= OBD_MD_FLBLKSZ; | |
494 | ||
495 | /* XXX LOV STACKING: the lsm that is passed to us from LOV does not | |
496 | * have valid lsm_oinfo data structs, so don't go touching that. | |
497 | * This needs to be fixed in a big way. | |
498 | */ | |
499 | lsm->lsm_oi = oa->o_oi; | |
500 | *ea = lsm; | |
501 | ||
7f1ae4c0 | 502 | if (oti) { |
d7e09d03 PT |
503 | oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg); |
504 | ||
505 | if (oa->o_valid & OBD_MD_FLCOOKIE) { | |
506 | if (!oti->oti_logcookies) | |
507 | oti_alloc_cookies(oti, 1); | |
508 | *oti->oti_logcookies = oa->o_lcookie; | |
509 | } | |
510 | } | |
511 | ||
f537dd2c | 512 | CDEBUG(D_HA, "transno: %lld\n", |
d7e09d03 PT |
513 | lustre_msg_get_transno(req->rq_repmsg)); |
514 | out_req: | |
515 | ptlrpc_req_finished(req); | |
516 | out: | |
517 | if (rc && !*ea) | |
518 | obd_free_memmd(exp, &lsm); | |
0a3bdb00 | 519 | return rc; |
d7e09d03 PT |
520 | } |
521 | ||
522 | int osc_punch_base(struct obd_export *exp, struct obd_info *oinfo, | |
523 | obd_enqueue_update_f upcall, void *cookie, | |
524 | struct ptlrpc_request_set *rqset) | |
525 | { | |
29ac6840 | 526 | struct ptlrpc_request *req; |
d7e09d03 | 527 | struct osc_setattr_args *sa; |
29ac6840 CH |
528 | struct ost_body *body; |
529 | int rc; | |
d7e09d03 PT |
530 | |
531 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_PUNCH); | |
7f1ae4c0 | 532 | if (!req) |
0a3bdb00 | 533 | return -ENOMEM; |
d7e09d03 | 534 | |
d7e09d03 PT |
535 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_PUNCH); |
536 | if (rc) { | |
537 | ptlrpc_request_free(req); | |
0a3bdb00 | 538 | return rc; |
d7e09d03 PT |
539 | } |
540 | req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */ | |
541 | ptlrpc_at_set_req_timeout(req); | |
542 | ||
543 | body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); | |
544 | LASSERT(body); | |
3b2f75fd | 545 | lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, |
546 | oinfo->oi_oa); | |
d7e09d03 PT |
547 | |
548 | ptlrpc_request_set_replen(req); | |
549 | ||
550 | req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_setattr_interpret; | |
e72f36e2 | 551 | CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args)); |
d7e09d03 | 552 | sa = ptlrpc_req_async_args(req); |
29ac6840 | 553 | sa->sa_oa = oinfo->oi_oa; |
d7e09d03 PT |
554 | sa->sa_upcall = upcall; |
555 | sa->sa_cookie = cookie; | |
556 | if (rqset == PTLRPCD_SET) | |
c5c4c6fa | 557 | ptlrpcd_add_req(req); |
d7e09d03 PT |
558 | else |
559 | ptlrpc_set_add_req(rqset, req); | |
560 | ||
0a3bdb00 | 561 | return 0; |
d7e09d03 PT |
562 | } |
563 | ||
d7e09d03 PT |
564 | static int osc_sync_interpret(const struct lu_env *env, |
565 | struct ptlrpc_request *req, | |
566 | void *arg, int rc) | |
567 | { | |
568 | struct osc_fsync_args *fa = arg; | |
569 | struct ost_body *body; | |
d7e09d03 PT |
570 | |
571 | if (rc) | |
26c4ea46 | 572 | goto out; |
d7e09d03 PT |
573 | |
574 | body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); | |
7f1ae4c0 | 575 | if (!body) { |
e72f36e2 | 576 | CERROR("can't unpack ost_body\n"); |
26c4ea46 TJ |
577 | rc = -EPROTO; |
578 | goto out; | |
d7e09d03 PT |
579 | } |
580 | ||
581 | *fa->fa_oi->oi_oa = body->oa; | |
582 | out: | |
583 | rc = fa->fa_upcall(fa->fa_cookie, rc); | |
0a3bdb00 | 584 | return rc; |
d7e09d03 PT |
585 | } |
586 | ||
587 | int osc_sync_base(struct obd_export *exp, struct obd_info *oinfo, | |
588 | obd_enqueue_update_f upcall, void *cookie, | |
589 | struct ptlrpc_request_set *rqset) | |
590 | { | |
591 | struct ptlrpc_request *req; | |
29ac6840 | 592 | struct ost_body *body; |
d7e09d03 | 593 | struct osc_fsync_args *fa; |
29ac6840 | 594 | int rc; |
d7e09d03 PT |
595 | |
596 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_SYNC); | |
7f1ae4c0 | 597 | if (!req) |
0a3bdb00 | 598 | return -ENOMEM; |
d7e09d03 | 599 | |
d7e09d03 PT |
600 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC); |
601 | if (rc) { | |
602 | ptlrpc_request_free(req); | |
0a3bdb00 | 603 | return rc; |
d7e09d03 PT |
604 | } |
605 | ||
606 | /* overload the size and blocks fields in the oa with start/end */ | |
607 | body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); | |
608 | LASSERT(body); | |
3b2f75fd | 609 | lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, |
610 | oinfo->oi_oa); | |
d7e09d03 PT |
611 | |
612 | ptlrpc_request_set_replen(req); | |
613 | req->rq_interpret_reply = osc_sync_interpret; | |
614 | ||
615 | CLASSERT(sizeof(*fa) <= sizeof(req->rq_async_args)); | |
616 | fa = ptlrpc_req_async_args(req); | |
617 | fa->fa_oi = oinfo; | |
618 | fa->fa_upcall = upcall; | |
619 | fa->fa_cookie = cookie; | |
620 | ||
621 | if (rqset == PTLRPCD_SET) | |
c5c4c6fa | 622 | ptlrpcd_add_req(req); |
d7e09d03 PT |
623 | else |
624 | ptlrpc_set_add_req(rqset, req); | |
625 | ||
0a3bdb00 | 626 | return 0; |
d7e09d03 PT |
627 | } |
628 | ||
d7e09d03 PT |
629 | /* Find and cancel locally locks matched by @mode in the resource found by |
630 | * @objid. Found locks are added into @cancel list. Returns the amount of | |
30aa9c52 OD |
631 | * locks added to @cancels list. |
632 | */ | |
d7e09d03 PT |
633 | static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa, |
634 | struct list_head *cancels, | |
52ee0d20 | 635 | enum ldlm_mode mode, __u64 lock_flags) |
d7e09d03 PT |
636 | { |
637 | struct ldlm_namespace *ns = exp->exp_obd->obd_namespace; | |
638 | struct ldlm_res_id res_id; | |
639 | struct ldlm_resource *res; | |
640 | int count; | |
d7e09d03 PT |
641 | |
642 | /* Return, i.e. cancel nothing, only if ELC is supported (flag in | |
643 | * export) but disabled through procfs (flag in NS). | |
644 | * | |
645 | * This distinguishes from a case when ELC is not supported originally, | |
646 | * when we still want to cancel locks in advance and just cancel them | |
30aa9c52 OD |
647 | * locally, without sending any RPC. |
648 | */ | |
d7e09d03 | 649 | if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns)) |
0a3bdb00 | 650 | return 0; |
d7e09d03 PT |
651 | |
652 | ostid_build_res_name(&oa->o_oi, &res_id); | |
653 | res = ldlm_resource_get(ns, NULL, &res_id, 0, 0); | |
7f1ae4c0 | 654 | if (!res) |
0a3bdb00 | 655 | return 0; |
d7e09d03 PT |
656 | |
657 | LDLM_RESOURCE_ADDREF(res); | |
658 | count = ldlm_cancel_resource_local(res, cancels, NULL, mode, | |
659 | lock_flags, 0, NULL); | |
660 | LDLM_RESOURCE_DELREF(res); | |
661 | ldlm_resource_putref(res); | |
0a3bdb00 | 662 | return count; |
d7e09d03 PT |
663 | } |
664 | ||
665 | static int osc_destroy_interpret(const struct lu_env *env, | |
666 | struct ptlrpc_request *req, void *data, | |
667 | int rc) | |
668 | { | |
669 | struct client_obd *cli = &req->rq_import->imp_obd->u.cli; | |
670 | ||
671 | atomic_dec(&cli->cl_destroy_in_flight); | |
672 | wake_up(&cli->cl_destroy_waitq); | |
673 | return 0; | |
674 | } | |
675 | ||
676 | static int osc_can_send_destroy(struct client_obd *cli) | |
677 | { | |
678 | if (atomic_inc_return(&cli->cl_destroy_in_flight) <= | |
679 | cli->cl_max_rpcs_in_flight) { | |
680 | /* The destroy request can be sent */ | |
681 | return 1; | |
682 | } | |
683 | if (atomic_dec_return(&cli->cl_destroy_in_flight) < | |
684 | cli->cl_max_rpcs_in_flight) { | |
685 | /* | |
686 | * The counter has been modified between the two atomic | |
687 | * operations. | |
688 | */ | |
689 | wake_up(&cli->cl_destroy_waitq); | |
690 | } | |
691 | return 0; | |
692 | } | |
693 | ||
74d4ec11 SB |
694 | static int osc_create(const struct lu_env *env, struct obd_export *exp, |
695 | struct obdo *oa, struct lov_stripe_md **ea, | |
696 | struct obd_trans_info *oti) | |
d7e09d03 PT |
697 | { |
698 | int rc = 0; | |
d7e09d03 PT |
699 | |
700 | LASSERT(oa); | |
701 | LASSERT(ea); | |
702 | LASSERT(oa->o_valid & OBD_MD_FLGROUP); | |
703 | ||
704 | if ((oa->o_valid & OBD_MD_FLFLAGS) && | |
705 | oa->o_flags == OBD_FL_RECREATE_OBJS) { | |
0a3bdb00 | 706 | return osc_real_create(exp, oa, ea, oti); |
d7e09d03 PT |
707 | } |
708 | ||
709 | if (!fid_seq_is_mdt(ostid_seq(&oa->o_oi))) | |
0a3bdb00 | 710 | return osc_real_create(exp, oa, ea, oti); |
d7e09d03 PT |
711 | |
712 | /* we should not get here anymore */ | |
713 | LBUG(); | |
714 | ||
0a3bdb00 | 715 | return rc; |
d7e09d03 PT |
716 | } |
717 | ||
718 | /* Destroy requests can be async always on the client, and we don't even really | |
719 | * care about the return code since the client cannot do anything at all about | |
720 | * a destroy failure. | |
721 | * When the MDS is unlinking a filename, it saves the file objects into a | |
722 | * recovery llog, and these object records are cancelled when the OST reports | |
723 | * they were destroyed and sync'd to disk (i.e. transaction committed). | |
724 | * If the client dies, or the OST is down when the object should be destroyed, | |
725 | * the records are not cancelled, and when the OST reconnects to the MDS next, | |
726 | * it will retrieve the llog unlink logs and then sends the log cancellation | |
30aa9c52 OD |
727 | * cookies to the MDS after committing destroy transactions. |
728 | */ | |
d7e09d03 PT |
729 | static int osc_destroy(const struct lu_env *env, struct obd_export *exp, |
730 | struct obdo *oa, struct lov_stripe_md *ea, | |
ef2e0f55 | 731 | struct obd_trans_info *oti, struct obd_export *md_export) |
d7e09d03 | 732 | { |
29ac6840 | 733 | struct client_obd *cli = &exp->exp_obd->u.cli; |
d7e09d03 | 734 | struct ptlrpc_request *req; |
29ac6840 | 735 | struct ost_body *body; |
d7e09d03 PT |
736 | LIST_HEAD(cancels); |
737 | int rc, count; | |
d7e09d03 PT |
738 | |
739 | if (!oa) { | |
740 | CDEBUG(D_INFO, "oa NULL\n"); | |
0a3bdb00 | 741 | return -EINVAL; |
d7e09d03 PT |
742 | } |
743 | ||
744 | count = osc_resource_get_unused(exp, oa, &cancels, LCK_PW, | |
745 | LDLM_FL_DISCARD_DATA); | |
746 | ||
747 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_DESTROY); | |
7f1ae4c0 | 748 | if (!req) { |
d7e09d03 | 749 | ldlm_lock_list_put(&cancels, l_bl_ast, count); |
0a3bdb00 | 750 | return -ENOMEM; |
d7e09d03 PT |
751 | } |
752 | ||
d7e09d03 PT |
753 | rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY, |
754 | 0, &cancels, count); | |
755 | if (rc) { | |
756 | ptlrpc_request_free(req); | |
0a3bdb00 | 757 | return rc; |
d7e09d03 PT |
758 | } |
759 | ||
760 | req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */ | |
761 | ptlrpc_at_set_req_timeout(req); | |
762 | ||
7f1ae4c0 | 763 | if (oti && oa->o_valid & OBD_MD_FLCOOKIE) |
d7e09d03 PT |
764 | oa->o_lcookie = *oti->oti_logcookies; |
765 | body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); | |
766 | LASSERT(body); | |
3b2f75fd | 767 | lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa); |
d7e09d03 | 768 | |
d7e09d03 PT |
769 | ptlrpc_request_set_replen(req); |
770 | ||
11d66e89 | 771 | /* If osc_destroy is for destroying the unlink orphan, |
d7e09d03 PT |
772 | * sent from MDT to OST, which should not be blocked here, |
773 | * because the process might be triggered by ptlrpcd, and | |
30aa9c52 OD |
774 | * it is not good to block ptlrpcd thread (b=16006 |
775 | **/ | |
d7e09d03 PT |
776 | if (!(oa->o_flags & OBD_FL_DELORPHAN)) { |
777 | req->rq_interpret_reply = osc_destroy_interpret; | |
778 | if (!osc_can_send_destroy(cli)) { | |
779 | struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, | |
780 | NULL); | |
781 | ||
782 | /* | |
783 | * Wait until the number of on-going destroy RPCs drops | |
784 | * under max_rpc_in_flight | |
785 | */ | |
786 | l_wait_event_exclusive(cli->cl_destroy_waitq, | |
787 | osc_can_send_destroy(cli), &lwi); | |
788 | } | |
789 | } | |
790 | ||
791 | /* Do not wait for response */ | |
c5c4c6fa | 792 | ptlrpcd_add_req(req); |
0a3bdb00 | 793 | return 0; |
d7e09d03 PT |
794 | } |
795 | ||
796 | static void osc_announce_cached(struct client_obd *cli, struct obdo *oa, | |
797 | long writing_bytes) | |
798 | { | |
21aef7d9 | 799 | u32 bits = OBD_MD_FLBLOCKS|OBD_MD_FLGRANT; |
d7e09d03 PT |
800 | |
801 | LASSERT(!(oa->o_valid & bits)); | |
802 | ||
803 | oa->o_valid |= bits; | |
804 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
805 | oa->o_dirty = cli->cl_dirty; | |
806 | if (unlikely(cli->cl_dirty - cli->cl_dirty_transit > | |
807 | cli->cl_dirty_max)) { | |
808 | CERROR("dirty %lu - %lu > dirty_max %lu\n", | |
809 | cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max); | |
810 | oa->o_undirty = 0; | |
c52f69c5 | 811 | } else if (unlikely(atomic_read(&obd_dirty_pages) - |
d7e09d03 PT |
812 | atomic_read(&obd_dirty_transit_pages) > |
813 | (long)(obd_max_dirty_pages + 1))) { | |
814 | /* The atomic_read() allowing the atomic_inc() are | |
815 | * not covered by a lock thus they may safely race and trip | |
30aa9c52 OD |
816 | * this CERROR() unless we add in a small fudge factor (+1). |
817 | */ | |
c52f69c5 | 818 | CERROR("dirty %d - %d > system dirty_max %d\n", |
d7e09d03 PT |
819 | atomic_read(&obd_dirty_pages), |
820 | atomic_read(&obd_dirty_transit_pages), | |
821 | obd_max_dirty_pages); | |
822 | oa->o_undirty = 0; | |
823 | } else if (unlikely(cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff)) { | |
824 | CERROR("dirty %lu - dirty_max %lu too big???\n", | |
825 | cli->cl_dirty, cli->cl_dirty_max); | |
826 | oa->o_undirty = 0; | |
827 | } else { | |
828 | long max_in_flight = (cli->cl_max_pages_per_rpc << | |
09cbfeaf | 829 | PAGE_SHIFT)* |
d7e09d03 PT |
830 | (cli->cl_max_rpcs_in_flight + 1); |
831 | oa->o_undirty = max(cli->cl_dirty_max, max_in_flight); | |
832 | } | |
833 | oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant; | |
834 | oa->o_dropped = cli->cl_lost_grant; | |
835 | cli->cl_lost_grant = 0; | |
836 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
1d8cb70c | 837 | CDEBUG(D_CACHE, "dirty: %llu undirty: %u dropped %u grant: %llu\n", |
d7e09d03 PT |
838 | oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant); |
839 | ||
840 | } | |
841 | ||
842 | void osc_update_next_shrink(struct client_obd *cli) | |
843 | { | |
844 | cli->cl_next_shrink_grant = | |
845 | cfs_time_shift(cli->cl_grant_shrink_interval); | |
72a87fca | 846 | CDEBUG(D_CACHE, "next time %ld to shrink grant\n", |
d7e09d03 PT |
847 | cli->cl_next_shrink_grant); |
848 | } | |
849 | ||
21aef7d9 | 850 | static void __osc_update_grant(struct client_obd *cli, u64 grant) |
d7e09d03 PT |
851 | { |
852 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
853 | cli->cl_avail_grant += grant; | |
854 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
855 | } | |
856 | ||
857 | static void osc_update_grant(struct client_obd *cli, struct ost_body *body) | |
858 | { | |
859 | if (body->oa.o_valid & OBD_MD_FLGRANT) { | |
b0f5aad5 | 860 | CDEBUG(D_CACHE, "got %llu extra grant\n", body->oa.o_grant); |
d7e09d03 PT |
861 | __osc_update_grant(cli, body->oa.o_grant); |
862 | } | |
863 | } | |
864 | ||
865 | static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp, | |
21aef7d9 | 866 | u32 keylen, void *key, u32 vallen, |
d7e09d03 PT |
867 | void *val, struct ptlrpc_request_set *set); |
868 | ||
869 | static int osc_shrink_grant_interpret(const struct lu_env *env, | |
870 | struct ptlrpc_request *req, | |
871 | void *aa, int rc) | |
872 | { | |
873 | struct client_obd *cli = &req->rq_import->imp_obd->u.cli; | |
f024bad4 | 874 | struct obdo *oa = ((struct osc_brw_async_args *)aa)->aa_oa; |
d7e09d03 PT |
875 | struct ost_body *body; |
876 | ||
877 | if (rc != 0) { | |
878 | __osc_update_grant(cli, oa->o_grant); | |
26c4ea46 | 879 | goto out; |
d7e09d03 PT |
880 | } |
881 | ||
882 | body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); | |
883 | LASSERT(body); | |
884 | osc_update_grant(cli, body); | |
885 | out: | |
2ba262fb | 886 | kmem_cache_free(obdo_cachep, oa); |
d7e09d03 PT |
887 | return rc; |
888 | } | |
889 | ||
890 | static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa) | |
891 | { | |
892 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
893 | oa->o_grant = cli->cl_avail_grant / 4; | |
894 | cli->cl_avail_grant -= oa->o_grant; | |
895 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
896 | if (!(oa->o_valid & OBD_MD_FLFLAGS)) { | |
897 | oa->o_valid |= OBD_MD_FLFLAGS; | |
898 | oa->o_flags = 0; | |
899 | } | |
900 | oa->o_flags |= OBD_FL_SHRINK_GRANT; | |
901 | osc_update_next_shrink(cli); | |
902 | } | |
903 | ||
904 | /* Shrink the current grant, either from some large amount to enough for a | |
905 | * full set of in-flight RPCs, or if we have already shrunk to that limit | |
906 | * then to enough for a single RPC. This avoids keeping more grant than | |
30aa9c52 OD |
907 | * needed, and avoids shrinking the grant piecemeal. |
908 | */ | |
d7e09d03 PT |
909 | static int osc_shrink_grant(struct client_obd *cli) |
910 | { | |
911 | __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) * | |
09cbfeaf | 912 | (cli->cl_max_pages_per_rpc << PAGE_SHIFT); |
d7e09d03 PT |
913 | |
914 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
915 | if (cli->cl_avail_grant <= target_bytes) | |
09cbfeaf | 916 | target_bytes = cli->cl_max_pages_per_rpc << PAGE_SHIFT; |
d7e09d03 PT |
917 | client_obd_list_unlock(&cli->cl_loi_list_lock); |
918 | ||
919 | return osc_shrink_grant_to_target(cli, target_bytes); | |
920 | } | |
921 | ||
922 | int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes) | |
923 | { | |
29ac6840 | 924 | int rc = 0; |
d7e09d03 | 925 | struct ost_body *body; |
d7e09d03 PT |
926 | |
927 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
928 | /* Don't shrink if we are already above or below the desired limit | |
929 | * We don't want to shrink below a single RPC, as that will negatively | |
30aa9c52 OD |
930 | * impact block allocation and long-term performance. |
931 | */ | |
09cbfeaf KS |
932 | if (target_bytes < cli->cl_max_pages_per_rpc << PAGE_SHIFT) |
933 | target_bytes = cli->cl_max_pages_per_rpc << PAGE_SHIFT; | |
d7e09d03 PT |
934 | |
935 | if (target_bytes >= cli->cl_avail_grant) { | |
936 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
0a3bdb00 | 937 | return 0; |
d7e09d03 PT |
938 | } |
939 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
940 | ||
7795178d | 941 | body = kzalloc(sizeof(*body), GFP_NOFS); |
d7e09d03 | 942 | if (!body) |
0a3bdb00 | 943 | return -ENOMEM; |
d7e09d03 PT |
944 | |
945 | osc_announce_cached(cli, &body->oa, 0); | |
946 | ||
947 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
948 | body->oa.o_grant = cli->cl_avail_grant - target_bytes; | |
949 | cli->cl_avail_grant = target_bytes; | |
950 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
951 | if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) { | |
952 | body->oa.o_valid |= OBD_MD_FLFLAGS; | |
953 | body->oa.o_flags = 0; | |
954 | } | |
955 | body->oa.o_flags |= OBD_FL_SHRINK_GRANT; | |
956 | osc_update_next_shrink(cli); | |
957 | ||
958 | rc = osc_set_info_async(NULL, cli->cl_import->imp_obd->obd_self_export, | |
959 | sizeof(KEY_GRANT_SHRINK), KEY_GRANT_SHRINK, | |
960 | sizeof(*body), body, NULL); | |
961 | if (rc != 0) | |
962 | __osc_update_grant(cli, body->oa.o_grant); | |
7795178d | 963 | kfree(body); |
0a3bdb00 | 964 | return rc; |
d7e09d03 PT |
965 | } |
966 | ||
967 | static int osc_should_shrink_grant(struct client_obd *client) | |
968 | { | |
a649ad1d GKH |
969 | unsigned long time = cfs_time_current(); |
970 | unsigned long next_shrink = client->cl_next_shrink_grant; | |
d7e09d03 PT |
971 | |
972 | if ((client->cl_import->imp_connect_data.ocd_connect_flags & | |
973 | OBD_CONNECT_GRANT_SHRINK) == 0) | |
974 | return 0; | |
975 | ||
976 | if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) { | |
977 | /* Get the current RPC size directly, instead of going via: | |
978 | * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export) | |
30aa9c52 OD |
979 | * Keep comment here so that it can be found by searching. |
980 | */ | |
09cbfeaf | 981 | int brw_size = client->cl_max_pages_per_rpc << PAGE_SHIFT; |
d7e09d03 PT |
982 | |
983 | if (client->cl_import->imp_state == LUSTRE_IMP_FULL && | |
984 | client->cl_avail_grant > brw_size) | |
985 | return 1; | |
71e8dd9a AM |
986 | |
987 | osc_update_next_shrink(client); | |
d7e09d03 PT |
988 | } |
989 | return 0; | |
990 | } | |
991 | ||
992 | static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data) | |
993 | { | |
994 | struct client_obd *client; | |
995 | ||
79910d7d | 996 | list_for_each_entry(client, &item->ti_obd_list, cl_grant_shrink_list) { |
d7e09d03 PT |
997 | if (osc_should_shrink_grant(client)) |
998 | osc_shrink_grant(client); | |
999 | } | |
1000 | return 0; | |
1001 | } | |
1002 | ||
1003 | static int osc_add_shrink_grant(struct client_obd *client) | |
1004 | { | |
1005 | int rc; | |
1006 | ||
1007 | rc = ptlrpc_add_timeout_client(client->cl_grant_shrink_interval, | |
1008 | TIMEOUT_GRANT, | |
1009 | osc_grant_shrink_grant_cb, NULL, | |
1010 | &client->cl_grant_shrink_list); | |
1011 | if (rc) { | |
1012 | CERROR("add grant client %s error %d\n", | |
79910d7d | 1013 | client->cl_import->imp_obd->obd_name, rc); |
d7e09d03 PT |
1014 | return rc; |
1015 | } | |
72a87fca | 1016 | CDEBUG(D_CACHE, "add grant client %s\n", |
d7e09d03 PT |
1017 | client->cl_import->imp_obd->obd_name); |
1018 | osc_update_next_shrink(client); | |
1019 | return 0; | |
1020 | } | |
1021 | ||
1022 | static int osc_del_shrink_grant(struct client_obd *client) | |
1023 | { | |
1024 | return ptlrpc_del_timeout_client(&client->cl_grant_shrink_list, | |
1025 | TIMEOUT_GRANT); | |
1026 | } | |
1027 | ||
1028 | static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd) | |
1029 | { | |
1030 | /* | |
1031 | * ocd_grant is the total grant amount we're expect to hold: if we've | |
1032 | * been evicted, it's the new avail_grant amount, cl_dirty will drop | |
1033 | * to 0 as inflight RPCs fail out; otherwise, it's avail_grant + dirty. | |
1034 | * | |
1035 | * race is tolerable here: if we're evicted, but imp_state already | |
1036 | * left EVICTED state, then cl_dirty must be 0 already. | |
1037 | */ | |
1038 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
1039 | if (cli->cl_import->imp_state == LUSTRE_IMP_EVICTED) | |
1040 | cli->cl_avail_grant = ocd->ocd_grant; | |
1041 | else | |
1042 | cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty; | |
1043 | ||
1044 | if (cli->cl_avail_grant < 0) { | |
1045 | CWARN("%s: available grant < 0: avail/ocd/dirty %ld/%u/%ld\n", | |
1046 | cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant, | |
1047 | ocd->ocd_grant, cli->cl_dirty); | |
1048 | /* workaround for servers which do not have the patch from | |
30aa9c52 OD |
1049 | * LU-2679 |
1050 | */ | |
d7e09d03 PT |
1051 | cli->cl_avail_grant = ocd->ocd_grant; |
1052 | } | |
1053 | ||
1054 | /* determine the appropriate chunk size used by osc_extent. */ | |
09cbfeaf | 1055 | cli->cl_chunkbits = max_t(int, PAGE_SHIFT, ocd->ocd_blocksize); |
d7e09d03 PT |
1056 | client_obd_list_unlock(&cli->cl_loi_list_lock); |
1057 | ||
2d00bd17 JP |
1058 | CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld chunk bits: %d\n", |
1059 | cli->cl_import->imp_obd->obd_name, | |
1060 | cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits); | |
d7e09d03 PT |
1061 | |
1062 | if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK && | |
1063 | list_empty(&cli->cl_grant_shrink_list)) | |
1064 | osc_add_shrink_grant(cli); | |
1065 | } | |
1066 | ||
1067 | /* We assume that the reason this OSC got a short read is because it read | |
1068 | * beyond the end of a stripe file; i.e. lustre is reading a sparse file | |
1069 | * via the LOV, and it _knows_ it's reading inside the file, it's just that | |
30aa9c52 OD |
1070 | * this stripe never got written at or beyond this stripe offset yet. |
1071 | */ | |
21aef7d9 | 1072 | static void handle_short_read(int nob_read, u32 page_count, |
d7e09d03 PT |
1073 | struct brw_page **pga) |
1074 | { | |
1075 | char *ptr; | |
1076 | int i = 0; | |
1077 | ||
1078 | /* skip bytes read OK */ | |
1079 | while (nob_read > 0) { | |
e72f36e2 | 1080 | LASSERT(page_count > 0); |
d7e09d03 PT |
1081 | |
1082 | if (pga[i]->count > nob_read) { | |
1083 | /* EOF inside this page */ | |
1084 | ptr = kmap(pga[i]->pg) + | |
1085 | (pga[i]->off & ~CFS_PAGE_MASK); | |
1086 | memset(ptr + nob_read, 0, pga[i]->count - nob_read); | |
1087 | kunmap(pga[i]->pg); | |
1088 | page_count--; | |
1089 | i++; | |
1090 | break; | |
1091 | } | |
1092 | ||
1093 | nob_read -= pga[i]->count; | |
1094 | page_count--; | |
1095 | i++; | |
1096 | } | |
1097 | ||
1098 | /* zero remaining pages */ | |
1099 | while (page_count-- > 0) { | |
1100 | ptr = kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK); | |
1101 | memset(ptr, 0, pga[i]->count); | |
1102 | kunmap(pga[i]->pg); | |
1103 | i++; | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | static int check_write_rcs(struct ptlrpc_request *req, | |
1108 | int requested_nob, int niocount, | |
21aef7d9 | 1109 | u32 page_count, struct brw_page **pga) |
d7e09d03 | 1110 | { |
29ac6840 CH |
1111 | int i; |
1112 | __u32 *remote_rcs; | |
d7e09d03 PT |
1113 | |
1114 | remote_rcs = req_capsule_server_sized_get(&req->rq_pill, &RMF_RCS, | |
1115 | sizeof(*remote_rcs) * | |
1116 | niocount); | |
7f1ae4c0 | 1117 | if (!remote_rcs) { |
d7e09d03 | 1118 | CDEBUG(D_INFO, "Missing/short RC vector on BRW_WRITE reply\n"); |
fbe7c6c7 | 1119 | return -EPROTO; |
d7e09d03 PT |
1120 | } |
1121 | ||
1122 | /* return error if any niobuf was in error */ | |
1123 | for (i = 0; i < niocount; i++) { | |
1124 | if ((int)remote_rcs[i] < 0) | |
e8291974 | 1125 | return remote_rcs[i]; |
d7e09d03 PT |
1126 | |
1127 | if (remote_rcs[i] != 0) { | |
1128 | CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n", | |
79910d7d | 1129 | i, remote_rcs[i], req); |
fbe7c6c7 | 1130 | return -EPROTO; |
d7e09d03 PT |
1131 | } |
1132 | } | |
1133 | ||
1134 | if (req->rq_bulk->bd_nob_transferred != requested_nob) { | |
1135 | CERROR("Unexpected # bytes transferred: %d (requested %d)\n", | |
1136 | req->rq_bulk->bd_nob_transferred, requested_nob); | |
fbe7c6c7 | 1137 | return -EPROTO; |
d7e09d03 PT |
1138 | } |
1139 | ||
fbe7c6c7 | 1140 | return 0; |
d7e09d03 PT |
1141 | } |
1142 | ||
1143 | static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2) | |
1144 | { | |
1145 | if (p1->flag != p2->flag) { | |
7cf1054b HE |
1146 | unsigned mask = ~(OBD_BRW_FROM_GRANT | OBD_BRW_NOCACHE | |
1147 | OBD_BRW_SYNC | OBD_BRW_ASYNC|OBD_BRW_NOQUOTA); | |
d7e09d03 PT |
1148 | |
1149 | /* warn if we try to combine flags that we don't know to be | |
30aa9c52 OD |
1150 | * safe to combine |
1151 | */ | |
d7e09d03 | 1152 | if (unlikely((p1->flag & mask) != (p2->flag & mask))) { |
2d00bd17 | 1153 | CWARN("Saw flags 0x%x and 0x%x in the same brw, please report this at http://bugs.whamcloud.com/\n", |
d7e09d03 PT |
1154 | p1->flag, p2->flag); |
1155 | } | |
1156 | return 0; | |
1157 | } | |
1158 | ||
1159 | return (p1->off + p1->count == p2->off); | |
1160 | } | |
1161 | ||
21aef7d9 | 1162 | static u32 osc_checksum_bulk(int nob, u32 pg_count, |
29ac6840 | 1163 | struct brw_page **pga, int opc, |
d133210f | 1164 | enum cksum_type cksum_type) |
d7e09d03 | 1165 | { |
29ac6840 CH |
1166 | __u32 cksum; |
1167 | int i = 0; | |
1168 | struct cfs_crypto_hash_desc *hdesc; | |
1169 | unsigned int bufsize; | |
1170 | int err; | |
1171 | unsigned char cfs_alg = cksum_obd2cfs(cksum_type); | |
d7e09d03 PT |
1172 | |
1173 | LASSERT(pg_count > 0); | |
1174 | ||
1175 | hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0); | |
1176 | if (IS_ERR(hdesc)) { | |
1177 | CERROR("Unable to initialize checksum hash %s\n", | |
1178 | cfs_crypto_hash_name(cfs_alg)); | |
1179 | return PTR_ERR(hdesc); | |
1180 | } | |
1181 | ||
1182 | while (nob > 0 && pg_count > 0) { | |
1183 | int count = pga[i]->count > nob ? nob : pga[i]->count; | |
1184 | ||
1185 | /* corrupt the data before we compute the checksum, to | |
30aa9c52 OD |
1186 | * simulate an OST->client data error |
1187 | */ | |
d7e09d03 PT |
1188 | if (i == 0 && opc == OST_READ && |
1189 | OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) { | |
1190 | unsigned char *ptr = kmap(pga[i]->pg); | |
1191 | int off = pga[i]->off & ~CFS_PAGE_MASK; | |
50ffcb7e | 1192 | |
d7e09d03 PT |
1193 | memcpy(ptr + off, "bad1", min(4, nob)); |
1194 | kunmap(pga[i]->pg); | |
1195 | } | |
1196 | cfs_crypto_hash_update_page(hdesc, pga[i]->pg, | |
79910d7d | 1197 | pga[i]->off & ~CFS_PAGE_MASK, |
d7e09d03 | 1198 | count); |
aa3bee0d GKH |
1199 | CDEBUG(D_PAGE, |
1200 | "page %p map %p index %lu flags %lx count %u priv %0lx: off %d\n", | |
1201 | pga[i]->pg, pga[i]->pg->mapping, pga[i]->pg->index, | |
1202 | (long)pga[i]->pg->flags, page_count(pga[i]->pg), | |
1203 | page_private(pga[i]->pg), | |
1204 | (int)(pga[i]->off & ~CFS_PAGE_MASK)); | |
d7e09d03 PT |
1205 | |
1206 | nob -= pga[i]->count; | |
1207 | pg_count--; | |
1208 | i++; | |
1209 | } | |
1210 | ||
1211 | bufsize = 4; | |
1212 | err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize); | |
1213 | ||
1214 | if (err) | |
1215 | cfs_crypto_hash_final(hdesc, NULL, NULL); | |
1216 | ||
1217 | /* For sending we only compute the wrong checksum instead | |
30aa9c52 OD |
1218 | * of corrupting the data so it is still correct on a redo |
1219 | */ | |
d7e09d03 PT |
1220 | if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND)) |
1221 | cksum++; | |
1222 | ||
1223 | return cksum; | |
1224 | } | |
1225 | ||
1d8cb70c GD |
1226 | static int osc_brw_prep_request(int cmd, struct client_obd *cli, |
1227 | struct obdo *oa, | |
21aef7d9 | 1228 | struct lov_stripe_md *lsm, u32 page_count, |
d7e09d03 PT |
1229 | struct brw_page **pga, |
1230 | struct ptlrpc_request **reqp, | |
ef2e0f55 | 1231 | int reserve, |
d7e09d03 PT |
1232 | int resend) |
1233 | { | |
29ac6840 | 1234 | struct ptlrpc_request *req; |
d7e09d03 | 1235 | struct ptlrpc_bulk_desc *desc; |
29ac6840 CH |
1236 | struct ost_body *body; |
1237 | struct obd_ioobj *ioobj; | |
1238 | struct niobuf_remote *niobuf; | |
d7e09d03 PT |
1239 | int niocount, i, requested_nob, opc, rc; |
1240 | struct osc_brw_async_args *aa; | |
29ac6840 | 1241 | struct req_capsule *pill; |
d7e09d03 PT |
1242 | struct brw_page *pg_prev; |
1243 | ||
d7e09d03 | 1244 | if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ)) |
0a3bdb00 | 1245 | return -ENOMEM; /* Recoverable */ |
d7e09d03 | 1246 | if (OBD_FAIL_CHECK(OBD_FAIL_OSC_BRW_PREP_REQ2)) |
0a3bdb00 | 1247 | return -EINVAL; /* Fatal */ |
d7e09d03 PT |
1248 | |
1249 | if ((cmd & OBD_BRW_WRITE) != 0) { | |
1250 | opc = OST_WRITE; | |
1251 | req = ptlrpc_request_alloc_pool(cli->cl_import, | |
aefd9d71 | 1252 | osc_rq_pool, |
d7e09d03 PT |
1253 | &RQF_OST_BRW_WRITE); |
1254 | } else { | |
1255 | opc = OST_READ; | |
1256 | req = ptlrpc_request_alloc(cli->cl_import, &RQF_OST_BRW_READ); | |
1257 | } | |
7f1ae4c0 | 1258 | if (!req) |
0a3bdb00 | 1259 | return -ENOMEM; |
d7e09d03 PT |
1260 | |
1261 | for (niocount = i = 1; i < page_count; i++) { | |
1262 | if (!can_merge_pages(pga[i - 1], pga[i])) | |
1263 | niocount++; | |
1264 | } | |
1265 | ||
1266 | pill = &req->rq_pill; | |
1267 | req_capsule_set_size(pill, &RMF_OBD_IOOBJ, RCL_CLIENT, | |
1268 | sizeof(*ioobj)); | |
1269 | req_capsule_set_size(pill, &RMF_NIOBUF_REMOTE, RCL_CLIENT, | |
1270 | niocount * sizeof(*niobuf)); | |
d7e09d03 PT |
1271 | |
1272 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, opc); | |
1273 | if (rc) { | |
1274 | ptlrpc_request_free(req); | |
0a3bdb00 | 1275 | return rc; |
d7e09d03 PT |
1276 | } |
1277 | req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */ | |
1278 | ptlrpc_at_set_req_timeout(req); | |
1279 | /* ask ptlrpc not to resend on EINPROGRESS since BRWs have their own | |
30aa9c52 OD |
1280 | * retry logic |
1281 | */ | |
d7e09d03 PT |
1282 | req->rq_no_retry_einprogress = 1; |
1283 | ||
1284 | desc = ptlrpc_prep_bulk_imp(req, page_count, | |
1285 | cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS, | |
1286 | opc == OST_WRITE ? BULK_GET_SOURCE : BULK_PUT_SINK, | |
1287 | OST_BULK_PORTAL); | |
1288 | ||
7f1ae4c0 | 1289 | if (!desc) { |
26c4ea46 TJ |
1290 | rc = -ENOMEM; |
1291 | goto out; | |
1292 | } | |
d7e09d03 PT |
1293 | /* NB request now owns desc and will free it when it gets freed */ |
1294 | ||
1295 | body = req_capsule_client_get(pill, &RMF_OST_BODY); | |
1296 | ioobj = req_capsule_client_get(pill, &RMF_OBD_IOOBJ); | |
1297 | niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE); | |
7f1ae4c0 | 1298 | LASSERT(body && ioobj && niobuf); |
d7e09d03 | 1299 | |
3b2f75fd | 1300 | lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa); |
d7e09d03 PT |
1301 | |
1302 | obdo_to_ioobj(oa, ioobj); | |
1303 | ioobj->ioo_bufcnt = niocount; | |
1304 | /* The high bits of ioo_max_brw tells server _maximum_ number of bulks | |
1305 | * that might be send for this request. The actual number is decided | |
1306 | * when the RPC is finally sent in ptlrpc_register_bulk(). It sends | |
1307 | * "max - 1" for old client compatibility sending "0", and also so the | |
30aa9c52 OD |
1308 | * the actual maximum is a power-of-two number, not one less. LU-1431 |
1309 | */ | |
d7e09d03 | 1310 | ioobj_max_brw_set(ioobj, desc->bd_md_max_brw); |
d7e09d03 PT |
1311 | LASSERT(page_count > 0); |
1312 | pg_prev = pga[0]; | |
1313 | for (requested_nob = i = 0; i < page_count; i++, niobuf++) { | |
1314 | struct brw_page *pg = pga[i]; | |
1315 | int poff = pg->off & ~CFS_PAGE_MASK; | |
1316 | ||
1317 | LASSERT(pg->count > 0); | |
1318 | /* make sure there is no gap in the middle of page array */ | |
1319 | LASSERTF(page_count == 1 || | |
09cbfeaf | 1320 | (ergo(i == 0, poff + pg->count == PAGE_SIZE) && |
d7e09d03 | 1321 | ergo(i > 0 && i < page_count - 1, |
09cbfeaf | 1322 | poff == 0 && pg->count == PAGE_SIZE) && |
d7e09d03 | 1323 | ergo(i == page_count - 1, poff == 0)), |
b0f5aad5 | 1324 | "i: %d/%d pg: %p off: %llu, count: %u\n", |
d7e09d03 PT |
1325 | i, page_count, pg, pg->off, pg->count); |
1326 | LASSERTF(i == 0 || pg->off > pg_prev->off, | |
2d00bd17 | 1327 | "i %d p_c %u pg %p [pri %lu ind %lu] off %llu prev_pg %p [pri %lu ind %lu] off %llu\n", |
d7e09d03 PT |
1328 | i, page_count, |
1329 | pg->pg, page_private(pg->pg), pg->pg->index, pg->off, | |
1330 | pg_prev->pg, page_private(pg_prev->pg), | |
1331 | pg_prev->pg->index, pg_prev->off); | |
1332 | LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) == | |
1333 | (pg->flag & OBD_BRW_SRVLOCK)); | |
1334 | ||
1335 | ptlrpc_prep_bulk_page_pin(desc, pg->pg, poff, pg->count); | |
1336 | requested_nob += pg->count; | |
1337 | ||
1338 | if (i > 0 && can_merge_pages(pg_prev, pg)) { | |
1339 | niobuf--; | |
1340 | niobuf->len += pg->count; | |
1341 | } else { | |
1342 | niobuf->offset = pg->off; | |
29ac6840 CH |
1343 | niobuf->len = pg->count; |
1344 | niobuf->flags = pg->flag; | |
d7e09d03 PT |
1345 | } |
1346 | pg_prev = pg; | |
1347 | } | |
1348 | ||
1349 | LASSERTF((void *)(niobuf - niocount) == | |
1350 | req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE), | |
1351 | "want %p - real %p\n", req_capsule_client_get(&req->rq_pill, | |
1352 | &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount)); | |
1353 | ||
1354 | osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0); | |
1355 | if (resend) { | |
1356 | if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) { | |
1357 | body->oa.o_valid |= OBD_MD_FLFLAGS; | |
1358 | body->oa.o_flags = 0; | |
1359 | } | |
1360 | body->oa.o_flags |= OBD_FL_RECOV_RESEND; | |
1361 | } | |
1362 | ||
1363 | if (osc_should_shrink_grant(cli)) | |
1364 | osc_shrink_grant_local(cli, &body->oa); | |
1365 | ||
1366 | /* size[REQ_REC_OFF] still sizeof (*body) */ | |
1367 | if (opc == OST_WRITE) { | |
1368 | if (cli->cl_checksum && | |
1369 | !sptlrpc_flavor_has_bulk(&req->rq_flvr)) { | |
1370 | /* store cl_cksum_type in a local variable since | |
30aa9c52 OD |
1371 | * it can be changed via lprocfs |
1372 | */ | |
d133210f | 1373 | enum cksum_type cksum_type = cli->cl_cksum_type; |
d7e09d03 PT |
1374 | |
1375 | if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) { | |
1376 | oa->o_flags &= OBD_FL_LOCAL_MASK; | |
1377 | body->oa.o_flags = 0; | |
1378 | } | |
1379 | body->oa.o_flags |= cksum_type_pack(cksum_type); | |
1380 | body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS; | |
1381 | body->oa.o_cksum = osc_checksum_bulk(requested_nob, | |
1382 | page_count, pga, | |
1383 | OST_WRITE, | |
1384 | cksum_type); | |
1385 | CDEBUG(D_PAGE, "checksum at write origin: %x\n", | |
1386 | body->oa.o_cksum); | |
1387 | /* save this in 'oa', too, for later checking */ | |
1388 | oa->o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS; | |
1389 | oa->o_flags |= cksum_type_pack(cksum_type); | |
1390 | } else { | |
1391 | /* clear out the checksum flag, in case this is a | |
30aa9c52 OD |
1392 | * resend but cl_checksum is no longer set. b=11238 |
1393 | */ | |
d7e09d03 PT |
1394 | oa->o_valid &= ~OBD_MD_FLCKSUM; |
1395 | } | |
1396 | oa->o_cksum = body->oa.o_cksum; | |
1397 | /* 1 RC per niobuf */ | |
1398 | req_capsule_set_size(pill, &RMF_RCS, RCL_SERVER, | |
1399 | sizeof(__u32) * niocount); | |
1400 | } else { | |
1401 | if (cli->cl_checksum && | |
1402 | !sptlrpc_flavor_has_bulk(&req->rq_flvr)) { | |
1403 | if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) | |
1404 | body->oa.o_flags = 0; | |
1405 | body->oa.o_flags |= cksum_type_pack(cli->cl_cksum_type); | |
1406 | body->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS; | |
1407 | } | |
1408 | } | |
1409 | ptlrpc_request_set_replen(req); | |
1410 | ||
1411 | CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args)); | |
1412 | aa = ptlrpc_req_async_args(req); | |
1413 | aa->aa_oa = oa; | |
1414 | aa->aa_requested_nob = requested_nob; | |
1415 | aa->aa_nio_count = niocount; | |
1416 | aa->aa_page_count = page_count; | |
1417 | aa->aa_resends = 0; | |
1418 | aa->aa_ppga = pga; | |
1419 | aa->aa_cli = cli; | |
1420 | INIT_LIST_HEAD(&aa->aa_oaps); | |
d7e09d03 PT |
1421 | |
1422 | *reqp = req; | |
0a3bdb00 | 1423 | return 0; |
d7e09d03 PT |
1424 | |
1425 | out: | |
1426 | ptlrpc_req_finished(req); | |
0a3bdb00 | 1427 | return rc; |
d7e09d03 PT |
1428 | } |
1429 | ||
1430 | static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer, | |
1431 | __u32 client_cksum, __u32 server_cksum, int nob, | |
21aef7d9 | 1432 | u32 page_count, struct brw_page **pga, |
d133210f | 1433 | enum cksum_type client_cksum_type) |
d7e09d03 PT |
1434 | { |
1435 | __u32 new_cksum; | |
1436 | char *msg; | |
d133210f | 1437 | enum cksum_type cksum_type; |
d7e09d03 PT |
1438 | |
1439 | if (server_cksum == client_cksum) { | |
1440 | CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum); | |
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | cksum_type = cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ? | |
1445 | oa->o_flags : 0); | |
1446 | new_cksum = osc_checksum_bulk(nob, page_count, pga, OST_WRITE, | |
1447 | cksum_type); | |
1448 | ||
1449 | if (cksum_type != client_cksum_type) | |
2d00bd17 JP |
1450 | msg = "the server did not use the checksum type specified in the original request - likely a protocol problem" |
1451 | ; | |
d7e09d03 | 1452 | else if (new_cksum == server_cksum) |
2d00bd17 JP |
1453 | msg = "changed on the client after we checksummed it - likely false positive due to mmap IO (bug 11742)" |
1454 | ; | |
d7e09d03 PT |
1455 | else if (new_cksum == client_cksum) |
1456 | msg = "changed in transit before arrival at OST"; | |
1457 | else | |
2d00bd17 JP |
1458 | msg = "changed in transit AND doesn't match the original - likely false positive due to mmap IO (bug 11742)" |
1459 | ; | |
d7e09d03 PT |
1460 | |
1461 | LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID | |
b0f5aad5 | 1462 | " object "DOSTID" extent [%llu-%llu]\n", |
d7e09d03 PT |
1463 | msg, libcfs_nid2str(peer->nid), |
1464 | oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0, | |
1465 | oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0, | |
1466 | oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0, | |
1467 | POSTID(&oa->o_oi), pga[0]->off, | |
1468 | pga[page_count-1]->off + pga[page_count-1]->count - 1); | |
2d00bd17 JP |
1469 | CERROR("original client csum %x (type %x), server csum %x (type %x), client csum now %x\n", |
1470 | client_cksum, client_cksum_type, | |
d7e09d03 PT |
1471 | server_cksum, cksum_type, new_cksum); |
1472 | return 1; | |
1473 | } | |
1474 | ||
1475 | /* Note rc enters this function as number of bytes transferred */ | |
1476 | static int osc_brw_fini_request(struct ptlrpc_request *req, int rc) | |
1477 | { | |
1478 | struct osc_brw_async_args *aa = (void *)&req->rq_async_args; | |
1479 | const lnet_process_id_t *peer = | |
1480 | &req->rq_import->imp_connection->c_peer; | |
1481 | struct client_obd *cli = aa->aa_cli; | |
1482 | struct ost_body *body; | |
1483 | __u32 client_cksum = 0; | |
d7e09d03 PT |
1484 | |
1485 | if (rc < 0 && rc != -EDQUOT) { | |
1486 | DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc); | |
0a3bdb00 | 1487 | return rc; |
d7e09d03 PT |
1488 | } |
1489 | ||
7f1ae4c0 | 1490 | LASSERTF(req->rq_repmsg, "rc = %d\n", rc); |
d7e09d03 | 1491 | body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); |
7f1ae4c0 | 1492 | if (!body) { |
d7e09d03 | 1493 | DEBUG_REQ(D_INFO, req, "Can't unpack body\n"); |
0a3bdb00 | 1494 | return -EPROTO; |
d7e09d03 PT |
1495 | } |
1496 | ||
1497 | /* set/clear over quota flag for a uid/gid */ | |
1498 | if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE && | |
1499 | body->oa.o_valid & (OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA)) { | |
1500 | unsigned int qid[MAXQUOTAS] = { body->oa.o_uid, body->oa.o_gid }; | |
1501 | ||
55f5a824 | 1502 | CDEBUG(D_QUOTA, "setdq for [%u %u] with valid %#llx, flags %x\n", |
d7e09d03 PT |
1503 | body->oa.o_uid, body->oa.o_gid, body->oa.o_valid, |
1504 | body->oa.o_flags); | |
1505 | osc_quota_setdq(cli, qid, body->oa.o_valid, body->oa.o_flags); | |
1506 | } | |
1507 | ||
1508 | osc_update_grant(cli, body); | |
1509 | ||
1510 | if (rc < 0) | |
0a3bdb00 | 1511 | return rc; |
d7e09d03 PT |
1512 | |
1513 | if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM) | |
1514 | client_cksum = aa->aa_oa->o_cksum; /* save for later */ | |
1515 | ||
1516 | if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) { | |
1517 | if (rc > 0) { | |
1518 | CERROR("Unexpected +ve rc %d\n", rc); | |
0a3bdb00 | 1519 | return -EPROTO; |
d7e09d03 PT |
1520 | } |
1521 | LASSERT(req->rq_bulk->bd_nob == aa->aa_requested_nob); | |
1522 | ||
1523 | if (sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk)) | |
0a3bdb00 | 1524 | return -EAGAIN; |
d7e09d03 PT |
1525 | |
1526 | if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum && | |
1527 | check_write_checksum(&body->oa, peer, client_cksum, | |
1528 | body->oa.o_cksum, aa->aa_requested_nob, | |
1529 | aa->aa_page_count, aa->aa_ppga, | |
1530 | cksum_type_unpack(aa->aa_oa->o_flags))) | |
0a3bdb00 | 1531 | return -EAGAIN; |
d7e09d03 | 1532 | |
1d8cb70c GD |
1533 | rc = check_write_rcs(req, aa->aa_requested_nob, |
1534 | aa->aa_nio_count, | |
d7e09d03 | 1535 | aa->aa_page_count, aa->aa_ppga); |
26c4ea46 | 1536 | goto out; |
d7e09d03 PT |
1537 | } |
1538 | ||
1539 | /* The rest of this function executes only for OST_READs */ | |
1540 | ||
1541 | /* if unwrap_bulk failed, return -EAGAIN to retry */ | |
1542 | rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc); | |
26c4ea46 TJ |
1543 | if (rc < 0) { |
1544 | rc = -EAGAIN; | |
1545 | goto out; | |
1546 | } | |
d7e09d03 PT |
1547 | |
1548 | if (rc > aa->aa_requested_nob) { | |
1549 | CERROR("Unexpected rc %d (%d requested)\n", rc, | |
1550 | aa->aa_requested_nob); | |
0a3bdb00 | 1551 | return -EPROTO; |
d7e09d03 PT |
1552 | } |
1553 | ||
1554 | if (rc != req->rq_bulk->bd_nob_transferred) { | |
e72f36e2 | 1555 | CERROR("Unexpected rc %d (%d transferred)\n", |
79910d7d | 1556 | rc, req->rq_bulk->bd_nob_transferred); |
fbe7c6c7 | 1557 | return -EPROTO; |
d7e09d03 PT |
1558 | } |
1559 | ||
1560 | if (rc < aa->aa_requested_nob) | |
1561 | handle_short_read(rc, aa->aa_page_count, aa->aa_ppga); | |
1562 | ||
1563 | if (body->oa.o_valid & OBD_MD_FLCKSUM) { | |
1564 | static int cksum_counter; | |
29ac6840 | 1565 | __u32 server_cksum = body->oa.o_cksum; |
80feb1ef DE |
1566 | char *via = ""; |
1567 | char *router = ""; | |
d133210f | 1568 | enum cksum_type cksum_type; |
d7e09d03 | 1569 | |
b2952d62 | 1570 | cksum_type = cksum_type_unpack(body->oa.o_valid&OBD_MD_FLFLAGS ? |
d7e09d03 PT |
1571 | body->oa.o_flags : 0); |
1572 | client_cksum = osc_checksum_bulk(rc, aa->aa_page_count, | |
1573 | aa->aa_ppga, OST_READ, | |
1574 | cksum_type); | |
1575 | ||
80feb1ef | 1576 | if (peer->nid != req->rq_bulk->bd_sender) { |
d7e09d03 PT |
1577 | via = " via "; |
1578 | router = libcfs_nid2str(req->rq_bulk->bd_sender); | |
1579 | } | |
1580 | ||
a2ff0f97 | 1581 | if (server_cksum != client_cksum) { |
2d00bd17 | 1582 | LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from %s%s%s inode " DFID " object " DOSTID " extent [%llu-%llu]\n", |
d7e09d03 PT |
1583 | req->rq_import->imp_obd->obd_name, |
1584 | libcfs_nid2str(peer->nid), | |
1585 | via, router, | |
1586 | body->oa.o_valid & OBD_MD_FLFID ? | |
2d00bd17 | 1587 | body->oa.o_parent_seq : (__u64)0, |
d7e09d03 | 1588 | body->oa.o_valid & OBD_MD_FLFID ? |
2d00bd17 | 1589 | body->oa.o_parent_oid : 0, |
d7e09d03 | 1590 | body->oa.o_valid & OBD_MD_FLFID ? |
2d00bd17 | 1591 | body->oa.o_parent_ver : 0, |
d7e09d03 PT |
1592 | POSTID(&body->oa.o_oi), |
1593 | aa->aa_ppga[0]->off, | |
1594 | aa->aa_ppga[aa->aa_page_count-1]->off + | |
1595 | aa->aa_ppga[aa->aa_page_count-1]->count - | |
2d00bd17 | 1596 | 1); |
d7e09d03 PT |
1597 | CERROR("client %x, server %x, cksum_type %x\n", |
1598 | client_cksum, server_cksum, cksum_type); | |
1599 | cksum_counter = 0; | |
1600 | aa->aa_oa->o_cksum = client_cksum; | |
1601 | rc = -EAGAIN; | |
1602 | } else { | |
1603 | cksum_counter++; | |
1604 | CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum); | |
1605 | rc = 0; | |
1606 | } | |
1607 | } else if (unlikely(client_cksum)) { | |
1608 | static int cksum_missed; | |
1609 | ||
1610 | cksum_missed++; | |
1611 | if ((cksum_missed & (-cksum_missed)) == cksum_missed) | |
1612 | CERROR("Checksum %u requested from %s but not sent\n", | |
1613 | cksum_missed, libcfs_nid2str(peer->nid)); | |
1614 | } else { | |
1615 | rc = 0; | |
1616 | } | |
1617 | out: | |
1618 | if (rc >= 0) | |
3b2f75fd | 1619 | lustre_get_wire_obdo(&req->rq_import->imp_connect_data, |
1620 | aa->aa_oa, &body->oa); | |
d7e09d03 | 1621 | |
0a3bdb00 | 1622 | return rc; |
d7e09d03 PT |
1623 | } |
1624 | ||
d7e09d03 PT |
1625 | static int osc_brw_redo_request(struct ptlrpc_request *request, |
1626 | struct osc_brw_async_args *aa, int rc) | |
1627 | { | |
1628 | struct ptlrpc_request *new_req; | |
1629 | struct osc_brw_async_args *new_aa; | |
1630 | struct osc_async_page *oap; | |
d7e09d03 PT |
1631 | |
1632 | DEBUG_REQ(rc == -EINPROGRESS ? D_RPCTRACE : D_ERROR, request, | |
1633 | "redo for recoverable error %d", rc); | |
1634 | ||
1635 | rc = osc_brw_prep_request(lustre_msg_get_opc(request->rq_reqmsg) == | |
b2952d62 | 1636 | OST_WRITE ? OBD_BRW_WRITE : OBD_BRW_READ, |
d7e09d03 PT |
1637 | aa->aa_cli, aa->aa_oa, |
1638 | NULL /* lsm unused by osc currently */, | |
1639 | aa->aa_page_count, aa->aa_ppga, | |
ef2e0f55 | 1640 | &new_req, 0, 1); |
d7e09d03 | 1641 | if (rc) |
0a3bdb00 | 1642 | return rc; |
d7e09d03 PT |
1643 | |
1644 | list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) { | |
7f1ae4c0 | 1645 | if (oap->oap_request) { |
d7e09d03 PT |
1646 | LASSERTF(request == oap->oap_request, |
1647 | "request %p != oap_request %p\n", | |
1648 | request, oap->oap_request); | |
1649 | if (oap->oap_interrupted) { | |
1650 | ptlrpc_req_finished(new_req); | |
0a3bdb00 | 1651 | return -EINTR; |
d7e09d03 PT |
1652 | } |
1653 | } | |
1654 | } | |
1655 | /* New request takes over pga and oaps from old request. | |
30aa9c52 OD |
1656 | * Note that copying a list_head doesn't work, need to move it... |
1657 | */ | |
d7e09d03 PT |
1658 | aa->aa_resends++; |
1659 | new_req->rq_interpret_reply = request->rq_interpret_reply; | |
1660 | new_req->rq_async_args = request->rq_async_args; | |
d7e09d03 | 1661 | /* cap resend delay to the current request timeout, this is similar to |
30aa9c52 OD |
1662 | * what ptlrpc does (see after_reply()) |
1663 | */ | |
d7e09d03 | 1664 | if (aa->aa_resends > new_req->rq_timeout) |
219e6de6 | 1665 | new_req->rq_sent = ktime_get_real_seconds() + new_req->rq_timeout; |
d7e09d03 | 1666 | else |
219e6de6 | 1667 | new_req->rq_sent = ktime_get_real_seconds() + aa->aa_resends; |
d7e09d03 PT |
1668 | new_req->rq_generation_set = 1; |
1669 | new_req->rq_import_generation = request->rq_import_generation; | |
1670 | ||
1671 | new_aa = ptlrpc_req_async_args(new_req); | |
1672 | ||
1673 | INIT_LIST_HEAD(&new_aa->aa_oaps); | |
1674 | list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps); | |
1675 | INIT_LIST_HEAD(&new_aa->aa_exts); | |
1676 | list_splice_init(&aa->aa_exts, &new_aa->aa_exts); | |
1677 | new_aa->aa_resends = aa->aa_resends; | |
1678 | ||
1679 | list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) { | |
1680 | if (oap->oap_request) { | |
1681 | ptlrpc_req_finished(oap->oap_request); | |
1682 | oap->oap_request = ptlrpc_request_addref(new_req); | |
1683 | } | |
1684 | } | |
1685 | ||
d7e09d03 PT |
1686 | /* XXX: This code will run into problem if we're going to support |
1687 | * to add a series of BRW RPCs into a self-defined ptlrpc_request_set | |
1688 | * and wait for all of them to be finished. We should inherit request | |
30aa9c52 OD |
1689 | * set from old request. |
1690 | */ | |
c5c4c6fa | 1691 | ptlrpcd_add_req(new_req); |
d7e09d03 PT |
1692 | |
1693 | DEBUG_REQ(D_INFO, new_req, "new request"); | |
0a3bdb00 | 1694 | return 0; |
d7e09d03 PT |
1695 | } |
1696 | ||
1697 | /* | |
1698 | * ugh, we want disk allocation on the target to happen in offset order. we'll | |
1699 | * follow sedgewicks advice and stick to the dead simple shellsort -- it'll do | |
1700 | * fine for our small page arrays and doesn't require allocation. its an | |
1701 | * insertion sort that swaps elements that are strides apart, shrinking the | |
1702 | * stride down until its '1' and the array is sorted. | |
1703 | */ | |
1704 | static void sort_brw_pages(struct brw_page **array, int num) | |
1705 | { | |
1706 | int stride, i, j; | |
1707 | struct brw_page *tmp; | |
1708 | ||
1709 | if (num == 1) | |
1710 | return; | |
1711 | for (stride = 1; stride < num ; stride = (stride * 3) + 1) | |
1712 | ; | |
1713 | ||
1714 | do { | |
1715 | stride /= 3; | |
1716 | for (i = stride ; i < num ; i++) { | |
1717 | tmp = array[i]; | |
1718 | j = i; | |
1719 | while (j >= stride && array[j - stride]->off > tmp->off) { | |
1720 | array[j] = array[j - stride]; | |
1721 | j -= stride; | |
1722 | } | |
1723 | array[j] = tmp; | |
1724 | } | |
1725 | } while (stride > 1); | |
1726 | } | |
1727 | ||
21aef7d9 | 1728 | static void osc_release_ppga(struct brw_page **ppga, u32 count) |
d7e09d03 | 1729 | { |
7f1ae4c0 | 1730 | LASSERT(ppga); |
7795178d | 1731 | kfree(ppga); |
d7e09d03 PT |
1732 | } |
1733 | ||
d7e09d03 PT |
1734 | static int brw_interpret(const struct lu_env *env, |
1735 | struct ptlrpc_request *req, void *data, int rc) | |
1736 | { | |
1737 | struct osc_brw_async_args *aa = data; | |
1738 | struct osc_extent *ext; | |
1739 | struct osc_extent *tmp; | |
29ac6840 | 1740 | struct cl_object *obj = NULL; |
d7e09d03 | 1741 | struct client_obd *cli = aa->aa_cli; |
d7e09d03 PT |
1742 | |
1743 | rc = osc_brw_fini_request(req, rc); | |
1744 | CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc); | |
1745 | /* When server return -EINPROGRESS, client should always retry | |
30aa9c52 OD |
1746 | * regardless of the number of times the bulk was resent already. |
1747 | */ | |
d7e09d03 PT |
1748 | if (osc_recoverable_error(rc)) { |
1749 | if (req->rq_import_generation != | |
1750 | req->rq_import->imp_generation) { | |
2d00bd17 | 1751 | CDEBUG(D_HA, "%s: resend cross eviction for object: " DOSTID ", rc = %d.\n", |
d7e09d03 PT |
1752 | req->rq_import->imp_obd->obd_name, |
1753 | POSTID(&aa->aa_oa->o_oi), rc); | |
1754 | } else if (rc == -EINPROGRESS || | |
1755 | client_should_resend(aa->aa_resends, aa->aa_cli)) { | |
1756 | rc = osc_brw_redo_request(req, aa, rc); | |
1757 | } else { | |
b0f5aad5 | 1758 | CERROR("%s: too many resent retries for object: %llu:%llu, rc = %d.\n", |
d7e09d03 PT |
1759 | req->rq_import->imp_obd->obd_name, |
1760 | POSTID(&aa->aa_oa->o_oi), rc); | |
1761 | } | |
1762 | ||
1763 | if (rc == 0) | |
0a3bdb00 | 1764 | return 0; |
d7e09d03 PT |
1765 | else if (rc == -EAGAIN || rc == -EINPROGRESS) |
1766 | rc = -EIO; | |
1767 | } | |
1768 | ||
d7e09d03 | 1769 | list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) { |
7f1ae4c0 | 1770 | if (!obj && rc == 0) { |
d7e09d03 PT |
1771 | obj = osc2cl(ext->oe_obj); |
1772 | cl_object_get(obj); | |
1773 | } | |
1774 | ||
1775 | list_del_init(&ext->oe_link); | |
1776 | osc_extent_finish(env, ext, 1, rc); | |
1777 | } | |
1778 | LASSERT(list_empty(&aa->aa_exts)); | |
1779 | LASSERT(list_empty(&aa->aa_oaps)); | |
1780 | ||
7f1ae4c0 | 1781 | if (obj) { |
d7e09d03 PT |
1782 | struct obdo *oa = aa->aa_oa; |
1783 | struct cl_attr *attr = &osc_env_info(env)->oti_attr; | |
1784 | unsigned long valid = 0; | |
1785 | ||
1786 | LASSERT(rc == 0); | |
1787 | if (oa->o_valid & OBD_MD_FLBLOCKS) { | |
1788 | attr->cat_blocks = oa->o_blocks; | |
1789 | valid |= CAT_BLOCKS; | |
1790 | } | |
1791 | if (oa->o_valid & OBD_MD_FLMTIME) { | |
1792 | attr->cat_mtime = oa->o_mtime; | |
1793 | valid |= CAT_MTIME; | |
1794 | } | |
1795 | if (oa->o_valid & OBD_MD_FLATIME) { | |
1796 | attr->cat_atime = oa->o_atime; | |
1797 | valid |= CAT_ATIME; | |
1798 | } | |
1799 | if (oa->o_valid & OBD_MD_FLCTIME) { | |
1800 | attr->cat_ctime = oa->o_ctime; | |
1801 | valid |= CAT_CTIME; | |
1802 | } | |
1803 | if (valid != 0) { | |
1804 | cl_object_attr_lock(obj); | |
1805 | cl_object_attr_set(env, obj, attr, valid); | |
1806 | cl_object_attr_unlock(obj); | |
1807 | } | |
1808 | cl_object_put(env, obj); | |
1809 | } | |
2ba262fb | 1810 | kmem_cache_free(obdo_cachep, aa->aa_oa); |
d7e09d03 PT |
1811 | |
1812 | cl_req_completion(env, aa->aa_clerq, rc < 0 ? rc : | |
1813 | req->rq_bulk->bd_nob_transferred); | |
1814 | osc_release_ppga(aa->aa_ppga, aa->aa_page_count); | |
1815 | ptlrpc_lprocfs_brw(req, req->rq_bulk->bd_nob_transferred); | |
1816 | ||
1817 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
1818 | /* We need to decrement before osc_ap_completion->osc_wake_cache_waiters | |
1819 | * is called so we know whether to go to sync BRWs or wait for more | |
30aa9c52 OD |
1820 | * RPCs to complete |
1821 | */ | |
d7e09d03 PT |
1822 | if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) |
1823 | cli->cl_w_in_flight--; | |
1824 | else | |
1825 | cli->cl_r_in_flight--; | |
1826 | osc_wake_cache_waiters(cli); | |
1827 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
1828 | ||
c5c4c6fa | 1829 | osc_io_unplug(env, cli, NULL); |
0a3bdb00 | 1830 | return rc; |
d7e09d03 PT |
1831 | } |
1832 | ||
d7e09d03 PT |
1833 | /** |
1834 | * Build an RPC by the list of extent @ext_list. The caller must ensure | |
1835 | * that the total pages in this list are NOT over max pages per RPC. | |
1836 | * Extents in the list must be in OES_RPC state. | |
1837 | */ | |
1838 | int osc_build_rpc(const struct lu_env *env, struct client_obd *cli, | |
c5c4c6fa | 1839 | struct list_head *ext_list, int cmd) |
d7e09d03 | 1840 | { |
29ac6840 CH |
1841 | struct ptlrpc_request *req = NULL; |
1842 | struct osc_extent *ext; | |
1843 | struct brw_page **pga = NULL; | |
1844 | struct osc_brw_async_args *aa = NULL; | |
1845 | struct obdo *oa = NULL; | |
1846 | struct osc_async_page *oap; | |
1847 | struct osc_async_page *tmp; | |
1848 | struct cl_req *clerq = NULL; | |
1849 | enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ; | |
1850 | struct ldlm_lock *lock = NULL; | |
1851 | struct cl_req_attr *crattr = NULL; | |
1852 | u64 starting_offset = OBD_OBJECT_EOF; | |
1853 | u64 ending_offset = 0; | |
1854 | int mpflag = 0; | |
1855 | int mem_tight = 0; | |
1856 | int page_count = 0; | |
1857 | int i; | |
1858 | int rc; | |
1859 | struct ost_body *body; | |
d7e09d03 | 1860 | LIST_HEAD(rpc_list); |
d7e09d03 | 1861 | |
d7e09d03 PT |
1862 | LASSERT(!list_empty(ext_list)); |
1863 | ||
1864 | /* add pages into rpc_list to build BRW rpc */ | |
1865 | list_for_each_entry(ext, ext_list, oe_link) { | |
1866 | LASSERT(ext->oe_state == OES_RPC); | |
1867 | mem_tight |= ext->oe_memalloc; | |
1868 | list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) { | |
1869 | ++page_count; | |
1870 | list_add_tail(&oap->oap_rpc_item, &rpc_list); | |
1871 | if (starting_offset > oap->oap_obj_off) | |
1872 | starting_offset = oap->oap_obj_off; | |
1873 | else | |
1874 | LASSERT(oap->oap_page_off == 0); | |
1875 | if (ending_offset < oap->oap_obj_off + oap->oap_count) | |
1876 | ending_offset = oap->oap_obj_off + | |
1877 | oap->oap_count; | |
1878 | else | |
1879 | LASSERT(oap->oap_page_off + oap->oap_count == | |
09cbfeaf | 1880 | PAGE_SIZE); |
d7e09d03 PT |
1881 | } |
1882 | } | |
1883 | ||
1884 | if (mem_tight) | |
1885 | mpflag = cfs_memory_pressure_get_and_set(); | |
1886 | ||
7795178d | 1887 | crattr = kzalloc(sizeof(*crattr), GFP_NOFS); |
3408e9ae | 1888 | if (!crattr) { |
26c4ea46 TJ |
1889 | rc = -ENOMEM; |
1890 | goto out; | |
1891 | } | |
cad6fafa | 1892 | |
7795178d | 1893 | pga = kcalloc(page_count, sizeof(*pga), GFP_NOFS); |
7f1ae4c0 | 1894 | if (!pga) { |
26c4ea46 TJ |
1895 | rc = -ENOMEM; |
1896 | goto out; | |
1897 | } | |
d7e09d03 | 1898 | |
c4418dac | 1899 | oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS); |
7f1ae4c0 | 1900 | if (!oa) { |
26c4ea46 TJ |
1901 | rc = -ENOMEM; |
1902 | goto out; | |
1903 | } | |
d7e09d03 PT |
1904 | |
1905 | i = 0; | |
1906 | list_for_each_entry(oap, &rpc_list, oap_rpc_item) { | |
1907 | struct cl_page *page = oap2cl_page(oap); | |
50ffcb7e | 1908 | |
7f1ae4c0 | 1909 | if (!clerq) { |
d7e09d03 | 1910 | clerq = cl_req_alloc(env, page, crt, |
cad6fafa | 1911 | 1 /* only 1-object rpcs for now */); |
26c4ea46 TJ |
1912 | if (IS_ERR(clerq)) { |
1913 | rc = PTR_ERR(clerq); | |
1914 | goto out; | |
1915 | } | |
d7e09d03 PT |
1916 | lock = oap->oap_ldlm_lock; |
1917 | } | |
1918 | if (mem_tight) | |
1919 | oap->oap_brw_flags |= OBD_BRW_MEMALLOC; | |
1920 | pga[i] = &oap->oap_brw_page; | |
1921 | pga[i]->off = oap->oap_obj_off + oap->oap_page_off; | |
1922 | CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n", | |
cad6fafa BJ |
1923 | pga[i]->pg, page_index(oap->oap_page), oap, |
1924 | pga[i]->flag); | |
d7e09d03 PT |
1925 | i++; |
1926 | cl_req_page_add(env, clerq, page); | |
1927 | } | |
1928 | ||
1929 | /* always get the data for the obdo for the rpc */ | |
7f1ae4c0 | 1930 | LASSERT(clerq); |
cad6fafa BJ |
1931 | crattr->cra_oa = oa; |
1932 | cl_req_attr_set(env, clerq, crattr, ~0ULL); | |
d7e09d03 PT |
1933 | if (lock) { |
1934 | oa->o_handle = lock->l_remote_handle; | |
1935 | oa->o_valid |= OBD_MD_FLHANDLE; | |
1936 | } | |
1937 | ||
1938 | rc = cl_req_prep(env, clerq); | |
1939 | if (rc != 0) { | |
1940 | CERROR("cl_req_prep failed: %d\n", rc); | |
26c4ea46 | 1941 | goto out; |
d7e09d03 PT |
1942 | } |
1943 | ||
1944 | sort_brw_pages(pga, page_count); | |
1945 | rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count, | |
79910d7d | 1946 | pga, &req, 1, 0); |
d7e09d03 PT |
1947 | if (rc != 0) { |
1948 | CERROR("prep_req failed: %d\n", rc); | |
26c4ea46 | 1949 | goto out; |
d7e09d03 PT |
1950 | } |
1951 | ||
d7e09d03 PT |
1952 | req->rq_interpret_reply = brw_interpret; |
1953 | ||
1954 | if (mem_tight != 0) | |
1955 | req->rq_memalloc = 1; | |
1956 | ||
1957 | /* Need to update the timestamps after the request is built in case | |
1958 | * we race with setattr (locally or in queue at OST). If OST gets | |
1959 | * later setattr before earlier BRW (as determined by the request xid), | |
1960 | * the OST will not use BRW timestamps. Sadly, there is no obvious | |
30aa9c52 OD |
1961 | * way to do this in a single call. bug 10150 |
1962 | */ | |
3ce08cd7 NY |
1963 | body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); |
1964 | crattr->cra_oa = &body->oa; | |
cad6fafa | 1965 | cl_req_attr_set(env, clerq, crattr, |
d7e09d03 PT |
1966 | OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME); |
1967 | ||
cad6fafa | 1968 | lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid); |
d7e09d03 PT |
1969 | |
1970 | CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args)); | |
1971 | aa = ptlrpc_req_async_args(req); | |
1972 | INIT_LIST_HEAD(&aa->aa_oaps); | |
1973 | list_splice_init(&rpc_list, &aa->aa_oaps); | |
1974 | INIT_LIST_HEAD(&aa->aa_exts); | |
1975 | list_splice_init(ext_list, &aa->aa_exts); | |
1976 | aa->aa_clerq = clerq; | |
1977 | ||
1978 | /* queued sync pages can be torn down while the pages | |
30aa9c52 OD |
1979 | * were between the pending list and the rpc |
1980 | */ | |
d7e09d03 PT |
1981 | tmp = NULL; |
1982 | list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) { | |
1983 | /* only one oap gets a request reference */ | |
7f1ae4c0 | 1984 | if (!tmp) |
d7e09d03 PT |
1985 | tmp = oap; |
1986 | if (oap->oap_interrupted && !req->rq_intr) { | |
1987 | CDEBUG(D_INODE, "oap %p in req %p interrupted\n", | |
79910d7d | 1988 | oap, req); |
d7e09d03 PT |
1989 | ptlrpc_mark_interrupted(req); |
1990 | } | |
1991 | } | |
7f1ae4c0 | 1992 | if (tmp) |
d7e09d03 PT |
1993 | tmp->oap_request = ptlrpc_request_addref(req); |
1994 | ||
1995 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
09cbfeaf | 1996 | starting_offset >>= PAGE_SHIFT; |
d7e09d03 PT |
1997 | if (cmd == OBD_BRW_READ) { |
1998 | cli->cl_r_in_flight++; | |
1999 | lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count); | |
2000 | lprocfs_oh_tally(&cli->cl_read_rpc_hist, cli->cl_r_in_flight); | |
2001 | lprocfs_oh_tally_log2(&cli->cl_read_offset_hist, | |
2002 | starting_offset + 1); | |
2003 | } else { | |
2004 | cli->cl_w_in_flight++; | |
2005 | lprocfs_oh_tally_log2(&cli->cl_write_page_hist, page_count); | |
2006 | lprocfs_oh_tally(&cli->cl_write_rpc_hist, cli->cl_w_in_flight); | |
2007 | lprocfs_oh_tally_log2(&cli->cl_write_offset_hist, | |
2008 | starting_offset + 1); | |
2009 | } | |
2010 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
2011 | ||
2012 | DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %dr/%dw in flight", | |
2013 | page_count, aa, cli->cl_r_in_flight, | |
2014 | cli->cl_w_in_flight); | |
2015 | ||
c5c4c6fa | 2016 | ptlrpcd_add_req(req); |
d7e09d03 | 2017 | rc = 0; |
d7e09d03 PT |
2018 | |
2019 | out: | |
2020 | if (mem_tight != 0) | |
2021 | cfs_memory_pressure_restore(mpflag); | |
2022 | ||
f999d098 | 2023 | kfree(crattr); |
cad6fafa | 2024 | |
d7e09d03 | 2025 | if (rc != 0) { |
7f1ae4c0 | 2026 | LASSERT(!req); |
d7e09d03 PT |
2027 | |
2028 | if (oa) | |
2ba262fb | 2029 | kmem_cache_free(obdo_cachep, oa); |
59e267c0 | 2030 | kfree(pga); |
d7e09d03 | 2031 | /* this should happen rarely and is pretty bad, it makes the |
30aa9c52 OD |
2032 | * pending list not follow the dirty order |
2033 | */ | |
d7e09d03 PT |
2034 | while (!list_empty(ext_list)) { |
2035 | ext = list_entry(ext_list->next, struct osc_extent, | |
79910d7d | 2036 | oe_link); |
d7e09d03 PT |
2037 | list_del_init(&ext->oe_link); |
2038 | osc_extent_finish(env, ext, 0, rc); | |
2039 | } | |
2040 | if (clerq && !IS_ERR(clerq)) | |
2041 | cl_req_completion(env, clerq, rc); | |
2042 | } | |
0a3bdb00 | 2043 | return rc; |
d7e09d03 PT |
2044 | } |
2045 | ||
2046 | static int osc_set_lock_data_with_check(struct ldlm_lock *lock, | |
2047 | struct ldlm_enqueue_info *einfo) | |
2048 | { | |
2049 | void *data = einfo->ei_cbdata; | |
2050 | int set = 0; | |
2051 | ||
d7e09d03 PT |
2052 | LASSERT(lock->l_blocking_ast == einfo->ei_cb_bl); |
2053 | LASSERT(lock->l_resource->lr_type == einfo->ei_type); | |
2054 | LASSERT(lock->l_completion_ast == einfo->ei_cb_cp); | |
2055 | LASSERT(lock->l_glimpse_ast == einfo->ei_cb_gl); | |
2056 | ||
2057 | lock_res_and_lock(lock); | |
2058 | spin_lock(&osc_ast_guard); | |
2059 | ||
7f1ae4c0 | 2060 | if (!lock->l_ast_data) |
d7e09d03 PT |
2061 | lock->l_ast_data = data; |
2062 | if (lock->l_ast_data == data) | |
2063 | set = 1; | |
2064 | ||
2065 | spin_unlock(&osc_ast_guard); | |
2066 | unlock_res_and_lock(lock); | |
2067 | ||
2068 | return set; | |
2069 | } | |
2070 | ||
2071 | static int osc_set_data_with_check(struct lustre_handle *lockh, | |
2072 | struct ldlm_enqueue_info *einfo) | |
2073 | { | |
2074 | struct ldlm_lock *lock = ldlm_handle2lock(lockh); | |
2075 | int set = 0; | |
2076 | ||
7f1ae4c0 | 2077 | if (lock) { |
d7e09d03 PT |
2078 | set = osc_set_lock_data_with_check(lock, einfo); |
2079 | LDLM_LOCK_PUT(lock); | |
2080 | } else | |
2081 | CERROR("lockh %p, data %p - client evicted?\n", | |
2082 | lockh, einfo->ei_cbdata); | |
2083 | return set; | |
2084 | } | |
2085 | ||
d7e09d03 PT |
2086 | /* find any ldlm lock of the inode in osc |
2087 | * return 0 not find | |
2088 | * 1 find one | |
30aa9c52 OD |
2089 | * < 0 error |
2090 | */ | |
d7e09d03 PT |
2091 | static int osc_find_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm, |
2092 | ldlm_iterator_t replace, void *data) | |
2093 | { | |
2094 | struct ldlm_res_id res_id; | |
2095 | struct obd_device *obd = class_exp2obd(exp); | |
2096 | int rc = 0; | |
2097 | ||
2098 | ostid_build_res_name(&lsm->lsm_oi, &res_id); | |
2099 | rc = ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data); | |
2100 | if (rc == LDLM_ITER_STOP) | |
fbe7c6c7 | 2101 | return 1; |
d7e09d03 | 2102 | if (rc == LDLM_ITER_CONTINUE) |
fbe7c6c7 JL |
2103 | return 0; |
2104 | return rc; | |
d7e09d03 PT |
2105 | } |
2106 | ||
2107 | static int osc_enqueue_fini(struct ptlrpc_request *req, struct ost_lvb *lvb, | |
2108 | obd_enqueue_update_f upcall, void *cookie, | |
2109 | __u64 *flags, int agl, int rc) | |
2110 | { | |
2111 | int intent = *flags & LDLM_FL_HAS_INTENT; | |
d7e09d03 PT |
2112 | |
2113 | if (intent) { | |
2114 | /* The request was created before ldlm_cli_enqueue call. */ | |
2115 | if (rc == ELDLM_LOCK_ABORTED) { | |
2116 | struct ldlm_reply *rep; | |
50ffcb7e | 2117 | |
d7e09d03 PT |
2118 | rep = req_capsule_server_get(&req->rq_pill, |
2119 | &RMF_DLM_REP); | |
2120 | ||
2d58de78 LW |
2121 | rep->lock_policy_res1 = |
2122 | ptlrpc_status_ntoh(rep->lock_policy_res1); | |
d7e09d03 PT |
2123 | if (rep->lock_policy_res1) |
2124 | rc = rep->lock_policy_res1; | |
2125 | } | |
2126 | } | |
2127 | ||
2128 | if ((intent != 0 && rc == ELDLM_LOCK_ABORTED && agl == 0) || | |
2129 | (rc == 0)) { | |
2130 | *flags |= LDLM_FL_LVB_READY; | |
1d8cb70c | 2131 | CDEBUG(D_INODE, "got kms %llu blocks %llu mtime %llu\n", |
d7e09d03 PT |
2132 | lvb->lvb_size, lvb->lvb_blocks, lvb->lvb_mtime); |
2133 | } | |
2134 | ||
2135 | /* Call the update callback. */ | |
2136 | rc = (*upcall)(cookie, rc); | |
0a3bdb00 | 2137 | return rc; |
d7e09d03 PT |
2138 | } |
2139 | ||
2140 | static int osc_enqueue_interpret(const struct lu_env *env, | |
2141 | struct ptlrpc_request *req, | |
2142 | struct osc_enqueue_args *aa, int rc) | |
2143 | { | |
2144 | struct ldlm_lock *lock; | |
2145 | struct lustre_handle handle; | |
2146 | __u32 mode; | |
2147 | struct ost_lvb *lvb; | |
2148 | __u32 lvb_len; | |
2149 | __u64 *flags = aa->oa_flags; | |
2150 | ||
2151 | /* Make a local copy of a lock handle and a mode, because aa->oa_* | |
30aa9c52 OD |
2152 | * might be freed anytime after lock upcall has been called. |
2153 | */ | |
d7e09d03 PT |
2154 | lustre_handle_copy(&handle, aa->oa_lockh); |
2155 | mode = aa->oa_ei->ei_mode; | |
2156 | ||
2157 | /* ldlm_cli_enqueue is holding a reference on the lock, so it must | |
30aa9c52 OD |
2158 | * be valid. |
2159 | */ | |
d7e09d03 PT |
2160 | lock = ldlm_handle2lock(&handle); |
2161 | ||
2162 | /* Take an additional reference so that a blocking AST that | |
2163 | * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed | |
2164 | * to arrive after an upcall has been executed by | |
30aa9c52 OD |
2165 | * osc_enqueue_fini(). |
2166 | */ | |
d7e09d03 PT |
2167 | ldlm_lock_addref(&handle, mode); |
2168 | ||
2169 | /* Let CP AST to grant the lock first. */ | |
2170 | OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1); | |
2171 | ||
2172 | if (aa->oa_agl && rc == ELDLM_LOCK_ABORTED) { | |
2173 | lvb = NULL; | |
2174 | lvb_len = 0; | |
2175 | } else { | |
2176 | lvb = aa->oa_lvb; | |
2177 | lvb_len = sizeof(*aa->oa_lvb); | |
2178 | } | |
2179 | ||
2180 | /* Complete obtaining the lock procedure. */ | |
2181 | rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_ei->ei_type, 1, | |
2182 | mode, flags, lvb, lvb_len, &handle, rc); | |
2183 | /* Complete osc stuff. */ | |
2184 | rc = osc_enqueue_fini(req, aa->oa_lvb, aa->oa_upcall, aa->oa_cookie, | |
2185 | flags, aa->oa_agl, rc); | |
2186 | ||
2187 | OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10); | |
2188 | ||
2189 | /* Release the lock for async request. */ | |
2190 | if (lustre_handle_is_used(&handle) && rc == ELDLM_OK) | |
2191 | /* | |
2192 | * Releases a reference taken by ldlm_cli_enqueue(), if it is | |
2193 | * not already released by | |
2194 | * ldlm_cli_enqueue_fini()->failed_lock_cleanup() | |
2195 | */ | |
2196 | ldlm_lock_decref(&handle, mode); | |
2197 | ||
7f1ae4c0 | 2198 | LASSERTF(lock, "lockh %p, req %p, aa %p - client evicted?\n", |
d7e09d03 PT |
2199 | aa->oa_lockh, req, aa); |
2200 | ldlm_lock_decref(&handle, mode); | |
2201 | LDLM_LOCK_PUT(lock); | |
2202 | return rc; | |
2203 | } | |
2204 | ||
d7e09d03 PT |
2205 | struct ptlrpc_request_set *PTLRPCD_SET = (void *)1; |
2206 | ||
2207 | /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock | |
2208 | * from the 2nd OSC before a lock from the 1st one. This does not deadlock with | |
2209 | * other synchronous requests, however keeping some locks and trying to obtain | |
2210 | * others may take a considerable amount of time in a case of ost failure; and | |
2211 | * when other sync requests do not get released lock from a client, the client | |
2212 | * is excluded from the cluster -- such scenarious make the life difficult, so | |
30aa9c52 OD |
2213 | * release locks just after they are obtained. |
2214 | */ | |
d7e09d03 PT |
2215 | int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id, |
2216 | __u64 *flags, ldlm_policy_data_t *policy, | |
2217 | struct ost_lvb *lvb, int kms_valid, | |
2218 | obd_enqueue_update_f upcall, void *cookie, | |
2219 | struct ldlm_enqueue_info *einfo, | |
2220 | struct lustre_handle *lockh, | |
2221 | struct ptlrpc_request_set *rqset, int async, int agl) | |
2222 | { | |
2223 | struct obd_device *obd = exp->exp_obd; | |
2224 | struct ptlrpc_request *req = NULL; | |
2225 | int intent = *flags & LDLM_FL_HAS_INTENT; | |
875332d4 | 2226 | __u64 match_lvb = (agl != 0 ? 0 : LDLM_FL_LVB_READY); |
52ee0d20 | 2227 | enum ldlm_mode mode; |
d7e09d03 | 2228 | int rc; |
d7e09d03 PT |
2229 | |
2230 | /* Filesystem lock extents are extended to page boundaries so that | |
30aa9c52 OD |
2231 | * dealing with the page cache is a little smoother. |
2232 | */ | |
d7e09d03 PT |
2233 | policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK; |
2234 | policy->l_extent.end |= ~CFS_PAGE_MASK; | |
2235 | ||
2236 | /* | |
2237 | * kms is not valid when either object is completely fresh (so that no | |
2238 | * locks are cached), or object was evicted. In the latter case cached | |
2239 | * lock cannot be used, because it would prime inode state with | |
2240 | * potentially stale LVB. | |
2241 | */ | |
2242 | if (!kms_valid) | |
2243 | goto no_match; | |
2244 | ||
2245 | /* Next, search for already existing extent locks that will cover us */ | |
2246 | /* If we're trying to read, we also search for an existing PW lock. The | |
2247 | * VFS and page cache already protect us locally, so lots of readers/ | |
2248 | * writers can share a single PW lock. | |
2249 | * | |
2250 | * There are problems with conversion deadlocks, so instead of | |
2251 | * converting a read lock to a write lock, we'll just enqueue a new | |
2252 | * one. | |
2253 | * | |
2254 | * At some point we should cancel the read lock instead of making them | |
2255 | * send us a blocking callback, but there are problems with canceling | |
30aa9c52 OD |
2256 | * locks out from other users right now, too. |
2257 | */ | |
d7e09d03 PT |
2258 | mode = einfo->ei_mode; |
2259 | if (einfo->ei_mode == LCK_PR) | |
2260 | mode |= LCK_PW; | |
2261 | mode = ldlm_lock_match(obd->obd_namespace, *flags | match_lvb, res_id, | |
2262 | einfo->ei_type, policy, mode, lockh, 0); | |
2263 | if (mode) { | |
2264 | struct ldlm_lock *matched = ldlm_handle2lock(lockh); | |
2265 | ||
2266 | if ((agl != 0) && !(matched->l_flags & LDLM_FL_LVB_READY)) { | |
2267 | /* For AGL, if enqueue RPC is sent but the lock is not | |
2268 | * granted, then skip to process this strpe. | |
30aa9c52 OD |
2269 | * Return -ECANCELED to tell the caller. |
2270 | */ | |
d7e09d03 PT |
2271 | ldlm_lock_decref(lockh, mode); |
2272 | LDLM_LOCK_PUT(matched); | |
0a3bdb00 | 2273 | return -ECANCELED; |
71e8dd9a AM |
2274 | } |
2275 | ||
2276 | if (osc_set_lock_data_with_check(matched, einfo)) { | |
d7e09d03 PT |
2277 | *flags |= LDLM_FL_LVB_READY; |
2278 | /* addref the lock only if not async requests and PW | |
30aa9c52 OD |
2279 | * lock is matched whereas we asked for PR. |
2280 | */ | |
d7e09d03 PT |
2281 | if (!rqset && einfo->ei_mode != mode) |
2282 | ldlm_lock_addref(lockh, LCK_PR); | |
2283 | if (intent) { | |
2284 | /* I would like to be able to ASSERT here that | |
2285 | * rss <= kms, but I can't, for reasons which | |
30aa9c52 OD |
2286 | * are explained in lov_enqueue() |
2287 | */ | |
d7e09d03 PT |
2288 | } |
2289 | ||
2290 | /* We already have a lock, and it's referenced. | |
2291 | * | |
2292 | * At this point, the cl_lock::cll_state is CLS_QUEUING, | |
30aa9c52 OD |
2293 | * AGL upcall may change it to CLS_HELD directly. |
2294 | */ | |
d7e09d03 PT |
2295 | (*upcall)(cookie, ELDLM_OK); |
2296 | ||
2297 | if (einfo->ei_mode != mode) | |
2298 | ldlm_lock_decref(lockh, LCK_PW); | |
2299 | else if (rqset) | |
2300 | /* For async requests, decref the lock. */ | |
2301 | ldlm_lock_decref(lockh, einfo->ei_mode); | |
2302 | LDLM_LOCK_PUT(matched); | |
0a3bdb00 | 2303 | return ELDLM_OK; |
d7e09d03 | 2304 | } |
71e8dd9a AM |
2305 | |
2306 | ldlm_lock_decref(lockh, mode); | |
2307 | LDLM_LOCK_PUT(matched); | |
d7e09d03 PT |
2308 | } |
2309 | ||
2310 | no_match: | |
2311 | if (intent) { | |
2312 | LIST_HEAD(cancels); | |
50ffcb7e | 2313 | |
d7e09d03 PT |
2314 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), |
2315 | &RQF_LDLM_ENQUEUE_LVB); | |
7f1ae4c0 | 2316 | if (!req) |
0a3bdb00 | 2317 | return -ENOMEM; |
d7e09d03 PT |
2318 | |
2319 | rc = ldlm_prep_enqueue_req(exp, req, &cancels, 0); | |
2320 | if (rc) { | |
2321 | ptlrpc_request_free(req); | |
0a3bdb00 | 2322 | return rc; |
d7e09d03 PT |
2323 | } |
2324 | ||
2325 | req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, | |
ec83e611 | 2326 | sizeof(*lvb)); |
d7e09d03 PT |
2327 | ptlrpc_request_set_replen(req); |
2328 | } | |
2329 | ||
2330 | /* users of osc_enqueue() can pass this flag for ldlm_lock_match() */ | |
2331 | *flags &= ~LDLM_FL_BLOCK_GRANTED; | |
2332 | ||
2333 | rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb, | |
2334 | sizeof(*lvb), LVB_T_OST, lockh, async); | |
2335 | if (rqset) { | |
2336 | if (!rc) { | |
2337 | struct osc_enqueue_args *aa; | |
50ffcb7e | 2338 | |
d7e09d03 PT |
2339 | CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args)); |
2340 | aa = ptlrpc_req_async_args(req); | |
2341 | aa->oa_ei = einfo; | |
2342 | aa->oa_exp = exp; | |
2343 | aa->oa_flags = flags; | |
2344 | aa->oa_upcall = upcall; | |
2345 | aa->oa_cookie = cookie; | |
2346 | aa->oa_lvb = lvb; | |
2347 | aa->oa_lockh = lockh; | |
2348 | aa->oa_agl = !!agl; | |
2349 | ||
2350 | req->rq_interpret_reply = | |
2351 | (ptlrpc_interpterer_t)osc_enqueue_interpret; | |
2352 | if (rqset == PTLRPCD_SET) | |
c5c4c6fa | 2353 | ptlrpcd_add_req(req); |
d7e09d03 PT |
2354 | else |
2355 | ptlrpc_set_add_req(rqset, req); | |
2356 | } else if (intent) { | |
2357 | ptlrpc_req_finished(req); | |
2358 | } | |
0a3bdb00 | 2359 | return rc; |
d7e09d03 PT |
2360 | } |
2361 | ||
2362 | rc = osc_enqueue_fini(req, lvb, upcall, cookie, flags, agl, rc); | |
2363 | if (intent) | |
2364 | ptlrpc_req_finished(req); | |
2365 | ||
0a3bdb00 | 2366 | return rc; |
d7e09d03 PT |
2367 | } |
2368 | ||
d7e09d03 PT |
2369 | int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id, |
2370 | __u32 type, ldlm_policy_data_t *policy, __u32 mode, | |
875332d4 | 2371 | __u64 *flags, void *data, struct lustre_handle *lockh, |
d7e09d03 PT |
2372 | int unref) |
2373 | { | |
2374 | struct obd_device *obd = exp->exp_obd; | |
875332d4 | 2375 | __u64 lflags = *flags; |
52ee0d20 | 2376 | enum ldlm_mode rc; |
d7e09d03 PT |
2377 | |
2378 | if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH)) | |
0a3bdb00 | 2379 | return -EIO; |
d7e09d03 PT |
2380 | |
2381 | /* Filesystem lock extents are extended to page boundaries so that | |
30aa9c52 OD |
2382 | * dealing with the page cache is a little smoother |
2383 | */ | |
d7e09d03 PT |
2384 | policy->l_extent.start -= policy->l_extent.start & ~CFS_PAGE_MASK; |
2385 | policy->l_extent.end |= ~CFS_PAGE_MASK; | |
2386 | ||
2387 | /* Next, search for already existing extent locks that will cover us */ | |
2388 | /* If we're trying to read, we also search for an existing PW lock. The | |
2389 | * VFS and page cache already protect us locally, so lots of readers/ | |
30aa9c52 OD |
2390 | * writers can share a single PW lock. |
2391 | */ | |
d7e09d03 PT |
2392 | rc = mode; |
2393 | if (mode == LCK_PR) | |
2394 | rc |= LCK_PW; | |
2395 | rc = ldlm_lock_match(obd->obd_namespace, lflags, | |
2396 | res_id, type, policy, rc, lockh, unref); | |
2397 | if (rc) { | |
7f1ae4c0 | 2398 | if (data) { |
d7e09d03 PT |
2399 | if (!osc_set_data_with_check(lockh, data)) { |
2400 | if (!(lflags & LDLM_FL_TEST_LOCK)) | |
2401 | ldlm_lock_decref(lockh, rc); | |
0a3bdb00 | 2402 | return 0; |
d7e09d03 PT |
2403 | } |
2404 | } | |
2405 | if (!(lflags & LDLM_FL_TEST_LOCK) && mode != rc) { | |
2406 | ldlm_lock_addref(lockh, LCK_PR); | |
2407 | ldlm_lock_decref(lockh, LCK_PW); | |
2408 | } | |
0a3bdb00 | 2409 | return rc; |
d7e09d03 | 2410 | } |
0a3bdb00 | 2411 | return rc; |
d7e09d03 PT |
2412 | } |
2413 | ||
2414 | int osc_cancel_base(struct lustre_handle *lockh, __u32 mode) | |
2415 | { | |
d7e09d03 PT |
2416 | if (unlikely(mode == LCK_GROUP)) |
2417 | ldlm_lock_decref_and_cancel(lockh, mode); | |
2418 | else | |
2419 | ldlm_lock_decref(lockh, mode); | |
2420 | ||
0a3bdb00 | 2421 | return 0; |
d7e09d03 PT |
2422 | } |
2423 | ||
d7e09d03 PT |
2424 | static int osc_statfs_interpret(const struct lu_env *env, |
2425 | struct ptlrpc_request *req, | |
2426 | struct osc_async_args *aa, int rc) | |
2427 | { | |
2428 | struct obd_statfs *msfs; | |
d7e09d03 PT |
2429 | |
2430 | if (rc == -EBADR) | |
2431 | /* The request has in fact never been sent | |
2432 | * due to issues at a higher level (LOV). | |
2433 | * Exit immediately since the caller is | |
2434 | * aware of the problem and takes care | |
30aa9c52 OD |
2435 | * of the clean up |
2436 | */ | |
defa220f | 2437 | return rc; |
d7e09d03 PT |
2438 | |
2439 | if ((rc == -ENOTCONN || rc == -EAGAIN) && | |
26c4ea46 TJ |
2440 | (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY)) { |
2441 | rc = 0; | |
2442 | goto out; | |
2443 | } | |
d7e09d03 PT |
2444 | |
2445 | if (rc != 0) | |
26c4ea46 | 2446 | goto out; |
d7e09d03 PT |
2447 | |
2448 | msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS); | |
7f1ae4c0 | 2449 | if (!msfs) { |
26c4ea46 TJ |
2450 | rc = -EPROTO; |
2451 | goto out; | |
d7e09d03 PT |
2452 | } |
2453 | ||
2454 | *aa->aa_oi->oi_osfs = *msfs; | |
2455 | out: | |
2456 | rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc); | |
0a3bdb00 | 2457 | return rc; |
d7e09d03 PT |
2458 | } |
2459 | ||
2460 | static int osc_statfs_async(struct obd_export *exp, | |
2461 | struct obd_info *oinfo, __u64 max_age, | |
2462 | struct ptlrpc_request_set *rqset) | |
2463 | { | |
29ac6840 | 2464 | struct obd_device *obd = class_exp2obd(exp); |
d7e09d03 PT |
2465 | struct ptlrpc_request *req; |
2466 | struct osc_async_args *aa; | |
29ac6840 | 2467 | int rc; |
d7e09d03 PT |
2468 | |
2469 | /* We could possibly pass max_age in the request (as an absolute | |
2470 | * timestamp or a "seconds.usec ago") so the target can avoid doing | |
2471 | * extra calls into the filesystem if that isn't necessary (e.g. | |
2472 | * during mount that would help a bit). Having relative timestamps | |
2473 | * is not so great if request processing is slow, while absolute | |
30aa9c52 OD |
2474 | * timestamps are not ideal because they need time synchronization. |
2475 | */ | |
d7e09d03 | 2476 | req = ptlrpc_request_alloc(obd->u.cli.cl_import, &RQF_OST_STATFS); |
7f1ae4c0 | 2477 | if (!req) |
0a3bdb00 | 2478 | return -ENOMEM; |
d7e09d03 PT |
2479 | |
2480 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS); | |
2481 | if (rc) { | |
2482 | ptlrpc_request_free(req); | |
0a3bdb00 | 2483 | return rc; |
d7e09d03 PT |
2484 | } |
2485 | ptlrpc_request_set_replen(req); | |
2486 | req->rq_request_portal = OST_CREATE_PORTAL; | |
2487 | ptlrpc_at_set_req_timeout(req); | |
2488 | ||
2489 | if (oinfo->oi_flags & OBD_STATFS_NODELAY) { | |
2490 | /* procfs requests not want stat in wait for avoid deadlock */ | |
2491 | req->rq_no_resend = 1; | |
2492 | req->rq_no_delay = 1; | |
2493 | } | |
2494 | ||
2495 | req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_statfs_interpret; | |
2496 | CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args)); | |
2497 | aa = ptlrpc_req_async_args(req); | |
2498 | aa->aa_oi = oinfo; | |
2499 | ||
2500 | ptlrpc_set_add_req(rqset, req); | |
0a3bdb00 | 2501 | return 0; |
d7e09d03 PT |
2502 | } |
2503 | ||
2504 | static int osc_statfs(const struct lu_env *env, struct obd_export *exp, | |
2505 | struct obd_statfs *osfs, __u64 max_age, __u32 flags) | |
2506 | { | |
29ac6840 CH |
2507 | struct obd_device *obd = class_exp2obd(exp); |
2508 | struct obd_statfs *msfs; | |
d7e09d03 | 2509 | struct ptlrpc_request *req; |
29ac6840 | 2510 | struct obd_import *imp = NULL; |
d7e09d03 | 2511 | int rc; |
d7e09d03 | 2512 | |
30aa9c52 OD |
2513 | /* Since the request might also come from lprocfs, so we need |
2514 | * sync this with client_disconnect_export Bug15684 | |
2515 | */ | |
d7e09d03 PT |
2516 | down_read(&obd->u.cli.cl_sem); |
2517 | if (obd->u.cli.cl_import) | |
2518 | imp = class_import_get(obd->u.cli.cl_import); | |
2519 | up_read(&obd->u.cli.cl_sem); | |
2520 | if (!imp) | |
0a3bdb00 | 2521 | return -ENODEV; |
d7e09d03 PT |
2522 | |
2523 | /* We could possibly pass max_age in the request (as an absolute | |
2524 | * timestamp or a "seconds.usec ago") so the target can avoid doing | |
2525 | * extra calls into the filesystem if that isn't necessary (e.g. | |
2526 | * during mount that would help a bit). Having relative timestamps | |
2527 | * is not so great if request processing is slow, while absolute | |
30aa9c52 OD |
2528 | * timestamps are not ideal because they need time synchronization. |
2529 | */ | |
d7e09d03 PT |
2530 | req = ptlrpc_request_alloc(imp, &RQF_OST_STATFS); |
2531 | ||
2532 | class_import_put(imp); | |
2533 | ||
7f1ae4c0 | 2534 | if (!req) |
0a3bdb00 | 2535 | return -ENOMEM; |
d7e09d03 PT |
2536 | |
2537 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_STATFS); | |
2538 | if (rc) { | |
2539 | ptlrpc_request_free(req); | |
0a3bdb00 | 2540 | return rc; |
d7e09d03 PT |
2541 | } |
2542 | ptlrpc_request_set_replen(req); | |
2543 | req->rq_request_portal = OST_CREATE_PORTAL; | |
2544 | ptlrpc_at_set_req_timeout(req); | |
2545 | ||
2546 | if (flags & OBD_STATFS_NODELAY) { | |
2547 | /* procfs requests not want stat in wait for avoid deadlock */ | |
2548 | req->rq_no_resend = 1; | |
2549 | req->rq_no_delay = 1; | |
2550 | } | |
2551 | ||
2552 | rc = ptlrpc_queue_wait(req); | |
2553 | if (rc) | |
26c4ea46 | 2554 | goto out; |
d7e09d03 PT |
2555 | |
2556 | msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS); | |
7f1ae4c0 | 2557 | if (!msfs) { |
26c4ea46 TJ |
2558 | rc = -EPROTO; |
2559 | goto out; | |
d7e09d03 PT |
2560 | } |
2561 | ||
2562 | *osfs = *msfs; | |
2563 | ||
d7e09d03 PT |
2564 | out: |
2565 | ptlrpc_req_finished(req); | |
2566 | return rc; | |
2567 | } | |
2568 | ||
2569 | /* Retrieve object striping information. | |
2570 | * | |
2571 | * @lmmu is a pointer to an in-core struct with lmm_ost_count indicating | |
2572 | * the maximum number of OST indices which will fit in the user buffer. | |
2573 | * lmm_magic must be LOV_MAGIC (we only use 1 slot here). | |
2574 | */ | |
ec2d71d0 OD |
2575 | static int osc_getstripe(struct lov_stripe_md *lsm, |
2576 | struct lov_user_md __user *lump) | |
d7e09d03 PT |
2577 | { |
2578 | /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */ | |
2579 | struct lov_user_md_v3 lum, *lumk; | |
2580 | struct lov_user_ost_data_v1 *lmm_objects; | |
2581 | int rc = 0, lum_size; | |
d7e09d03 PT |
2582 | |
2583 | if (!lsm) | |
0a3bdb00 | 2584 | return -ENODATA; |
d7e09d03 PT |
2585 | |
2586 | /* we only need the header part from user space to get lmm_magic and | |
30aa9c52 OD |
2587 | * lmm_stripe_count, (the header part is common to v1 and v3) |
2588 | */ | |
d7e09d03 PT |
2589 | lum_size = sizeof(struct lov_user_md_v1); |
2590 | if (copy_from_user(&lum, lump, lum_size)) | |
0a3bdb00 | 2591 | return -EFAULT; |
d7e09d03 PT |
2592 | |
2593 | if ((lum.lmm_magic != LOV_USER_MAGIC_V1) && | |
2594 | (lum.lmm_magic != LOV_USER_MAGIC_V3)) | |
0a3bdb00 | 2595 | return -EINVAL; |
d7e09d03 PT |
2596 | |
2597 | /* lov_user_md_vX and lov_mds_md_vX must have the same size */ | |
2598 | LASSERT(sizeof(struct lov_user_md_v1) == sizeof(struct lov_mds_md_v1)); | |
2599 | LASSERT(sizeof(struct lov_user_md_v3) == sizeof(struct lov_mds_md_v3)); | |
2600 | LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lumk->lmm_objects[0])); | |
2601 | ||
2602 | /* we can use lov_mds_md_size() to compute lum_size | |
30aa9c52 OD |
2603 | * because lov_user_md_vX and lov_mds_md_vX have the same size |
2604 | */ | |
d7e09d03 PT |
2605 | if (lum.lmm_stripe_count > 0) { |
2606 | lum_size = lov_mds_md_size(lum.lmm_stripe_count, lum.lmm_magic); | |
7795178d | 2607 | lumk = kzalloc(lum_size, GFP_NOFS); |
d7e09d03 | 2608 | if (!lumk) |
0a3bdb00 | 2609 | return -ENOMEM; |
d7e09d03 PT |
2610 | |
2611 | if (lum.lmm_magic == LOV_USER_MAGIC_V1) | |
2612 | lmm_objects = | |
2613 | &(((struct lov_user_md_v1 *)lumk)->lmm_objects[0]); | |
2614 | else | |
2615 | lmm_objects = &(lumk->lmm_objects[0]); | |
2616 | lmm_objects->l_ost_oi = lsm->lsm_oi; | |
2617 | } else { | |
2618 | lum_size = lov_mds_md_size(0, lum.lmm_magic); | |
2619 | lumk = &lum; | |
2620 | } | |
2621 | ||
2622 | lumk->lmm_oi = lsm->lsm_oi; | |
2623 | lumk->lmm_stripe_count = 1; | |
2624 | ||
2625 | if (copy_to_user(lump, lumk, lum_size)) | |
2626 | rc = -EFAULT; | |
2627 | ||
2628 | if (lumk != &lum) | |
7795178d | 2629 | kfree(lumk); |
d7e09d03 | 2630 | |
0a3bdb00 | 2631 | return rc; |
d7e09d03 PT |
2632 | } |
2633 | ||
d7e09d03 | 2634 | static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len, |
e09bee34 | 2635 | void *karg, void __user *uarg) |
d7e09d03 PT |
2636 | { |
2637 | struct obd_device *obd = exp->exp_obd; | |
2638 | struct obd_ioctl_data *data = karg; | |
2639 | int err = 0; | |
d7e09d03 PT |
2640 | |
2641 | if (!try_module_get(THIS_MODULE)) { | |
19b2056f JN |
2642 | CERROR("%s: cannot get module '%s'\n", obd->obd_name, |
2643 | module_name(THIS_MODULE)); | |
d7e09d03 PT |
2644 | return -EINVAL; |
2645 | } | |
2646 | switch (cmd) { | |
2647 | case OBD_IOC_LOV_GET_CONFIG: { | |
2648 | char *buf; | |
2649 | struct lov_desc *desc; | |
2650 | struct obd_uuid uuid; | |
2651 | ||
2652 | buf = NULL; | |
2653 | len = 0; | |
b7856753 | 2654 | if (obd_ioctl_getdata(&buf, &len, uarg)) { |
26c4ea46 TJ |
2655 | err = -EINVAL; |
2656 | goto out; | |
2657 | } | |
d7e09d03 PT |
2658 | |
2659 | data = (struct obd_ioctl_data *)buf; | |
2660 | ||
2661 | if (sizeof(*desc) > data->ioc_inllen1) { | |
2662 | obd_ioctl_freedata(buf, len); | |
26c4ea46 TJ |
2663 | err = -EINVAL; |
2664 | goto out; | |
d7e09d03 PT |
2665 | } |
2666 | ||
2667 | if (data->ioc_inllen2 < sizeof(uuid)) { | |
2668 | obd_ioctl_freedata(buf, len); | |
26c4ea46 TJ |
2669 | err = -EINVAL; |
2670 | goto out; | |
d7e09d03 PT |
2671 | } |
2672 | ||
2673 | desc = (struct lov_desc *)data->ioc_inlbuf1; | |
2674 | desc->ld_tgt_count = 1; | |
2675 | desc->ld_active_tgt_count = 1; | |
2676 | desc->ld_default_stripe_count = 1; | |
2677 | desc->ld_default_stripe_size = 0; | |
2678 | desc->ld_default_stripe_offset = 0; | |
2679 | desc->ld_pattern = 0; | |
2680 | memcpy(&desc->ld_uuid, &obd->obd_uuid, sizeof(uuid)); | |
2681 | ||
2682 | memcpy(data->ioc_inlbuf2, &obd->obd_uuid, sizeof(uuid)); | |
2683 | ||
b7856753 | 2684 | err = copy_to_user(uarg, buf, len); |
d7e09d03 PT |
2685 | if (err) |
2686 | err = -EFAULT; | |
2687 | obd_ioctl_freedata(buf, len); | |
26c4ea46 | 2688 | goto out; |
d7e09d03 PT |
2689 | } |
2690 | case LL_IOC_LOV_SETSTRIPE: | |
2691 | err = obd_alloc_memmd(exp, karg); | |
2692 | if (err > 0) | |
2693 | err = 0; | |
26c4ea46 | 2694 | goto out; |
d7e09d03 PT |
2695 | case LL_IOC_LOV_GETSTRIPE: |
2696 | err = osc_getstripe(karg, uarg); | |
26c4ea46 | 2697 | goto out; |
d7e09d03 PT |
2698 | case OBD_IOC_CLIENT_RECOVER: |
2699 | err = ptlrpc_recover_import(obd->u.cli.cl_import, | |
2700 | data->ioc_inlbuf1, 0); | |
2701 | if (err > 0) | |
2702 | err = 0; | |
26c4ea46 | 2703 | goto out; |
d7e09d03 PT |
2704 | case IOC_OSC_SET_ACTIVE: |
2705 | err = ptlrpc_set_import_active(obd->u.cli.cl_import, | |
2706 | data->ioc_offset); | |
26c4ea46 | 2707 | goto out; |
d7e09d03 | 2708 | case OBD_IOC_POLL_QUOTACHECK: |
167a47c5 | 2709 | err = osc_quota_poll_check(exp, karg); |
26c4ea46 | 2710 | goto out; |
d7e09d03 PT |
2711 | case OBD_IOC_PING_TARGET: |
2712 | err = ptlrpc_obd_ping(obd); | |
26c4ea46 | 2713 | goto out; |
d7e09d03 PT |
2714 | default: |
2715 | CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n", | |
2716 | cmd, current_comm()); | |
26c4ea46 TJ |
2717 | err = -ENOTTY; |
2718 | goto out; | |
d7e09d03 PT |
2719 | } |
2720 | out: | |
2721 | module_put(THIS_MODULE); | |
2722 | return err; | |
2723 | } | |
2724 | ||
2725 | static int osc_get_info(const struct lu_env *env, struct obd_export *exp, | |
21aef7d9 | 2726 | u32 keylen, void *key, __u32 *vallen, void *val, |
d7e09d03 PT |
2727 | struct lov_stripe_md *lsm) |
2728 | { | |
d7e09d03 | 2729 | if (!vallen || !val) |
0a3bdb00 | 2730 | return -EFAULT; |
d7e09d03 PT |
2731 | |
2732 | if (KEY_IS(KEY_LOCK_TO_STRIPE)) { | |
2733 | __u32 *stripe = val; | |
2734 | *vallen = sizeof(*stripe); | |
2735 | *stripe = 0; | |
0a3bdb00 | 2736 | return 0; |
d7e09d03 PT |
2737 | } else if (KEY_IS(KEY_LAST_ID)) { |
2738 | struct ptlrpc_request *req; | |
29ac6840 CH |
2739 | u64 *reply; |
2740 | char *tmp; | |
2741 | int rc; | |
d7e09d03 PT |
2742 | |
2743 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), | |
2744 | &RQF_OST_GET_INFO_LAST_ID); | |
7f1ae4c0 | 2745 | if (!req) |
0a3bdb00 | 2746 | return -ENOMEM; |
d7e09d03 PT |
2747 | |
2748 | req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY, | |
2749 | RCL_CLIENT, keylen); | |
2750 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO); | |
2751 | if (rc) { | |
2752 | ptlrpc_request_free(req); | |
0a3bdb00 | 2753 | return rc; |
d7e09d03 PT |
2754 | } |
2755 | ||
2756 | tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY); | |
2757 | memcpy(tmp, key, keylen); | |
2758 | ||
2759 | req->rq_no_delay = req->rq_no_resend = 1; | |
2760 | ptlrpc_request_set_replen(req); | |
2761 | rc = ptlrpc_queue_wait(req); | |
2762 | if (rc) | |
26c4ea46 | 2763 | goto out; |
d7e09d03 PT |
2764 | |
2765 | reply = req_capsule_server_get(&req->rq_pill, &RMF_OBD_ID); | |
7f1ae4c0 | 2766 | if (!reply) { |
26c4ea46 TJ |
2767 | rc = -EPROTO; |
2768 | goto out; | |
2769 | } | |
d7e09d03 | 2770 | |
21aef7d9 | 2771 | *((u64 *)val) = *reply; |
c71d2645 | 2772 | out: |
d7e09d03 | 2773 | ptlrpc_req_finished(req); |
0a3bdb00 | 2774 | return rc; |
d7e09d03 | 2775 | } else if (KEY_IS(KEY_FIEMAP)) { |
167a47c5 | 2776 | struct ll_fiemap_info_key *fm_key = key; |
29ac6840 CH |
2777 | struct ldlm_res_id res_id; |
2778 | ldlm_policy_data_t policy; | |
2779 | struct lustre_handle lockh; | |
52ee0d20 | 2780 | enum ldlm_mode mode = 0; |
29ac6840 CH |
2781 | struct ptlrpc_request *req; |
2782 | struct ll_user_fiemap *reply; | |
2783 | char *tmp; | |
2784 | int rc; | |
9d865439 AB |
2785 | |
2786 | if (!(fm_key->fiemap.fm_flags & FIEMAP_FLAG_SYNC)) | |
2787 | goto skip_locking; | |
2788 | ||
2789 | policy.l_extent.start = fm_key->fiemap.fm_start & | |
2790 | CFS_PAGE_MASK; | |
2791 | ||
2792 | if (OBD_OBJECT_EOF - fm_key->fiemap.fm_length <= | |
09cbfeaf | 2793 | fm_key->fiemap.fm_start + PAGE_SIZE - 1) |
9d865439 AB |
2794 | policy.l_extent.end = OBD_OBJECT_EOF; |
2795 | else | |
2796 | policy.l_extent.end = (fm_key->fiemap.fm_start + | |
2797 | fm_key->fiemap.fm_length + | |
09cbfeaf | 2798 | PAGE_SIZE - 1) & CFS_PAGE_MASK; |
9d865439 AB |
2799 | |
2800 | ostid_build_res_name(&fm_key->oa.o_oi, &res_id); | |
2801 | mode = ldlm_lock_match(exp->exp_obd->obd_namespace, | |
2802 | LDLM_FL_BLOCK_GRANTED | | |
2803 | LDLM_FL_LVB_READY, | |
2804 | &res_id, LDLM_EXTENT, &policy, | |
2805 | LCK_PR | LCK_PW, &lockh, 0); | |
2806 | if (mode) { /* lock is cached on client */ | |
2807 | if (mode != LCK_PR) { | |
2808 | ldlm_lock_addref(&lockh, LCK_PR); | |
2809 | ldlm_lock_decref(&lockh, LCK_PW); | |
2810 | } | |
2811 | } else { /* no cached lock, needs acquire lock on server side */ | |
2812 | fm_key->oa.o_valid |= OBD_MD_FLFLAGS; | |
2813 | fm_key->oa.o_flags |= OBD_FL_SRVLOCK; | |
2814 | } | |
d7e09d03 | 2815 | |
9d865439 | 2816 | skip_locking: |
d7e09d03 PT |
2817 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), |
2818 | &RQF_OST_GET_INFO_FIEMAP); | |
7f1ae4c0 | 2819 | if (!req) { |
26c4ea46 TJ |
2820 | rc = -ENOMEM; |
2821 | goto drop_lock; | |
2822 | } | |
d7e09d03 PT |
2823 | |
2824 | req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY, | |
2825 | RCL_CLIENT, keylen); | |
2826 | req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, | |
2827 | RCL_CLIENT, *vallen); | |
2828 | req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, | |
2829 | RCL_SERVER, *vallen); | |
2830 | ||
2831 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO); | |
2832 | if (rc) { | |
2833 | ptlrpc_request_free(req); | |
26c4ea46 | 2834 | goto drop_lock; |
d7e09d03 PT |
2835 | } |
2836 | ||
2837 | tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY); | |
2838 | memcpy(tmp, key, keylen); | |
2839 | tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_VAL); | |
2840 | memcpy(tmp, val, *vallen); | |
2841 | ||
2842 | ptlrpc_request_set_replen(req); | |
2843 | rc = ptlrpc_queue_wait(req); | |
2844 | if (rc) | |
26c4ea46 | 2845 | goto fini_req; |
d7e09d03 PT |
2846 | |
2847 | reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL); | |
7f1ae4c0 | 2848 | if (!reply) { |
26c4ea46 TJ |
2849 | rc = -EPROTO; |
2850 | goto fini_req; | |
2851 | } | |
d7e09d03 PT |
2852 | |
2853 | memcpy(val, reply, *vallen); | |
9d865439 | 2854 | fini_req: |
d7e09d03 | 2855 | ptlrpc_req_finished(req); |
9d865439 AB |
2856 | drop_lock: |
2857 | if (mode) | |
2858 | ldlm_lock_decref(&lockh, LCK_PR); | |
0a3bdb00 | 2859 | return rc; |
d7e09d03 PT |
2860 | } |
2861 | ||
0a3bdb00 | 2862 | return -EINVAL; |
d7e09d03 PT |
2863 | } |
2864 | ||
2865 | static int osc_set_info_async(const struct lu_env *env, struct obd_export *exp, | |
21aef7d9 | 2866 | u32 keylen, void *key, u32 vallen, |
d7e09d03 PT |
2867 | void *val, struct ptlrpc_request_set *set) |
2868 | { | |
2869 | struct ptlrpc_request *req; | |
29ac6840 CH |
2870 | struct obd_device *obd = exp->exp_obd; |
2871 | struct obd_import *imp = class_exp2cliimp(exp); | |
2872 | char *tmp; | |
2873 | int rc; | |
d7e09d03 PT |
2874 | |
2875 | OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_SHUTDOWN, 10); | |
2876 | ||
2877 | if (KEY_IS(KEY_CHECKSUM)) { | |
2878 | if (vallen != sizeof(int)) | |
0a3bdb00 | 2879 | return -EINVAL; |
d7e09d03 | 2880 | exp->exp_obd->u.cli.cl_checksum = (*(int *)val) ? 1 : 0; |
0a3bdb00 | 2881 | return 0; |
d7e09d03 PT |
2882 | } |
2883 | ||
2884 | if (KEY_IS(KEY_SPTLRPC_CONF)) { | |
2885 | sptlrpc_conf_client_adapt(obd); | |
0a3bdb00 | 2886 | return 0; |
d7e09d03 PT |
2887 | } |
2888 | ||
2889 | if (KEY_IS(KEY_FLUSH_CTX)) { | |
2890 | sptlrpc_import_flush_my_ctx(imp); | |
0a3bdb00 | 2891 | return 0; |
d7e09d03 PT |
2892 | } |
2893 | ||
2894 | if (KEY_IS(KEY_CACHE_SET)) { | |
2895 | struct client_obd *cli = &obd->u.cli; | |
2896 | ||
7f1ae4c0 | 2897 | LASSERT(!cli->cl_cache); /* only once */ |
167a47c5 | 2898 | cli->cl_cache = val; |
d7e09d03 PT |
2899 | atomic_inc(&cli->cl_cache->ccc_users); |
2900 | cli->cl_lru_left = &cli->cl_cache->ccc_lru_left; | |
2901 | ||
2902 | /* add this osc into entity list */ | |
2903 | LASSERT(list_empty(&cli->cl_lru_osc)); | |
2904 | spin_lock(&cli->cl_cache->ccc_lru_lock); | |
2905 | list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru); | |
2906 | spin_unlock(&cli->cl_cache->ccc_lru_lock); | |
2907 | ||
0a3bdb00 | 2908 | return 0; |
d7e09d03 PT |
2909 | } |
2910 | ||
2911 | if (KEY_IS(KEY_CACHE_LRU_SHRINK)) { | |
2912 | struct client_obd *cli = &obd->u.cli; | |
2913 | int nr = atomic_read(&cli->cl_lru_in_list) >> 1; | |
2914 | int target = *(int *)val; | |
2915 | ||
2916 | nr = osc_lru_shrink(cli, min(nr, target)); | |
2917 | *(int *)val -= nr; | |
0a3bdb00 | 2918 | return 0; |
d7e09d03 PT |
2919 | } |
2920 | ||
2921 | if (!set && !KEY_IS(KEY_GRANT_SHRINK)) | |
0a3bdb00 | 2922 | return -EINVAL; |
d7e09d03 PT |
2923 | |
2924 | /* We pass all other commands directly to OST. Since nobody calls osc | |
30aa9c52 OD |
2925 | * methods directly and everybody is supposed to go through LOV, we |
2926 | * assume lov checked invalid values for us. | |
2927 | * The only recognised values so far are evict_by_nid and mds_conn. | |
2928 | * Even if something bad goes through, we'd get a -EINVAL from OST | |
2929 | * anyway. | |
2930 | */ | |
d7e09d03 PT |
2931 | |
2932 | req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ? | |
2933 | &RQF_OST_SET_GRANT_INFO : | |
2934 | &RQF_OBD_SET_INFO); | |
7f1ae4c0 | 2935 | if (!req) |
0a3bdb00 | 2936 | return -ENOMEM; |
d7e09d03 PT |
2937 | |
2938 | req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY, | |
2939 | RCL_CLIENT, keylen); | |
2940 | if (!KEY_IS(KEY_GRANT_SHRINK)) | |
2941 | req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL, | |
2942 | RCL_CLIENT, vallen); | |
2943 | rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO); | |
2944 | if (rc) { | |
2945 | ptlrpc_request_free(req); | |
0a3bdb00 | 2946 | return rc; |
d7e09d03 PT |
2947 | } |
2948 | ||
2949 | tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY); | |
2950 | memcpy(tmp, key, keylen); | |
2951 | tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ? | |
2952 | &RMF_OST_BODY : | |
2953 | &RMF_SETINFO_VAL); | |
2954 | memcpy(tmp, val, vallen); | |
2955 | ||
2956 | if (KEY_IS(KEY_GRANT_SHRINK)) { | |
f024bad4 | 2957 | struct osc_brw_async_args *aa; |
d7e09d03 PT |
2958 | struct obdo *oa; |
2959 | ||
2960 | CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args)); | |
2961 | aa = ptlrpc_req_async_args(req); | |
c4418dac | 2962 | oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS); |
d7e09d03 PT |
2963 | if (!oa) { |
2964 | ptlrpc_req_finished(req); | |
0a3bdb00 | 2965 | return -ENOMEM; |
d7e09d03 PT |
2966 | } |
2967 | *oa = ((struct ost_body *)val)->oa; | |
2968 | aa->aa_oa = oa; | |
2969 | req->rq_interpret_reply = osc_shrink_grant_interpret; | |
2970 | } | |
2971 | ||
2972 | ptlrpc_request_set_replen(req); | |
2973 | if (!KEY_IS(KEY_GRANT_SHRINK)) { | |
7f1ae4c0 | 2974 | LASSERT(set); |
d7e09d03 PT |
2975 | ptlrpc_set_add_req(set, req); |
2976 | ptlrpc_check_set(NULL, set); | |
c5c4c6fa OW |
2977 | } else { |
2978 | ptlrpcd_add_req(req); | |
2979 | } | |
d7e09d03 | 2980 | |
0a3bdb00 | 2981 | return 0; |
d7e09d03 PT |
2982 | } |
2983 | ||
d7e09d03 PT |
2984 | static int osc_reconnect(const struct lu_env *env, |
2985 | struct obd_export *exp, struct obd_device *obd, | |
2986 | struct obd_uuid *cluuid, | |
2987 | struct obd_connect_data *data, | |
2988 | void *localdata) | |
2989 | { | |
2990 | struct client_obd *cli = &obd->u.cli; | |
2991 | ||
7f1ae4c0 | 2992 | if (data && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) { |
d7e09d03 PT |
2993 | long lost_grant; |
2994 | ||
2995 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
2996 | data->ocd_grant = (cli->cl_avail_grant + cli->cl_dirty) ?: | |
2997 | 2 * cli_brw_size(obd); | |
2998 | lost_grant = cli->cl_lost_grant; | |
2999 | cli->cl_lost_grant = 0; | |
3000 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
3001 | ||
2d00bd17 JP |
3002 | CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d, lost: %ld.\n", |
3003 | data->ocd_connect_flags, | |
d7e09d03 PT |
3004 | data->ocd_version, data->ocd_grant, lost_grant); |
3005 | } | |
3006 | ||
0a3bdb00 | 3007 | return 0; |
d7e09d03 PT |
3008 | } |
3009 | ||
3010 | static int osc_disconnect(struct obd_export *exp) | |
3011 | { | |
3012 | struct obd_device *obd = class_exp2obd(exp); | |
d7e09d03 PT |
3013 | int rc; |
3014 | ||
d7e09d03 PT |
3015 | rc = client_disconnect_export(exp); |
3016 | /** | |
3017 | * Initially we put del_shrink_grant before disconnect_export, but it | |
3018 | * causes the following problem if setup (connect) and cleanup | |
3019 | * (disconnect) are tangled together. | |
3020 | * connect p1 disconnect p2 | |
3021 | * ptlrpc_connect_import | |
3022 | * ............... class_manual_cleanup | |
3023 | * osc_disconnect | |
3024 | * del_shrink_grant | |
3025 | * ptlrpc_connect_interrupt | |
3026 | * init_grant_shrink | |
3027 | * add this client to shrink list | |
3028 | * cleanup_osc | |
3029 | * Bang! pinger trigger the shrink. | |
3030 | * So the osc should be disconnected from the shrink list, after we | |
3031 | * are sure the import has been destroyed. BUG18662 | |
3032 | */ | |
7f1ae4c0 | 3033 | if (!obd->u.cli.cl_import) |
d7e09d03 PT |
3034 | osc_del_shrink_grant(&obd->u.cli); |
3035 | return rc; | |
3036 | } | |
3037 | ||
3038 | static int osc_import_event(struct obd_device *obd, | |
3039 | struct obd_import *imp, | |
3040 | enum obd_import_event event) | |
3041 | { | |
3042 | struct client_obd *cli; | |
3043 | int rc = 0; | |
3044 | ||
d7e09d03 PT |
3045 | LASSERT(imp->imp_obd == obd); |
3046 | ||
3047 | switch (event) { | |
3048 | case IMP_EVENT_DISCON: { | |
3049 | cli = &obd->u.cli; | |
3050 | client_obd_list_lock(&cli->cl_loi_list_lock); | |
3051 | cli->cl_avail_grant = 0; | |
3052 | cli->cl_lost_grant = 0; | |
3053 | client_obd_list_unlock(&cli->cl_loi_list_lock); | |
3054 | break; | |
3055 | } | |
3056 | case IMP_EVENT_INACTIVE: { | |
3057 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL); | |
3058 | break; | |
3059 | } | |
3060 | case IMP_EVENT_INVALIDATE: { | |
3061 | struct ldlm_namespace *ns = obd->obd_namespace; | |
29ac6840 CH |
3062 | struct lu_env *env; |
3063 | int refcheck; | |
d7e09d03 PT |
3064 | |
3065 | env = cl_env_get(&refcheck); | |
3066 | if (!IS_ERR(env)) { | |
3067 | /* Reset grants */ | |
3068 | cli = &obd->u.cli; | |
3069 | /* all pages go to failing rpcs due to the invalid | |
30aa9c52 OD |
3070 | * import |
3071 | */ | |
c5c4c6fa | 3072 | osc_io_unplug(env, cli, NULL); |
d7e09d03 PT |
3073 | |
3074 | ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY); | |
3075 | cl_env_put(env, &refcheck); | |
3076 | } else | |
3077 | rc = PTR_ERR(env); | |
3078 | break; | |
3079 | } | |
3080 | case IMP_EVENT_ACTIVE: { | |
3081 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL); | |
3082 | break; | |
3083 | } | |
3084 | case IMP_EVENT_OCD: { | |
3085 | struct obd_connect_data *ocd = &imp->imp_connect_data; | |
3086 | ||
3087 | if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT) | |
3088 | osc_init_grant(&obd->u.cli, ocd); | |
3089 | ||
3090 | /* See bug 7198 */ | |
3091 | if (ocd->ocd_connect_flags & OBD_CONNECT_REQPORTAL) | |
b2952d62 | 3092 | imp->imp_client->cli_request_portal = OST_REQUEST_PORTAL; |
d7e09d03 PT |
3093 | |
3094 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL); | |
3095 | break; | |
3096 | } | |
3097 | case IMP_EVENT_DEACTIVATE: { | |
3098 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DEACTIVATE, NULL); | |
3099 | break; | |
3100 | } | |
3101 | case IMP_EVENT_ACTIVATE: { | |
3102 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVATE, NULL); | |
3103 | break; | |
3104 | } | |
3105 | default: | |
3106 | CERROR("Unknown import event %d\n", event); | |
3107 | LBUG(); | |
3108 | } | |
0a3bdb00 | 3109 | return rc; |
d7e09d03 PT |
3110 | } |
3111 | ||
3112 | /** | |
3113 | * Determine whether the lock can be canceled before replaying the lock | |
3114 | * during recovery, see bug16774 for detailed information. | |
3115 | * | |
3116 | * \retval zero the lock can't be canceled | |
3117 | * \retval other ok to cancel | |
3118 | */ | |
3119 | static int osc_cancel_for_recovery(struct ldlm_lock *lock) | |
3120 | { | |
3121 | check_res_locked(lock->l_resource); | |
3122 | ||
3123 | /* | |
3124 | * Cancel all unused extent lock in granted mode LCK_PR or LCK_CR. | |
3125 | * | |
3126 | * XXX as a future improvement, we can also cancel unused write lock | |
3127 | * if it doesn't have dirty data and active mmaps. | |
3128 | */ | |
3129 | if (lock->l_resource->lr_type == LDLM_EXTENT && | |
3130 | (lock->l_granted_mode == LCK_PR || | |
3131 | lock->l_granted_mode == LCK_CR) && | |
3132 | (osc_dlm_lock_pageref(lock) == 0)) | |
0a3bdb00 | 3133 | return 1; |
d7e09d03 | 3134 | |
0a3bdb00 | 3135 | return 0; |
d7e09d03 PT |
3136 | } |
3137 | ||
3138 | static int brw_queue_work(const struct lu_env *env, void *data) | |
3139 | { | |
3140 | struct client_obd *cli = data; | |
3141 | ||
3142 | CDEBUG(D_CACHE, "Run writeback work for client obd %p.\n", cli); | |
3143 | ||
c5c4c6fa | 3144 | osc_io_unplug(env, cli, NULL); |
0a3bdb00 | 3145 | return 0; |
d7e09d03 PT |
3146 | } |
3147 | ||
3148 | int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg) | |
3149 | { | |
ea7893bb | 3150 | struct lprocfs_static_vars lvars = { NULL }; |
29ac6840 CH |
3151 | struct client_obd *cli = &obd->u.cli; |
3152 | void *handler; | |
3153 | int rc; | |
aefd9d71 LX |
3154 | int adding; |
3155 | int added; | |
3156 | int req_count; | |
d7e09d03 PT |
3157 | |
3158 | rc = ptlrpcd_addref(); | |
3159 | if (rc) | |
0a3bdb00 | 3160 | return rc; |
d7e09d03 PT |
3161 | |
3162 | rc = client_obd_setup(obd, lcfg); | |
3163 | if (rc) | |
26c4ea46 | 3164 | goto out_ptlrpcd; |
d7e09d03 PT |
3165 | |
3166 | handler = ptlrpcd_alloc_work(cli->cl_import, brw_queue_work, cli); | |
26c4ea46 TJ |
3167 | if (IS_ERR(handler)) { |
3168 | rc = PTR_ERR(handler); | |
3169 | goto out_client_setup; | |
3170 | } | |
d7e09d03 PT |
3171 | cli->cl_writeback_work = handler; |
3172 | ||
3173 | rc = osc_quota_setup(obd); | |
3174 | if (rc) | |
26c4ea46 | 3175 | goto out_ptlrpcd_work; |
d7e09d03 PT |
3176 | |
3177 | cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL; | |
3178 | lprocfs_osc_init_vars(&lvars); | |
9b801302 | 3179 | if (lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars) == 0) { |
d7e09d03 PT |
3180 | lproc_osc_attach_seqstat(obd); |
3181 | sptlrpc_lprocfs_cliobd_attach(obd); | |
3182 | ptlrpc_lprocfs_register_obd(obd); | |
3183 | } | |
3184 | ||
aefd9d71 LX |
3185 | /* |
3186 | * We try to control the total number of requests with a upper limit | |
3187 | * osc_reqpool_maxreqcount. There might be some race which will cause | |
3188 | * over-limit allocation, but it is fine. | |
3189 | */ | |
3190 | req_count = atomic_read(&osc_pool_req_count); | |
3191 | if (req_count < osc_reqpool_maxreqcount) { | |
3192 | adding = cli->cl_max_rpcs_in_flight + 2; | |
3193 | if (req_count + adding > osc_reqpool_maxreqcount) | |
3194 | adding = osc_reqpool_maxreqcount - req_count; | |
3195 | ||
3196 | added = ptlrpc_add_rqs_to_pool(osc_rq_pool, adding); | |
3197 | atomic_add(added, &osc_pool_req_count); | |
3198 | } | |
d7e09d03 PT |
3199 | |
3200 | INIT_LIST_HEAD(&cli->cl_grant_shrink_list); | |
3201 | ns_register_cancel(obd->obd_namespace, osc_cancel_for_recovery); | |
0a3bdb00 | 3202 | return rc; |
d7e09d03 PT |
3203 | |
3204 | out_ptlrpcd_work: | |
3205 | ptlrpcd_destroy_work(handler); | |
3206 | out_client_setup: | |
3207 | client_obd_cleanup(obd); | |
3208 | out_ptlrpcd: | |
3209 | ptlrpcd_decref(); | |
0a3bdb00 | 3210 | return rc; |
d7e09d03 PT |
3211 | } |
3212 | ||
3213 | static int osc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) | |
3214 | { | |
d7e09d03 PT |
3215 | switch (stage) { |
3216 | case OBD_CLEANUP_EARLY: { | |
3217 | struct obd_import *imp; | |
50ffcb7e | 3218 | |
d7e09d03 PT |
3219 | imp = obd->u.cli.cl_import; |
3220 | CDEBUG(D_HA, "Deactivating import %s\n", obd->obd_name); | |
3221 | /* ptlrpc_abort_inflight to stop an mds_lov_synchronize */ | |
3222 | ptlrpc_deactivate_import(imp); | |
3223 | spin_lock(&imp->imp_lock); | |
3224 | imp->imp_pingable = 0; | |
3225 | spin_unlock(&imp->imp_lock); | |
3226 | break; | |
3227 | } | |
3228 | case OBD_CLEANUP_EXPORTS: { | |
3229 | struct client_obd *cli = &obd->u.cli; | |
3230 | /* LU-464 | |
3231 | * for echo client, export may be on zombie list, wait for | |
3232 | * zombie thread to cull it, because cli.cl_import will be | |
3233 | * cleared in client_disconnect_export(): | |
3234 | * class_export_destroy() -> obd_cleanup() -> | |
3235 | * echo_device_free() -> echo_client_cleanup() -> | |
3236 | * obd_disconnect() -> osc_disconnect() -> | |
3237 | * client_disconnect_export() | |
3238 | */ | |
3239 | obd_zombie_barrier(); | |
3240 | if (cli->cl_writeback_work) { | |
3241 | ptlrpcd_destroy_work(cli->cl_writeback_work); | |
3242 | cli->cl_writeback_work = NULL; | |
3243 | } | |
3244 | obd_cleanup_client_import(obd); | |
3245 | ptlrpc_lprocfs_unregister_obd(obd); | |
3246 | lprocfs_obd_cleanup(obd); | |
d7e09d03 PT |
3247 | break; |
3248 | } | |
3249 | } | |
41f8d410 | 3250 | return 0; |
d7e09d03 PT |
3251 | } |
3252 | ||
f51e5a20 | 3253 | static int osc_cleanup(struct obd_device *obd) |
d7e09d03 PT |
3254 | { |
3255 | struct client_obd *cli = &obd->u.cli; | |
3256 | int rc; | |
3257 | ||
d7e09d03 | 3258 | /* lru cleanup */ |
7f1ae4c0 | 3259 | if (cli->cl_cache) { |
d7e09d03 PT |
3260 | LASSERT(atomic_read(&cli->cl_cache->ccc_users) > 0); |
3261 | spin_lock(&cli->cl_cache->ccc_lru_lock); | |
3262 | list_del_init(&cli->cl_lru_osc); | |
3263 | spin_unlock(&cli->cl_cache->ccc_lru_lock); | |
3264 | cli->cl_lru_left = NULL; | |
3265 | atomic_dec(&cli->cl_cache->ccc_users); | |
3266 | cli->cl_cache = NULL; | |
3267 | } | |
3268 | ||
3269 | /* free memory of osc quota cache */ | |
3270 | osc_quota_cleanup(obd); | |
3271 | ||
3272 | rc = client_obd_cleanup(obd); | |
3273 | ||
3274 | ptlrpcd_decref(); | |
0a3bdb00 | 3275 | return rc; |
d7e09d03 PT |
3276 | } |
3277 | ||
3278 | int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg) | |
3279 | { | |
ea7893bb | 3280 | struct lprocfs_static_vars lvars = { NULL }; |
d7e09d03 PT |
3281 | int rc = 0; |
3282 | ||
3283 | lprocfs_osc_init_vars(&lvars); | |
3284 | ||
3285 | switch (lcfg->lcfg_command) { | |
3286 | default: | |
3287 | rc = class_process_proc_param(PARAM_OSC, lvars.obd_vars, | |
3288 | lcfg, obd); | |
3289 | if (rc > 0) | |
3290 | rc = 0; | |
3291 | break; | |
3292 | } | |
3293 | ||
fbe7c6c7 | 3294 | return rc; |
d7e09d03 PT |
3295 | } |
3296 | ||
21aef7d9 | 3297 | static int osc_process_config(struct obd_device *obd, u32 len, void *buf) |
d7e09d03 PT |
3298 | { |
3299 | return osc_process_config_base(obd, buf); | |
3300 | } | |
3301 | ||
f51e5a20 | 3302 | static struct obd_ops osc_obd_ops = { |
a13b1f32 DC |
3303 | .owner = THIS_MODULE, |
3304 | .setup = osc_setup, | |
3305 | .precleanup = osc_precleanup, | |
3306 | .cleanup = osc_cleanup, | |
3307 | .add_conn = client_import_add_conn, | |
3308 | .del_conn = client_import_del_conn, | |
3309 | .connect = client_connect_import, | |
3310 | .reconnect = osc_reconnect, | |
3311 | .disconnect = osc_disconnect, | |
3312 | .statfs = osc_statfs, | |
3313 | .statfs_async = osc_statfs_async, | |
3314 | .packmd = osc_packmd, | |
3315 | .unpackmd = osc_unpackmd, | |
3316 | .create = osc_create, | |
3317 | .destroy = osc_destroy, | |
3318 | .getattr = osc_getattr, | |
3319 | .getattr_async = osc_getattr_async, | |
3320 | .setattr = osc_setattr, | |
3321 | .setattr_async = osc_setattr_async, | |
3322 | .find_cbdata = osc_find_cbdata, | |
3323 | .iocontrol = osc_iocontrol, | |
3324 | .get_info = osc_get_info, | |
3325 | .set_info_async = osc_set_info_async, | |
3326 | .import_event = osc_import_event, | |
3327 | .process_config = osc_process_config, | |
3328 | .quotactl = osc_quotactl, | |
3329 | .quotacheck = osc_quotacheck, | |
d7e09d03 PT |
3330 | }; |
3331 | ||
3332 | extern struct lu_kmem_descr osc_caches[]; | |
3333 | extern spinlock_t osc_ast_guard; | |
3334 | extern struct lock_class_key osc_ast_guard_class; | |
3335 | ||
b47ea4bb | 3336 | static int __init osc_init(void) |
d7e09d03 | 3337 | { |
ea7893bb | 3338 | struct lprocfs_static_vars lvars = { NULL }; |
aefd9d71 LX |
3339 | unsigned int reqpool_size; |
3340 | unsigned int reqsize; | |
d7e09d03 | 3341 | int rc; |
d7e09d03 PT |
3342 | |
3343 | /* print an address of _any_ initialized kernel symbol from this | |
3344 | * module, to allow debugging with gdb that doesn't support data | |
30aa9c52 OD |
3345 | * symbols from modules. |
3346 | */ | |
d7e09d03 PT |
3347 | CDEBUG(D_INFO, "Lustre OSC module (%p).\n", &osc_caches); |
3348 | ||
3349 | rc = lu_kmem_init(osc_caches); | |
a55e0f44 | 3350 | if (rc) |
0a3bdb00 | 3351 | return rc; |
d7e09d03 PT |
3352 | |
3353 | lprocfs_osc_init_vars(&lvars); | |
3354 | ||
2962b440 | 3355 | rc = class_register_type(&osc_obd_ops, NULL, |
d7e09d03 | 3356 | LUSTRE_OSC_NAME, &osc_device_type); |
aefd9d71 LX |
3357 | if (rc) |
3358 | goto out_kmem; | |
d7e09d03 PT |
3359 | |
3360 | spin_lock_init(&osc_ast_guard); | |
3361 | lockdep_set_class(&osc_ast_guard, &osc_ast_guard_class); | |
3362 | ||
aefd9d71 LX |
3363 | /* This is obviously too much memory, only prevent overflow here */ |
3364 | if (osc_reqpool_mem_max >= 1 << 12 || osc_reqpool_mem_max == 0) { | |
3365 | rc = -EINVAL; | |
3366 | goto out_type; | |
3367 | } | |
3368 | ||
3369 | reqpool_size = osc_reqpool_mem_max << 20; | |
3370 | ||
3371 | reqsize = 1; | |
3372 | while (reqsize < OST_MAXREQSIZE) | |
3373 | reqsize = reqsize << 1; | |
3374 | ||
3375 | /* | |
3376 | * We don't enlarge the request count in OSC pool according to | |
3377 | * cl_max_rpcs_in_flight. The allocation from the pool will only be | |
3378 | * tried after normal allocation failed. So a small OSC pool won't | |
3379 | * cause much performance degression in most of cases. | |
3380 | */ | |
3381 | osc_reqpool_maxreqcount = reqpool_size / reqsize; | |
3382 | ||
3383 | atomic_set(&osc_pool_req_count, 0); | |
3384 | osc_rq_pool = ptlrpc_init_rq_pool(0, OST_MAXREQSIZE, | |
3385 | ptlrpc_add_rqs_to_pool); | |
3386 | ||
3387 | if (osc_rq_pool) | |
3388 | return 0; | |
3389 | ||
3390 | rc = -ENOMEM; | |
3391 | ||
3392 | out_type: | |
3393 | class_unregister_type(LUSTRE_OSC_NAME); | |
3394 | out_kmem: | |
3395 | lu_kmem_fini(osc_caches); | |
0a3bdb00 | 3396 | return rc; |
d7e09d03 PT |
3397 | } |
3398 | ||
3399 | static void /*__exit*/ osc_exit(void) | |
3400 | { | |
3401 | class_unregister_type(LUSTRE_OSC_NAME); | |
3402 | lu_kmem_fini(osc_caches); | |
aefd9d71 | 3403 | ptlrpc_free_rq_pool(osc_rq_pool); |
d7e09d03 PT |
3404 | } |
3405 | ||
a0455471 | 3406 | MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>"); |
d7e09d03 PT |
3407 | MODULE_DESCRIPTION("Lustre Object Storage Client (OSC)"); |
3408 | MODULE_LICENSE("GPL"); | |
6960736c | 3409 | MODULE_VERSION(LUSTRE_VERSION_STRING); |
d7e09d03 | 3410 | |
6960736c GKH |
3411 | module_init(osc_init); |
3412 | module_exit(osc_exit); |