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 | |
6a5b99a4 | 18 | * http://www.gnu.org/licenses/gpl-2.0.html |
d7e09d03 | 19 | * |
d7e09d03 PT |
20 | * GPL HEADER END |
21 | */ | |
22 | /* | |
23 | * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Use is subject to license terms. | |
25 | * | |
1dc563a6 | 26 | * Copyright (c) 2011, 2015, Intel Corporation. |
d7e09d03 PT |
27 | */ |
28 | /* | |
29 | * This file is part of Lustre, http://www.lustre.org/ | |
30 | * Lustre is a trademark of Sun Microsystems, Inc. | |
31 | */ | |
32 | ||
33 | #define DEBUG_SUBSYSTEM S_MDC | |
34 | ||
35 | # include <linux/module.h> | |
36 | # include <linux/pagemap.h> | |
37 | # include <linux/miscdevice.h> | |
38 | # include <linux/init.h> | |
39 | # include <linux/utsname.h> | |
40 | ||
05932307 | 41 | #include "../include/lustre_acl.h" |
8877d3bf | 42 | #include "../include/lustre/lustre_ioctl.h" |
05932307 | 43 | #include "../include/obd_class.h" |
8e9dfe8a | 44 | #include "../include/lustre_lmv.h" |
05932307 GKH |
45 | #include "../include/lustre_fid.h" |
46 | #include "../include/lprocfs_status.h" | |
47 | #include "../include/lustre_param.h" | |
48 | #include "../include/lustre_log.h" | |
e2780478 | 49 | #include "../include/lustre_kernelcomm.h" |
d7e09d03 PT |
50 | |
51 | #include "mdc_internal.h" | |
52 | ||
53 | #define REQUEST_MINOR 244 | |
54 | ||
d7e09d03 PT |
55 | static int mdc_cleanup(struct obd_device *obd); |
56 | ||
d7e09d03 PT |
57 | static inline int mdc_queue_wait(struct ptlrpc_request *req) |
58 | { | |
59 | struct client_obd *cli = &req->rq_import->imp_obd->u.cli; | |
60 | int rc; | |
61 | ||
1d5d5ec1 | 62 | /* obd_get_request_slot() ensures that this client has no more |
d7e09d03 | 63 | * than cl_max_rpcs_in_flight RPCs simultaneously inf light |
1df232ee OD |
64 | * against an MDT. |
65 | */ | |
1d5d5ec1 | 66 | rc = obd_get_request_slot(cli); |
d7e09d03 PT |
67 | if (rc != 0) |
68 | return rc; | |
69 | ||
70 | rc = ptlrpc_queue_wait(req); | |
1d5d5ec1 | 71 | obd_put_request_slot(cli); |
d7e09d03 PT |
72 | |
73 | return rc; | |
74 | } | |
75 | ||
ef2e0f55 | 76 | static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid) |
d7e09d03 PT |
77 | { |
78 | struct ptlrpc_request *req; | |
79 | struct mdt_body *body; | |
80 | int rc; | |
d7e09d03 | 81 | |
ef2e0f55 OD |
82 | req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), |
83 | &RQF_MDS_GETSTATUS, | |
d7e09d03 | 84 | LUSTRE_MDS_VERSION, MDS_GETSTATUS); |
34e3ff96 | 85 | if (!req) |
0a3bdb00 | 86 | return -ENOMEM; |
d7e09d03 | 87 | |
ef2e0f55 OD |
88 | mdc_pack_body(req, NULL, 0, 0, -1, 0); |
89 | req->rq_send_state = LUSTRE_IMP_FULL; | |
d7e09d03 PT |
90 | |
91 | ptlrpc_request_set_replen(req); | |
92 | ||
93 | rc = ptlrpc_queue_wait(req); | |
94 | if (rc) | |
d5fdc207 | 95 | goto out; |
d7e09d03 PT |
96 | |
97 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
34e3ff96 | 98 | if (!body) { |
d5fdc207 JL |
99 | rc = -EPROTO; |
100 | goto out; | |
101 | } | |
d7e09d03 | 102 | |
2e1b5b8b | 103 | *rootfid = body->mbo_fid1; |
d7e09d03 | 104 | CDEBUG(D_NET, |
b0f5aad5 | 105 | "root fid="DFID", last_committed=%llu\n", |
d7e09d03 PT |
106 | PFID(rootfid), |
107 | lustre_msg_get_last_committed(req->rq_repmsg)); | |
d7e09d03 PT |
108 | out: |
109 | ptlrpc_req_finished(req); | |
110 | return rc; | |
111 | } | |
112 | ||
d7e09d03 PT |
113 | /* |
114 | * This function now is known to always saying that it will receive 4 buffers | |
115 | * from server. Even for cases when acl_size and md_size is zero, RPC header | |
116 | * will contain 4 fields and RPC itself will contain zero size fields. This is | |
117 | * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed | |
118 | * and thus zero, it shrinks it, making zero size. The same story about | |
119 | * md_size. And this is course of problem when client waits for smaller number | |
120 | * of fields. This issue will be fixed later when client gets aware of RPC | |
121 | * layouts. --umka | |
122 | */ | |
123 | static int mdc_getattr_common(struct obd_export *exp, | |
124 | struct ptlrpc_request *req) | |
125 | { | |
126 | struct req_capsule *pill = &req->rq_pill; | |
127 | struct mdt_body *body; | |
128 | void *eadata; | |
129 | int rc; | |
d7e09d03 PT |
130 | |
131 | /* Request message already built. */ | |
132 | rc = ptlrpc_queue_wait(req); | |
133 | if (rc != 0) | |
0a3bdb00 | 134 | return rc; |
d7e09d03 PT |
135 | |
136 | /* sanity check for the reply */ | |
137 | body = req_capsule_server_get(pill, &RMF_MDT_BODY); | |
34e3ff96 | 138 | if (!body) |
0a3bdb00 | 139 | return -EPROTO; |
d7e09d03 | 140 | |
2e1b5b8b | 141 | CDEBUG(D_NET, "mode: %o\n", body->mbo_mode); |
d7e09d03 | 142 | |
483eec0d | 143 | mdc_update_max_ea_from_body(exp, body); |
2e1b5b8b | 144 | if (body->mbo_eadatasize != 0) { |
d7e09d03 | 145 | eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD, |
2e1b5b8b | 146 | body->mbo_eadatasize); |
34e3ff96 | 147 | if (!eadata) |
0a3bdb00 | 148 | return -EPROTO; |
d7e09d03 PT |
149 | } |
150 | ||
0a3bdb00 | 151 | return 0; |
d7e09d03 PT |
152 | } |
153 | ||
60ebee3b | 154 | static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data, |
22e0bc6a | 155 | struct ptlrpc_request **request) |
d7e09d03 PT |
156 | { |
157 | struct ptlrpc_request *req; | |
158 | int rc; | |
d7e09d03 PT |
159 | |
160 | /* Single MDS without an LMV case */ | |
161 | if (op_data->op_flags & MF_GET_MDT_IDX) { | |
162 | op_data->op_mds = 0; | |
0a3bdb00 | 163 | return 0; |
d7e09d03 PT |
164 | } |
165 | *request = NULL; | |
166 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR); | |
34e3ff96 | 167 | if (!req) |
0a3bdb00 | 168 | return -ENOMEM; |
d7e09d03 | 169 | |
d7e09d03 PT |
170 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR); |
171 | if (rc) { | |
172 | ptlrpc_request_free(req); | |
0a3bdb00 | 173 | return rc; |
d7e09d03 PT |
174 | } |
175 | ||
ef2e0f55 OD |
176 | mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid, |
177 | op_data->op_mode, -1, 0); | |
d7e09d03 PT |
178 | |
179 | req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, | |
180 | op_data->op_mode); | |
d7e09d03 PT |
181 | ptlrpc_request_set_replen(req); |
182 | ||
183 | rc = mdc_getattr_common(exp, req); | |
184 | if (rc) | |
185 | ptlrpc_req_finished(req); | |
186 | else | |
187 | *request = req; | |
0a3bdb00 | 188 | return rc; |
d7e09d03 PT |
189 | } |
190 | ||
60ebee3b | 191 | static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data, |
22e0bc6a | 192 | struct ptlrpc_request **request) |
d7e09d03 PT |
193 | { |
194 | struct ptlrpc_request *req; | |
195 | int rc; | |
d7e09d03 PT |
196 | |
197 | *request = NULL; | |
198 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), | |
199 | &RQF_MDS_GETATTR_NAME); | |
34e3ff96 | 200 | if (!req) |
0a3bdb00 | 201 | return -ENOMEM; |
d7e09d03 | 202 | |
d7e09d03 PT |
203 | req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, |
204 | op_data->op_namelen + 1); | |
205 | ||
206 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME); | |
207 | if (rc) { | |
208 | ptlrpc_request_free(req); | |
0a3bdb00 | 209 | return rc; |
d7e09d03 PT |
210 | } |
211 | ||
ef2e0f55 OD |
212 | mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid, |
213 | op_data->op_mode, op_data->op_suppgids[0], 0); | |
d7e09d03 PT |
214 | |
215 | if (op_data->op_name) { | |
216 | char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME); | |
7436d070 | 217 | |
d7e09d03 PT |
218 | LASSERT(strnlen(op_data->op_name, op_data->op_namelen) == |
219 | op_data->op_namelen); | |
220 | memcpy(name, op_data->op_name, op_data->op_namelen); | |
221 | } | |
222 | ||
223 | req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, | |
224 | op_data->op_mode); | |
225 | ptlrpc_request_set_replen(req); | |
226 | ||
227 | rc = mdc_getattr_common(exp, req); | |
228 | if (rc) | |
229 | ptlrpc_req_finished(req); | |
230 | else | |
231 | *request = req; | |
0a3bdb00 | 232 | return rc; |
d7e09d03 PT |
233 | } |
234 | ||
301af906 SM |
235 | static int mdc_xattr_common(struct obd_export *exp, |
236 | const struct req_format *fmt, | |
d7e09d03 | 237 | const struct lu_fid *fid, |
ef2e0f55 | 238 | int opcode, u64 valid, |
d7e09d03 PT |
239 | const char *xattr_name, const char *input, |
240 | int input_size, int output_size, int flags, | |
241 | __u32 suppgid, struct ptlrpc_request **request) | |
242 | { | |
243 | struct ptlrpc_request *req; | |
244 | int xattr_namelen = 0; | |
245 | char *tmp; | |
246 | int rc; | |
d7e09d03 PT |
247 | |
248 | *request = NULL; | |
249 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt); | |
34e3ff96 | 250 | if (!req) |
0a3bdb00 | 251 | return -ENOMEM; |
d7e09d03 | 252 | |
d7e09d03 PT |
253 | if (xattr_name) { |
254 | xattr_namelen = strlen(xattr_name) + 1; | |
255 | req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, | |
256 | xattr_namelen); | |
257 | } | |
258 | if (input_size) { | |
259 | LASSERT(input); | |
260 | req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, | |
261 | input_size); | |
262 | } | |
263 | ||
e93a3082 AP |
264 | /* Flush local XATTR locks to get rid of a possible cancel RPC */ |
265 | if (opcode == MDS_REINT && fid_is_sane(fid) && | |
266 | exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) { | |
267 | LIST_HEAD(cancels); | |
268 | int count; | |
269 | ||
270 | /* Without that packing would fail */ | |
271 | if (input_size == 0) | |
272 | req_capsule_set_size(&req->rq_pill, &RMF_EADATA, | |
273 | RCL_CLIENT, 0); | |
274 | ||
275 | count = mdc_resource_get_unused(exp, fid, | |
276 | &cancels, LCK_EX, | |
277 | MDS_INODELOCK_XATTR); | |
278 | ||
279 | rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count); | |
280 | if (rc) { | |
281 | ptlrpc_request_free(req); | |
282 | return rc; | |
283 | } | |
284 | } else { | |
285 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode); | |
286 | if (rc) { | |
287 | ptlrpc_request_free(req); | |
288 | return rc; | |
289 | } | |
d7e09d03 PT |
290 | } |
291 | ||
292 | if (opcode == MDS_REINT) { | |
293 | struct mdt_rec_setxattr *rec; | |
294 | ||
295 | CLASSERT(sizeof(struct mdt_rec_setxattr) == | |
296 | sizeof(struct mdt_rec_reint)); | |
297 | rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT); | |
298 | rec->sx_opcode = REINT_SETXATTR; | |
4b1a25f0 PT |
299 | rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid()); |
300 | rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid()); | |
d7e09d03 PT |
301 | rec->sx_cap = cfs_curproc_cap_pack(); |
302 | rec->sx_suppgid1 = suppgid; | |
303 | rec->sx_suppgid2 = -1; | |
304 | rec->sx_fid = *fid; | |
305 | rec->sx_valid = valid | OBD_MD_FLCTIME; | |
14e3f92a | 306 | rec->sx_time = ktime_get_real_seconds(); |
d7e09d03 PT |
307 | rec->sx_size = output_size; |
308 | rec->sx_flags = flags; | |
309 | ||
d7e09d03 | 310 | } else { |
ef2e0f55 | 311 | mdc_pack_body(req, fid, valid, output_size, suppgid, flags); |
d7e09d03 PT |
312 | } |
313 | ||
314 | if (xattr_name) { | |
315 | tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME); | |
316 | memcpy(tmp, xattr_name, xattr_namelen); | |
317 | } | |
318 | if (input_size) { | |
319 | tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA); | |
320 | memcpy(tmp, input, input_size); | |
321 | } | |
322 | ||
323 | if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER)) | |
324 | req_capsule_set_size(&req->rq_pill, &RMF_EADATA, | |
325 | RCL_SERVER, output_size); | |
326 | ptlrpc_request_set_replen(req); | |
327 | ||
328 | /* make rpc */ | |
329 | if (opcode == MDS_REINT) | |
330 | mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL); | |
331 | ||
332 | rc = ptlrpc_queue_wait(req); | |
333 | ||
334 | if (opcode == MDS_REINT) | |
335 | mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL); | |
336 | ||
337 | if (rc) | |
338 | ptlrpc_req_finished(req); | |
339 | else | |
340 | *request = req; | |
0a3bdb00 | 341 | return rc; |
d7e09d03 PT |
342 | } |
343 | ||
60ebee3b | 344 | static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid, |
ef2e0f55 OD |
345 | u64 valid, const char *xattr_name, |
346 | const char *input, int input_size, int output_size, | |
347 | int flags, __u32 suppgid, | |
348 | struct ptlrpc_request **request) | |
d7e09d03 PT |
349 | { |
350 | return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR, | |
ef2e0f55 | 351 | fid, MDS_REINT, valid, xattr_name, |
d7e09d03 PT |
352 | input, input_size, output_size, flags, |
353 | suppgid, request); | |
354 | } | |
355 | ||
60ebee3b | 356 | static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid, |
ef2e0f55 OD |
357 | u64 valid, const char *xattr_name, |
358 | const char *input, int input_size, int output_size, | |
359 | int flags, struct ptlrpc_request **request) | |
d7e09d03 PT |
360 | { |
361 | return mdc_xattr_common(exp, &RQF_MDS_GETXATTR, | |
ef2e0f55 | 362 | fid, MDS_GETXATTR, valid, xattr_name, |
d7e09d03 PT |
363 | input, input_size, output_size, flags, |
364 | -1, request); | |
365 | } | |
366 | ||
367 | #ifdef CONFIG_FS_POSIX_ACL | |
368 | static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md) | |
369 | { | |
370 | struct req_capsule *pill = &req->rq_pill; | |
371 | struct mdt_body *body = md->body; | |
372 | struct posix_acl *acl; | |
373 | void *buf; | |
374 | int rc; | |
d7e09d03 | 375 | |
2e1b5b8b | 376 | if (!body->mbo_aclsize) |
0a3bdb00 | 377 | return 0; |
d7e09d03 | 378 | |
2e1b5b8b | 379 | buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize); |
d7e09d03 PT |
380 | |
381 | if (!buf) | |
0a3bdb00 | 382 | return -EPROTO; |
d7e09d03 | 383 | |
2e1b5b8b | 384 | acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize); |
34e3ff96 | 385 | if (!acl) |
f4289400 CM |
386 | return 0; |
387 | ||
d7e09d03 PT |
388 | if (IS_ERR(acl)) { |
389 | rc = PTR_ERR(acl); | |
390 | CERROR("convert xattr to acl: %d\n", rc); | |
0a3bdb00 | 391 | return rc; |
d7e09d03 PT |
392 | } |
393 | ||
0d4d717f | 394 | rc = posix_acl_valid(&init_user_ns, acl); |
d7e09d03 PT |
395 | if (rc) { |
396 | CERROR("validate acl: %d\n", rc); | |
397 | posix_acl_release(acl); | |
0a3bdb00 | 398 | return rc; |
d7e09d03 PT |
399 | } |
400 | ||
401 | md->posix_acl = acl; | |
0a3bdb00 | 402 | return 0; |
d7e09d03 PT |
403 | } |
404 | #else | |
405 | #define mdc_unpack_acl(req, md) 0 | |
406 | #endif | |
407 | ||
358855b2 SB |
408 | static int mdc_get_lustre_md(struct obd_export *exp, |
409 | struct ptlrpc_request *req, | |
410 | struct obd_export *dt_exp, | |
411 | struct obd_export *md_exp, | |
412 | struct lustre_md *md) | |
d7e09d03 PT |
413 | { |
414 | struct req_capsule *pill = &req->rq_pill; | |
415 | int rc; | |
d7e09d03 PT |
416 | |
417 | LASSERT(md); | |
418 | memset(md, 0, sizeof(*md)); | |
419 | ||
420 | md->body = req_capsule_server_get(pill, &RMF_MDT_BODY); | |
d7e09d03 | 421 | |
2e1b5b8b | 422 | if (md->body->mbo_valid & OBD_MD_FLEASIZE) { |
d7e09d03 PT |
423 | int lmmsize; |
424 | struct lov_mds_md *lmm; | |
425 | ||
2e1b5b8b | 426 | if (!S_ISREG(md->body->mbo_mode)) { |
ee990b33 SM |
427 | CDEBUG(D_INFO, |
428 | "OBD_MD_FLEASIZE set, should be a regular file, but is not\n"); | |
d5fdc207 JL |
429 | rc = -EPROTO; |
430 | goto out; | |
d7e09d03 PT |
431 | } |
432 | ||
2e1b5b8b | 433 | if (md->body->mbo_eadatasize == 0) { |
ee990b33 SM |
434 | CDEBUG(D_INFO, |
435 | "OBD_MD_FLEASIZE set, but eadatasize 0\n"); | |
d5fdc207 JL |
436 | rc = -EPROTO; |
437 | goto out; | |
d7e09d03 | 438 | } |
2e1b5b8b | 439 | lmmsize = md->body->mbo_eadatasize; |
d7e09d03 | 440 | lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize); |
d5fdc207 JL |
441 | if (!lmm) { |
442 | rc = -EPROTO; | |
443 | goto out; | |
444 | } | |
d7e09d03 PT |
445 | |
446 | rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize); | |
447 | if (rc < 0) | |
d5fdc207 | 448 | goto out; |
d7e09d03 PT |
449 | |
450 | if (rc < sizeof(*md->lsm)) { | |
ee990b33 SM |
451 | CDEBUG(D_INFO, |
452 | "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n", | |
d7e09d03 | 453 | rc, (int)sizeof(*md->lsm)); |
d5fdc207 JL |
454 | rc = -EPROTO; |
455 | goto out; | |
d7e09d03 PT |
456 | } |
457 | ||
2e1b5b8b | 458 | } else if (md->body->mbo_valid & OBD_MD_FLDIREA) { |
d7e09d03 PT |
459 | int lmvsize; |
460 | struct lov_mds_md *lmv; | |
461 | ||
2e1b5b8b | 462 | if (!S_ISDIR(md->body->mbo_mode)) { |
ee990b33 SM |
463 | CDEBUG(D_INFO, |
464 | "OBD_MD_FLDIREA set, should be a directory, but is not\n"); | |
d5fdc207 JL |
465 | rc = -EPROTO; |
466 | goto out; | |
d7e09d03 PT |
467 | } |
468 | ||
2e1b5b8b | 469 | if (md->body->mbo_eadatasize == 0) { |
ee990b33 SM |
470 | CDEBUG(D_INFO, |
471 | "OBD_MD_FLDIREA is set, but eadatasize 0\n"); | |
0a3bdb00 | 472 | return -EPROTO; |
d7e09d03 | 473 | } |
2e1b5b8b JH |
474 | if (md->body->mbo_valid & OBD_MD_MEA) { |
475 | lmvsize = md->body->mbo_eadatasize; | |
d7e09d03 PT |
476 | lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD, |
477 | lmvsize); | |
d5fdc207 JL |
478 | if (!lmv) { |
479 | rc = -EPROTO; | |
480 | goto out; | |
481 | } | |
d7e09d03 | 482 | |
f6718d1d | 483 | rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv, |
d7e09d03 PT |
484 | lmvsize); |
485 | if (rc < 0) | |
d5fdc207 | 486 | goto out; |
d7e09d03 | 487 | |
f6718d1d | 488 | if (rc < sizeof(*md->lmv)) { |
ee990b33 | 489 | CDEBUG(D_INFO, |
f6718d1d | 490 | "size too small: rc < sizeof(*md->lmv) (%d < %d)\n", |
491 | rc, (int)sizeof(*md->lmv)); | |
d5fdc207 JL |
492 | rc = -EPROTO; |
493 | goto out; | |
d7e09d03 PT |
494 | } |
495 | } | |
496 | } | |
497 | rc = 0; | |
498 | ||
2e1b5b8b | 499 | if (md->body->mbo_valid & OBD_MD_FLACL) { |
d7e09d03 PT |
500 | /* for ACL, it's possible that FLACL is set but aclsize is zero. |
501 | * only when aclsize != 0 there's an actual segment for ACL | |
502 | * in reply buffer. | |
503 | */ | |
2e1b5b8b | 504 | if (md->body->mbo_aclsize) { |
d7e09d03 PT |
505 | rc = mdc_unpack_acl(req, md); |
506 | if (rc) | |
d5fdc207 | 507 | goto out; |
d7e09d03 PT |
508 | #ifdef CONFIG_FS_POSIX_ACL |
509 | } else { | |
510 | md->posix_acl = NULL; | |
511 | #endif | |
512 | } | |
513 | } | |
d7e09d03 | 514 | |
d7e09d03 PT |
515 | out: |
516 | if (rc) { | |
d7e09d03 PT |
517 | #ifdef CONFIG_FS_POSIX_ACL |
518 | posix_acl_release(md->posix_acl); | |
519 | #endif | |
520 | if (md->lsm) | |
521 | obd_free_memmd(dt_exp, &md->lsm); | |
522 | } | |
523 | return rc; | |
524 | } | |
525 | ||
358855b2 | 526 | static int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md) |
d7e09d03 | 527 | { |
0a3bdb00 | 528 | return 0; |
d7e09d03 PT |
529 | } |
530 | ||
531 | /** | |
532 | * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING | |
533 | * RPC chains. | |
534 | */ | |
535 | void mdc_replay_open(struct ptlrpc_request *req) | |
536 | { | |
537 | struct md_open_data *mod = req->rq_cb_data; | |
538 | struct ptlrpc_request *close_req; | |
539 | struct obd_client_handle *och; | |
540 | struct lustre_handle old; | |
541 | struct mdt_body *body; | |
d7e09d03 | 542 | |
34e3ff96 | 543 | if (!mod) { |
d7e09d03 PT |
544 | DEBUG_REQ(D_ERROR, req, |
545 | "Can't properly replay without open data."); | |
d7e09d03 PT |
546 | return; |
547 | } | |
548 | ||
549 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
d7e09d03 PT |
550 | |
551 | och = mod->mod_och; | |
34e3ff96 | 552 | if (och) { |
d7e09d03 PT |
553 | struct lustre_handle *file_fh; |
554 | ||
555 | LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC); | |
556 | ||
557 | file_fh = &och->och_fh; | |
55f5a824 | 558 | CDEBUG(D_HA, "updating handle from %#llx to %#llx\n", |
2e1b5b8b | 559 | file_fh->cookie, body->mbo_handle.cookie); |
d7e09d03 | 560 | old = *file_fh; |
2e1b5b8b | 561 | *file_fh = body->mbo_handle; |
d7e09d03 PT |
562 | } |
563 | close_req = mod->mod_close_req; | |
34e3ff96 | 564 | if (close_req) { |
d7e09d03 PT |
565 | __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg); |
566 | struct mdt_ioepoch *epoch; | |
567 | ||
568 | LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING); | |
569 | epoch = req_capsule_client_get(&close_req->rq_pill, | |
570 | &RMF_MDT_EPOCH); | |
571 | LASSERT(epoch); | |
572 | ||
34e3ff96 | 573 | if (och) |
d7e09d03 PT |
574 | LASSERT(!memcmp(&old, &epoch->handle, sizeof(old))); |
575 | DEBUG_REQ(D_HA, close_req, "updating close body with new fh"); | |
2e1b5b8b | 576 | epoch->handle = body->mbo_handle; |
d7e09d03 | 577 | } |
d7e09d03 PT |
578 | } |
579 | ||
580 | void mdc_commit_open(struct ptlrpc_request *req) | |
581 | { | |
582 | struct md_open_data *mod = req->rq_cb_data; | |
7436d070 | 583 | |
34e3ff96 | 584 | if (!mod) |
d7e09d03 PT |
585 | return; |
586 | ||
587 | /** | |
588 | * No need to touch md_open_data::mod_och, it holds a reference on | |
589 | * \var mod and will zero references to each other, \var mod will be | |
590 | * freed after that when md_open_data::mod_och will put the reference. | |
591 | */ | |
592 | ||
593 | /** | |
594 | * Do not let open request to disappear as it still may be needed | |
595 | * for close rpc to happen (it may happen on evict only, otherwise | |
596 | * ptlrpc_request::rq_replay does not let mdc_commit_open() to be | |
597 | * called), just mark this rpc as committed to distinguish these 2 | |
598 | * cases, see mdc_close() for details. The open request reference will | |
599 | * be put along with freeing \var mod. | |
600 | */ | |
601 | ptlrpc_request_addref(req); | |
602 | spin_lock(&req->rq_lock); | |
603 | req->rq_committed = 1; | |
604 | spin_unlock(&req->rq_lock); | |
605 | req->rq_cb_data = NULL; | |
606 | obd_mod_put(mod); | |
607 | } | |
608 | ||
609 | int mdc_set_open_replay_data(struct obd_export *exp, | |
610 | struct obd_client_handle *och, | |
63d42578 | 611 | struct lookup_intent *it) |
d7e09d03 PT |
612 | { |
613 | struct md_open_data *mod; | |
614 | struct mdt_rec_create *rec; | |
615 | struct mdt_body *body; | |
8bf86fd9 | 616 | struct ptlrpc_request *open_req = it->it_request; |
d7e09d03 | 617 | struct obd_import *imp = open_req->rq_import; |
d7e09d03 PT |
618 | |
619 | if (!open_req->rq_replay) | |
0a3bdb00 | 620 | return 0; |
d7e09d03 PT |
621 | |
622 | rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT); | |
623 | body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY); | |
34e3ff96 | 624 | LASSERT(rec); |
d7e09d03 PT |
625 | /* Incoming message in my byte order (it's been swabbed). */ |
626 | /* Outgoing messages always in my byte order. */ | |
34e3ff96 | 627 | LASSERT(body); |
d7e09d03 PT |
628 | |
629 | /* Only if the import is replayable, we set replay_open data */ | |
630 | if (och && imp->imp_replayable) { | |
631 | mod = obd_mod_alloc(); | |
34e3ff96 | 632 | if (!mod) { |
d7e09d03 PT |
633 | DEBUG_REQ(D_ERROR, open_req, |
634 | "Can't allocate md_open_data"); | |
0a3bdb00 | 635 | return 0; |
d7e09d03 PT |
636 | } |
637 | ||
638 | /** | |
639 | * Take a reference on \var mod, to be freed on mdc_close(). | |
640 | * It protects \var mod from being freed on eviction (commit | |
641 | * callback is called despite rq_replay flag). | |
642 | * Another reference for \var och. | |
643 | */ | |
644 | obd_mod_get(mod); | |
645 | obd_mod_get(mod); | |
646 | ||
647 | spin_lock(&open_req->rq_lock); | |
648 | och->och_mod = mod; | |
649 | mod->mod_och = och; | |
63d42578 HZ |
650 | mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) || |
651 | it_disposition(it, DISP_OPEN_STRIPE); | |
d7e09d03 PT |
652 | mod->mod_open_req = open_req; |
653 | open_req->rq_cb_data = mod; | |
654 | open_req->rq_commit_cb = mdc_commit_open; | |
655 | spin_unlock(&open_req->rq_lock); | |
656 | } | |
657 | ||
2e1b5b8b JH |
658 | rec->cr_fid2 = body->mbo_fid1; |
659 | rec->cr_ioepoch = body->mbo_ioepoch; | |
660 | rec->cr_old_handle.cookie = body->mbo_handle.cookie; | |
d7e09d03 | 661 | open_req->rq_replay_cb = mdc_replay_open; |
2e1b5b8b | 662 | if (!fid_is_sane(&body->mbo_fid1)) { |
ee990b33 SM |
663 | DEBUG_REQ(D_ERROR, open_req, |
664 | "Saving replay request with insane fid"); | |
d7e09d03 PT |
665 | LBUG(); |
666 | } | |
667 | ||
668 | DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data"); | |
0a3bdb00 | 669 | return 0; |
d7e09d03 PT |
670 | } |
671 | ||
63d42578 HZ |
672 | static void mdc_free_open(struct md_open_data *mod) |
673 | { | |
674 | int committed = 0; | |
675 | ||
676 | if (mod->mod_is_create == 0 && | |
677 | imp_connect_disp_stripe(mod->mod_open_req->rq_import)) | |
678 | committed = 1; | |
679 | ||
37199135 AB |
680 | /* |
681 | * No reason to asssert here if the open request has | |
682 | * rq_replay == 1. It means that mdc_close failed, and | |
683 | * close request wasn`t sent. It is not fatal to client. | |
684 | * The worst thing is eviction if the client gets open lock | |
685 | */ | |
686 | DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, | |
687 | "free open request rq_replay = %d\n", | |
688 | mod->mod_open_req->rq_replay); | |
63d42578 HZ |
689 | |
690 | ptlrpc_request_committed(mod->mod_open_req, committed); | |
691 | if (mod->mod_close_req) | |
692 | ptlrpc_request_committed(mod->mod_close_req, committed); | |
693 | } | |
694 | ||
358855b2 SB |
695 | static int mdc_clear_open_replay_data(struct obd_export *exp, |
696 | struct obd_client_handle *och) | |
d7e09d03 PT |
697 | { |
698 | struct md_open_data *mod = och->och_mod; | |
d7e09d03 PT |
699 | |
700 | /** | |
701 | * It is possible to not have \var mod in a case of eviction between | |
702 | * lookup and ll_file_open(). | |
703 | **/ | |
34e3ff96 | 704 | if (!mod) |
0a3bdb00 | 705 | return 0; |
d7e09d03 PT |
706 | |
707 | LASSERT(mod != LP_POISON); | |
34e3ff96 | 708 | LASSERT(mod->mod_open_req); |
63d42578 | 709 | mdc_free_open(mod); |
d7e09d03 PT |
710 | |
711 | mod->mod_och = NULL; | |
712 | och->och_mod = NULL; | |
713 | obd_mod_put(mod); | |
714 | ||
0a3bdb00 | 715 | return 0; |
d7e09d03 PT |
716 | } |
717 | ||
718 | /* Prepares the request for the replay by the given reply */ | |
719 | static void mdc_close_handle_reply(struct ptlrpc_request *req, | |
720 | struct md_op_data *op_data, int rc) { | |
721 | struct mdt_body *repbody; | |
722 | struct mdt_ioepoch *epoch; | |
723 | ||
724 | if (req && rc == -EAGAIN) { | |
725 | repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
726 | epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH); | |
727 | ||
728 | epoch->flags |= MF_SOM_AU; | |
2e1b5b8b | 729 | if (repbody->mbo_valid & OBD_MD_FLGETATTRLOCK) |
d7e09d03 PT |
730 | op_data->op_flags |= MF_GETATTR_LOCK; |
731 | } | |
732 | } | |
733 | ||
f6219c16 LC |
734 | static int mdc_close(struct obd_export *exp, struct md_op_data *op_data, |
735 | struct md_open_data *mod, struct ptlrpc_request **request) | |
d7e09d03 PT |
736 | { |
737 | struct obd_device *obd = class_exp2obd(exp); | |
738 | struct ptlrpc_request *req; | |
48d23e61 JX |
739 | struct req_format *req_fmt; |
740 | int rc; | |
741 | int saved_rc = 0; | |
742 | ||
48d23e61 JX |
743 | req_fmt = &RQF_MDS_CLOSE; |
744 | if (op_data->op_bias & MDS_HSM_RELEASE) { | |
745 | req_fmt = &RQF_MDS_RELEASE_CLOSE; | |
746 | ||
747 | /* allocate a FID for volatile file */ | |
8f18c8a4 | 748 | rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data); |
48d23e61 JX |
749 | if (rc < 0) { |
750 | CERROR("%s: "DFID" failed to allocate FID: %d\n", | |
751 | obd->obd_name, PFID(&op_data->op_fid1), rc); | |
752 | /* save the errcode and proceed to close */ | |
753 | saved_rc = rc; | |
754 | } | |
755 | } | |
d7e09d03 PT |
756 | |
757 | *request = NULL; | |
37199135 AB |
758 | if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE)) |
759 | req = NULL; | |
760 | else | |
761 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt); | |
d7e09d03 PT |
762 | |
763 | /* Ensure that this close's handle is fixed up during replay. */ | |
34e3ff96 OD |
764 | if (likely(mod)) { |
765 | LASSERTF(mod->mod_open_req && | |
d7e09d03 PT |
766 | mod->mod_open_req->rq_type != LI_POISON, |
767 | "POISONED open %p!\n", mod->mod_open_req); | |
768 | ||
769 | mod->mod_close_req = req; | |
770 | ||
771 | DEBUG_REQ(D_HA, mod->mod_open_req, "matched open"); | |
772 | /* We no longer want to preserve this open for replay even | |
1df232ee OD |
773 | * though the open was committed. b=3632, b=3633 |
774 | */ | |
d7e09d03 PT |
775 | spin_lock(&mod->mod_open_req->rq_lock); |
776 | mod->mod_open_req->rq_replay = 0; | |
777 | spin_unlock(&mod->mod_open_req->rq_lock); | |
778 | } else { | |
e5e663ae SM |
779 | CDEBUG(D_HA, |
780 | "couldn't find open req; expecting close error\n"); | |
d7e09d03 | 781 | } |
37199135 AB |
782 | if (!req) { |
783 | /* | |
784 | * TODO: repeat close after errors | |
785 | */ | |
786 | CWARN("%s: close of FID "DFID" failed, file reference will be dropped when this client unmounts or is evicted\n", | |
787 | obd->obd_name, PFID(&op_data->op_fid1)); | |
788 | rc = -ENOMEM; | |
789 | goto out; | |
790 | } | |
791 | ||
792 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE); | |
793 | if (rc) { | |
794 | ptlrpc_request_free(req); | |
795 | goto out; | |
796 | } | |
797 | ||
798 | /* | |
799 | * To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a | |
800 | * portal whose threads are not taking any DLM locks and are therefore | |
801 | * always progressing | |
802 | */ | |
803 | req->rq_request_portal = MDS_READPAGE_PORTAL; | |
804 | ptlrpc_at_set_req_timeout(req); | |
d7e09d03 PT |
805 | |
806 | mdc_close_pack(req, op_data); | |
807 | ||
808 | req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, | |
44779340 | 809 | obd->u.cli.cl_default_mds_easize); |
d7e09d03 | 810 | req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER, |
44779340 | 811 | obd->u.cli.cl_default_mds_cookiesize); |
d7e09d03 PT |
812 | |
813 | ptlrpc_request_set_replen(req); | |
814 | ||
815 | mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL); | |
816 | rc = ptlrpc_queue_wait(req); | |
817 | mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL); | |
818 | ||
34e3ff96 | 819 | if (!req->rq_repmsg) { |
d7e09d03 PT |
820 | CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req, |
821 | req->rq_status); | |
822 | if (rc == 0) | |
823 | rc = req->rq_status ?: -EIO; | |
824 | } else if (rc == 0 || rc == -EAGAIN) { | |
825 | struct mdt_body *body; | |
826 | ||
827 | rc = lustre_msg_get_status(req->rq_repmsg); | |
828 | if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) { | |
ee990b33 SM |
829 | DEBUG_REQ(D_ERROR, req, |
830 | "type == PTL_RPC_MSG_ERR, err = %d", rc); | |
d7e09d03 PT |
831 | if (rc > 0) |
832 | rc = -rc; | |
833 | } | |
834 | body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); | |
34e3ff96 | 835 | if (!body) |
d7e09d03 PT |
836 | rc = -EPROTO; |
837 | } else if (rc == -ESTALE) { | |
838 | /** | |
839 | * it can be allowed error after 3633 if open was committed and | |
840 | * server failed before close was sent. Let's check if mod | |
841 | * exists and return no error in that case | |
842 | */ | |
843 | if (mod) { | |
844 | DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc); | |
d7e09d03 PT |
845 | if (mod->mod_open_req->rq_committed) |
846 | rc = 0; | |
847 | } | |
848 | } | |
849 | ||
37199135 | 850 | out: |
d7e09d03 PT |
851 | if (mod) { |
852 | if (rc != 0) | |
853 | mod->mod_close_req = NULL; | |
854 | /* Since now, mod is accessed through open_req only, | |
1df232ee OD |
855 | * thus close req does not keep a reference on mod anymore. |
856 | */ | |
d7e09d03 PT |
857 | obd_mod_put(mod); |
858 | } | |
859 | *request = req; | |
860 | mdc_close_handle_reply(req, op_data, rc); | |
48d23e61 | 861 | return rc < 0 ? rc : saved_rc; |
d7e09d03 PT |
862 | } |
863 | ||
f6219c16 LC |
864 | static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data, |
865 | struct md_open_data *mod) | |
d7e09d03 PT |
866 | { |
867 | struct obd_device *obd = class_exp2obd(exp); | |
868 | struct ptlrpc_request *req; | |
869 | int rc; | |
d7e09d03 PT |
870 | |
871 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), | |
872 | &RQF_MDS_DONE_WRITING); | |
34e3ff96 | 873 | if (!req) |
0a3bdb00 | 874 | return -ENOMEM; |
d7e09d03 | 875 | |
d7e09d03 PT |
876 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING); |
877 | if (rc) { | |
878 | ptlrpc_request_free(req); | |
0a3bdb00 | 879 | return rc; |
d7e09d03 PT |
880 | } |
881 | ||
34e3ff96 OD |
882 | if (mod) { |
883 | LASSERTF(mod->mod_open_req && | |
d7e09d03 PT |
884 | mod->mod_open_req->rq_type != LI_POISON, |
885 | "POISONED setattr %p!\n", mod->mod_open_req); | |
886 | ||
887 | mod->mod_close_req = req; | |
888 | DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr"); | |
889 | /* We no longer want to preserve this setattr for replay even | |
1df232ee OD |
890 | * though the open was committed. b=3632, b=3633 |
891 | */ | |
d7e09d03 PT |
892 | spin_lock(&mod->mod_open_req->rq_lock); |
893 | mod->mod_open_req->rq_replay = 0; | |
894 | spin_unlock(&mod->mod_open_req->rq_lock); | |
895 | } | |
896 | ||
897 | mdc_close_pack(req, op_data); | |
898 | ptlrpc_request_set_replen(req); | |
899 | ||
900 | mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL); | |
901 | rc = ptlrpc_queue_wait(req); | |
902 | mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL); | |
903 | ||
904 | if (rc == -ESTALE) { | |
905 | /** | |
906 | * it can be allowed error after 3633 if open or setattr were | |
907 | * committed and server failed before close was sent. | |
908 | * Let's check if mod exists and return no error in that case | |
909 | */ | |
910 | if (mod) { | |
d7e09d03 PT |
911 | if (mod->mod_open_req->rq_committed) |
912 | rc = 0; | |
913 | } | |
914 | } | |
915 | ||
916 | if (mod) { | |
917 | if (rc != 0) | |
918 | mod->mod_close_req = NULL; | |
34e3ff96 | 919 | LASSERT(mod->mod_open_req); |
63d42578 HZ |
920 | mdc_free_open(mod); |
921 | ||
d7e09d03 | 922 | /* Since now, mod is accessed through setattr req only, |
1df232ee OD |
923 | * thus DW req does not keep a reference on mod anymore. |
924 | */ | |
d7e09d03 PT |
925 | obd_mod_put(mod); |
926 | } | |
927 | ||
928 | mdc_close_handle_reply(req, op_data, rc); | |
929 | ptlrpc_req_finished(req); | |
0a3bdb00 | 930 | return rc; |
d7e09d03 PT |
931 | } |
932 | ||
4f76f0ec | 933 | static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid, |
934 | u64 offset, struct page **pages, int npages, | |
935 | struct ptlrpc_request **request) | |
936 | { | |
937 | struct ptlrpc_bulk_desc *desc; | |
938 | struct ptlrpc_request *req; | |
939 | wait_queue_head_t waitq; | |
940 | struct l_wait_info lwi; | |
941 | int resends = 0; | |
942 | int rc; | |
943 | int i; | |
944 | ||
945 | *request = NULL; | |
946 | init_waitqueue_head(&waitq); | |
947 | ||
948 | restart_bulk: | |
949 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE); | |
950 | if (!req) | |
951 | return -ENOMEM; | |
952 | ||
953 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE); | |
954 | if (rc) { | |
955 | ptlrpc_request_free(req); | |
956 | return rc; | |
957 | } | |
958 | ||
959 | req->rq_request_portal = MDS_READPAGE_PORTAL; | |
960 | ptlrpc_at_set_req_timeout(req); | |
961 | ||
962 | desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK, | |
963 | MDS_BULK_PORTAL); | |
964 | if (!desc) { | |
965 | ptlrpc_request_free(req); | |
966 | return -ENOMEM; | |
967 | } | |
968 | ||
969 | /* NB req now owns desc and will free it when it gets freed */ | |
970 | for (i = 0; i < npages; i++) | |
971 | ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_SIZE); | |
972 | ||
973 | mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid); | |
974 | ||
975 | ptlrpc_request_set_replen(req); | |
976 | rc = ptlrpc_queue_wait(req); | |
977 | if (rc) { | |
978 | ptlrpc_req_finished(req); | |
979 | if (rc != -ETIMEDOUT) | |
980 | return rc; | |
981 | ||
982 | resends++; | |
983 | if (!client_should_resend(resends, &exp->exp_obd->u.cli)) { | |
984 | CERROR("%s: too many resend retries: rc = %d\n", | |
985 | exp->exp_obd->obd_name, -EIO); | |
986 | return -EIO; | |
987 | } | |
988 | lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, | |
989 | NULL); | |
990 | l_wait_event(waitq, 0, &lwi); | |
991 | ||
992 | goto restart_bulk; | |
993 | } | |
994 | ||
995 | rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, | |
996 | req->rq_bulk->bd_nob_transferred); | |
997 | if (rc < 0) { | |
998 | ptlrpc_req_finished(req); | |
999 | return rc; | |
1000 | } | |
1001 | ||
1002 | if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) { | |
1003 | CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n", | |
1004 | exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred, | |
1005 | PAGE_SIZE * npages); | |
1006 | ptlrpc_req_finished(req); | |
1007 | return -EPROTO; | |
1008 | } | |
1009 | ||
1010 | *request = req; | |
1011 | return 0; | |
1012 | } | |
1013 | ||
1014 | static void mdc_release_page(struct page *page, int remove) | |
1015 | { | |
1016 | if (remove) { | |
1017 | lock_page(page); | |
1018 | if (likely(page->mapping)) | |
1019 | truncate_complete_page(page->mapping, page); | |
1020 | unlock_page(page); | |
1021 | } | |
1022 | put_page(page); | |
1023 | } | |
1024 | ||
1025 | static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash, | |
1026 | __u64 *start, __u64 *end, int hash64) | |
1027 | { | |
1028 | /* | |
1029 | * Complement of hash is used as an index so that | |
1030 | * radix_tree_gang_lookup() can be used to find a page with starting | |
1031 | * hash _smaller_ than one we are looking for. | |
1032 | */ | |
1033 | unsigned long offset = hash_x_index(*hash, hash64); | |
1034 | struct page *page; | |
1035 | int found; | |
1036 | ||
1037 | spin_lock_irq(&mapping->tree_lock); | |
1038 | found = radix_tree_gang_lookup(&mapping->page_tree, | |
1039 | (void **)&page, offset, 1); | |
1040 | if (found > 0 && !radix_tree_exceptional_entry(page)) { | |
1041 | struct lu_dirpage *dp; | |
1042 | ||
1043 | get_page(page); | |
1044 | spin_unlock_irq(&mapping->tree_lock); | |
1045 | /* | |
1046 | * In contrast to find_lock_page() we are sure that directory | |
1047 | * page cannot be truncated (while DLM lock is held) and, | |
1048 | * hence, can avoid restart. | |
1049 | * | |
1050 | * In fact, page cannot be locked here at all, because | |
1051 | * mdc_read_page_remote does synchronous io. | |
1052 | */ | |
1053 | wait_on_page_locked(page); | |
1054 | if (PageUptodate(page)) { | |
1055 | dp = kmap(page); | |
1056 | if (BITS_PER_LONG == 32 && hash64) { | |
1057 | *start = le64_to_cpu(dp->ldp_hash_start) >> 32; | |
1058 | *end = le64_to_cpu(dp->ldp_hash_end) >> 32; | |
1059 | *hash = *hash >> 32; | |
1060 | } else { | |
1061 | *start = le64_to_cpu(dp->ldp_hash_start); | |
1062 | *end = le64_to_cpu(dp->ldp_hash_end); | |
1063 | } | |
1064 | if (unlikely(*start == 1 && *hash == 0)) | |
1065 | *hash = *start; | |
1066 | else | |
1067 | LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n", | |
1068 | *start, *end, *hash); | |
1069 | CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx], hash %#llx\n", | |
1070 | offset, *start, *end, *hash); | |
1071 | if (*hash > *end) { | |
1072 | kunmap(page); | |
1073 | mdc_release_page(page, 0); | |
1074 | page = NULL; | |
1075 | } else if (*end != *start && *hash == *end) { | |
1076 | /* | |
1077 | * upon hash collision, remove this page, | |
1078 | * otherwise put page reference, and | |
1079 | * mdc_read_page_remote() will issue RPC to | |
1080 | * fetch the page we want. | |
1081 | */ | |
1082 | kunmap(page); | |
1083 | mdc_release_page(page, | |
1084 | le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE); | |
1085 | page = NULL; | |
1086 | } | |
1087 | } else { | |
1088 | put_page(page); | |
1089 | page = ERR_PTR(-EIO); | |
1090 | } | |
1091 | } else { | |
1092 | spin_unlock_irq(&mapping->tree_lock); | |
1093 | page = NULL; | |
1094 | } | |
1095 | return page; | |
1096 | } | |
1097 | ||
1098 | /* | |
1099 | * Adjust a set of pages, each page containing an array of lu_dirpages, | |
1100 | * so that each page can be used as a single logical lu_dirpage. | |
1101 | * | |
1102 | * A lu_dirpage is laid out as follows, where s = ldp_hash_start, | |
1103 | * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a | |
1104 | * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end | |
1105 | * value is used as a cookie to request the next lu_dirpage in a | |
1106 | * directory listing that spans multiple pages (two in this example): | |
1107 | * ________ | |
1108 | * | | | |
1109 | * .|--------v------- -----. | |
1110 | * |s|e|f|p|ent|ent| ... |ent| | |
1111 | * '--|-------------- -----' Each PAGE contains a single | |
1112 | * '------. lu_dirpage. | |
1113 | * .---------v------- -----. | |
1114 | * |s|e|f|p|ent| 0 | ... | 0 | | |
1115 | * '----------------- -----' | |
1116 | * | |
1117 | * However, on hosts where the native VM page size (PAGE_SIZE) is | |
1118 | * larger than LU_PAGE_SIZE, a single host page may contain multiple | |
1119 | * lu_dirpages. After reading the lu_dirpages from the MDS, the | |
1120 | * ldp_hash_end of the first lu_dirpage refers to the one immediately | |
1121 | * after it in the same PAGE (arrows simplified for brevity, but | |
1122 | * in general e0==s1, e1==s2, etc.): | |
1123 | * | |
1124 | * .-------------------- -----. | |
1125 | * |s0|e0|f0|p|ent|ent| ... |ent| | |
1126 | * |---v---------------- -----| | |
1127 | * |s1|e1|f1|p|ent|ent| ... |ent| | |
1128 | * |---v---------------- -----| Here, each PAGE contains | |
1129 | * ... multiple lu_dirpages. | |
1130 | * |---v---------------- -----| | |
1131 | * |s'|e'|f'|p|ent|ent| ... |ent| | |
1132 | * '---|---------------- -----' | |
1133 | * v | |
1134 | * .----------------------------. | |
1135 | * | next PAGE | | |
1136 | * | |
1137 | * This structure is transformed into a single logical lu_dirpage as follows: | |
1138 | * | |
1139 | * - Replace e0 with e' so the request for the next lu_dirpage gets the page | |
1140 | * labeled 'next PAGE'. | |
1141 | * | |
1142 | * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether | |
1143 | * a hash collision with the next page exists. | |
1144 | * | |
1145 | * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span | |
1146 | * to the first entry of the next lu_dirpage. | |
1147 | */ | |
1148 | #if PAGE_SIZE > LU_PAGE_SIZE | |
1149 | static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs) | |
1150 | { | |
1151 | int i; | |
1152 | ||
1153 | for (i = 0; i < cfs_pgs; i++) { | |
88b090e3 | 1154 | struct lu_dirpage *dp = kmap(pages[i]); |
4f76f0ec | 1155 | __u64 hash_end = le64_to_cpu(dp->ldp_hash_end); |
1156 | __u32 flags = le32_to_cpu(dp->ldp_flags); | |
4f76f0ec | 1157 | struct lu_dirpage *first = dp; |
1158 | struct lu_dirent *end_dirent = NULL; | |
1159 | struct lu_dirent *ent; | |
1160 | ||
1161 | while (--lu_pgs > 0) { | |
1162 | ent = lu_dirent_start(dp); | |
1163 | for (end_dirent = ent; ent; | |
1164 | end_dirent = ent, ent = lu_dirent_next(ent)); | |
1165 | ||
1166 | /* Advance dp to next lu_dirpage. */ | |
1167 | dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE); | |
1168 | ||
1169 | /* Check if we've reached the end of the CFS_PAGE. */ | |
1170 | if (!((unsigned long)dp & ~PAGE_MASK)) | |
1171 | break; | |
1172 | ||
1173 | /* Save the hash and flags of this lu_dirpage. */ | |
1174 | hash_end = le64_to_cpu(dp->ldp_hash_end); | |
1175 | flags = le32_to_cpu(dp->ldp_flags); | |
1176 | ||
1177 | /* Check if lu_dirpage contains no entries. */ | |
1178 | if (!end_dirent) | |
1179 | break; | |
1180 | ||
1181 | /* | |
1182 | * Enlarge the end entry lde_reclen from 0 to | |
1183 | * first entry of next lu_dirpage. | |
1184 | */ | |
1185 | LASSERT(!le16_to_cpu(end_dirent->lde_reclen)); | |
1186 | end_dirent->lde_reclen = | |
1187 | cpu_to_le16((char *)(dp->ldp_entries) - | |
1188 | (char *)end_dirent); | |
1189 | } | |
1190 | ||
1191 | first->ldp_hash_end = hash_end; | |
1192 | first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE); | |
1193 | first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE); | |
1194 | ||
1195 | kunmap(pages[i]); | |
1196 | } | |
1197 | LASSERTF(lu_pgs == 0, "left = %d", lu_pgs); | |
1198 | } | |
1199 | #else | |
1200 | #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0) | |
1201 | #endif /* PAGE_SIZE > LU_PAGE_SIZE */ | |
1202 | ||
1203 | /* parameters for readdir page */ | |
1204 | struct readpage_param { | |
1205 | struct md_op_data *rp_mod; | |
1206 | __u64 rp_off; | |
1207 | int rp_hash64; | |
1208 | struct obd_export *rp_exp; | |
1209 | struct md_callback *rp_cb; | |
1210 | }; | |
1211 | ||
1212 | /** | |
1213 | * Read pages from server. | |
1214 | * | |
1215 | * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains | |
1216 | * a header lu_dirpage which describes the start/end hash, and whether this | |
1217 | * page is empty (contains no dir entry) or hash collide with next page. | |
1218 | * After client receives reply, several pages will be integrated into dir page | |
1219 | * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the | |
1220 | * lu_dirpage for this integrated page will be adjusted. | |
1221 | **/ | |
1222 | static int mdc_read_page_remote(void *data, struct page *page0) | |
1223 | { | |
1224 | struct readpage_param *rp = data; | |
1225 | struct page **page_pool; | |
1226 | struct page *page; | |
1227 | struct lu_dirpage *dp; | |
1228 | int rd_pgs = 0; /* number of pages read actually */ | |
1229 | int npages; | |
1230 | struct md_op_data *op_data = rp->rp_mod; | |
1231 | struct ptlrpc_request *req; | |
1232 | int max_pages = op_data->op_max_pages; | |
1233 | struct inode *inode; | |
1234 | struct lu_fid *fid; | |
1235 | int i; | |
1236 | int rc; | |
1237 | ||
1238 | LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES); | |
1239 | inode = op_data->op_data; | |
1240 | fid = &op_data->op_fid1; | |
1241 | LASSERT(inode); | |
1242 | ||
1243 | page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS); | |
1244 | if (page_pool) { | |
1245 | page_pool[0] = page0; | |
1246 | } else { | |
1247 | page_pool = &page0; | |
1248 | max_pages = 1; | |
1249 | } | |
1250 | ||
1251 | for (npages = 1; npages < max_pages; npages++) { | |
1252 | page = page_cache_alloc_cold(inode->i_mapping); | |
1253 | if (!page) | |
1254 | break; | |
1255 | page_pool[npages] = page; | |
1256 | } | |
1257 | ||
1258 | rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req); | |
1259 | if (!rc) { | |
1260 | int lu_pgs = req->rq_bulk->bd_nob_transferred; | |
1261 | ||
1262 | rd_pgs = (req->rq_bulk->bd_nob_transferred + | |
1263 | PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1264 | lu_pgs >>= LU_PAGE_SHIFT; | |
1265 | LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK)); | |
1266 | ||
1267 | CDEBUG(D_INODE, "read %d(%d)/%d pages\n", rd_pgs, lu_pgs, | |
1268 | op_data->op_npages); | |
1269 | ||
1270 | mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs); | |
1271 | ||
1272 | SetPageUptodate(page0); | |
1273 | } | |
1274 | ||
1275 | unlock_page(page0); | |
1276 | ptlrpc_req_finished(req); | |
1277 | CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages); | |
1278 | for (i = 1; i < npages; i++) { | |
1279 | unsigned long offset; | |
1280 | __u64 hash; | |
1281 | int ret; | |
1282 | ||
1283 | page = page_pool[i]; | |
1284 | ||
1285 | if (rc < 0 || i >= rd_pgs) { | |
1286 | put_page(page); | |
1287 | continue; | |
1288 | } | |
1289 | ||
1290 | SetPageUptodate(page); | |
1291 | ||
1292 | dp = kmap(page); | |
1293 | hash = le64_to_cpu(dp->ldp_hash_start); | |
1294 | kunmap(page); | |
1295 | ||
1296 | offset = hash_x_index(hash, rp->rp_hash64); | |
1297 | ||
1298 | prefetchw(&page->flags); | |
1299 | ret = add_to_page_cache_lru(page, inode->i_mapping, offset, | |
1300 | GFP_KERNEL); | |
1301 | if (!ret) | |
1302 | unlock_page(page); | |
1303 | else | |
1304 | CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: rc = %d\n", | |
1305 | offset, ret); | |
1306 | put_page(page); | |
1307 | } | |
1308 | ||
1309 | if (page_pool != &page0) | |
1310 | kfree(page_pool); | |
1311 | ||
1312 | return rc; | |
1313 | } | |
1314 | ||
1315 | /** | |
1316 | * Read dir page from cache first, if it can not find it, read it from | |
1317 | * server and add into the cache. | |
1318 | * | |
1319 | * \param[in] exp MDC export | |
1320 | * \param[in] op_data client MD stack parameters, transferring parameters | |
1321 | * between different layers on client MD stack. | |
1322 | * \param[in] cb_op callback required for ldlm lock enqueue during | |
1323 | * read page | |
1324 | * \param[in] hash_offset the hash offset of the page to be read | |
1325 | * \param[in] ppage the page to be read | |
1326 | * | |
1327 | * retval = 0 get the page successfully | |
1328 | * errno(<0) get the page failed | |
1329 | */ | |
1330 | static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data, | |
1331 | struct md_callback *cb_op, __u64 hash_offset, | |
1332 | struct page **ppage) | |
1333 | { | |
1334 | struct lookup_intent it = { .it_op = IT_READDIR }; | |
1335 | struct page *page; | |
1336 | struct inode *dir = op_data->op_data; | |
1337 | struct address_space *mapping; | |
1338 | struct lu_dirpage *dp; | |
1339 | __u64 start = 0; | |
1340 | __u64 end = 0; | |
1341 | struct lustre_handle lockh; | |
1342 | struct ptlrpc_request *enq_req = NULL; | |
1343 | struct readpage_param rp_param; | |
1344 | int rc; | |
1345 | ||
1346 | *ppage = NULL; | |
1347 | ||
1348 | LASSERT(dir); | |
1349 | mapping = dir->i_mapping; | |
1350 | ||
70a251f6 | 1351 | rc = mdc_intent_lock(exp, op_data, &it, &enq_req, |
4f76f0ec | 1352 | cb_op->md_blocking_ast, 0); |
1353 | if (enq_req) | |
1354 | ptlrpc_req_finished(enq_req); | |
1355 | ||
1356 | if (rc < 0) { | |
1357 | CERROR("%s: "DFID" lock enqueue fails: rc = %d\n", | |
1358 | exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc); | |
1359 | return rc; | |
1360 | } | |
1361 | ||
1362 | rc = 0; | |
bc30c172 JH |
1363 | lockh.cookie = it.it_lock_handle; |
1364 | mdc_set_lock_data(exp, &lockh, dir, NULL); | |
4f76f0ec | 1365 | |
1366 | rp_param.rp_off = hash_offset; | |
1367 | rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64; | |
1368 | page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end, | |
1369 | rp_param.rp_hash64); | |
1370 | if (IS_ERR(page)) { | |
1371 | CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n", | |
1372 | exp->exp_obd->obd_name, PFID(&op_data->op_fid1), | |
1373 | rp_param.rp_off, PTR_ERR(page)); | |
1374 | rc = PTR_ERR(page); | |
1375 | goto out_unlock; | |
1376 | } else if (page) { | |
1377 | /* | |
1378 | * XXX nikita: not entirely correct handling of a corner case: | |
1379 | * suppose hash chain of entries with hash value HASH crosses | |
1380 | * border between pages P0 and P1. First both P0 and P1 are | |
1381 | * cached, seekdir() is called for some entry from the P0 part | |
1382 | * of the chain. Later P0 goes out of cache. telldir(HASH) | |
1383 | * happens and finds P1, as it starts with matching hash | |
1384 | * value. Remaining entries from P0 part of the chain are | |
1385 | * skipped. (Is that really a bug?) | |
1386 | * | |
1387 | * Possible solutions: 0. don't cache P1 is such case, handle | |
1388 | * it as an "overflow" page. 1. invalidate all pages at | |
1389 | * once. 2. use HASH|1 as an index for P1. | |
1390 | */ | |
1391 | goto hash_collision; | |
1392 | } | |
1393 | ||
1394 | rp_param.rp_exp = exp; | |
1395 | rp_param.rp_mod = op_data; | |
1396 | page = read_cache_page(mapping, | |
1397 | hash_x_index(rp_param.rp_off, | |
1398 | rp_param.rp_hash64), | |
1399 | mdc_read_page_remote, &rp_param); | |
1400 | if (IS_ERR(page)) { | |
1401 | CERROR("%s: read cache page: "DFID" at %llu: rc %ld\n", | |
1402 | exp->exp_obd->obd_name, PFID(&op_data->op_fid1), | |
1403 | rp_param.rp_off, PTR_ERR(page)); | |
1404 | rc = PTR_ERR(page); | |
1405 | goto out_unlock; | |
1406 | } | |
1407 | ||
1408 | wait_on_page_locked(page); | |
1409 | (void)kmap(page); | |
1410 | if (!PageUptodate(page)) { | |
1411 | CERROR("%s: page not updated: "DFID" at %llu: rc %d\n", | |
1412 | exp->exp_obd->obd_name, PFID(&op_data->op_fid1), | |
1413 | rp_param.rp_off, -5); | |
1414 | goto fail; | |
1415 | } | |
1416 | if (!PageChecked(page)) | |
1417 | SetPageChecked(page); | |
1418 | if (PageError(page)) { | |
1419 | CERROR("%s: page error: "DFID" at %llu: rc %d\n", | |
1420 | exp->exp_obd->obd_name, PFID(&op_data->op_fid1), | |
1421 | rp_param.rp_off, -5); | |
1422 | goto fail; | |
1423 | } | |
1424 | ||
1425 | hash_collision: | |
1426 | dp = page_address(page); | |
1427 | if (BITS_PER_LONG == 32 && rp_param.rp_hash64) { | |
1428 | start = le64_to_cpu(dp->ldp_hash_start) >> 32; | |
1429 | end = le64_to_cpu(dp->ldp_hash_end) >> 32; | |
1430 | rp_param.rp_off = hash_offset >> 32; | |
1431 | } else { | |
1432 | start = le64_to_cpu(dp->ldp_hash_start); | |
1433 | end = le64_to_cpu(dp->ldp_hash_end); | |
1434 | rp_param.rp_off = hash_offset; | |
1435 | } | |
1436 | if (end == start) { | |
1437 | LASSERT(start == rp_param.rp_off); | |
1438 | CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end); | |
1439 | #if BITS_PER_LONG == 32 | |
1440 | CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n", | |
1441 | le64_to_cpu(dp->ldp_hash_start), | |
1442 | le64_to_cpu(dp->ldp_hash_end), hash_offset); | |
1443 | #endif | |
1444 | /* | |
1445 | * Fetch whole overflow chain... | |
1446 | * | |
1447 | * XXX not yet. | |
1448 | */ | |
1449 | goto fail; | |
1450 | } | |
1451 | *ppage = page; | |
1452 | out_unlock: | |
4f76f0ec | 1453 | ldlm_lock_decref(&lockh, it.it_lock_mode); |
4f76f0ec | 1454 | return rc; |
1455 | fail: | |
1456 | kunmap(page); | |
1457 | mdc_release_page(page, 1); | |
1458 | rc = -EIO; | |
1459 | goto out_unlock; | |
1460 | } | |
1461 | ||
d7e09d03 PT |
1462 | static int mdc_statfs(const struct lu_env *env, |
1463 | struct obd_export *exp, struct obd_statfs *osfs, | |
1464 | __u64 max_age, __u32 flags) | |
1465 | { | |
1466 | struct obd_device *obd = class_exp2obd(exp); | |
1467 | struct ptlrpc_request *req; | |
1468 | struct obd_statfs *msfs; | |
1469 | struct obd_import *imp = NULL; | |
1470 | int rc; | |
d7e09d03 PT |
1471 | |
1472 | /* | |
1473 | * Since the request might also come from lprocfs, so we need | |
1474 | * sync this with client_disconnect_export Bug15684 | |
1475 | */ | |
1476 | down_read(&obd->u.cli.cl_sem); | |
1477 | if (obd->u.cli.cl_import) | |
1478 | imp = class_import_get(obd->u.cli.cl_import); | |
1479 | up_read(&obd->u.cli.cl_sem); | |
1480 | if (!imp) | |
0a3bdb00 | 1481 | return -ENODEV; |
d7e09d03 PT |
1482 | |
1483 | req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS, | |
1484 | LUSTRE_MDS_VERSION, MDS_STATFS); | |
34e3ff96 | 1485 | if (!req) { |
d5fdc207 JL |
1486 | rc = -ENOMEM; |
1487 | goto output; | |
1488 | } | |
d7e09d03 PT |
1489 | |
1490 | ptlrpc_request_set_replen(req); | |
1491 | ||
1492 | if (flags & OBD_STATFS_NODELAY) { | |
1493 | /* procfs requests not want stay in wait for avoid deadlock */ | |
1494 | req->rq_no_resend = 1; | |
1495 | req->rq_no_delay = 1; | |
1496 | } | |
1497 | ||
1498 | rc = ptlrpc_queue_wait(req); | |
1499 | if (rc) { | |
1500 | /* check connection error first */ | |
1501 | if (imp->imp_connect_error) | |
1502 | rc = imp->imp_connect_error; | |
d5fdc207 | 1503 | goto out; |
d7e09d03 PT |
1504 | } |
1505 | ||
1506 | msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS); | |
34e3ff96 | 1507 | if (!msfs) { |
d5fdc207 JL |
1508 | rc = -EPROTO; |
1509 | goto out; | |
1510 | } | |
d7e09d03 PT |
1511 | |
1512 | *osfs = *msfs; | |
d7e09d03 PT |
1513 | out: |
1514 | ptlrpc_req_finished(req); | |
1515 | output: | |
1516 | class_import_put(imp); | |
1517 | return rc; | |
1518 | } | |
1519 | ||
1520 | static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf) | |
1521 | { | |
1522 | __u32 keylen, vallen; | |
1523 | void *key; | |
1524 | int rc; | |
1525 | ||
1526 | if (gf->gf_pathlen > PATH_MAX) | |
0a3bdb00 | 1527 | return -ENAMETOOLONG; |
d7e09d03 | 1528 | if (gf->gf_pathlen < 2) |
0a3bdb00 | 1529 | return -EOVERFLOW; |
d7e09d03 PT |
1530 | |
1531 | /* Key is KEY_FID2PATH + getinfo_fid2path description */ | |
1532 | keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf); | |
7b81779d | 1533 | key = kzalloc(keylen, GFP_NOFS); |
bb144d09 | 1534 | if (!key) |
0a3bdb00 | 1535 | return -ENOMEM; |
d7e09d03 PT |
1536 | memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH)); |
1537 | memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf)); | |
1538 | ||
b0f5aad5 | 1539 | CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n", |
d7e09d03 PT |
1540 | PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno); |
1541 | ||
d5fdc207 JL |
1542 | if (!fid_is_sane(&gf->gf_fid)) { |
1543 | rc = -EINVAL; | |
1544 | goto out; | |
1545 | } | |
d7e09d03 PT |
1546 | |
1547 | /* Val is struct getinfo_fid2path result plus path */ | |
1548 | vallen = sizeof(*gf) + gf->gf_pathlen; | |
1549 | ||
1550 | rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL); | |
1551 | if (rc != 0 && rc != -EREMOTE) | |
d5fdc207 | 1552 | goto out; |
d7e09d03 | 1553 | |
d5fdc207 JL |
1554 | if (vallen <= sizeof(*gf)) { |
1555 | rc = -EPROTO; | |
1556 | goto out; | |
1557 | } else if (vallen > sizeof(*gf) + gf->gf_pathlen) { | |
1558 | rc = -EOVERFLOW; | |
1559 | goto out; | |
1560 | } | |
d7e09d03 | 1561 | |
b0f5aad5 | 1562 | CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n", |
d7e09d03 PT |
1563 | PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path); |
1564 | ||
1565 | out: | |
7b81779d | 1566 | kfree(key); |
d7e09d03 PT |
1567 | return rc; |
1568 | } | |
1569 | ||
1570 | static int mdc_ioc_hsm_progress(struct obd_export *exp, | |
1571 | struct hsm_progress_kernel *hpk) | |
1572 | { | |
1573 | struct obd_import *imp = class_exp2cliimp(exp); | |
1574 | struct hsm_progress_kernel *req_hpk; | |
1575 | struct ptlrpc_request *req; | |
1576 | int rc; | |
d7e09d03 PT |
1577 | |
1578 | req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS, | |
1579 | LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS); | |
34e3ff96 | 1580 | if (!req) { |
d5fdc207 JL |
1581 | rc = -ENOMEM; |
1582 | goto out; | |
1583 | } | |
d7e09d03 | 1584 | |
341f1f0a | 1585 | mdc_pack_body(req, NULL, 0, 0, -1, 0); |
d7e09d03 PT |
1586 | |
1587 | /* Copy hsm_progress struct */ | |
1588 | req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS); | |
34e3ff96 | 1589 | if (!req_hpk) { |
d5fdc207 JL |
1590 | rc = -EPROTO; |
1591 | goto out; | |
1592 | } | |
d7e09d03 PT |
1593 | |
1594 | *req_hpk = *hpk; | |
2d58de78 | 1595 | req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval); |
d7e09d03 PT |
1596 | |
1597 | ptlrpc_request_set_replen(req); | |
1598 | ||
1599 | rc = mdc_queue_wait(req); | |
d7e09d03 PT |
1600 | out: |
1601 | ptlrpc_req_finished(req); | |
1602 | return rc; | |
1603 | } | |
1604 | ||
1605 | static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives) | |
1606 | { | |
1607 | __u32 *archive_mask; | |
1608 | struct ptlrpc_request *req; | |
1609 | int rc; | |
d7e09d03 PT |
1610 | |
1611 | req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER, | |
1612 | LUSTRE_MDS_VERSION, | |
1613 | MDS_HSM_CT_REGISTER); | |
34e3ff96 | 1614 | if (!req) { |
d5fdc207 JL |
1615 | rc = -ENOMEM; |
1616 | goto out; | |
1617 | } | |
d7e09d03 | 1618 | |
341f1f0a | 1619 | mdc_pack_body(req, NULL, 0, 0, -1, 0); |
d7e09d03 PT |
1620 | |
1621 | /* Copy hsm_progress struct */ | |
1622 | archive_mask = req_capsule_client_get(&req->rq_pill, | |
1623 | &RMF_MDS_HSM_ARCHIVE); | |
34e3ff96 | 1624 | if (!archive_mask) { |
d5fdc207 JL |
1625 | rc = -EPROTO; |
1626 | goto out; | |
1627 | } | |
d7e09d03 PT |
1628 | |
1629 | *archive_mask = archives; | |
1630 | ||
1631 | ptlrpc_request_set_replen(req); | |
1632 | ||
1633 | rc = mdc_queue_wait(req); | |
d7e09d03 PT |
1634 | out: |
1635 | ptlrpc_req_finished(req); | |
1636 | return rc; | |
1637 | } | |
1638 | ||
1639 | static int mdc_ioc_hsm_current_action(struct obd_export *exp, | |
1640 | struct md_op_data *op_data) | |
1641 | { | |
1642 | struct hsm_current_action *hca = op_data->op_data; | |
1643 | struct hsm_current_action *req_hca; | |
1644 | struct ptlrpc_request *req; | |
1645 | int rc; | |
d7e09d03 PT |
1646 | |
1647 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), | |
1648 | &RQF_MDS_HSM_ACTION); | |
34e3ff96 | 1649 | if (!req) |
0a3bdb00 | 1650 | return -ENOMEM; |
d7e09d03 | 1651 | |
d7e09d03 PT |
1652 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION); |
1653 | if (rc) { | |
1654 | ptlrpc_request_free(req); | |
0a3bdb00 | 1655 | return rc; |
d7e09d03 PT |
1656 | } |
1657 | ||
341f1f0a | 1658 | mdc_pack_body(req, &op_data->op_fid1, 0, 0, |
ef2e0f55 | 1659 | op_data->op_suppgids[0], 0); |
d7e09d03 PT |
1660 | |
1661 | ptlrpc_request_set_replen(req); | |
1662 | ||
1663 | rc = mdc_queue_wait(req); | |
1664 | if (rc) | |
d5fdc207 | 1665 | goto out; |
d7e09d03 PT |
1666 | |
1667 | req_hca = req_capsule_server_get(&req->rq_pill, | |
1668 | &RMF_MDS_HSM_CURRENT_ACTION); | |
34e3ff96 | 1669 | if (!req_hca) { |
d5fdc207 JL |
1670 | rc = -EPROTO; |
1671 | goto out; | |
1672 | } | |
d7e09d03 PT |
1673 | |
1674 | *hca = *req_hca; | |
1675 | ||
d7e09d03 PT |
1676 | out: |
1677 | ptlrpc_req_finished(req); | |
1678 | return rc; | |
1679 | } | |
1680 | ||
1681 | static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp) | |
1682 | { | |
1683 | struct ptlrpc_request *req; | |
1684 | int rc; | |
d7e09d03 PT |
1685 | |
1686 | req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER, | |
1687 | LUSTRE_MDS_VERSION, | |
1688 | MDS_HSM_CT_UNREGISTER); | |
34e3ff96 | 1689 | if (!req) { |
d5fdc207 JL |
1690 | rc = -ENOMEM; |
1691 | goto out; | |
1692 | } | |
d7e09d03 | 1693 | |
341f1f0a | 1694 | mdc_pack_body(req, NULL, 0, 0, -1, 0); |
d7e09d03 PT |
1695 | |
1696 | ptlrpc_request_set_replen(req); | |
1697 | ||
1698 | rc = mdc_queue_wait(req); | |
d7e09d03 PT |
1699 | out: |
1700 | ptlrpc_req_finished(req); | |
1701 | return rc; | |
1702 | } | |
1703 | ||
1704 | static int mdc_ioc_hsm_state_get(struct obd_export *exp, | |
1705 | struct md_op_data *op_data) | |
1706 | { | |
1707 | struct hsm_user_state *hus = op_data->op_data; | |
1708 | struct hsm_user_state *req_hus; | |
1709 | struct ptlrpc_request *req; | |
1710 | int rc; | |
d7e09d03 PT |
1711 | |
1712 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), | |
1713 | &RQF_MDS_HSM_STATE_GET); | |
34e3ff96 | 1714 | if (!req) |
0a3bdb00 | 1715 | return -ENOMEM; |
d7e09d03 | 1716 | |
d7e09d03 PT |
1717 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET); |
1718 | if (rc != 0) { | |
1719 | ptlrpc_request_free(req); | |
0a3bdb00 | 1720 | return rc; |
d7e09d03 PT |
1721 | } |
1722 | ||
341f1f0a | 1723 | mdc_pack_body(req, &op_data->op_fid1, 0, 0, |
ef2e0f55 | 1724 | op_data->op_suppgids[0], 0); |
d7e09d03 PT |
1725 | |
1726 | ptlrpc_request_set_replen(req); | |
1727 | ||
1728 | rc = mdc_queue_wait(req); | |
1729 | if (rc) | |
d5fdc207 | 1730 | goto out; |
d7e09d03 PT |
1731 | |
1732 | req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE); | |
34e3ff96 | 1733 | if (!req_hus) { |
d5fdc207 JL |
1734 | rc = -EPROTO; |
1735 | goto out; | |
1736 | } | |
d7e09d03 PT |
1737 | |
1738 | *hus = *req_hus; | |
1739 | ||
d7e09d03 PT |
1740 | out: |
1741 | ptlrpc_req_finished(req); | |
1742 | return rc; | |
1743 | } | |
1744 | ||
1745 | static int mdc_ioc_hsm_state_set(struct obd_export *exp, | |
1746 | struct md_op_data *op_data) | |
1747 | { | |
1748 | struct hsm_state_set *hss = op_data->op_data; | |
1749 | struct hsm_state_set *req_hss; | |
1750 | struct ptlrpc_request *req; | |
1751 | int rc; | |
d7e09d03 PT |
1752 | |
1753 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), | |
1754 | &RQF_MDS_HSM_STATE_SET); | |
34e3ff96 | 1755 | if (!req) |
0a3bdb00 | 1756 | return -ENOMEM; |
d7e09d03 | 1757 | |
d7e09d03 PT |
1758 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET); |
1759 | if (rc) { | |
1760 | ptlrpc_request_free(req); | |
0a3bdb00 | 1761 | return rc; |
d7e09d03 PT |
1762 | } |
1763 | ||
341f1f0a | 1764 | mdc_pack_body(req, &op_data->op_fid1, 0, 0, |
ef2e0f55 | 1765 | op_data->op_suppgids[0], 0); |
d7e09d03 PT |
1766 | |
1767 | /* Copy states */ | |
1768 | req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET); | |
34e3ff96 | 1769 | if (!req_hss) { |
d5fdc207 JL |
1770 | rc = -EPROTO; |
1771 | goto out; | |
1772 | } | |
d7e09d03 PT |
1773 | *req_hss = *hss; |
1774 | ||
1775 | ptlrpc_request_set_replen(req); | |
1776 | ||
1777 | rc = mdc_queue_wait(req); | |
d7e09d03 PT |
1778 | out: |
1779 | ptlrpc_req_finished(req); | |
1780 | return rc; | |
1781 | } | |
1782 | ||
1783 | static int mdc_ioc_hsm_request(struct obd_export *exp, | |
1784 | struct hsm_user_request *hur) | |
1785 | { | |
1786 | struct obd_import *imp = class_exp2cliimp(exp); | |
1787 | struct ptlrpc_request *req; | |
1788 | struct hsm_request *req_hr; | |
1789 | struct hsm_user_item *req_hui; | |
1790 | char *req_opaque; | |
1791 | int rc; | |
d7e09d03 PT |
1792 | |
1793 | req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST); | |
34e3ff96 | 1794 | if (!req) { |
d5fdc207 JL |
1795 | rc = -ENOMEM; |
1796 | goto out; | |
1797 | } | |
d7e09d03 PT |
1798 | |
1799 | req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT, | |
1800 | hur->hur_request.hr_itemcount | |
1801 | * sizeof(struct hsm_user_item)); | |
1802 | req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT, | |
1803 | hur->hur_request.hr_data_len); | |
1804 | ||
1805 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST); | |
1806 | if (rc) { | |
1807 | ptlrpc_request_free(req); | |
0a3bdb00 | 1808 | return rc; |
d7e09d03 PT |
1809 | } |
1810 | ||
341f1f0a | 1811 | mdc_pack_body(req, NULL, 0, 0, -1, 0); |
d7e09d03 PT |
1812 | |
1813 | /* Copy hsm_request struct */ | |
1814 | req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST); | |
34e3ff96 | 1815 | if (!req_hr) { |
d5fdc207 JL |
1816 | rc = -EPROTO; |
1817 | goto out; | |
1818 | } | |
d7e09d03 PT |
1819 | *req_hr = hur->hur_request; |
1820 | ||
1821 | /* Copy hsm_user_item structs */ | |
1822 | req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM); | |
34e3ff96 | 1823 | if (!req_hui) { |
d5fdc207 JL |
1824 | rc = -EPROTO; |
1825 | goto out; | |
1826 | } | |
d7e09d03 PT |
1827 | memcpy(req_hui, hur->hur_user_item, |
1828 | hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item)); | |
1829 | ||
1830 | /* Copy opaque field */ | |
1831 | req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA); | |
34e3ff96 | 1832 | if (!req_opaque) { |
d5fdc207 JL |
1833 | rc = -EPROTO; |
1834 | goto out; | |
1835 | } | |
d7e09d03 PT |
1836 | memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len); |
1837 | ||
1838 | ptlrpc_request_set_replen(req); | |
1839 | ||
1840 | rc = mdc_queue_wait(req); | |
d7e09d03 PT |
1841 | out: |
1842 | ptlrpc_req_finished(req); | |
1843 | return rc; | |
1844 | } | |
1845 | ||
1846 | static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags) | |
1847 | { | |
1848 | struct kuc_hdr *lh = (struct kuc_hdr *)buf; | |
1849 | ||
18e042f0 | 1850 | LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE); |
d7e09d03 PT |
1851 | |
1852 | lh->kuc_magic = KUC_MAGIC; | |
1853 | lh->kuc_transport = KUC_TRANSPORT_CHANGELOG; | |
1854 | lh->kuc_flags = flags; | |
1855 | lh->kuc_msgtype = CL_RECORD; | |
1856 | lh->kuc_msglen = len; | |
1857 | return lh; | |
1858 | } | |
1859 | ||
1860 | #define D_CHANGELOG 0 | |
1861 | ||
1862 | struct changelog_show { | |
1863 | __u64 cs_startrec; | |
1864 | __u32 cs_flags; | |
1865 | struct file *cs_fp; | |
1866 | char *cs_buf; | |
1867 | struct obd_device *cs_obd; | |
1868 | }; | |
1869 | ||
e377988e | 1870 | static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh, |
d7e09d03 PT |
1871 | struct llog_rec_hdr *hdr, void *data) |
1872 | { | |
1873 | struct changelog_show *cs = data; | |
1874 | struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr; | |
1875 | struct kuc_hdr *lh; | |
1876 | int len, rc; | |
d7e09d03 | 1877 | |
e377988e AD |
1878 | if (rec->cr_hdr.lrh_type != CHANGELOG_REC) { |
1879 | rc = -EINVAL; | |
1880 | CERROR("%s: not a changelog rec %x/%d: rc = %d\n", | |
1881 | cs->cs_obd->obd_name, rec->cr_hdr.lrh_type, | |
1882 | rec->cr.cr_type, rc); | |
0a3bdb00 | 1883 | return rc; |
d7e09d03 PT |
1884 | } |
1885 | ||
1886 | if (rec->cr.cr_index < cs->cs_startrec) { | |
1887 | /* Skip entries earlier than what we are interested in */ | |
b0f5aad5 | 1888 | CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n", |
d7e09d03 | 1889 | rec->cr.cr_index, cs->cs_startrec); |
0a3bdb00 | 1890 | return 0; |
d7e09d03 PT |
1891 | } |
1892 | ||
b0f5aad5 | 1893 | CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID |
d7e09d03 PT |
1894 | " %.*s\n", rec->cr.cr_index, rec->cr.cr_type, |
1895 | changelog_type2str(rec->cr.cr_type), rec->cr.cr_time, | |
1896 | rec->cr.cr_flags & CLF_FLAGMASK, | |
1897 | PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), | |
1898 | rec->cr.cr_namelen, changelog_rec_name(&rec->cr)); | |
1899 | ||
1900 | len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen; | |
1901 | ||
1902 | /* Set up the message */ | |
1903 | lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags); | |
1904 | memcpy(lh + 1, &rec->cr, len - sizeof(*lh)); | |
1905 | ||
1906 | rc = libcfs_kkuc_msg_put(cs->cs_fp, lh); | |
301af906 | 1907 | CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len, rc); |
d7e09d03 | 1908 | |
0a3bdb00 | 1909 | return rc; |
d7e09d03 PT |
1910 | } |
1911 | ||
1912 | static int mdc_changelog_send_thread(void *csdata) | |
1913 | { | |
1914 | struct changelog_show *cs = csdata; | |
1915 | struct llog_ctxt *ctxt = NULL; | |
1916 | struct llog_handle *llh = NULL; | |
1917 | struct kuc_hdr *kuch; | |
1918 | int rc; | |
1919 | ||
b0f5aad5 | 1920 | CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n", |
d7e09d03 PT |
1921 | cs->cs_fp, cs->cs_startrec); |
1922 | ||
7b81779d | 1923 | cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS); |
bb144d09 | 1924 | if (!cs->cs_buf) { |
d5fdc207 JL |
1925 | rc = -ENOMEM; |
1926 | goto out; | |
1927 | } | |
d7e09d03 PT |
1928 | |
1929 | /* Set up the remote catalog handle */ | |
1930 | ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT); | |
34e3ff96 | 1931 | if (!ctxt) { |
d5fdc207 JL |
1932 | rc = -ENOENT; |
1933 | goto out; | |
1934 | } | |
d7e09d03 PT |
1935 | rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG, |
1936 | LLOG_OPEN_EXISTS); | |
1937 | if (rc) { | |
1938 | CERROR("%s: fail to open changelog catalog: rc = %d\n", | |
1939 | cs->cs_obd->obd_name, rc); | |
d5fdc207 | 1940 | goto out; |
d7e09d03 PT |
1941 | } |
1942 | rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL); | |
1943 | if (rc) { | |
1944 | CERROR("llog_init_handle failed %d\n", rc); | |
d5fdc207 | 1945 | goto out; |
d7e09d03 PT |
1946 | } |
1947 | ||
e377988e | 1948 | rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0); |
d7e09d03 PT |
1949 | |
1950 | /* Send EOF no matter what our result */ | |
4a87df3e CP |
1951 | kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags); |
1952 | if (kuch) { | |
d7e09d03 PT |
1953 | kuch->kuc_msgtype = CL_EOF; |
1954 | libcfs_kkuc_msg_put(cs->cs_fp, kuch); | |
1955 | } | |
1956 | ||
1957 | out: | |
1958 | fput(cs->cs_fp); | |
1959 | if (llh) | |
1960 | llog_cat_close(NULL, llh); | |
1961 | if (ctxt) | |
1962 | llog_ctxt_put(ctxt); | |
6fa7cbab | 1963 | kfree(cs->cs_buf); |
7b81779d | 1964 | kfree(cs); |
d7e09d03 PT |
1965 | return rc; |
1966 | } | |
1967 | ||
1968 | static int mdc_ioc_changelog_send(struct obd_device *obd, | |
1969 | struct ioc_changelog *icc) | |
1970 | { | |
1971 | struct changelog_show *cs; | |
060c2820 | 1972 | struct task_struct *task; |
d7e09d03 PT |
1973 | int rc; |
1974 | ||
1975 | /* Freed in mdc_changelog_send_thread */ | |
7b81779d | 1976 | cs = kzalloc(sizeof(*cs), GFP_NOFS); |
d7e09d03 PT |
1977 | if (!cs) |
1978 | return -ENOMEM; | |
1979 | ||
1980 | cs->cs_obd = obd; | |
1981 | cs->cs_startrec = icc->icc_recno; | |
1982 | /* matching fput in mdc_changelog_send_thread */ | |
1983 | cs->cs_fp = fget(icc->icc_id); | |
1984 | cs->cs_flags = icc->icc_flags; | |
1985 | ||
1986 | /* | |
1987 | * New thread because we should return to user app before | |
1988 | * writing into our pipe | |
1989 | */ | |
060c2820 JH |
1990 | task = kthread_run(mdc_changelog_send_thread, cs, |
1991 | "mdc_clg_send_thread"); | |
1992 | if (IS_ERR(task)) { | |
1993 | rc = PTR_ERR(task); | |
1994 | CERROR("%s: can't start changelog thread: rc = %d\n", | |
1995 | obd->obd_name, rc); | |
1996 | kfree(cs); | |
1997 | } else { | |
1998 | rc = 0; | |
1999 | CDEBUG(D_CHANGELOG, "%s: started changelog thread\n", | |
2000 | obd->obd_name); | |
d7e09d03 PT |
2001 | } |
2002 | ||
2003 | CERROR("Failed to start changelog thread: %d\n", rc); | |
d7e09d03 PT |
2004 | return rc; |
2005 | } | |
2006 | ||
2007 | static int mdc_ioc_hsm_ct_start(struct obd_export *exp, | |
2008 | struct lustre_kernelcomm *lk); | |
2009 | ||
2010 | static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp, | |
2011 | struct obd_quotactl *oqctl) | |
2012 | { | |
2013 | struct client_obd *cli = &exp->exp_obd->u.cli; | |
2014 | struct ptlrpc_request *req; | |
2015 | struct obd_quotactl *body; | |
2016 | int rc; | |
d7e09d03 PT |
2017 | |
2018 | req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), | |
2019 | &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION, | |
2020 | MDS_QUOTACHECK); | |
34e3ff96 | 2021 | if (!req) |
0a3bdb00 | 2022 | return -ENOMEM; |
d7e09d03 PT |
2023 | |
2024 | body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL); | |
2025 | *body = *oqctl; | |
2026 | ||
2027 | ptlrpc_request_set_replen(req); | |
2028 | ||
2029 | /* the next poll will find -ENODATA, that means quotacheck is | |
1df232ee OD |
2030 | * going on |
2031 | */ | |
d7e09d03 PT |
2032 | cli->cl_qchk_stat = -ENODATA; |
2033 | rc = ptlrpc_queue_wait(req); | |
2034 | if (rc) | |
2035 | cli->cl_qchk_stat = rc; | |
2036 | ptlrpc_req_finished(req); | |
0a3bdb00 | 2037 | return rc; |
d7e09d03 PT |
2038 | } |
2039 | ||
2040 | static int mdc_quota_poll_check(struct obd_export *exp, | |
2041 | struct if_quotacheck *qchk) | |
2042 | { | |
2043 | struct client_obd *cli = &exp->exp_obd->u.cli; | |
2044 | int rc; | |
d7e09d03 PT |
2045 | |
2046 | qchk->obd_uuid = cli->cl_target_uuid; | |
2047 | memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME)); | |
2048 | ||
2049 | rc = cli->cl_qchk_stat; | |
2050 | /* the client is not the previous one */ | |
2051 | if (rc == CL_NOT_QUOTACHECKED) | |
2052 | rc = -EINTR; | |
0a3bdb00 | 2053 | return rc; |
d7e09d03 PT |
2054 | } |
2055 | ||
2056 | static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp, | |
2057 | struct obd_quotactl *oqctl) | |
2058 | { | |
2059 | struct ptlrpc_request *req; | |
2060 | struct obd_quotactl *oqc; | |
2061 | int rc; | |
d7e09d03 PT |
2062 | |
2063 | req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), | |
2064 | &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION, | |
2065 | MDS_QUOTACTL); | |
34e3ff96 | 2066 | if (!req) |
0a3bdb00 | 2067 | return -ENOMEM; |
d7e09d03 PT |
2068 | |
2069 | oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL); | |
2070 | *oqc = *oqctl; | |
2071 | ||
2072 | ptlrpc_request_set_replen(req); | |
2073 | ptlrpc_at_set_req_timeout(req); | |
2074 | req->rq_no_resend = 1; | |
2075 | ||
2076 | rc = ptlrpc_queue_wait(req); | |
2077 | if (rc) | |
2078 | CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc); | |
2079 | ||
4a87df3e CP |
2080 | if (req->rq_repmsg) { |
2081 | oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL); | |
2082 | if (oqc) { | |
2083 | *oqctl = *oqc; | |
2084 | } else if (!rc) { | |
ffdac6ce | 2085 | CERROR("Can't unpack obd_quotactl\n"); |
4a87df3e CP |
2086 | rc = -EPROTO; |
2087 | } | |
d7e09d03 | 2088 | } else if (!rc) { |
4a87df3e | 2089 | CERROR("Can't unpack obd_quotactl\n"); |
d7e09d03 PT |
2090 | rc = -EPROTO; |
2091 | } | |
2092 | ptlrpc_req_finished(req); | |
2093 | ||
0a3bdb00 | 2094 | return rc; |
d7e09d03 PT |
2095 | } |
2096 | ||
2097 | static int mdc_ioc_swap_layouts(struct obd_export *exp, | |
2098 | struct md_op_data *op_data) | |
2099 | { | |
2100 | LIST_HEAD(cancels); | |
2101 | struct ptlrpc_request *req; | |
2102 | int rc, count; | |
2103 | struct mdc_swap_layouts *msl, *payload; | |
d7e09d03 PT |
2104 | |
2105 | msl = op_data->op_data; | |
2106 | ||
2107 | /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the | |
2108 | * first thing it will do is to cancel the 2 layout | |
2109 | * locks hold by this client. | |
2110 | * So the client must cancel its layout locks on the 2 fids | |
2111 | * with the request RPC to avoid extra RPC round trips | |
2112 | */ | |
2113 | count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels, | |
29792c81 JH |
2114 | LCK_CR, MDS_INODELOCK_LAYOUT | |
2115 | MDS_INODELOCK_XATTR); | |
d7e09d03 | 2116 | count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels, |
29792c81 JH |
2117 | LCK_CR, MDS_INODELOCK_LAYOUT | |
2118 | MDS_INODELOCK_XATTR); | |
d7e09d03 PT |
2119 | |
2120 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), | |
2121 | &RQF_MDS_SWAP_LAYOUTS); | |
34e3ff96 | 2122 | if (!req) { |
d7e09d03 | 2123 | ldlm_lock_list_put(&cancels, l_bl_ast, count); |
0a3bdb00 | 2124 | return -ENOMEM; |
d7e09d03 PT |
2125 | } |
2126 | ||
d7e09d03 PT |
2127 | rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count); |
2128 | if (rc) { | |
2129 | ptlrpc_request_free(req); | |
0a3bdb00 | 2130 | return rc; |
d7e09d03 PT |
2131 | } |
2132 | ||
2133 | mdc_swap_layouts_pack(req, op_data); | |
2134 | ||
2135 | payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS); | |
2136 | LASSERT(payload); | |
2137 | ||
2138 | *payload = *msl; | |
2139 | ||
2140 | ptlrpc_request_set_replen(req); | |
2141 | ||
2142 | rc = ptlrpc_queue_wait(req); | |
d7e09d03 | 2143 | |
d7e09d03 PT |
2144 | ptlrpc_req_finished(req); |
2145 | return rc; | |
2146 | } | |
2147 | ||
2148 | static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len, | |
e09bee34 | 2149 | void *karg, void __user *uarg) |
d7e09d03 PT |
2150 | { |
2151 | struct obd_device *obd = exp->exp_obd; | |
2152 | struct obd_ioctl_data *data = karg; | |
2153 | struct obd_import *imp = obd->u.cli.cl_import; | |
d7e09d03 | 2154 | int rc; |
d7e09d03 PT |
2155 | |
2156 | if (!try_module_get(THIS_MODULE)) { | |
19b2056f JN |
2157 | CERROR("%s: cannot get module '%s'\n", obd->obd_name, |
2158 | module_name(THIS_MODULE)); | |
d7e09d03 PT |
2159 | return -EINVAL; |
2160 | } | |
2161 | switch (cmd) { | |
2162 | case OBD_IOC_CHANGELOG_SEND: | |
2163 | rc = mdc_ioc_changelog_send(obd, karg); | |
d5fdc207 | 2164 | goto out; |
d7e09d03 PT |
2165 | case OBD_IOC_CHANGELOG_CLEAR: { |
2166 | struct ioc_changelog *icc = karg; | |
1a4cd3e9 SM |
2167 | struct changelog_setinfo cs = { |
2168 | .cs_recno = icc->icc_recno, | |
2169 | .cs_id = icc->icc_id | |
2170 | }; | |
2171 | ||
d7e09d03 PT |
2172 | rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR), |
2173 | KEY_CHANGELOG_CLEAR, sizeof(cs), &cs, | |
2174 | NULL); | |
d5fdc207 | 2175 | goto out; |
d7e09d03 PT |
2176 | } |
2177 | case OBD_IOC_FID2PATH: | |
2178 | rc = mdc_ioc_fid2path(exp, karg); | |
d5fdc207 | 2179 | goto out; |
d7e09d03 PT |
2180 | case LL_IOC_HSM_CT_START: |
2181 | rc = mdc_ioc_hsm_ct_start(exp, karg); | |
78eb9092 TL |
2182 | /* ignore if it was already registered on this MDS. */ |
2183 | if (rc == -EEXIST) | |
2184 | rc = 0; | |
d5fdc207 | 2185 | goto out; |
d7e09d03 PT |
2186 | case LL_IOC_HSM_PROGRESS: |
2187 | rc = mdc_ioc_hsm_progress(exp, karg); | |
d5fdc207 | 2188 | goto out; |
d7e09d03 PT |
2189 | case LL_IOC_HSM_STATE_GET: |
2190 | rc = mdc_ioc_hsm_state_get(exp, karg); | |
d5fdc207 | 2191 | goto out; |
d7e09d03 PT |
2192 | case LL_IOC_HSM_STATE_SET: |
2193 | rc = mdc_ioc_hsm_state_set(exp, karg); | |
d5fdc207 | 2194 | goto out; |
d7e09d03 PT |
2195 | case LL_IOC_HSM_ACTION: |
2196 | rc = mdc_ioc_hsm_current_action(exp, karg); | |
d5fdc207 | 2197 | goto out; |
d7e09d03 PT |
2198 | case LL_IOC_HSM_REQUEST: |
2199 | rc = mdc_ioc_hsm_request(exp, karg); | |
d5fdc207 | 2200 | goto out; |
d7e09d03 PT |
2201 | case OBD_IOC_CLIENT_RECOVER: |
2202 | rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0); | |
2203 | if (rc < 0) | |
d5fdc207 JL |
2204 | goto out; |
2205 | rc = 0; | |
2206 | goto out; | |
d7e09d03 PT |
2207 | case IOC_OSC_SET_ACTIVE: |
2208 | rc = ptlrpc_set_import_active(imp, data->ioc_offset); | |
d5fdc207 | 2209 | goto out; |
d7e09d03 PT |
2210 | case OBD_IOC_POLL_QUOTACHECK: |
2211 | rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg); | |
d5fdc207 | 2212 | goto out; |
d7e09d03 PT |
2213 | case OBD_IOC_PING_TARGET: |
2214 | rc = ptlrpc_obd_ping(obd); | |
d5fdc207 | 2215 | goto out; |
d7e09d03 PT |
2216 | /* |
2217 | * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by | |
2218 | * LMV instead of MDC. But when the cluster is upgraded from 1.8, | |
2219 | * there'd be no LMV layer thus we might be called here. Eventually | |
2220 | * this code should be removed. | |
2221 | * bz20731, LU-592. | |
2222 | */ | |
2223 | case IOC_OBD_STATFS: { | |
2224 | struct obd_statfs stat_buf = {0}; | |
2225 | ||
9797fb0e | 2226 | if (*((__u32 *)data->ioc_inlbuf2) != 0) { |
d5fdc207 JL |
2227 | rc = -ENODEV; |
2228 | goto out; | |
2229 | } | |
d7e09d03 PT |
2230 | |
2231 | /* copy UUID */ | |
2232 | if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd), | |
ef4356bf | 2233 | min_t(size_t, data->ioc_plen2, |
22e0bc6a | 2234 | sizeof(struct obd_uuid)))) { |
d5fdc207 JL |
2235 | rc = -EFAULT; |
2236 | goto out; | |
2237 | } | |
d7e09d03 PT |
2238 | |
2239 | rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf, | |
2240 | cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), | |
2241 | 0); | |
2242 | if (rc != 0) | |
d5fdc207 | 2243 | goto out; |
d7e09d03 PT |
2244 | |
2245 | if (copy_to_user(data->ioc_pbuf1, &stat_buf, | |
ef4356bf | 2246 | min_t(size_t, data->ioc_plen1, |
22e0bc6a | 2247 | sizeof(stat_buf)))) { |
d5fdc207 JL |
2248 | rc = -EFAULT; |
2249 | goto out; | |
2250 | } | |
d7e09d03 | 2251 | |
d5fdc207 JL |
2252 | rc = 0; |
2253 | goto out; | |
d7e09d03 PT |
2254 | } |
2255 | case OBD_IOC_QUOTACTL: { | |
2256 | struct if_quotactl *qctl = karg; | |
2257 | struct obd_quotactl *oqctl; | |
2258 | ||
7b81779d | 2259 | oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS); |
bb144d09 | 2260 | if (!oqctl) { |
d5fdc207 JL |
2261 | rc = -ENOMEM; |
2262 | goto out; | |
2263 | } | |
d7e09d03 PT |
2264 | |
2265 | QCTL_COPY(oqctl, qctl); | |
2266 | rc = obd_quotactl(exp, oqctl); | |
2267 | if (rc == 0) { | |
2268 | QCTL_COPY(qctl, oqctl); | |
2269 | qctl->qc_valid = QC_MDTIDX; | |
2270 | qctl->obd_uuid = obd->u.cli.cl_target_uuid; | |
2271 | } | |
c1f3d689 | 2272 | |
7b81779d | 2273 | kfree(oqctl); |
d5fdc207 | 2274 | goto out; |
d7e09d03 | 2275 | } |
c1f3d689 JH |
2276 | case LL_IOC_GET_CONNECT_FLAGS: |
2277 | if (copy_to_user(uarg, exp_connect_flags_ptr(exp), | |
d5fdc207 JL |
2278 | sizeof(*exp_connect_flags_ptr(exp)))) { |
2279 | rc = -EFAULT; | |
2280 | goto out; | |
2281 | } | |
c1f3d689 | 2282 | |
d5fdc207 JL |
2283 | rc = 0; |
2284 | goto out; | |
c1f3d689 | 2285 | case LL_IOC_LOV_SWAP_LAYOUTS: |
d7e09d03 | 2286 | rc = mdc_ioc_swap_layouts(exp, karg); |
d5fdc207 | 2287 | goto out; |
d7e09d03 | 2288 | default: |
c1f3d689 | 2289 | CERROR("unrecognised ioctl: cmd = %#x\n", cmd); |
d5fdc207 JL |
2290 | rc = -ENOTTY; |
2291 | goto out; | |
d7e09d03 PT |
2292 | } |
2293 | out: | |
2294 | module_put(THIS_MODULE); | |
2295 | ||
2296 | return rc; | |
2297 | } | |
2298 | ||
f6219c16 LC |
2299 | static int mdc_get_info_rpc(struct obd_export *exp, |
2300 | u32 keylen, void *key, | |
2301 | int vallen, void *val) | |
d7e09d03 PT |
2302 | { |
2303 | struct obd_import *imp = class_exp2cliimp(exp); | |
2304 | struct ptlrpc_request *req; | |
2305 | char *tmp; | |
2306 | int rc = -EINVAL; | |
d7e09d03 PT |
2307 | |
2308 | req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO); | |
34e3ff96 | 2309 | if (!req) |
0a3bdb00 | 2310 | return -ENOMEM; |
d7e09d03 PT |
2311 | |
2312 | req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY, | |
2313 | RCL_CLIENT, keylen); | |
2314 | req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN, | |
2315 | RCL_CLIENT, sizeof(__u32)); | |
2316 | ||
2317 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO); | |
2318 | if (rc) { | |
2319 | ptlrpc_request_free(req); | |
0a3bdb00 | 2320 | return rc; |
d7e09d03 PT |
2321 | } |
2322 | ||
2323 | tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY); | |
2324 | memcpy(tmp, key, keylen); | |
2325 | tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN); | |
2326 | memcpy(tmp, &vallen, sizeof(__u32)); | |
2327 | ||
2328 | req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL, | |
2329 | RCL_SERVER, vallen); | |
2330 | ptlrpc_request_set_replen(req); | |
2331 | ||
2332 | rc = ptlrpc_queue_wait(req); | |
2333 | /* -EREMOTE means the get_info result is partial, and it needs to | |
1df232ee OD |
2334 | * continue on another MDT, see fid2path part in lmv_iocontrol |
2335 | */ | |
d7e09d03 PT |
2336 | if (rc == 0 || rc == -EREMOTE) { |
2337 | tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL); | |
2338 | memcpy(val, tmp, vallen); | |
2339 | if (ptlrpc_rep_need_swab(req)) { | |
2340 | if (KEY_IS(KEY_FID2PATH)) | |
2341 | lustre_swab_fid2path(val); | |
2342 | } | |
2343 | } | |
2344 | ptlrpc_req_finished(req); | |
2345 | ||
0a3bdb00 | 2346 | return rc; |
d7e09d03 PT |
2347 | } |
2348 | ||
2349 | static void lustre_swab_hai(struct hsm_action_item *h) | |
2350 | { | |
2351 | __swab32s(&h->hai_len); | |
2352 | __swab32s(&h->hai_action); | |
2353 | lustre_swab_lu_fid(&h->hai_fid); | |
2354 | lustre_swab_lu_fid(&h->hai_dfid); | |
2355 | __swab64s(&h->hai_cookie); | |
2356 | __swab64s(&h->hai_extent.offset); | |
2357 | __swab64s(&h->hai_extent.length); | |
2358 | __swab64s(&h->hai_gid); | |
2359 | } | |
2360 | ||
2361 | static void lustre_swab_hal(struct hsm_action_list *h) | |
2362 | { | |
2363 | struct hsm_action_item *hai; | |
2364 | int i; | |
2365 | ||
2366 | __swab32s(&h->hal_version); | |
2367 | __swab32s(&h->hal_count); | |
2368 | __swab32s(&h->hal_archive_id); | |
2369 | __swab64s(&h->hal_flags); | |
18ecab0d | 2370 | hai = hai_first(h); |
18dfaebf | 2371 | for (i = 0; i < h->hal_count; i++, hai = hai_next(hai)) |
d7e09d03 | 2372 | lustre_swab_hai(hai); |
d7e09d03 PT |
2373 | } |
2374 | ||
2375 | static void lustre_swab_kuch(struct kuc_hdr *l) | |
2376 | { | |
2377 | __swab16s(&l->kuc_magic); | |
2378 | /* __u8 l->kuc_transport */ | |
2379 | __swab16s(&l->kuc_msgtype); | |
2380 | __swab16s(&l->kuc_msglen); | |
2381 | } | |
2382 | ||
2383 | static int mdc_ioc_hsm_ct_start(struct obd_export *exp, | |
2384 | struct lustre_kernelcomm *lk) | |
2385 | { | |
2386 | struct obd_import *imp = class_exp2cliimp(exp); | |
2387 | __u32 archive = lk->lk_data; | |
2388 | int rc = 0; | |
2389 | ||
2390 | if (lk->lk_group != KUC_GRP_HSM) { | |
2391 | CERROR("Bad copytool group %d\n", lk->lk_group); | |
2392 | return -EINVAL; | |
2393 | } | |
2394 | ||
2395 | CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd, | |
2396 | lk->lk_uid, lk->lk_group, lk->lk_flags); | |
2397 | ||
2398 | if (lk->lk_flags & LK_FLG_STOP) { | |
d7e09d03 | 2399 | /* Unregister with the coordinator */ |
78eb9092 | 2400 | rc = mdc_ioc_hsm_ct_unregister(imp); |
d7e09d03 | 2401 | } else { |
78eb9092 | 2402 | rc = mdc_ioc_hsm_ct_register(imp, archive); |
d7e09d03 PT |
2403 | } |
2404 | ||
2405 | return rc; | |
2406 | } | |
2407 | ||
2408 | /** | |
2409 | * Send a message to any listening copytools | |
2410 | * @param val KUC message (kuc_hdr + hsm_action_list) | |
2411 | * @param len total length of message | |
2412 | */ | |
2413 | static int mdc_hsm_copytool_send(int len, void *val) | |
2414 | { | |
2415 | struct kuc_hdr *lh = (struct kuc_hdr *)val; | |
2416 | struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1); | |
d7e09d03 PT |
2417 | |
2418 | if (len < sizeof(*lh) + sizeof(*hal)) { | |
2419 | CERROR("Short HSM message %d < %d\n", len, | |
9797fb0e | 2420 | (int)(sizeof(*lh) + sizeof(*hal))); |
0a3bdb00 | 2421 | return -EPROTO; |
d7e09d03 PT |
2422 | } |
2423 | if (lh->kuc_magic == __swab16(KUC_MAGIC)) { | |
2424 | lustre_swab_kuch(lh); | |
2425 | lustre_swab_hal(hal); | |
2426 | } else if (lh->kuc_magic != KUC_MAGIC) { | |
2427 | CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC); | |
0a3bdb00 | 2428 | return -EPROTO; |
d7e09d03 PT |
2429 | } |
2430 | ||
ee990b33 SM |
2431 | CDEBUG(D_HSM, |
2432 | "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n", | |
d7e09d03 PT |
2433 | lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype, |
2434 | lh->kuc_msglen, hal->hal_count, hal->hal_fsname); | |
2435 | ||
2436 | /* Broadcast to HSM listeners */ | |
da2a7272 | 2437 | return libcfs_kkuc_group_put(KUC_GRP_HSM, lh); |
d7e09d03 PT |
2438 | } |
2439 | ||
2440 | /** | |
2441 | * callback function passed to kuc for re-registering each HSM copytool | |
2442 | * running on MDC, after MDT shutdown/recovery. | |
406fc913 | 2443 | * @param data copytool registration data |
d7e09d03 PT |
2444 | * @param cb_arg callback argument (obd_import) |
2445 | */ | |
406fc913 | 2446 | static int mdc_hsm_ct_reregister(void *data, void *cb_arg) |
d7e09d03 | 2447 | { |
406fc913 | 2448 | struct kkuc_ct_data *kcd = data; |
d7e09d03 | 2449 | struct obd_import *imp = (struct obd_import *)cb_arg; |
d7e09d03 PT |
2450 | int rc; |
2451 | ||
406fc913 HD |
2452 | if (!kcd || kcd->kcd_magic != KKUC_CT_DATA_MAGIC) |
2453 | return -EPROTO; | |
2454 | ||
2455 | if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid)) | |
2456 | return 0; | |
2457 | ||
2458 | CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n", | |
2459 | imp->imp_obd->obd_name, kcd->kcd_archive); | |
2460 | rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive); | |
d7e09d03 PT |
2461 | |
2462 | /* ignore error if the copytool is already registered */ | |
406fc913 | 2463 | return (rc == -EEXIST) ? 0 : rc; |
d7e09d03 PT |
2464 | } |
2465 | ||
f6219c16 LC |
2466 | static int mdc_set_info_async(const struct lu_env *env, |
2467 | struct obd_export *exp, | |
2468 | u32 keylen, void *key, | |
2469 | u32 vallen, void *val, | |
2470 | struct ptlrpc_request_set *set) | |
d7e09d03 PT |
2471 | { |
2472 | struct obd_import *imp = class_exp2cliimp(exp); | |
2473 | int rc; | |
d7e09d03 PT |
2474 | |
2475 | if (KEY_IS(KEY_READ_ONLY)) { | |
2476 | if (vallen != sizeof(int)) | |
0a3bdb00 | 2477 | return -EINVAL; |
d7e09d03 PT |
2478 | |
2479 | spin_lock(&imp->imp_lock); | |
2480 | if (*((int *)val)) { | |
2481 | imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY; | |
2482 | imp->imp_connect_data.ocd_connect_flags |= | |
2483 | OBD_CONNECT_RDONLY; | |
2484 | } else { | |
2485 | imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY; | |
2486 | imp->imp_connect_data.ocd_connect_flags &= | |
2487 | ~OBD_CONNECT_RDONLY; | |
2488 | } | |
2489 | spin_unlock(&imp->imp_lock); | |
2490 | ||
2491 | rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION, | |
2492 | keylen, key, vallen, val, set); | |
0a3bdb00 | 2493 | return rc; |
d7e09d03 PT |
2494 | } |
2495 | if (KEY_IS(KEY_SPTLRPC_CONF)) { | |
2496 | sptlrpc_conf_client_adapt(exp->exp_obd); | |
0a3bdb00 | 2497 | return 0; |
d7e09d03 PT |
2498 | } |
2499 | if (KEY_IS(KEY_FLUSH_CTX)) { | |
2500 | sptlrpc_import_flush_my_ctx(imp); | |
0a3bdb00 | 2501 | return 0; |
d7e09d03 | 2502 | } |
d7e09d03 PT |
2503 | if (KEY_IS(KEY_CHANGELOG_CLEAR)) { |
2504 | rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION, | |
2505 | keylen, key, vallen, val, set); | |
0a3bdb00 | 2506 | return rc; |
d7e09d03 PT |
2507 | } |
2508 | if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) { | |
2509 | rc = mdc_hsm_copytool_send(vallen, val); | |
0a3bdb00 | 2510 | return rc; |
d7e09d03 PT |
2511 | } |
2512 | ||
2513 | CERROR("Unknown key %s\n", (char *)key); | |
0a3bdb00 | 2514 | return -EINVAL; |
d7e09d03 PT |
2515 | } |
2516 | ||
f6219c16 LC |
2517 | static int mdc_get_info(const struct lu_env *env, struct obd_export *exp, |
2518 | __u32 keylen, void *key, __u32 *vallen, void *val, | |
2519 | struct lov_stripe_md *lsm) | |
d7e09d03 PT |
2520 | { |
2521 | int rc = -EINVAL; | |
2522 | ||
2523 | if (KEY_IS(KEY_MAX_EASIZE)) { | |
2524 | int mdsize, *max_easize; | |
2525 | ||
2526 | if (*vallen != sizeof(int)) | |
0a3bdb00 | 2527 | return -EINVAL; |
44779340 | 2528 | mdsize = *(int *)val; |
d7e09d03 PT |
2529 | if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize) |
2530 | exp->exp_obd->u.cli.cl_max_mds_easize = mdsize; | |
2531 | max_easize = val; | |
2532 | *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize; | |
0a3bdb00 | 2533 | return 0; |
44779340 BB |
2534 | } else if (KEY_IS(KEY_DEFAULT_EASIZE)) { |
2535 | int *default_easize; | |
2536 | ||
2537 | if (*vallen != sizeof(int)) | |
2538 | return -EINVAL; | |
2539 | default_easize = val; | |
2540 | *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize; | |
2541 | return 0; | |
d7e09d03 PT |
2542 | } else if (KEY_IS(KEY_CONN_DATA)) { |
2543 | struct obd_import *imp = class_exp2cliimp(exp); | |
2544 | struct obd_connect_data *data = val; | |
2545 | ||
2546 | if (*vallen != sizeof(*data)) | |
0a3bdb00 | 2547 | return -EINVAL; |
d7e09d03 PT |
2548 | |
2549 | *data = imp->imp_connect_data; | |
0a3bdb00 | 2550 | return 0; |
d7e09d03 PT |
2551 | } else if (KEY_IS(KEY_TGT_COUNT)) { |
2552 | *((int *)val) = 1; | |
0a3bdb00 | 2553 | return 0; |
d7e09d03 PT |
2554 | } |
2555 | ||
2556 | rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val); | |
2557 | ||
0a3bdb00 | 2558 | return rc; |
d7e09d03 PT |
2559 | } |
2560 | ||
f6219c16 | 2561 | static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid, |
ef2e0f55 | 2562 | struct ptlrpc_request **request) |
d7e09d03 PT |
2563 | { |
2564 | struct ptlrpc_request *req; | |
2565 | int rc; | |
d7e09d03 PT |
2566 | |
2567 | *request = NULL; | |
2568 | req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC); | |
34e3ff96 | 2569 | if (!req) |
0a3bdb00 | 2570 | return -ENOMEM; |
d7e09d03 | 2571 | |
d7e09d03 PT |
2572 | rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC); |
2573 | if (rc) { | |
2574 | ptlrpc_request_free(req); | |
0a3bdb00 | 2575 | return rc; |
d7e09d03 PT |
2576 | } |
2577 | ||
ef2e0f55 | 2578 | mdc_pack_body(req, fid, 0, 0, -1, 0); |
d7e09d03 PT |
2579 | |
2580 | ptlrpc_request_set_replen(req); | |
2581 | ||
2582 | rc = ptlrpc_queue_wait(req); | |
2583 | if (rc) | |
2584 | ptlrpc_req_finished(req); | |
2585 | else | |
2586 | *request = req; | |
0a3bdb00 | 2587 | return rc; |
d7e09d03 PT |
2588 | } |
2589 | ||
2590 | static int mdc_import_event(struct obd_device *obd, struct obd_import *imp, | |
2591 | enum obd_import_event event) | |
2592 | { | |
2593 | int rc = 0; | |
2594 | ||
2595 | LASSERT(imp->imp_obd == obd); | |
2596 | ||
2597 | switch (event) { | |
2598 | case IMP_EVENT_DISCON: { | |
2599 | #if 0 | |
2600 | /* XXX Pass event up to OBDs stack. used only for FLD now */ | |
2601 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL); | |
2602 | #endif | |
2603 | break; | |
2604 | } | |
2605 | case IMP_EVENT_INACTIVE: { | |
2606 | struct client_obd *cli = &obd->u.cli; | |
2607 | /* | |
2608 | * Flush current sequence to make client obtain new one | |
2609 | * from server in case of disconnect/reconnect. | |
2610 | */ | |
34e3ff96 | 2611 | if (cli->cl_seq) |
d7e09d03 PT |
2612 | seq_client_flush(cli->cl_seq); |
2613 | ||
2614 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL); | |
2615 | break; | |
2616 | } | |
2617 | case IMP_EVENT_INVALIDATE: { | |
2618 | struct ldlm_namespace *ns = obd->obd_namespace; | |
2619 | ||
2620 | ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY); | |
2621 | ||
2622 | break; | |
2623 | } | |
2624 | case IMP_EVENT_ACTIVE: | |
2625 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL); | |
78eb9092 | 2626 | /* redo the kuc registration after reconnecting */ |
d7e09d03 | 2627 | if (rc == 0) |
dd40ca46 SB |
2628 | /* re-register HSM agents */ |
2629 | rc = libcfs_kkuc_group_foreach(KUC_GRP_HSM, | |
2630 | mdc_hsm_ct_reregister, | |
2631 | (void *)imp); | |
d7e09d03 PT |
2632 | break; |
2633 | case IMP_EVENT_OCD: | |
2634 | rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL); | |
2635 | break; | |
2636 | case IMP_EVENT_DEACTIVATE: | |
2637 | case IMP_EVENT_ACTIVATE: | |
2638 | break; | |
2639 | default: | |
2640 | CERROR("Unknown import event %x\n", event); | |
2641 | LBUG(); | |
2642 | } | |
0a3bdb00 | 2643 | return rc; |
d7e09d03 PT |
2644 | } |
2645 | ||
8f18c8a4 | 2646 | int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp, |
2647 | struct lu_fid *fid, struct md_op_data *op_data) | |
d7e09d03 PT |
2648 | { |
2649 | struct client_obd *cli = &exp->exp_obd->u.cli; | |
2650 | struct lu_client_seq *seq = cli->cl_seq; | |
29aaf496 | 2651 | |
8f18c8a4 | 2652 | return seq_client_alloc_fid(env, seq, fid); |
d7e09d03 PT |
2653 | } |
2654 | ||
f6219c16 | 2655 | static struct obd_uuid *mdc_get_uuid(struct obd_export *exp) |
982ec91f | 2656 | { |
d7e09d03 | 2657 | struct client_obd *cli = &exp->exp_obd->u.cli; |
7436d070 | 2658 | |
d7e09d03 PT |
2659 | return &cli->cl_target_uuid; |
2660 | } | |
2661 | ||
2662 | /** | |
2663 | * Determine whether the lock can be canceled before replaying it during | |
2664 | * recovery, non zero value will be return if the lock can be canceled, | |
2665 | * or zero returned for not | |
2666 | */ | |
7d443334 | 2667 | static int mdc_cancel_weight(struct ldlm_lock *lock) |
d7e09d03 PT |
2668 | { |
2669 | if (lock->l_resource->lr_type != LDLM_IBITS) | |
0a3bdb00 | 2670 | return 0; |
d7e09d03 PT |
2671 | |
2672 | /* FIXME: if we ever get into a situation where there are too many | |
2673 | * opened files with open locks on a single node, then we really | |
1df232ee OD |
2674 | * should replay these open locks to reget it |
2675 | */ | |
d7e09d03 | 2676 | if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN) |
0a3bdb00 | 2677 | return 0; |
d7e09d03 | 2678 | |
0a3bdb00 | 2679 | return 1; |
d7e09d03 PT |
2680 | } |
2681 | ||
2682 | static int mdc_resource_inode_free(struct ldlm_resource *res) | |
2683 | { | |
2684 | if (res->lr_lvb_inode) | |
2685 | res->lr_lvb_inode = NULL; | |
2686 | ||
2687 | return 0; | |
2688 | } | |
2689 | ||
f6219c16 | 2690 | static struct ldlm_valblock_ops inode_lvbo = { |
805e517a | 2691 | .lvbo_free = mdc_resource_inode_free, |
d7e09d03 PT |
2692 | }; |
2693 | ||
903af118 JH |
2694 | static int mdc_llog_init(struct obd_device *obd) |
2695 | { | |
2696 | struct obd_llog_group *olg = &obd->obd_olg; | |
2697 | struct llog_ctxt *ctxt; | |
2698 | int rc; | |
2699 | ||
2700 | rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd, | |
2701 | &llog_client_ops); | |
2702 | if (rc) | |
2703 | return rc; | |
2704 | ||
2705 | ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT); | |
2706 | llog_initiator_connect(ctxt); | |
2707 | llog_ctxt_put(ctxt); | |
2708 | ||
2709 | return 0; | |
2710 | } | |
2711 | ||
2712 | static void mdc_llog_finish(struct obd_device *obd) | |
2713 | { | |
2714 | struct llog_ctxt *ctxt; | |
2715 | ||
2716 | ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT); | |
2717 | if (ctxt) | |
2718 | llog_cleanup(NULL, ctxt); | |
2719 | } | |
2720 | ||
d7e09d03 PT |
2721 | static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg) |
2722 | { | |
2723 | struct client_obd *cli = &obd->u.cli; | |
ea7893bb | 2724 | struct lprocfs_static_vars lvars = { NULL }; |
d7e09d03 | 2725 | int rc; |
d7e09d03 | 2726 | |
7b81779d | 2727 | cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS); |
d7e09d03 | 2728 | if (!cli->cl_rpc_lock) |
0a3bdb00 | 2729 | return -ENOMEM; |
d7e09d03 PT |
2730 | mdc_init_rpc_lock(cli->cl_rpc_lock); |
2731 | ||
cc3b7758 SP |
2732 | rc = ptlrpcd_addref(); |
2733 | if (rc < 0) | |
2734 | goto err_rpc_lock; | |
d7e09d03 | 2735 | |
7b81779d | 2736 | cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS); |
d5fdc207 JL |
2737 | if (!cli->cl_close_lock) { |
2738 | rc = -ENOMEM; | |
cc3b7758 | 2739 | goto err_ptlrpcd_decref; |
d5fdc207 | 2740 | } |
d7e09d03 PT |
2741 | mdc_init_rpc_lock(cli->cl_close_lock); |
2742 | ||
2743 | rc = client_obd_setup(obd, cfg); | |
2744 | if (rc) | |
d5fdc207 | 2745 | goto err_close_lock; |
d7e09d03 | 2746 | lprocfs_mdc_init_vars(&lvars); |
9b801302 | 2747 | lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars); |
d7e09d03 PT |
2748 | sptlrpc_lprocfs_cliobd_attach(obd); |
2749 | ptlrpc_lprocfs_register_obd(obd); | |
2750 | ||
7d443334 | 2751 | ns_register_cancel(obd->obd_namespace, mdc_cancel_weight); |
d7e09d03 PT |
2752 | |
2753 | obd->obd_namespace->ns_lvbo = &inode_lvbo; | |
2754 | ||
903af118 | 2755 | rc = mdc_llog_init(obd); |
d7e09d03 PT |
2756 | if (rc) { |
2757 | mdc_cleanup(obd); | |
2758 | CERROR("failed to setup llogging subsystems\n"); | |
2759 | } | |
2760 | ||
0a3bdb00 | 2761 | return rc; |
d7e09d03 PT |
2762 | |
2763 | err_close_lock: | |
7b81779d | 2764 | kfree(cli->cl_close_lock); |
cc3b7758 SP |
2765 | err_ptlrpcd_decref: |
2766 | ptlrpcd_decref(); | |
d7e09d03 | 2767 | err_rpc_lock: |
7b81779d | 2768 | kfree(cli->cl_rpc_lock); |
0a3bdb00 | 2769 | return rc; |
d7e09d03 PT |
2770 | } |
2771 | ||
2772 | /* Initialize the default and maximum LOV EA and cookie sizes. This allows | |
44779340 BB |
2773 | * us to make MDS RPCs with large enough reply buffers to hold a default |
2774 | * sized EA and cookie without having to calculate this (via a call into the | |
2775 | * LOV + OSCs) each time we make an RPC. The maximum size is also tracked | |
2776 | * but not used to avoid wastefully vmalloc()'ing large reply buffers when | |
2777 | * a large number of stripes is possible. If a larger reply buffer is | |
2778 | * required it will be reallocated in the ptlrpc layer due to overflow. | |
2779 | */ | |
d7e09d03 | 2780 | static int mdc_init_ea_size(struct obd_export *exp, int easize, |
44779340 | 2781 | int def_easize, int cookiesize, int def_cookiesize) |
d7e09d03 PT |
2782 | { |
2783 | struct obd_device *obd = exp->exp_obd; | |
2784 | struct client_obd *cli = &obd->u.cli; | |
d7e09d03 PT |
2785 | |
2786 | if (cli->cl_max_mds_easize < easize) | |
2787 | cli->cl_max_mds_easize = easize; | |
2788 | ||
2789 | if (cli->cl_default_mds_easize < def_easize) | |
2790 | cli->cl_default_mds_easize = def_easize; | |
2791 | ||
2792 | if (cli->cl_max_mds_cookiesize < cookiesize) | |
2793 | cli->cl_max_mds_cookiesize = cookiesize; | |
2794 | ||
44779340 BB |
2795 | if (cli->cl_default_mds_cookiesize < def_cookiesize) |
2796 | cli->cl_default_mds_cookiesize = def_cookiesize; | |
2797 | ||
0a3bdb00 | 2798 | return 0; |
d7e09d03 PT |
2799 | } |
2800 | ||
2801 | static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) | |
2802 | { | |
d7e09d03 PT |
2803 | switch (stage) { |
2804 | case OBD_CLEANUP_EARLY: | |
2805 | break; | |
2806 | case OBD_CLEANUP_EXPORTS: | |
2807 | /* Failsafe, ok if racy */ | |
2808 | if (obd->obd_type->typ_refcnt <= 1) | |
17328956 | 2809 | libcfs_kkuc_group_rem(0, KUC_GRP_HSM); |
d7e09d03 PT |
2810 | |
2811 | obd_cleanup_client_import(obd); | |
2812 | ptlrpc_lprocfs_unregister_obd(obd); | |
2813 | lprocfs_obd_cleanup(obd); | |
2814 | ||
903af118 | 2815 | mdc_llog_finish(obd); |
d7e09d03 PT |
2816 | break; |
2817 | } | |
eb967126 | 2818 | return 0; |
d7e09d03 PT |
2819 | } |
2820 | ||
2821 | static int mdc_cleanup(struct obd_device *obd) | |
2822 | { | |
2823 | struct client_obd *cli = &obd->u.cli; | |
2824 | ||
7b81779d JL |
2825 | kfree(cli->cl_rpc_lock); |
2826 | kfree(cli->cl_close_lock); | |
d7e09d03 PT |
2827 | |
2828 | ptlrpcd_decref(); | |
2829 | ||
2830 | return client_obd_cleanup(obd); | |
2831 | } | |
2832 | ||
21aef7d9 | 2833 | static int mdc_process_config(struct obd_device *obd, u32 len, void *buf) |
d7e09d03 PT |
2834 | { |
2835 | struct lustre_cfg *lcfg = buf; | |
ea7893bb | 2836 | struct lprocfs_static_vars lvars = { NULL }; |
d7e09d03 PT |
2837 | int rc = 0; |
2838 | ||
2839 | lprocfs_mdc_init_vars(&lvars); | |
2840 | switch (lcfg->lcfg_command) { | |
2841 | default: | |
2842 | rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, | |
2843 | lcfg, obd); | |
2844 | if (rc > 0) | |
2845 | rc = 0; | |
2846 | break; | |
2847 | } | |
eb44520b | 2848 | return rc; |
d7e09d03 PT |
2849 | } |
2850 | ||
f6219c16 | 2851 | static struct obd_ops mdc_obd_ops = { |
a13b1f32 DC |
2852 | .owner = THIS_MODULE, |
2853 | .setup = mdc_setup, | |
2854 | .precleanup = mdc_precleanup, | |
2855 | .cleanup = mdc_cleanup, | |
2856 | .add_conn = client_import_add_conn, | |
2857 | .del_conn = client_import_del_conn, | |
2858 | .connect = client_connect_import, | |
2859 | .disconnect = client_disconnect_export, | |
2860 | .iocontrol = mdc_iocontrol, | |
2861 | .set_info_async = mdc_set_info_async, | |
2862 | .statfs = mdc_statfs, | |
2863 | .fid_init = client_fid_init, | |
2864 | .fid_fini = client_fid_fini, | |
2865 | .fid_alloc = mdc_fid_alloc, | |
2866 | .import_event = mdc_import_event, | |
2867 | .get_info = mdc_get_info, | |
2868 | .process_config = mdc_process_config, | |
2869 | .get_uuid = mdc_get_uuid, | |
2870 | .quotactl = mdc_quotactl, | |
2871 | .quotacheck = mdc_quotacheck | |
d7e09d03 PT |
2872 | }; |
2873 | ||
f6219c16 | 2874 | static struct md_ops mdc_md_ops = { |
df18a80a DC |
2875 | .getstatus = mdc_getstatus, |
2876 | .null_inode = mdc_null_inode, | |
df18a80a DC |
2877 | .close = mdc_close, |
2878 | .create = mdc_create, | |
2879 | .done_writing = mdc_done_writing, | |
2880 | .enqueue = mdc_enqueue, | |
2881 | .getattr = mdc_getattr, | |
2882 | .getattr_name = mdc_getattr_name, | |
2883 | .intent_lock = mdc_intent_lock, | |
2884 | .link = mdc_link, | |
df18a80a DC |
2885 | .rename = mdc_rename, |
2886 | .setattr = mdc_setattr, | |
2887 | .setxattr = mdc_setxattr, | |
2888 | .getxattr = mdc_getxattr, | |
2889 | .sync = mdc_sync, | |
4f76f0ec | 2890 | .read_page = mdc_read_page, |
df18a80a DC |
2891 | .unlink = mdc_unlink, |
2892 | .cancel_unused = mdc_cancel_unused, | |
2893 | .init_ea_size = mdc_init_ea_size, | |
2894 | .set_lock_data = mdc_set_lock_data, | |
2895 | .lock_match = mdc_lock_match, | |
2896 | .get_lustre_md = mdc_get_lustre_md, | |
2897 | .free_lustre_md = mdc_free_lustre_md, | |
2898 | .set_open_replay_data = mdc_set_open_replay_data, | |
2899 | .clear_open_replay_data = mdc_clear_open_replay_data, | |
df18a80a DC |
2900 | .intent_getattr_async = mdc_intent_getattr_async, |
2901 | .revalidate_lock = mdc_revalidate_lock | |
d7e09d03 PT |
2902 | }; |
2903 | ||
f6219c16 | 2904 | static int __init mdc_init(void) |
d7e09d03 | 2905 | { |
ea7893bb | 2906 | struct lprocfs_static_vars lvars = { NULL }; |
7436d070 | 2907 | |
d7e09d03 PT |
2908 | lprocfs_mdc_init_vars(&lvars); |
2909 | ||
2962b440 | 2910 | return class_register_type(&mdc_obd_ops, &mdc_md_ops, |
d7e09d03 | 2911 | LUSTRE_MDC_NAME, NULL); |
d7e09d03 PT |
2912 | } |
2913 | ||
2914 | static void /*__exit*/ mdc_exit(void) | |
2915 | { | |
2916 | class_unregister_type(LUSTRE_MDC_NAME); | |
2917 | } | |
2918 | ||
a0455471 | 2919 | MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>"); |
d7e09d03 | 2920 | MODULE_DESCRIPTION("Lustre Metadata Client"); |
5b0e50b9 | 2921 | MODULE_VERSION(LUSTRE_VERSION_STRING); |
d7e09d03 PT |
2922 | MODULE_LICENSE("GPL"); |
2923 | ||
2924 | module_init(mdc_init); | |
2925 | module_exit(mdc_exit); |