1 /*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37 #define DEBUG_SUBSYSTEM S_MDC
38
39 # include <linux/module.h>
40 # include <linux/kernel.h>
41
42 #include "../include/obd_class.h"
43 #include "mdc_internal.h"
44 #include "../include/lustre_fid.h"
45
46 /* mdc_setattr does its own semaphore handling */
mdc_reint(struct ptlrpc_request * request,struct mdc_rpc_lock * rpc_lock,int level)47 static int mdc_reint(struct ptlrpc_request *request,
48 struct mdc_rpc_lock *rpc_lock,
49 int level)
50 {
51 int rc;
52
53 request->rq_send_state = level;
54
55 mdc_get_rpc_lock(rpc_lock, NULL);
56 rc = ptlrpc_queue_wait(request);
57 mdc_put_rpc_lock(rpc_lock, NULL);
58 if (rc)
59 CDEBUG(D_INFO, "error in handling %d\n", rc);
60 else if (!req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY))
61 rc = -EPROTO;
62
63 return rc;
64 }
65
66 /* Find and cancel locally locks matched by inode @bits & @mode in the resource
67 * found by @fid. Found locks are added into @cancel list. Returns the amount of
68 * locks added to @cancels list. */
mdc_resource_get_unused(struct obd_export * exp,const struct lu_fid * fid,struct list_head * cancels,ldlm_mode_t mode,__u64 bits)69 int mdc_resource_get_unused(struct obd_export *exp, const struct lu_fid *fid,
70 struct list_head *cancels, ldlm_mode_t mode,
71 __u64 bits)
72 {
73 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
74 ldlm_policy_data_t policy = {};
75 struct ldlm_res_id res_id;
76 struct ldlm_resource *res;
77 int count;
78
79 /* Return, i.e. cancel nothing, only if ELC is supported (flag in
80 * export) but disabled through procfs (flag in NS).
81 *
82 * This distinguishes from a case when ELC is not supported originally,
83 * when we still want to cancel locks in advance and just cancel them
84 * locally, without sending any RPC. */
85 if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
86 return 0;
87
88 fid_build_reg_res_name(fid, &res_id);
89 res = ldlm_resource_get(exp->exp_obd->obd_namespace,
90 NULL, &res_id, 0, 0);
91 if (res == NULL)
92 return 0;
93 LDLM_RESOURCE_ADDREF(res);
94 /* Initialize ibits lock policy. */
95 policy.l_inodebits.bits = bits;
96 count = ldlm_cancel_resource_local(res, cancels, &policy,
97 mode, 0, 0, NULL);
98 LDLM_RESOURCE_DELREF(res);
99 ldlm_resource_putref(res);
100 return count;
101 }
102
mdc_setattr(struct obd_export * exp,struct md_op_data * op_data,void * ea,int ealen,void * ea2,int ea2len,struct ptlrpc_request ** request,struct md_open_data ** mod)103 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
104 void *ea, int ealen, void *ea2, int ea2len,
105 struct ptlrpc_request **request, struct md_open_data **mod)
106 {
107 LIST_HEAD(cancels);
108 struct ptlrpc_request *req;
109 struct mdc_rpc_lock *rpc_lock;
110 struct obd_device *obd = exp->exp_obd;
111 int count = 0, rc;
112 __u64 bits;
113
114 LASSERT(op_data != NULL);
115
116 bits = MDS_INODELOCK_UPDATE;
117 if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
118 bits |= MDS_INODELOCK_LOOKUP;
119 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
120 (fid_is_sane(&op_data->op_fid1)) &&
121 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
122 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
123 &cancels, LCK_EX, bits);
124 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
125 &RQF_MDS_REINT_SETATTR);
126 if (req == NULL) {
127 ldlm_lock_list_put(&cancels, l_bl_ast, count);
128 return -ENOMEM;
129 }
130 if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
131 req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
132 0);
133 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
134 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
135 ea2len);
136
137 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
138 if (rc) {
139 ptlrpc_request_free(req);
140 return rc;
141 }
142
143 rpc_lock = obd->u.cli.cl_rpc_lock;
144
145 if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
146 CDEBUG(D_INODE, "setting mtime %ld, ctime %ld\n",
147 LTIME_S(op_data->op_attr.ia_mtime),
148 LTIME_S(op_data->op_attr.ia_ctime));
149 mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
150
151 ptlrpc_request_set_replen(req);
152 if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
153 req->rq_import->imp_replayable) {
154 LASSERT(*mod == NULL);
155
156 *mod = obd_mod_alloc();
157 if (*mod == NULL) {
158 DEBUG_REQ(D_ERROR, req, "Can't allocate md_open_data");
159 } else {
160 req->rq_replay = 1;
161 req->rq_cb_data = *mod;
162 (*mod)->mod_open_req = req;
163 req->rq_commit_cb = mdc_commit_open;
164 (*mod)->mod_is_create = true;
165 /**
166 * Take an extra reference on \var mod, it protects \var
167 * mod from being freed on eviction (commit callback is
168 * called despite rq_replay flag).
169 * Will be put on mdc_done_writing().
170 */
171 obd_mod_get(*mod);
172 }
173 }
174
175 rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
176
177 /* Save the obtained info in the original RPC for the replay case. */
178 if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
179 struct mdt_ioepoch *epoch;
180 struct mdt_body *body;
181
182 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
183 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
184 LASSERT(epoch != NULL);
185 LASSERT(body != NULL);
186 epoch->handle = body->handle;
187 epoch->ioepoch = body->ioepoch;
188 req->rq_replay_cb = mdc_replay_open;
189 /** bug 3633, open may be committed and estale answer is not error */
190 } else if (rc == -ESTALE && (op_data->op_flags & MF_SOM_CHANGE)) {
191 rc = 0;
192 } else if (rc == -ERESTARTSYS) {
193 rc = 0;
194 }
195 *request = req;
196 if (rc && req->rq_commit_cb) {
197 /* Put an extra reference on \var mod on error case. */
198 if (mod != NULL && *mod != NULL)
199 obd_mod_put(*mod);
200 req->rq_commit_cb(req);
201 }
202 return rc;
203 }
204
mdc_create(struct obd_export * exp,struct md_op_data * op_data,const void * data,int datalen,int mode,__u32 uid,__u32 gid,cfs_cap_t cap_effective,__u64 rdev,struct ptlrpc_request ** request)205 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
206 const void *data, int datalen, int mode, __u32 uid, __u32 gid,
207 cfs_cap_t cap_effective, __u64 rdev,
208 struct ptlrpc_request **request)
209 {
210 struct ptlrpc_request *req;
211 int level, rc;
212 int count, resends = 0;
213 struct obd_import *import = exp->exp_obd->u.cli.cl_import;
214 int generation = import->imp_generation;
215 LIST_HEAD(cancels);
216
217 /* For case if upper layer did not alloc fid, do it now. */
218 if (!fid_is_sane(&op_data->op_fid2)) {
219 /*
220 * mdc_fid_alloc() may return errno 1 in case of switch to new
221 * sequence, handle this.
222 */
223 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
224 if (rc < 0) {
225 CERROR("Can't alloc new fid, rc %d\n", rc);
226 return rc;
227 }
228 }
229
230 rebuild:
231 count = 0;
232 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
233 (fid_is_sane(&op_data->op_fid1)))
234 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
235 &cancels, LCK_EX,
236 MDS_INODELOCK_UPDATE);
237
238 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
239 &RQF_MDS_REINT_CREATE_RMT_ACL);
240 if (req == NULL) {
241 ldlm_lock_list_put(&cancels, l_bl_ast, count);
242 return -ENOMEM;
243 }
244 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
245 op_data->op_namelen + 1);
246 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
247 data && datalen ? datalen : 0);
248
249 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
250 if (rc) {
251 ptlrpc_request_free(req);
252 return rc;
253 }
254
255 /*
256 * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
257 * tgt, for symlinks or lov MD data.
258 */
259 mdc_create_pack(req, op_data, data, datalen, mode, uid,
260 gid, cap_effective, rdev);
261
262 ptlrpc_request_set_replen(req);
263
264 /* ask ptlrpc not to resend on EINPROGRESS since we have our own retry
265 * logic here */
266 req->rq_no_retry_einprogress = 1;
267
268 if (resends) {
269 req->rq_generation_set = 1;
270 req->rq_import_generation = generation;
271 req->rq_sent = ktime_get_real_seconds() + resends;
272 }
273 level = LUSTRE_IMP_FULL;
274 resend:
275 rc = mdc_reint(req, exp->exp_obd->u.cli.cl_rpc_lock, level);
276
277 /* Resend if we were told to. */
278 if (rc == -ERESTARTSYS) {
279 level = LUSTRE_IMP_RECOVER;
280 goto resend;
281 } else if (rc == -EINPROGRESS) {
282 /* Retry create infinitely until succeed or get other
283 * error code. */
284 ptlrpc_req_finished(req);
285 resends++;
286
287 CDEBUG(D_HA, "%s: resend:%d create on "DFID"/"DFID"\n",
288 exp->exp_obd->obd_name, resends,
289 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
290
291 if (generation == import->imp_generation) {
292 goto rebuild;
293 } else {
294 CDEBUG(D_HA, "resend cross eviction\n");
295 return -EIO;
296 }
297 }
298
299 *request = req;
300 return rc;
301 }
302
mdc_unlink(struct obd_export * exp,struct md_op_data * op_data,struct ptlrpc_request ** request)303 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
304 struct ptlrpc_request **request)
305 {
306 LIST_HEAD(cancels);
307 struct obd_device *obd = class_exp2obd(exp);
308 struct ptlrpc_request *req = *request;
309 int count = 0, rc;
310
311 LASSERT(req == NULL);
312
313 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
314 (fid_is_sane(&op_data->op_fid1)) &&
315 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
316 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
317 &cancels, LCK_EX,
318 MDS_INODELOCK_UPDATE);
319 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
320 (fid_is_sane(&op_data->op_fid3)) &&
321 !OBD_FAIL_CHECK(OBD_FAIL_LDLM_BL_CALLBACK_NET))
322 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
323 &cancels, LCK_EX,
324 MDS_INODELOCK_FULL);
325 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
326 &RQF_MDS_REINT_UNLINK);
327 if (req == NULL) {
328 ldlm_lock_list_put(&cancels, l_bl_ast, count);
329 return -ENOMEM;
330 }
331 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
332 op_data->op_namelen + 1);
333
334 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
335 if (rc) {
336 ptlrpc_request_free(req);
337 return rc;
338 }
339
340 mdc_unlink_pack(req, op_data);
341
342 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
343 obd->u.cli.cl_default_mds_easize);
344 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
345 obd->u.cli.cl_default_mds_cookiesize);
346 ptlrpc_request_set_replen(req);
347
348 *request = req;
349
350 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
351 if (rc == -ERESTARTSYS)
352 rc = 0;
353 return rc;
354 }
355
mdc_link(struct obd_export * exp,struct md_op_data * op_data,struct ptlrpc_request ** request)356 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
357 struct ptlrpc_request **request)
358 {
359 LIST_HEAD(cancels);
360 struct obd_device *obd = exp->exp_obd;
361 struct ptlrpc_request *req;
362 int count = 0, rc;
363
364 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
365 (fid_is_sane(&op_data->op_fid2)))
366 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
367 &cancels, LCK_EX,
368 MDS_INODELOCK_UPDATE);
369 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
370 (fid_is_sane(&op_data->op_fid1)))
371 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
372 &cancels, LCK_EX,
373 MDS_INODELOCK_UPDATE);
374
375 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
376 if (req == NULL) {
377 ldlm_lock_list_put(&cancels, l_bl_ast, count);
378 return -ENOMEM;
379 }
380 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
381 op_data->op_namelen + 1);
382
383 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
384 if (rc) {
385 ptlrpc_request_free(req);
386 return rc;
387 }
388
389 mdc_link_pack(req, op_data);
390 ptlrpc_request_set_replen(req);
391
392 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
393 *request = req;
394 if (rc == -ERESTARTSYS)
395 rc = 0;
396
397 return rc;
398 }
399
mdc_rename(struct obd_export * exp,struct md_op_data * op_data,const char * old,int oldlen,const char * new,int newlen,struct ptlrpc_request ** request)400 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
401 const char *old, int oldlen, const char *new, int newlen,
402 struct ptlrpc_request **request)
403 {
404 LIST_HEAD(cancels);
405 struct obd_device *obd = exp->exp_obd;
406 struct ptlrpc_request *req;
407 int count = 0, rc;
408
409 if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
410 (fid_is_sane(&op_data->op_fid1)))
411 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
412 &cancels, LCK_EX,
413 MDS_INODELOCK_UPDATE);
414 if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
415 (fid_is_sane(&op_data->op_fid2)))
416 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
417 &cancels, LCK_EX,
418 MDS_INODELOCK_UPDATE);
419 if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
420 (fid_is_sane(&op_data->op_fid3)))
421 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
422 &cancels, LCK_EX,
423 MDS_INODELOCK_LOOKUP);
424 if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
425 (fid_is_sane(&op_data->op_fid4)))
426 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
427 &cancels, LCK_EX,
428 MDS_INODELOCK_FULL);
429
430 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
431 &RQF_MDS_REINT_RENAME);
432 if (req == NULL) {
433 ldlm_lock_list_put(&cancels, l_bl_ast, count);
434 return -ENOMEM;
435 }
436
437 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
438 req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT, newlen+1);
439
440 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
441 if (rc) {
442 ptlrpc_request_free(req);
443 return rc;
444 }
445
446 if (exp_connect_cancelset(exp) && req)
447 ldlm_cli_cancel_list(&cancels, count, req, 0);
448
449 mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
450
451 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
452 obd->u.cli.cl_default_mds_easize);
453 req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
454 obd->u.cli.cl_default_mds_cookiesize);
455 ptlrpc_request_set_replen(req);
456
457 rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
458 *request = req;
459 if (rc == -ERESTARTSYS)
460 rc = 0;
461
462 return rc;
463 }
464