4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_MDC
38 #include "../include/lustre_net.h"
39 #include "../include/lustre/lustre_idl.h"
40 #include "mdc_internal.h"
43 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
48 b->uid = from_kuid(&init_user_ns, current_uid());
49 b->gid = from_kgid(&init_user_ns, current_gid());
50 b->fsuid = from_kuid(&init_user_ns, current_fsuid());
51 b->fsgid = from_kgid(&init_user_ns, current_fsgid());
52 b->capability = cfs_curproc_cap_pack();
55 void mdc_pack_capa(struct ptlrpc_request *req,
56 const struct req_msg_field *field,
59 struct req_capsule *pill = &req->rq_pill;
60 struct lustre_capa *c;
63 LASSERT(req_capsule_get_size(pill, field, RCL_CLIENT) == 0);
67 c = req_capsule_client_get(pill, field);
70 DEBUG_CAPA(D_SEC, c, "pack");
73 void mdc_is_subdir_pack(struct ptlrpc_request *req, const struct lu_fid *pfid,
74 const struct lu_fid *cfid, int flags)
76 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
81 b->valid = OBD_MD_FLID;
88 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
89 struct md_op_data *op_data)
91 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
94 __mdc_pack_body(b, op_data->op_suppgids[0]);
95 b->fid1 = op_data->op_fid1;
96 b->fid2 = op_data->op_fid2;
97 b->valid |= OBD_MD_FLID;
99 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
100 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
103 void mdc_pack_body(struct ptlrpc_request *req,
104 const struct lu_fid *fid, struct obd_capa *oc,
105 __u64 valid, int ea_size, __u32 suppgid, int flags)
107 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
111 b->eadatasize = ea_size;
113 __mdc_pack_body(b, suppgid);
116 b->valid |= OBD_MD_FLID;
117 mdc_pack_capa(req, &RMF_CAPA1, oc);
121 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff,
122 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
124 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
127 b->valid |= OBD_MD_FLID;
128 b->size = pgoff; /* !! */
129 b->nlink = size; /* !! */
130 __mdc_pack_body(b, -1);
131 b->mode = LUDA_FID | LUDA_TYPE;
133 mdc_pack_capa(req, &RMF_CAPA1, oc);
136 /* packing of MDS records */
137 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
138 const void *data, int datalen, __u32 mode,
139 __u32 uid, __u32 gid, cfs_cap_t cap_effective, __u64 rdev)
141 struct mdt_rec_create *rec;
145 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
146 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
149 rec->cr_opcode = REINT_CREATE;
152 rec->cr_cap = cap_effective;
153 rec->cr_fid1 = op_data->op_fid1;
154 rec->cr_fid2 = op_data->op_fid2;
157 rec->cr_time = op_data->op_mod_time;
158 rec->cr_suppgid1 = op_data->op_suppgids[0];
159 rec->cr_suppgid2 = op_data->op_suppgids[1];
160 flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
161 if (op_data->op_bias & MDS_CREATE_VOLATILE)
162 flags |= MDS_OPEN_VOLATILE;
163 set_mrc_cr_flags(rec, flags);
164 rec->cr_bias = op_data->op_bias;
165 rec->cr_umask = current_umask();
167 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
169 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
170 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
173 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
174 memcpy(tmp, data, datalen);
178 static __u64 mds_pack_open_flags(__u64 flags, __u32 mode)
180 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
181 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
182 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK |
183 MDS_OPEN_BY_FID | MDS_OPEN_LEASE |
186 cr_flags |= MDS_OPEN_CREAT;
188 cr_flags |= MDS_OPEN_EXCL;
190 cr_flags |= MDS_OPEN_TRUNC;
191 if (flags & O_APPEND)
192 cr_flags |= MDS_OPEN_APPEND;
194 cr_flags |= MDS_OPEN_SYNC;
195 if (flags & O_DIRECTORY)
196 cr_flags |= MDS_OPEN_DIRECTORY;
197 if (flags & __FMODE_EXEC)
198 cr_flags |= MDS_FMODE_EXEC;
199 if (cl_is_lov_delay_create(flags))
200 cr_flags |= MDS_OPEN_DELAY_CREATE;
202 if (flags & O_NONBLOCK)
203 cr_flags |= MDS_OPEN_NORESTORE;
208 /* packing of MDS records */
209 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
210 __u32 mode, __u64 rdev, __u64 flags, const void *lmm,
213 struct mdt_rec_create *rec;
217 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
218 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
220 /* XXX do something about time, uid, gid */
221 rec->cr_opcode = REINT_OPEN;
222 rec->cr_fsuid = from_kuid(&init_user_ns, current_fsuid());
223 rec->cr_fsgid = from_kgid(&init_user_ns, current_fsgid());
224 rec->cr_cap = cfs_curproc_cap_pack();
225 rec->cr_fid1 = op_data->op_fid1;
226 rec->cr_fid2 = op_data->op_fid2;
229 cr_flags = mds_pack_open_flags(flags, mode);
231 rec->cr_time = op_data->op_mod_time;
232 rec->cr_suppgid1 = op_data->op_suppgids[0];
233 rec->cr_suppgid2 = op_data->op_suppgids[1];
234 rec->cr_bias = op_data->op_bias;
235 rec->cr_umask = current_umask();
236 rec->cr_old_handle = op_data->op_handle;
238 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
239 /* the next buffer is child capa, which is used for replay,
240 * will be packed from the data in reply message. */
242 if (op_data->op_name) {
243 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
244 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
245 if (op_data->op_bias & MDS_CREATE_VOLATILE)
246 cr_flags |= MDS_OPEN_VOLATILE;
250 cr_flags |= MDS_OPEN_HAS_EA;
251 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
252 memcpy(tmp, lmm, lmmlen);
254 set_mrc_cr_flags(rec, cr_flags);
257 static inline __u64 attr_pack(unsigned int ia_valid)
261 if (ia_valid & ATTR_MODE)
262 sa_valid |= MDS_ATTR_MODE;
263 if (ia_valid & ATTR_UID)
264 sa_valid |= MDS_ATTR_UID;
265 if (ia_valid & ATTR_GID)
266 sa_valid |= MDS_ATTR_GID;
267 if (ia_valid & ATTR_SIZE)
268 sa_valid |= MDS_ATTR_SIZE;
269 if (ia_valid & ATTR_ATIME)
270 sa_valid |= MDS_ATTR_ATIME;
271 if (ia_valid & ATTR_MTIME)
272 sa_valid |= MDS_ATTR_MTIME;
273 if (ia_valid & ATTR_CTIME)
274 sa_valid |= MDS_ATTR_CTIME;
275 if (ia_valid & ATTR_ATIME_SET)
276 sa_valid |= MDS_ATTR_ATIME_SET;
277 if (ia_valid & ATTR_MTIME_SET)
278 sa_valid |= MDS_ATTR_MTIME_SET;
279 if (ia_valid & ATTR_FORCE)
280 sa_valid |= MDS_ATTR_FORCE;
281 if (ia_valid & ATTR_ATTR_FLAG)
282 sa_valid |= MDS_ATTR_ATTR_FLAG;
283 if (ia_valid & ATTR_KILL_SUID)
284 sa_valid |= MDS_ATTR_KILL_SUID;
285 if (ia_valid & ATTR_KILL_SGID)
286 sa_valid |= MDS_ATTR_KILL_SGID;
287 if (ia_valid & ATTR_CTIME_SET)
288 sa_valid |= MDS_ATTR_CTIME_SET;
289 if (ia_valid & ATTR_FROM_OPEN)
290 sa_valid |= MDS_ATTR_FROM_OPEN;
291 if (ia_valid & ATTR_BLOCKS)
292 sa_valid |= MDS_ATTR_BLOCKS;
293 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
294 /* NFSD hack (see bug 5781) */
295 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
299 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
300 struct md_op_data *op_data)
302 rec->sa_opcode = REINT_SETATTR;
303 rec->sa_fsuid = from_kuid(&init_user_ns, current_fsuid());
304 rec->sa_fsgid = from_kgid(&init_user_ns, current_fsgid());
305 rec->sa_cap = cfs_curproc_cap_pack();
306 rec->sa_suppgid = -1;
308 rec->sa_fid = op_data->op_fid1;
309 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
310 rec->sa_mode = op_data->op_attr.ia_mode;
311 rec->sa_uid = from_kuid(&init_user_ns, op_data->op_attr.ia_uid);
312 rec->sa_gid = from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
313 rec->sa_size = op_data->op_attr.ia_size;
314 rec->sa_blocks = op_data->op_attr_blocks;
315 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
316 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
317 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
319 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
320 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
321 in_group_p(op_data->op_attr.ia_gid))
323 from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
325 rec->sa_suppgid = op_data->op_suppgids[0];
327 rec->sa_bias = op_data->op_bias;
330 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
331 struct md_op_data *op_data)
333 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
334 epoch->ioepoch = op_data->op_ioepoch;
335 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
338 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
339 void *ea, int ealen, void *ea2, int ea2len)
341 struct mdt_rec_setattr *rec;
342 struct mdt_ioepoch *epoch;
343 struct lov_user_md *lum = NULL;
345 CLASSERT(sizeof(struct mdt_rec_reint) ==
346 sizeof(struct mdt_rec_setattr));
347 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
348 mdc_setattr_pack_rec(rec, op_data);
350 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
352 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
353 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
354 mdc_ioepoch_pack(epoch, op_data);
360 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
361 if (ea == NULL) { /* Remove LOV EA */
362 lum->lmm_magic = LOV_USER_MAGIC_V1;
363 lum->lmm_stripe_size = 0;
364 lum->lmm_stripe_count = 0;
365 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
367 memcpy(lum, ea, ealen);
373 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
377 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
379 struct mdt_rec_unlink *rec;
382 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
383 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
384 LASSERT(rec != NULL);
386 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
387 REINT_RMENTRY : REINT_UNLINK;
388 rec->ul_fsuid = op_data->op_fsuid;
389 rec->ul_fsgid = op_data->op_fsgid;
390 rec->ul_cap = op_data->op_cap;
391 rec->ul_mode = op_data->op_mode;
392 rec->ul_suppgid1 = op_data->op_suppgids[0];
393 rec->ul_suppgid2 = -1;
394 rec->ul_fid1 = op_data->op_fid1;
395 rec->ul_fid2 = op_data->op_fid2;
396 rec->ul_time = op_data->op_mod_time;
397 rec->ul_bias = op_data->op_bias;
399 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
401 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
402 LASSERT(tmp != NULL);
403 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
406 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
408 struct mdt_rec_link *rec;
411 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
412 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
413 LASSERT(rec != NULL);
415 rec->lk_opcode = REINT_LINK;
416 rec->lk_fsuid = op_data->op_fsuid; /* current->fsuid; */
417 rec->lk_fsgid = op_data->op_fsgid; /* current->fsgid; */
418 rec->lk_cap = op_data->op_cap; /* current->cap_effective; */
419 rec->lk_suppgid1 = op_data->op_suppgids[0];
420 rec->lk_suppgid2 = op_data->op_suppgids[1];
421 rec->lk_fid1 = op_data->op_fid1;
422 rec->lk_fid2 = op_data->op_fid2;
423 rec->lk_time = op_data->op_mod_time;
424 rec->lk_bias = op_data->op_bias;
426 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
427 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
429 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
430 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
433 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
434 const char *old, int oldlen, const char *new, int newlen)
436 struct mdt_rec_rename *rec;
439 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
440 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
442 /* XXX do something about time, uid, gid */
443 rec->rn_opcode = REINT_RENAME;
444 rec->rn_fsuid = op_data->op_fsuid;
445 rec->rn_fsgid = op_data->op_fsgid;
446 rec->rn_cap = op_data->op_cap;
447 rec->rn_suppgid1 = op_data->op_suppgids[0];
448 rec->rn_suppgid2 = op_data->op_suppgids[1];
449 rec->rn_fid1 = op_data->op_fid1;
450 rec->rn_fid2 = op_data->op_fid2;
451 rec->rn_time = op_data->op_mod_time;
452 rec->rn_mode = op_data->op_mode;
453 rec->rn_bias = op_data->op_bias;
455 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
456 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
458 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
459 LOGL0(old, oldlen, tmp);
462 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
463 LOGL0(new, newlen, tmp);
467 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
468 struct md_op_data *op_data, int ea_size)
470 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
474 if (op_data->op_bias & MDS_CHECK_SPLIT)
475 b->valid |= OBD_MD_FLCKSPLIT;
476 if (op_data->op_bias & MDS_CROSS_REF)
477 b->valid |= OBD_MD_FLCROSSREF;
478 b->eadatasize = ea_size;
480 __mdc_pack_body(b, op_data->op_suppgids[0]);
482 b->fid1 = op_data->op_fid1;
483 b->fid2 = op_data->op_fid2;
484 b->valid |= OBD_MD_FLID;
486 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
488 if (op_data->op_name) {
489 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
491 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
496 static void mdc_hsm_release_pack(struct ptlrpc_request *req,
497 struct md_op_data *op_data)
499 if (op_data->op_bias & MDS_HSM_RELEASE) {
500 struct close_data *data;
501 struct ldlm_lock *lock;
503 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
504 LASSERT(data != NULL);
506 lock = ldlm_handle2lock(&op_data->op_lease_handle);
508 data->cd_handle = lock->l_remote_handle;
511 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
513 data->cd_data_version = op_data->op_data_version;
514 data->cd_fid = op_data->op_fid2;
518 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
520 struct mdt_ioepoch *epoch;
521 struct mdt_rec_setattr *rec;
523 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
524 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
526 mdc_setattr_pack_rec(rec, op_data);
527 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
528 mdc_ioepoch_pack(epoch, op_data);
529 mdc_hsm_release_pack(req, op_data);
532 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
536 client_obd_list_lock(&cli->cl_loi_list_lock);
537 rc = list_empty(&mcw->mcw_entry);
538 client_obd_list_unlock(&cli->cl_loi_list_lock);
542 /* We record requests in flight in cli->cl_r_in_flight here.
543 * There is only one write rpc possible in mdc anyway. If this to change
544 * in the future - the code may need to be revisited. */
545 int mdc_enter_request(struct client_obd *cli)
548 struct mdc_cache_waiter mcw;
549 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
551 client_obd_list_lock(&cli->cl_loi_list_lock);
552 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
553 list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
554 init_waitqueue_head(&mcw.mcw_waitq);
555 client_obd_list_unlock(&cli->cl_loi_list_lock);
556 rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw),
559 client_obd_list_lock(&cli->cl_loi_list_lock);
560 if (list_empty(&mcw.mcw_entry))
561 cli->cl_r_in_flight--;
562 list_del_init(&mcw.mcw_entry);
563 client_obd_list_unlock(&cli->cl_loi_list_lock);
566 cli->cl_r_in_flight++;
567 client_obd_list_unlock(&cli->cl_loi_list_lock);
572 void mdc_exit_request(struct client_obd *cli)
574 struct list_head *l, *tmp;
575 struct mdc_cache_waiter *mcw;
577 client_obd_list_lock(&cli->cl_loi_list_lock);
578 cli->cl_r_in_flight--;
579 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
580 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
581 /* No free request slots anymore */
585 mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
586 list_del_init(&mcw->mcw_entry);
587 cli->cl_r_in_flight++;
588 wake_up(&mcw->mcw_waitq);
590 /* Empty waiting list? Decrease reqs in-flight number */
592 client_obd_list_unlock(&cli->cl_loi_list_lock);