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, 2013, 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
40 # include <liblustre.h>
42 #include <lustre_net.h>
43 #include <lustre/lustre_idl.h>
45 #include <cl_object.h>
47 #include "mdc_internal.h"
50 /* some liblustre hackings here */
56 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
61 b->uid = current_uid();
62 b->gid = current_gid();
63 b->fsuid = current_fsuid();
64 b->fsgid = current_fsgid();
65 b->capability = cfs_curproc_cap_pack();
68 void mdc_pack_capa(struct ptlrpc_request *req, const struct req_msg_field *field,
71 struct req_capsule *pill = &req->rq_pill;
72 struct lustre_capa *c;
75 LASSERT(req_capsule_get_size(pill, field, RCL_CLIENT) == 0);
79 c = req_capsule_client_get(pill, field);
82 DEBUG_CAPA(D_SEC, c, "pack");
85 void mdc_is_subdir_pack(struct ptlrpc_request *req, const struct lu_fid *pfid,
86 const struct lu_fid *cfid, int flags)
88 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
93 b->valid = OBD_MD_FLID;
100 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
101 struct md_op_data *op_data)
103 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
106 __mdc_pack_body(b, op_data->op_suppgids[0]);
107 b->fid1 = op_data->op_fid1;
108 b->fid2 = op_data->op_fid2;
109 b->valid |= OBD_MD_FLID;
111 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
112 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
115 void mdc_pack_body(struct ptlrpc_request *req,
116 const struct lu_fid *fid, struct obd_capa *oc,
117 __u64 valid, int ea_size, __u32 suppgid, int flags)
119 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
123 b->eadatasize = ea_size;
125 __mdc_pack_body(b, suppgid);
128 b->valid |= OBD_MD_FLID;
129 mdc_pack_capa(req, &RMF_CAPA1, oc);
133 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff,
134 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
136 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
139 b->valid |= OBD_MD_FLID;
140 b->size = pgoff; /* !! */
141 b->nlink = size; /* !! */
142 __mdc_pack_body(b, -1);
143 b->mode = LUDA_FID | LUDA_TYPE;
145 mdc_pack_capa(req, &RMF_CAPA1, oc);
148 /* packing of MDS records */
149 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
150 const void *data, int datalen, __u32 mode,
151 __u32 uid, __u32 gid, cfs_cap_t cap_effective, __u64 rdev)
153 struct mdt_rec_create *rec;
157 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
158 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
161 rec->cr_opcode = REINT_CREATE;
164 rec->cr_cap = cap_effective;
165 rec->cr_fid1 = op_data->op_fid1;
166 rec->cr_fid2 = op_data->op_fid2;
169 rec->cr_time = op_data->op_mod_time;
170 rec->cr_suppgid1 = op_data->op_suppgids[0];
171 rec->cr_suppgid2 = op_data->op_suppgids[1];
172 flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
173 if (op_data->op_bias & MDS_CREATE_VOLATILE)
174 flags |= MDS_OPEN_VOLATILE;
175 set_mrc_cr_flags(rec, flags);
176 rec->cr_bias = op_data->op_bias;
177 rec->cr_umask = current_umask();
179 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
181 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
182 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
185 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
186 memcpy(tmp, data, datalen);
190 static __u64 mds_pack_open_flags(__u64 flags, __u32 mode)
192 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
193 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
194 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK |
195 MDS_OPEN_BY_FID | MDS_OPEN_LEASE |
198 cr_flags |= MDS_OPEN_CREAT;
200 cr_flags |= MDS_OPEN_EXCL;
202 cr_flags |= MDS_OPEN_TRUNC;
203 if (flags & O_APPEND)
204 cr_flags |= MDS_OPEN_APPEND;
206 cr_flags |= MDS_OPEN_SYNC;
207 if (flags & O_DIRECTORY)
208 cr_flags |= MDS_OPEN_DIRECTORY;
210 if (flags & FMODE_EXEC)
211 cr_flags |= MDS_FMODE_EXEC;
213 if (cl_is_lov_delay_create(flags))
214 cr_flags |= MDS_OPEN_DELAY_CREATE;
216 if ((flags & O_NOACCESS) || (flags & O_NONBLOCK))
217 cr_flags |= MDS_OPEN_NORESTORE;
222 /* packing of MDS records */
223 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
224 __u32 mode, __u64 rdev, __u64 flags, const void *lmm,
227 struct mdt_rec_create *rec;
231 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
232 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
234 /* XXX do something about time, uid, gid */
235 rec->cr_opcode = REINT_OPEN;
236 rec->cr_fsuid = current_fsuid();
237 rec->cr_fsgid = current_fsgid();
238 rec->cr_cap = cfs_curproc_cap_pack();
240 cr_flags = mds_pack_open_flags(flags, mode);
242 rec->cr_umask = current_umask();
243 if (op_data != NULL) {
244 rec->cr_fid1 = op_data->op_fid1;
245 rec->cr_fid2 = op_data->op_fid2;
246 rec->cr_time = op_data->op_mod_time;
247 rec->cr_suppgid1 = op_data->op_suppgids[0];
248 rec->cr_suppgid2 = op_data->op_suppgids[1];
249 rec->cr_bias = op_data->op_bias;
250 rec->cr_old_handle = op_data->op_handle;
252 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
253 /* the next buffer is child capa, which is used for replay,
254 * will be packed from the data in reply message. */
256 if (op_data->op_name) {
257 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
258 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
259 if (op_data->op_bias & MDS_CREATE_VOLATILE)
260 cr_flags |= MDS_OPEN_VOLATILE;
263 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
264 rec->cr_fid2 = op_data->op_fid2;
269 cr_flags |= MDS_OPEN_HAS_EA;
270 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
271 memcpy(tmp, lmm, lmmlen);
273 set_mrc_cr_flags(rec, cr_flags);
276 static inline __u64 attr_pack(unsigned int ia_valid) {
279 if (ia_valid & ATTR_MODE)
280 sa_valid |= MDS_ATTR_MODE;
281 if (ia_valid & ATTR_UID)
282 sa_valid |= MDS_ATTR_UID;
283 if (ia_valid & ATTR_GID)
284 sa_valid |= MDS_ATTR_GID;
285 if (ia_valid & ATTR_SIZE)
286 sa_valid |= MDS_ATTR_SIZE;
287 if (ia_valid & ATTR_ATIME)
288 sa_valid |= MDS_ATTR_ATIME;
289 if (ia_valid & ATTR_MTIME)
290 sa_valid |= MDS_ATTR_MTIME;
291 if (ia_valid & ATTR_CTIME)
292 sa_valid |= MDS_ATTR_CTIME;
293 if (ia_valid & ATTR_ATIME_SET)
294 sa_valid |= MDS_ATTR_ATIME_SET;
295 if (ia_valid & ATTR_MTIME_SET)
296 sa_valid |= MDS_ATTR_MTIME_SET;
297 if (ia_valid & ATTR_FORCE)
298 sa_valid |= MDS_ATTR_FORCE;
299 if (ia_valid & ATTR_ATTR_FLAG)
300 sa_valid |= MDS_ATTR_ATTR_FLAG;
301 if (ia_valid & ATTR_KILL_SUID)
302 sa_valid |= MDS_ATTR_KILL_SUID;
303 if (ia_valid & ATTR_KILL_SGID)
304 sa_valid |= MDS_ATTR_KILL_SGID;
305 if (ia_valid & ATTR_CTIME_SET)
306 sa_valid |= MDS_ATTR_CTIME_SET;
307 if (ia_valid & ATTR_FROM_OPEN)
308 sa_valid |= MDS_ATTR_FROM_OPEN;
309 if (ia_valid & ATTR_BLOCKS)
310 sa_valid |= MDS_ATTR_BLOCKS;
311 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
312 /* NFSD hack (see bug 5781) */
313 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
317 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
318 struct md_op_data *op_data)
320 rec->sa_opcode = REINT_SETATTR;
321 rec->sa_fsuid = current_fsuid();
322 rec->sa_fsgid = current_fsgid();
323 rec->sa_cap = cfs_curproc_cap_pack();
324 rec->sa_suppgid = -1;
326 rec->sa_fid = op_data->op_fid1;
327 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
328 rec->sa_mode = op_data->op_attr.ia_mode;
329 rec->sa_uid = op_data->op_attr.ia_uid;
330 rec->sa_gid = op_data->op_attr.ia_gid;
331 rec->sa_size = op_data->op_attr.ia_size;
332 rec->sa_blocks = op_data->op_attr_blocks;
333 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
334 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
335 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
336 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
337 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
338 in_group_p(op_data->op_attr.ia_gid))
339 rec->sa_suppgid = op_data->op_attr.ia_gid;
341 rec->sa_suppgid = op_data->op_suppgids[0];
343 rec->sa_bias = op_data->op_bias;
346 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
347 struct md_op_data *op_data)
349 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
350 epoch->ioepoch = op_data->op_ioepoch;
351 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
354 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
355 void *ea, int ealen, void *ea2, int ea2len)
357 struct mdt_rec_setattr *rec;
358 struct mdt_ioepoch *epoch;
359 struct lov_user_md *lum = NULL;
361 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
362 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
363 mdc_setattr_pack_rec(rec, op_data);
365 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
367 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
368 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
369 mdc_ioepoch_pack(epoch, op_data);
375 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
376 if (ea == NULL) { /* Remove LOV EA */
377 lum->lmm_magic = LOV_USER_MAGIC_V1;
378 lum->lmm_stripe_size = 0;
379 lum->lmm_stripe_count = 0;
380 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
382 memcpy(lum, ea, ealen);
388 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
392 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
394 struct mdt_rec_unlink *rec;
397 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
398 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
399 LASSERT(rec != NULL);
401 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
402 REINT_RMENTRY : REINT_UNLINK;
403 rec->ul_fsuid = op_data->op_fsuid;
404 rec->ul_fsgid = op_data->op_fsgid;
405 rec->ul_cap = op_data->op_cap;
406 rec->ul_mode = op_data->op_mode;
407 rec->ul_suppgid1= op_data->op_suppgids[0];
408 rec->ul_suppgid2= -1;
409 rec->ul_fid1 = op_data->op_fid1;
410 rec->ul_fid2 = op_data->op_fid2;
411 rec->ul_time = op_data->op_mod_time;
412 rec->ul_bias = op_data->op_bias;
414 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
416 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
417 LASSERT(tmp != NULL);
418 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
421 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
423 struct mdt_rec_link *rec;
426 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
427 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
428 LASSERT (rec != NULL);
430 rec->lk_opcode = REINT_LINK;
431 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
432 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
433 rec->lk_cap = op_data->op_cap;//current->cap_effective;
434 rec->lk_suppgid1 = op_data->op_suppgids[0];
435 rec->lk_suppgid2 = op_data->op_suppgids[1];
436 rec->lk_fid1 = op_data->op_fid1;
437 rec->lk_fid2 = op_data->op_fid2;
438 rec->lk_time = op_data->op_mod_time;
439 rec->lk_bias = op_data->op_bias;
441 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
442 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
444 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
445 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
448 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
449 const char *old, int oldlen, const char *new, int newlen)
451 struct mdt_rec_rename *rec;
454 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
455 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
457 /* XXX do something about time, uid, gid */
458 rec->rn_opcode = REINT_RENAME;
459 rec->rn_fsuid = op_data->op_fsuid;
460 rec->rn_fsgid = op_data->op_fsgid;
461 rec->rn_cap = op_data->op_cap;
462 rec->rn_suppgid1 = op_data->op_suppgids[0];
463 rec->rn_suppgid2 = op_data->op_suppgids[1];
464 rec->rn_fid1 = op_data->op_fid1;
465 rec->rn_fid2 = op_data->op_fid2;
466 rec->rn_time = op_data->op_mod_time;
467 rec->rn_mode = op_data->op_mode;
468 rec->rn_bias = op_data->op_bias;
470 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
471 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
473 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
474 LOGL0(old, oldlen, tmp);
477 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
478 LOGL0(new, newlen, tmp);
482 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
483 struct md_op_data *op_data, int ea_size)
485 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
489 if (op_data->op_bias & MDS_CHECK_SPLIT)
490 b->valid |= OBD_MD_FLCKSPLIT;
491 if (op_data->op_bias & MDS_CROSS_REF)
492 b->valid |= OBD_MD_FLCROSSREF;
493 b->eadatasize = ea_size;
495 __mdc_pack_body(b, op_data->op_suppgids[0]);
497 b->fid1 = op_data->op_fid1;
498 b->fid2 = op_data->op_fid2;
499 b->valid |= OBD_MD_FLID;
501 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
503 if (op_data->op_name) {
504 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
505 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
510 static void mdc_hsm_release_pack(struct ptlrpc_request *req,
511 struct md_op_data *op_data)
513 if (op_data->op_bias & MDS_HSM_RELEASE) {
514 struct close_data *data;
515 struct ldlm_lock *lock;
517 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
518 LASSERT(data != NULL);
520 lock = ldlm_handle2lock(&op_data->op_lease_handle);
522 data->cd_handle = lock->l_remote_handle;
525 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
527 data->cd_data_version = op_data->op_data_version;
528 data->cd_fid = op_data->op_fid2;
532 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
534 struct mdt_ioepoch *epoch;
535 struct mdt_rec_setattr *rec;
537 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
538 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
540 mdc_setattr_pack_rec(rec, op_data);
541 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
542 mdc_ioepoch_pack(epoch, op_data);
543 mdc_hsm_release_pack(req, op_data);
546 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
550 client_obd_list_lock(&cli->cl_loi_list_lock);
551 rc = cfs_list_empty(&mcw->mcw_entry);
552 client_obd_list_unlock(&cli->cl_loi_list_lock);
556 /* We record requests in flight in cli->cl_r_in_flight here.
557 * There is only one write rpc possible in mdc anyway. If this to change
558 * in the future - the code may need to be revisited. */
559 int mdc_enter_request(struct client_obd *cli)
562 struct mdc_cache_waiter mcw;
563 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
565 client_obd_list_lock(&cli->cl_loi_list_lock);
566 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
567 cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
568 init_waitqueue_head(&mcw.mcw_waitq);
569 client_obd_list_unlock(&cli->cl_loi_list_lock);
570 rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
572 client_obd_list_lock(&cli->cl_loi_list_lock);
573 if (cfs_list_empty(&mcw.mcw_entry))
574 cli->cl_r_in_flight--;
575 cfs_list_del_init(&mcw.mcw_entry);
576 client_obd_list_unlock(&cli->cl_loi_list_lock);
579 cli->cl_r_in_flight++;
580 client_obd_list_unlock(&cli->cl_loi_list_lock);
585 void mdc_exit_request(struct client_obd *cli)
588 struct mdc_cache_waiter *mcw;
590 client_obd_list_lock(&cli->cl_loi_list_lock);
591 cli->cl_r_in_flight--;
592 cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
593 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
594 /* No free request slots anymore */
598 mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
599 cfs_list_del_init(&mcw->mcw_entry);
600 cli->cl_r_in_flight++;
601 wake_up(&mcw->mcw_waitq);
603 /* Empty waiting list? Decrease reqs in-flight number */
605 client_obd_list_unlock(&cli->cl_loi_list_lock);