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>
44 #include <obd_class.h>
46 #include <cl_object.h>
48 #include "mdc_internal.h"
51 /* some liblustre hackings here */
57 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
62 b->uid = current_uid();
63 b->gid = current_gid();
64 b->fsuid = current_fsuid();
65 b->fsgid = current_fsgid();
66 b->capability = cfs_curproc_cap_pack();
69 void mdc_pack_capa(struct ptlrpc_request *req, const struct req_msg_field *field,
72 struct req_capsule *pill = &req->rq_pill;
73 struct lustre_capa *c;
76 LASSERT(req_capsule_get_size(pill, field, RCL_CLIENT) == 0);
80 c = req_capsule_client_get(pill, field);
83 DEBUG_CAPA(D_SEC, c, "pack");
86 void mdc_is_subdir_pack(struct ptlrpc_request *req, const struct lu_fid *pfid,
87 const struct lu_fid *cfid, int flags)
89 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
94 b->valid = OBD_MD_FLID;
101 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
102 struct md_op_data *op_data)
104 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
107 __mdc_pack_body(b, op_data->op_suppgids[0]);
108 b->fid1 = op_data->op_fid1;
109 b->fid2 = op_data->op_fid2;
110 b->valid |= OBD_MD_FLID;
112 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
113 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
116 void mdc_pack_body(struct ptlrpc_request *req,
117 const struct lu_fid *fid, struct obd_capa *oc,
118 __u64 valid, int ea_size, __u32 suppgid, int flags)
120 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
124 b->eadatasize = ea_size;
126 __mdc_pack_body(b, suppgid);
129 b->valid |= OBD_MD_FLID;
130 mdc_pack_capa(req, &RMF_CAPA1, oc);
134 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff,
135 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
137 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
140 b->valid |= OBD_MD_FLID;
141 b->size = pgoff; /* !! */
142 b->nlink = size; /* !! */
143 __mdc_pack_body(b, -1);
144 b->mode = LUDA_FID | LUDA_TYPE;
146 mdc_pack_capa(req, &RMF_CAPA1, oc);
149 /* packing of MDS records */
150 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
151 const void *data, int datalen, __u32 mode,
152 __u32 uid, __u32 gid, cfs_cap_t cap_effective, __u64 rdev)
154 struct mdt_rec_create *rec;
158 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
159 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
162 rec->cr_opcode = REINT_CREATE;
165 rec->cr_cap = cap_effective;
166 rec->cr_fid1 = op_data->op_fid1;
167 rec->cr_fid2 = op_data->op_fid2;
170 rec->cr_time = op_data->op_mod_time;
171 rec->cr_suppgid1 = op_data->op_suppgids[0];
172 rec->cr_suppgid2 = op_data->op_suppgids[1];
173 flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
174 if (op_data->op_bias & MDS_CREATE_VOLATILE)
175 flags |= MDS_OPEN_VOLATILE;
176 set_mrc_cr_flags(rec, flags);
177 rec->cr_bias = op_data->op_bias;
178 rec->cr_umask = current_umask();
180 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
182 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
183 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
186 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
187 memcpy(tmp, data, datalen);
191 static __u64 mds_pack_open_flags(__u64 flags, __u32 mode)
193 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
194 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
195 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK |
196 MDS_OPEN_BY_FID | MDS_OPEN_LEASE |
199 cr_flags |= MDS_OPEN_CREAT;
201 cr_flags |= MDS_OPEN_EXCL;
203 cr_flags |= MDS_OPEN_TRUNC;
204 if (flags & O_APPEND)
205 cr_flags |= MDS_OPEN_APPEND;
207 cr_flags |= MDS_OPEN_SYNC;
208 if (flags & O_DIRECTORY)
209 cr_flags |= MDS_OPEN_DIRECTORY;
211 if (flags & FMODE_EXEC)
212 cr_flags |= MDS_FMODE_EXEC;
214 if (cl_is_lov_delay_create(flags))
215 cr_flags |= MDS_OPEN_DELAY_CREATE;
217 if ((flags & O_NOACCESS) || (flags & O_NONBLOCK))
218 cr_flags |= MDS_OPEN_NORESTORE;
223 /* packing of MDS records */
224 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
225 __u32 mode, __u64 rdev, __u64 flags, const void *lmm,
228 struct mdt_rec_create *rec;
232 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
233 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
235 /* XXX do something about time, uid, gid */
236 rec->cr_opcode = REINT_OPEN;
237 rec->cr_fsuid = current_fsuid();
238 rec->cr_fsgid = current_fsgid();
239 rec->cr_cap = cfs_curproc_cap_pack();
241 cr_flags = mds_pack_open_flags(flags, mode);
243 rec->cr_umask = current_umask();
244 if (op_data != NULL) {
245 rec->cr_fid1 = op_data->op_fid1;
246 rec->cr_fid2 = op_data->op_fid2;
247 rec->cr_time = op_data->op_mod_time;
248 rec->cr_suppgid1 = op_data->op_suppgids[0];
249 rec->cr_suppgid2 = op_data->op_suppgids[1];
250 rec->cr_bias = op_data->op_bias;
251 rec->cr_old_handle = op_data->op_handle;
253 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
254 /* the next buffer is child capa, which is used for replay,
255 * will be packed from the data in reply message. */
257 if (op_data->op_name) {
258 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
259 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
260 if (op_data->op_bias & MDS_CREATE_VOLATILE)
261 cr_flags |= MDS_OPEN_VOLATILE;
264 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
265 rec->cr_fid2 = op_data->op_fid2;
270 cr_flags |= MDS_OPEN_HAS_EA;
271 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
272 memcpy(tmp, lmm, lmmlen);
274 set_mrc_cr_flags(rec, cr_flags);
277 static inline __u64 attr_pack(unsigned int ia_valid) {
280 if (ia_valid & ATTR_MODE)
281 sa_valid |= MDS_ATTR_MODE;
282 if (ia_valid & ATTR_UID)
283 sa_valid |= MDS_ATTR_UID;
284 if (ia_valid & ATTR_GID)
285 sa_valid |= MDS_ATTR_GID;
286 if (ia_valid & ATTR_SIZE)
287 sa_valid |= MDS_ATTR_SIZE;
288 if (ia_valid & ATTR_ATIME)
289 sa_valid |= MDS_ATTR_ATIME;
290 if (ia_valid & ATTR_MTIME)
291 sa_valid |= MDS_ATTR_MTIME;
292 if (ia_valid & ATTR_CTIME)
293 sa_valid |= MDS_ATTR_CTIME;
294 if (ia_valid & ATTR_ATIME_SET)
295 sa_valid |= MDS_ATTR_ATIME_SET;
296 if (ia_valid & ATTR_MTIME_SET)
297 sa_valid |= MDS_ATTR_MTIME_SET;
298 if (ia_valid & ATTR_FORCE)
299 sa_valid |= MDS_ATTR_FORCE;
300 if (ia_valid & ATTR_ATTR_FLAG)
301 sa_valid |= MDS_ATTR_ATTR_FLAG;
302 if (ia_valid & ATTR_KILL_SUID)
303 sa_valid |= MDS_ATTR_KILL_SUID;
304 if (ia_valid & ATTR_KILL_SGID)
305 sa_valid |= MDS_ATTR_KILL_SGID;
306 if (ia_valid & ATTR_CTIME_SET)
307 sa_valid |= MDS_ATTR_CTIME_SET;
308 if (ia_valid & ATTR_FROM_OPEN)
309 sa_valid |= MDS_ATTR_FROM_OPEN;
310 if (ia_valid & ATTR_BLOCKS)
311 sa_valid |= MDS_ATTR_BLOCKS;
312 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
313 /* NFSD hack (see bug 5781) */
314 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
318 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
319 struct md_op_data *op_data)
321 rec->sa_opcode = REINT_SETATTR;
322 rec->sa_fsuid = current_fsuid();
323 rec->sa_fsgid = current_fsgid();
324 rec->sa_cap = cfs_curproc_cap_pack();
325 rec->sa_suppgid = -1;
327 rec->sa_fid = op_data->op_fid1;
328 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
329 rec->sa_mode = op_data->op_attr.ia_mode;
330 rec->sa_uid = op_data->op_attr.ia_uid;
331 rec->sa_gid = op_data->op_attr.ia_gid;
332 rec->sa_size = op_data->op_attr.ia_size;
333 rec->sa_blocks = op_data->op_attr_blocks;
334 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
335 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
336 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
337 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
338 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
339 in_group_p(op_data->op_attr.ia_gid))
340 rec->sa_suppgid = op_data->op_attr.ia_gid;
342 rec->sa_suppgid = op_data->op_suppgids[0];
344 rec->sa_bias = op_data->op_bias;
347 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
348 struct md_op_data *op_data)
350 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
351 epoch->ioepoch = op_data->op_ioepoch;
352 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
355 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
356 void *ea, int ealen, void *ea2, int ea2len)
358 struct mdt_rec_setattr *rec;
359 struct mdt_ioepoch *epoch;
360 struct lov_user_md *lum = NULL;
362 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
363 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
364 mdc_setattr_pack_rec(rec, op_data);
366 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
368 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
369 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
370 mdc_ioepoch_pack(epoch, op_data);
376 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
377 if (ea == NULL) { /* Remove LOV EA */
378 lum->lmm_magic = LOV_USER_MAGIC_V1;
379 lum->lmm_stripe_size = 0;
380 lum->lmm_stripe_count = 0;
381 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
383 memcpy(lum, ea, ealen);
389 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
393 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
395 struct mdt_rec_unlink *rec;
398 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
399 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
400 LASSERT(rec != NULL);
402 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
403 REINT_RMENTRY : REINT_UNLINK;
404 rec->ul_fsuid = op_data->op_fsuid;
405 rec->ul_fsgid = op_data->op_fsgid;
406 rec->ul_cap = op_data->op_cap;
407 rec->ul_mode = op_data->op_mode;
408 rec->ul_suppgid1= op_data->op_suppgids[0];
409 rec->ul_suppgid2= -1;
410 rec->ul_fid1 = op_data->op_fid1;
411 rec->ul_fid2 = op_data->op_fid2;
412 rec->ul_time = op_data->op_mod_time;
413 rec->ul_bias = op_data->op_bias;
415 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
417 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
418 LASSERT(tmp != NULL);
419 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
422 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
424 struct mdt_rec_link *rec;
427 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
428 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
429 LASSERT (rec != NULL);
431 rec->lk_opcode = REINT_LINK;
432 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
433 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
434 rec->lk_cap = op_data->op_cap;//current->cap_effective;
435 rec->lk_suppgid1 = op_data->op_suppgids[0];
436 rec->lk_suppgid2 = op_data->op_suppgids[1];
437 rec->lk_fid1 = op_data->op_fid1;
438 rec->lk_fid2 = op_data->op_fid2;
439 rec->lk_time = op_data->op_mod_time;
440 rec->lk_bias = op_data->op_bias;
442 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
443 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
445 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
446 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
449 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
450 const char *old, int oldlen, const char *new, int newlen)
452 struct mdt_rec_rename *rec;
455 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
456 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
458 /* XXX do something about time, uid, gid */
459 rec->rn_opcode = op_data->op_cli_flags & CLI_MIGRATE ?
460 REINT_MIGRATE : REINT_RENAME;
461 rec->rn_fsuid = op_data->op_fsuid;
462 rec->rn_fsgid = op_data->op_fsgid;
463 rec->rn_cap = op_data->op_cap;
464 rec->rn_suppgid1 = op_data->op_suppgids[0];
465 rec->rn_suppgid2 = op_data->op_suppgids[1];
466 rec->rn_fid1 = op_data->op_fid1;
467 rec->rn_fid2 = op_data->op_fid2;
468 rec->rn_time = op_data->op_mod_time;
469 rec->rn_mode = op_data->op_mode;
470 rec->rn_bias = op_data->op_bias;
472 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
473 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
475 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
476 LOGL0(old, oldlen, tmp);
479 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
480 LOGL0(new, newlen, tmp);
484 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
485 struct md_op_data *op_data, int ea_size)
487 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
491 if (op_data->op_bias & MDS_CHECK_SPLIT)
492 b->valid |= OBD_MD_FLCKSPLIT;
493 if (op_data->op_bias & MDS_CROSS_REF)
494 b->valid |= OBD_MD_FLCROSSREF;
495 b->eadatasize = ea_size;
497 __mdc_pack_body(b, op_data->op_suppgids[0]);
499 b->fid1 = op_data->op_fid1;
500 b->fid2 = op_data->op_fid2;
501 b->valid |= OBD_MD_FLID;
503 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
505 if (op_data->op_name) {
506 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
507 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
512 static void mdc_hsm_release_pack(struct ptlrpc_request *req,
513 struct md_op_data *op_data)
515 if (op_data->op_bias & MDS_HSM_RELEASE) {
516 struct close_data *data;
517 struct ldlm_lock *lock;
519 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
520 LASSERT(data != NULL);
522 lock = ldlm_handle2lock(&op_data->op_lease_handle);
524 data->cd_handle = lock->l_remote_handle;
527 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
529 data->cd_data_version = op_data->op_data_version;
530 data->cd_fid = op_data->op_fid2;
534 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
536 struct mdt_ioepoch *epoch;
537 struct mdt_rec_setattr *rec;
539 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
540 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
542 mdc_setattr_pack_rec(rec, op_data);
543 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
544 mdc_ioepoch_pack(epoch, op_data);
545 mdc_hsm_release_pack(req, op_data);
548 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
552 client_obd_list_lock(&cli->cl_loi_list_lock);
553 rc = cfs_list_empty(&mcw->mcw_entry);
554 client_obd_list_unlock(&cli->cl_loi_list_lock);
558 /* We record requests in flight in cli->cl_r_in_flight here.
559 * There is only one write rpc possible in mdc anyway. If this to change
560 * in the future - the code may need to be revisited. */
561 int mdc_enter_request(struct client_obd *cli)
564 struct mdc_cache_waiter mcw;
565 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
567 client_obd_list_lock(&cli->cl_loi_list_lock);
568 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
569 cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
570 init_waitqueue_head(&mcw.mcw_waitq);
571 client_obd_list_unlock(&cli->cl_loi_list_lock);
572 rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
574 client_obd_list_lock(&cli->cl_loi_list_lock);
575 if (cfs_list_empty(&mcw.mcw_entry))
576 cli->cl_r_in_flight--;
577 cfs_list_del_init(&mcw.mcw_entry);
578 client_obd_list_unlock(&cli->cl_loi_list_lock);
581 cli->cl_r_in_flight++;
582 client_obd_list_unlock(&cli->cl_loi_list_lock);
587 void mdc_exit_request(struct client_obd *cli)
590 struct mdc_cache_waiter *mcw;
592 client_obd_list_lock(&cli->cl_loi_list_lock);
593 cli->cl_r_in_flight--;
594 cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
595 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
596 /* No free request slots anymore */
600 mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
601 cfs_list_del_init(&mcw->mcw_entry);
602 cli->cl_r_in_flight++;
603 wake_up(&mcw->mcw_waitq);
605 /* Empty waiting list? Decrease reqs in-flight number */
607 client_obd_list_unlock(&cli->cl_loi_list_lock);