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
40 # include <liblustre.h>
42 #include <lustre_net.h>
43 #include <lustre/lustre_idl.h>
44 #include "mdc_internal.h"
47 /* some liblustre hackings here */
53 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
58 b->uid = cfs_curproc_uid();
59 b->gid = cfs_curproc_gid();
60 b->fsuid = cfs_curproc_fsuid();
61 b->fsgid = cfs_curproc_fsgid();
62 b->capability = cfs_curproc_cap_pack();
65 void mdc_pack_capa(struct ptlrpc_request *req, const struct req_msg_field *field,
68 struct req_capsule *pill = &req->rq_pill;
69 struct lustre_capa *c;
72 LASSERT(req_capsule_get_size(pill, field, RCL_CLIENT) == 0);
76 c = req_capsule_client_get(pill, field);
79 DEBUG_CAPA(D_SEC, c, "pack");
82 void mdc_is_subdir_pack(struct ptlrpc_request *req, const struct lu_fid *pfid,
83 const struct lu_fid *cfid, int flags)
85 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
90 b->valid = OBD_MD_FLID;
97 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
98 struct md_op_data *op_data)
100 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
103 __mdc_pack_body(b, op_data->op_suppgids[0]);
104 b->fid1 = op_data->op_fid1;
105 b->fid2 = op_data->op_fid2;
106 b->valid |= OBD_MD_FLID;
108 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
109 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
112 void mdc_pack_body(struct ptlrpc_request *req,
113 const struct lu_fid *fid, struct obd_capa *oc,
114 __u64 valid, int ea_size, __u32 suppgid, int flags)
116 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
120 b->eadatasize = ea_size;
122 __mdc_pack_body(b, suppgid);
125 b->valid |= OBD_MD_FLID;
126 mdc_pack_capa(req, &RMF_CAPA1, oc);
130 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff,
131 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
133 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
136 b->valid |= OBD_MD_FLID;
137 b->size = pgoff; /* !! */
138 b->nlink = size; /* !! */
139 __mdc_pack_body(b, -1);
140 b->mode = LUDA_FID | LUDA_TYPE;
142 mdc_pack_capa(req, &RMF_CAPA1, oc);
145 /* packing of MDS records */
146 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
147 const void *data, int datalen, __u32 mode,
148 __u32 uid, __u32 gid, cfs_cap_t cap_effective, __u64 rdev)
150 struct mdt_rec_create *rec;
154 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
155 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
158 rec->cr_opcode = REINT_CREATE;
161 rec->cr_cap = cap_effective;
162 rec->cr_fid1 = op_data->op_fid1;
163 rec->cr_fid2 = op_data->op_fid2;
166 rec->cr_time = op_data->op_mod_time;
167 rec->cr_suppgid1 = op_data->op_suppgids[0];
168 rec->cr_suppgid2 = op_data->op_suppgids[1];
169 flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
170 if (op_data->op_bias & MDS_CREATE_VOLATILE)
171 flags |= MDS_OPEN_VOLATILE;
172 set_mrc_cr_flags(rec, flags);
173 rec->cr_bias = op_data->op_bias;
174 rec->cr_umask = cfs_curproc_umask();
176 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
178 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
179 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
182 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
183 memcpy(tmp, data, datalen);
187 static __u64 mds_pack_open_flags(__u32 flags, __u32 mode)
189 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
190 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
191 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK |
194 cr_flags |= MDS_OPEN_CREAT;
196 cr_flags |= MDS_OPEN_EXCL;
198 cr_flags |= MDS_OPEN_TRUNC;
199 if (flags & O_APPEND)
200 cr_flags |= MDS_OPEN_APPEND;
202 cr_flags |= MDS_OPEN_SYNC;
203 if (flags & O_DIRECTORY)
204 cr_flags |= MDS_OPEN_DIRECTORY;
206 if (flags & FMODE_EXEC)
207 cr_flags |= MDS_FMODE_EXEC;
209 if (flags & O_LOV_DELAY_CREATE)
210 cr_flags |= MDS_OPEN_DELAY_CREATE;
212 if ((flags & O_NOACCESS) || (flags & O_NONBLOCK))
213 cr_flags |= MDS_OPEN_NORESTORE;
218 /* packing of MDS records */
219 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
220 __u32 mode, __u64 rdev, __u32 flags, const void *lmm,
223 struct mdt_rec_create *rec;
227 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
228 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
230 /* XXX do something about time, uid, gid */
231 rec->cr_opcode = REINT_OPEN;
232 rec->cr_fsuid = cfs_curproc_fsuid();
233 rec->cr_fsgid = cfs_curproc_fsgid();
234 rec->cr_cap = cfs_curproc_cap_pack();
235 if (op_data != NULL) {
236 rec->cr_fid1 = op_data->op_fid1;
237 rec->cr_fid2 = op_data->op_fid2;
240 cr_flags = mds_pack_open_flags(flags, mode);
242 rec->cr_time = op_data->op_mod_time;
243 rec->cr_suppgid1 = op_data->op_suppgids[0];
244 rec->cr_suppgid2 = op_data->op_suppgids[1];
245 rec->cr_bias = op_data->op_bias;
246 rec->cr_umask = cfs_curproc_umask();
248 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
249 /* the next buffer is child capa, which is used for replay,
250 * will be packed from the data in reply message. */
252 if (op_data->op_name) {
253 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
254 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
255 if (op_data->op_bias & MDS_CREATE_VOLATILE)
256 cr_flags |= MDS_OPEN_VOLATILE;
260 cr_flags |= MDS_OPEN_HAS_EA;
262 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
263 rec->cr_fid2 = op_data->op_fid2;
265 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
266 memcpy(tmp, lmm, lmmlen);
268 set_mrc_cr_flags(rec, cr_flags);
271 static inline __u64 attr_pack(unsigned int ia_valid) {
274 if (ia_valid & ATTR_MODE)
275 sa_valid |= MDS_ATTR_MODE;
276 if (ia_valid & ATTR_UID)
277 sa_valid |= MDS_ATTR_UID;
278 if (ia_valid & ATTR_GID)
279 sa_valid |= MDS_ATTR_GID;
280 if (ia_valid & ATTR_SIZE)
281 sa_valid |= MDS_ATTR_SIZE;
282 if (ia_valid & ATTR_ATIME)
283 sa_valid |= MDS_ATTR_ATIME;
284 if (ia_valid & ATTR_MTIME)
285 sa_valid |= MDS_ATTR_MTIME;
286 if (ia_valid & ATTR_CTIME)
287 sa_valid |= MDS_ATTR_CTIME;
288 if (ia_valid & ATTR_ATIME_SET)
289 sa_valid |= MDS_ATTR_ATIME_SET;
290 if (ia_valid & ATTR_MTIME_SET)
291 sa_valid |= MDS_ATTR_MTIME_SET;
292 if (ia_valid & ATTR_FORCE)
293 sa_valid |= MDS_ATTR_FORCE;
294 if (ia_valid & ATTR_ATTR_FLAG)
295 sa_valid |= MDS_ATTR_ATTR_FLAG;
296 if (ia_valid & ATTR_KILL_SUID)
297 sa_valid |= MDS_ATTR_KILL_SUID;
298 if (ia_valid & ATTR_KILL_SGID)
299 sa_valid |= MDS_ATTR_KILL_SGID;
300 if (ia_valid & ATTR_CTIME_SET)
301 sa_valid |= MDS_ATTR_CTIME_SET;
302 if (ia_valid & ATTR_FROM_OPEN)
303 sa_valid |= MDS_ATTR_FROM_OPEN;
304 if (ia_valid & ATTR_BLOCKS)
305 sa_valid |= MDS_ATTR_BLOCKS;
306 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
307 /* NFSD hack (see bug 5781) */
308 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
312 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
313 struct md_op_data *op_data)
315 rec->sa_opcode = REINT_SETATTR;
316 rec->sa_fsuid = cfs_curproc_fsuid();
317 rec->sa_fsgid = cfs_curproc_fsgid();
318 rec->sa_cap = cfs_curproc_cap_pack();
319 rec->sa_suppgid = -1;
321 rec->sa_fid = op_data->op_fid1;
322 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
323 rec->sa_mode = op_data->op_attr.ia_mode;
324 rec->sa_uid = op_data->op_attr.ia_uid;
325 rec->sa_gid = op_data->op_attr.ia_gid;
326 rec->sa_size = op_data->op_attr.ia_size;
327 rec->sa_blocks = op_data->op_attr_blocks;
328 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
329 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
330 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
331 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
332 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
333 cfs_curproc_is_in_groups(op_data->op_attr.ia_gid))
334 rec->sa_suppgid = op_data->op_attr.ia_gid;
336 rec->sa_suppgid = op_data->op_suppgids[0];
338 rec->sa_bias = op_data->op_bias;
341 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
342 struct md_op_data *op_data)
344 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
345 epoch->ioepoch = op_data->op_ioepoch;
346 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
349 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
350 void *ea, int ealen, void *ea2, int ea2len)
352 struct mdt_rec_setattr *rec;
353 struct mdt_ioepoch *epoch;
354 struct lov_user_md *lum = NULL;
356 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
357 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
358 mdc_setattr_pack_rec(rec, op_data);
360 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
362 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
363 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
364 mdc_ioepoch_pack(epoch, op_data);
370 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
371 if (ea == NULL) { /* Remove LOV EA */
372 lum->lmm_magic = LOV_USER_MAGIC_V1;
373 lum->lmm_stripe_size = 0;
374 lum->lmm_stripe_count = 0;
375 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
377 memcpy(lum, ea, ealen);
383 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
387 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
389 struct mdt_rec_unlink *rec;
392 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
393 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
394 LASSERT(rec != NULL);
396 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
397 REINT_RMENTRY : REINT_UNLINK;
398 rec->ul_fsuid = op_data->op_fsuid;
399 rec->ul_fsgid = op_data->op_fsgid;
400 rec->ul_cap = op_data->op_cap;
401 rec->ul_mode = op_data->op_mode;
402 rec->ul_suppgid1= op_data->op_suppgids[0];
403 rec->ul_suppgid2= -1;
404 rec->ul_fid1 = op_data->op_fid1;
405 rec->ul_fid2 = op_data->op_fid2;
406 rec->ul_time = op_data->op_mod_time;
407 rec->ul_bias = op_data->op_bias;
409 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
411 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
412 LASSERT(tmp != NULL);
413 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
416 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
418 struct mdt_rec_link *rec;
421 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
422 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
423 LASSERT (rec != NULL);
425 rec->lk_opcode = REINT_LINK;
426 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
427 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
428 rec->lk_cap = op_data->op_cap;//current->cap_effective;
429 rec->lk_suppgid1 = op_data->op_suppgids[0];
430 rec->lk_suppgid2 = op_data->op_suppgids[1];
431 rec->lk_fid1 = op_data->op_fid1;
432 rec->lk_fid2 = op_data->op_fid2;
433 rec->lk_time = op_data->op_mod_time;
434 rec->lk_bias = op_data->op_bias;
436 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
437 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
439 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
440 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
443 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
444 const char *old, int oldlen, const char *new, int newlen)
446 struct mdt_rec_rename *rec;
449 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
450 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
452 /* XXX do something about time, uid, gid */
453 rec->rn_opcode = REINT_RENAME;
454 rec->rn_fsuid = op_data->op_fsuid;
455 rec->rn_fsgid = op_data->op_fsgid;
456 rec->rn_cap = op_data->op_cap;
457 rec->rn_suppgid1 = op_data->op_suppgids[0];
458 rec->rn_suppgid2 = op_data->op_suppgids[1];
459 rec->rn_fid1 = op_data->op_fid1;
460 rec->rn_fid2 = op_data->op_fid2;
461 rec->rn_time = op_data->op_mod_time;
462 rec->rn_mode = op_data->op_mode;
463 rec->rn_bias = op_data->op_bias;
465 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
466 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
468 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
469 LOGL0(old, oldlen, tmp);
472 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
473 LOGL0(new, newlen, tmp);
477 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
478 struct md_op_data *op_data, int ea_size)
480 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
484 if (op_data->op_bias & MDS_CHECK_SPLIT)
485 b->valid |= OBD_MD_FLCKSPLIT;
486 if (op_data->op_bias & MDS_CROSS_REF)
487 b->valid |= OBD_MD_FLCROSSREF;
488 b->eadatasize = ea_size;
490 __mdc_pack_body(b, op_data->op_suppgids[0]);
492 b->fid1 = op_data->op_fid1;
493 b->fid2 = op_data->op_fid2;
494 b->valid |= OBD_MD_FLID;
496 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
498 if (op_data->op_name) {
499 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
500 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
505 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
507 struct mdt_ioepoch *epoch;
508 struct mdt_rec_setattr *rec;
510 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
511 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
513 mdc_setattr_pack_rec(rec, op_data);
514 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
515 mdc_ioepoch_pack(epoch, op_data);
518 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
522 client_obd_list_lock(&cli->cl_loi_list_lock);
523 rc = cfs_list_empty(&mcw->mcw_entry);
524 client_obd_list_unlock(&cli->cl_loi_list_lock);
528 /* We record requests in flight in cli->cl_r_in_flight here.
529 * There is only one write rpc possible in mdc anyway. If this to change
530 * in the future - the code may need to be revisited. */
531 int mdc_enter_request(struct client_obd *cli)
534 struct mdc_cache_waiter mcw;
535 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
537 client_obd_list_lock(&cli->cl_loi_list_lock);
538 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
539 cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
540 cfs_waitq_init(&mcw.mcw_waitq);
541 client_obd_list_unlock(&cli->cl_loi_list_lock);
542 rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
544 client_obd_list_lock(&cli->cl_loi_list_lock);
545 if (cfs_list_empty(&mcw.mcw_entry))
546 cli->cl_r_in_flight--;
547 cfs_list_del_init(&mcw.mcw_entry);
548 client_obd_list_unlock(&cli->cl_loi_list_lock);
551 cli->cl_r_in_flight++;
552 client_obd_list_unlock(&cli->cl_loi_list_lock);
557 void mdc_exit_request(struct client_obd *cli)
560 struct mdc_cache_waiter *mcw;
562 client_obd_list_lock(&cli->cl_loi_list_lock);
563 cli->cl_r_in_flight--;
564 cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
565 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
566 /* No free request slots anymore */
570 mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
571 cfs_list_del_init(&mcw->mcw_entry);
572 cli->cl_r_in_flight++;
573 cfs_waitq_signal(&mcw->mcw_waitq);
575 /* Empty waiting list? Decrease reqs in-flight number */
577 client_obd_list_unlock(&cli->cl_loi_list_lock);