1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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 = current->uid;
59 b->gid = current->gid;
60 b->fsuid = current->fsuid;
61 b->fsgid = current->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_pack_body(struct ptlrpc_request *req,
98 const struct lu_fid *fid, struct obd_capa *oc,
99 __u64 valid, int ea_size, __u32 suppgid, int flags)
101 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
105 b->eadatasize = ea_size;
107 __mdc_pack_body(b, suppgid);
110 b->valid |= OBD_MD_FLID;
111 mdc_pack_capa(req, &RMF_CAPA1, oc);
115 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff,
116 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
118 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
121 b->valid |= OBD_MD_FLID;
122 b->size = pgoff; /* !! */
123 b->nlink = size; /* !! */
124 __mdc_pack_body(b, -1);
125 b->mode = LUDA_FID | LUDA_TYPE;
127 mdc_pack_capa(req, &RMF_CAPA1, oc);
130 /* packing of MDS records */
131 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
132 const void *data, int datalen, __u32 mode,
133 __u32 uid, __u32 gid, cfs_cap_t cap_effective, __u64 rdev)
135 struct mdt_rec_create *rec;
138 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
139 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
142 rec->cr_opcode = REINT_CREATE;
145 rec->cr_cap = cap_effective;
146 rec->cr_fid1 = op_data->op_fid1;
147 rec->cr_fid2 = op_data->op_fid2;
150 rec->cr_time = op_data->op_mod_time;
151 rec->cr_suppgid1 = op_data->op_suppgids[0];
152 rec->cr_suppgid2 = op_data->op_suppgids[1];
153 rec->cr_flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
154 rec->cr_bias = op_data->op_bias;
156 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
158 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
159 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
162 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
163 memcpy(tmp, data, datalen);
167 static __u32 mds_pack_open_flags(__u32 flags, __u32 mode)
169 __u32 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
170 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
171 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK));
173 cr_flags |= MDS_OPEN_CREAT;
175 cr_flags |= MDS_OPEN_EXCL;
177 cr_flags |= MDS_OPEN_TRUNC;
178 if (flags & O_APPEND)
179 cr_flags |= MDS_OPEN_APPEND;
181 cr_flags |= MDS_OPEN_SYNC;
182 if (flags & O_DIRECTORY)
183 cr_flags |= MDS_OPEN_DIRECTORY;
184 if (mode & M_JOIN_FILE)
185 cr_flags |= MDS_OPEN_JOIN_FILE;
187 if (flags & FMODE_EXEC)
188 cr_flags |= MDS_FMODE_EXEC;
190 if (flags & O_LOV_DELAY_CREATE)
191 cr_flags |= MDS_OPEN_DELAY_CREATE;
196 /* packing of MDS records */
197 void mdc_join_pack(struct ptlrpc_request *req,
198 struct md_op_data *op_data,
201 struct mdt_rec_join *rec;
203 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_JOINFILE);
204 LASSERT(rec != NULL);
205 rec->jr_fid = op_data->op_fid2;
206 rec->jr_headsize = head_size;
209 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
210 __u32 mode, __u64 rdev, __u32 flags, const void *lmm,
213 struct mdt_rec_create *rec;
216 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
217 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
219 /* XXX do something about time, uid, gid */
220 rec->cr_opcode = REINT_OPEN;
221 rec->cr_fsuid = current->fsuid;
222 rec->cr_fsgid = current->fsgid;
223 rec->cr_cap = cfs_curproc_cap_pack();
224 if (op_data != NULL) {
225 rec->cr_fid1 = op_data->op_fid1;
226 rec->cr_fid2 = op_data->op_fid2;
229 rec->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;
236 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
237 /* the next buffer is child capa, which is used for replay,
238 * will be packed from the data in reply message. */
240 if (op_data->op_name) {
241 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
242 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
246 rec->cr_flags |= MDS_OPEN_HAS_EA;
248 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
249 rec->cr_fid2 = op_data->op_fid2;
251 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
252 memcpy (tmp, lmm, lmmlen);
256 static inline __u64 attr_pack(unsigned int ia_valid) {
259 if (ia_valid & ATTR_MODE)
260 sa_valid |= MDS_ATTR_MODE;
261 if (ia_valid & ATTR_UID)
262 sa_valid |= MDS_ATTR_UID;
263 if (ia_valid & ATTR_GID)
264 sa_valid |= MDS_ATTR_GID;
265 if (ia_valid & ATTR_SIZE)
266 sa_valid |= MDS_ATTR_SIZE;
267 if (ia_valid & ATTR_ATIME)
268 sa_valid |= MDS_ATTR_ATIME;
269 if (ia_valid & ATTR_MTIME)
270 sa_valid |= MDS_ATTR_MTIME;
271 if (ia_valid & ATTR_CTIME)
272 sa_valid |= MDS_ATTR_CTIME;
273 if (ia_valid & ATTR_ATIME_SET)
274 sa_valid |= MDS_ATTR_ATIME_SET;
275 if (ia_valid & ATTR_MTIME_SET)
276 sa_valid |= MDS_ATTR_MTIME_SET;
277 if (ia_valid & ATTR_FORCE)
278 sa_valid |= MDS_ATTR_FORCE;
279 if (ia_valid & ATTR_ATTR_FLAG)
280 sa_valid |= MDS_ATTR_ATTR_FLAG;
281 if (ia_valid & ATTR_KILL_SUID)
282 sa_valid |= MDS_ATTR_KILL_SUID;
283 if (ia_valid & ATTR_KILL_SGID)
284 sa_valid |= MDS_ATTR_KILL_SGID;
285 if (ia_valid & ATTR_CTIME_SET)
286 sa_valid |= MDS_ATTR_CTIME_SET;
287 if (ia_valid & ATTR_FROM_OPEN)
288 sa_valid |= MDS_ATTR_FROM_OPEN;
289 if (ia_valid & ATTR_BLOCKS)
290 sa_valid |= MDS_ATTR_BLOCKS;
291 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
292 /* NFSD hack (see bug 5781) */
293 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
297 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
298 struct md_op_data *op_data)
300 rec->sa_opcode = REINT_SETATTR;
301 rec->sa_fsuid = current->fsuid;
302 rec->sa_fsgid = current->fsgid;
303 rec->sa_cap = cfs_curproc_cap_pack();
304 rec->sa_suppgid = -1;
306 rec->sa_fid = op_data->op_fid1;
307 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
308 rec->sa_mode = op_data->op_attr.ia_mode;
309 rec->sa_uid = op_data->op_attr.ia_uid;
310 rec->sa_gid = op_data->op_attr.ia_gid;
311 rec->sa_size = op_data->op_attr.ia_size;
312 rec->sa_blocks = op_data->op_attr_blocks;
313 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
314 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
315 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
316 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
317 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
318 in_group_p(op_data->op_attr.ia_gid))
319 rec->sa_suppgid = op_data->op_attr.ia_gid;
321 rec->sa_suppgid = op_data->op_suppgids[0];
324 static void mdc_epoch_pack(struct mdt_epoch *epoch, struct md_op_data *op_data)
326 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
327 epoch->ioepoch = op_data->op_ioepoch;
328 epoch->flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
331 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
332 void *ea, int ealen, void *ea2, int ea2len)
334 struct mdt_rec_setattr *rec;
335 struct mdt_epoch *epoch;
337 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
338 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
339 mdc_setattr_pack_rec(rec, op_data);
341 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
343 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
344 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
345 mdc_epoch_pack(epoch, op_data);
351 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_EADATA), ea, ealen);
356 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
360 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
362 struct mdt_rec_unlink *rec;
365 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
366 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
367 LASSERT (rec != NULL);
369 rec->ul_opcode = REINT_UNLINK;
370 rec->ul_fsuid = op_data->op_fsuid;
371 rec->ul_fsgid = op_data->op_fsgid;
372 rec->ul_cap = op_data->op_cap;
373 rec->ul_mode = op_data->op_mode;
374 rec->ul_suppgid1= op_data->op_suppgids[0];
375 rec->ul_suppgid2= -1;
376 rec->ul_fid1 = op_data->op_fid1;
377 rec->ul_fid2 = op_data->op_fid2;
378 rec->ul_time = op_data->op_mod_time;
379 rec->ul_bias = op_data->op_bias;
381 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
383 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
384 LASSERT(tmp != NULL);
385 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
388 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
390 struct mdt_rec_link *rec;
393 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
394 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
395 LASSERT (rec != NULL);
397 rec->lk_opcode = REINT_LINK;
398 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
399 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
400 rec->lk_cap = op_data->op_cap;//current->cap_effective;
401 rec->lk_suppgid1 = op_data->op_suppgids[0];
402 rec->lk_suppgid2 = op_data->op_suppgids[1];
403 rec->lk_fid1 = op_data->op_fid1;
404 rec->lk_fid2 = op_data->op_fid2;
405 rec->lk_time = op_data->op_mod_time;
406 rec->lk_bias = op_data->op_bias;
408 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
409 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
411 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
412 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
415 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
416 const char *old, int oldlen, const char *new, int newlen)
418 struct mdt_rec_rename *rec;
421 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
422 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
424 /* XXX do something about time, uid, gid */
425 rec->rn_opcode = REINT_RENAME;
426 rec->rn_fsuid = op_data->op_fsuid;
427 rec->rn_fsgid = op_data->op_fsgid;
428 rec->rn_cap = op_data->op_cap;
429 rec->rn_suppgid1 = op_data->op_suppgids[0];
430 rec->rn_suppgid2 = op_data->op_suppgids[1];
431 rec->rn_fid1 = op_data->op_fid1;
432 rec->rn_fid2 = op_data->op_fid2;
433 rec->rn_time = op_data->op_mod_time;
434 rec->rn_mode = op_data->op_mode;
435 rec->rn_bias = op_data->op_bias;
437 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
438 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
440 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
441 LOGL0(old, oldlen, tmp);
444 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
445 LOGL0(new, newlen, tmp);
449 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
450 struct md_op_data *op_data)
452 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
455 b->fsuid = current->fsuid;
456 b->fsgid = current->fsgid;
457 b->capability = cfs_curproc_cap_pack();
459 if (op_data->op_bias & MDS_CHECK_SPLIT)
460 b->valid |= OBD_MD_FLCKSPLIT;
461 if (op_data->op_bias & MDS_CROSS_REF)
462 b->valid |= OBD_MD_FLCROSSREF;
463 b->flags = flags | MDS_BFLAG_EXT_FLAGS;
464 b->suppgid = op_data->op_suppgids[0];
466 b->fid1 = op_data->op_fid1;
467 b->fid2 = op_data->op_fid2;
468 b->valid |= OBD_MD_FLID;
470 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
472 if (op_data->op_name) {
473 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
474 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
479 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
481 struct mdt_epoch *epoch;
482 struct mdt_rec_setattr *rec;
484 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
485 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
487 mdc_setattr_pack_rec(rec, op_data);
488 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
489 mdc_epoch_pack(epoch, op_data);
492 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
496 client_obd_list_lock(&cli->cl_loi_list_lock);
497 rc = list_empty(&mcw->mcw_entry);
498 client_obd_list_unlock(&cli->cl_loi_list_lock);
502 /* We record requests in flight in cli->cl_r_in_flight here.
503 * There is only one write rpc possible in mdc anyway. If this to change
504 * in the future - the code may need to be revisited. */
505 void mdc_enter_request(struct client_obd *cli)
507 struct mdc_cache_waiter mcw;
508 struct l_wait_info lwi = { 0 };
510 client_obd_list_lock(&cli->cl_loi_list_lock);
511 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
512 list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
513 cfs_waitq_init(&mcw.mcw_waitq);
514 client_obd_list_unlock(&cli->cl_loi_list_lock);
515 l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
517 cli->cl_r_in_flight++;
518 client_obd_list_unlock(&cli->cl_loi_list_lock);
522 void mdc_exit_request(struct client_obd *cli)
524 struct list_head *l, *tmp;
525 struct mdc_cache_waiter *mcw;
527 client_obd_list_lock(&cli->cl_loi_list_lock);
528 cli->cl_r_in_flight--;
529 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
531 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
532 /* No free request slots anymore */
536 mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
537 list_del_init(&mcw->mcw_entry);
538 cli->cl_r_in_flight++;
539 cfs_waitq_signal(&mcw->mcw_waitq);
541 /* Empty waiting list? Decrease reqs in-flight number */
543 client_obd_list_unlock(&cli->cl_loi_list_lock);