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/lustre_idl.h>
43 #include <lustre_net.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 = current->cap_effective;
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 mdc_pack_capa(req, &RMF_CAPA1, oc);
128 /* packing of MDS records */
129 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
130 const void *data, int datalen, __u32 mode,
131 __u32 uid, __u32 gid, __u32 cap_effective, __u64 rdev)
133 struct mdt_rec_create *rec;
136 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
137 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
140 rec->cr_opcode = REINT_CREATE;
143 rec->cr_cap = cap_effective;
144 rec->cr_fid1 = op_data->op_fid1;
145 rec->cr_fid2 = op_data->op_fid2;
148 rec->cr_time = op_data->op_mod_time;
149 rec->cr_suppgid1 = op_data->op_suppgids[0];
150 rec->cr_suppgid2 = op_data->op_suppgids[1];
151 rec->cr_flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
152 rec->cr_bias = op_data->op_bias;
154 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
156 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
157 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
160 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
161 memcpy(tmp, data, datalen);
165 static __u32 mds_pack_open_flags(__u32 flags)
167 __u32 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
168 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
169 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK));
171 cr_flags |= MDS_OPEN_CREAT;
173 cr_flags |= MDS_OPEN_EXCL;
175 cr_flags |= MDS_OPEN_TRUNC;
176 if (flags & O_APPEND)
177 cr_flags |= MDS_OPEN_APPEND;
179 cr_flags |= MDS_OPEN_SYNC;
180 if (flags & O_DIRECTORY)
181 cr_flags |= MDS_OPEN_DIRECTORY;
182 if (flags & O_JOIN_FILE)
183 cr_flags |= MDS_OPEN_JOIN_FILE;
185 if (flags & FMODE_EXEC)
186 cr_flags |= MDS_FMODE_EXEC;
188 if (flags & O_LOV_DELAY_CREATE)
189 cr_flags |= MDS_OPEN_DELAY_CREATE;
194 /* packing of MDS records */
195 void mdc_join_pack(struct ptlrpc_request *req,
196 struct md_op_data *op_data,
199 struct mdt_rec_join *rec;
201 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_JOINFILE);
202 LASSERT(rec != NULL);
203 rec->jr_fid = op_data->op_fid2;
204 rec->jr_headsize = head_size;
207 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
208 __u32 mode, __u64 rdev, __u32 flags, const void *lmm,
211 struct mdt_rec_create *rec;
214 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
215 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
217 /* XXX do something about time, uid, gid */
218 rec->cr_opcode = REINT_OPEN;
219 rec->cr_fsuid = current->fsuid;
220 rec->cr_fsgid = current->fsgid;
221 rec->cr_cap = current->cap_effective;
222 if (op_data != NULL) {
223 rec->cr_fid1 = op_data->op_fid1;
224 rec->cr_fid2 = op_data->op_fid2;
227 rec->cr_flags = mds_pack_open_flags(flags);
229 rec->cr_time = op_data->op_mod_time;
230 rec->cr_suppgid1 = op_data->op_suppgids[0];
231 rec->cr_suppgid2 = op_data->op_suppgids[1];
232 rec->cr_bias = op_data->op_bias;
234 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
235 /* the next buffer is child capa, which is used for replay,
236 * will be packed from the data in reply message. */
238 if (op_data->op_name) {
239 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
240 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
244 rec->cr_flags |= MDS_OPEN_HAS_EA;
246 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
247 rec->cr_fid2 = op_data->op_fid2;
249 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
250 memcpy (tmp, lmm, lmmlen);
254 static inline __u64 attr_pack(unsigned int ia_valid) {
257 if (ia_valid & ATTR_MODE)
258 sa_valid |= MDS_ATTR_MODE;
259 if (ia_valid & ATTR_UID)
260 sa_valid |= MDS_ATTR_UID;
261 if (ia_valid & ATTR_GID)
262 sa_valid |= MDS_ATTR_GID;
263 if (ia_valid & ATTR_SIZE)
264 sa_valid |= MDS_ATTR_SIZE;
265 if (ia_valid & ATTR_ATIME)
266 sa_valid |= MDS_ATTR_ATIME;
267 if (ia_valid & ATTR_MTIME)
268 sa_valid |= MDS_ATTR_MTIME;
269 if (ia_valid & ATTR_CTIME)
270 sa_valid |= MDS_ATTR_CTIME;
271 if (ia_valid & ATTR_ATIME_SET)
272 sa_valid |= MDS_ATTR_ATIME_SET;
273 if (ia_valid & ATTR_MTIME_SET)
274 sa_valid |= MDS_ATTR_MTIME_SET;
275 if (ia_valid & ATTR_FORCE)
276 sa_valid |= MDS_ATTR_FORCE;
277 if (ia_valid & ATTR_ATTR_FLAG)
278 sa_valid |= MDS_ATTR_ATTR_FLAG;
279 if (ia_valid & ATTR_KILL_SUID)
280 sa_valid |= MDS_ATTR_KILL_SUID;
281 if (ia_valid & ATTR_KILL_SGID)
282 sa_valid |= MDS_ATTR_KILL_SGID;
283 if (ia_valid & ATTR_CTIME_SET)
284 sa_valid |= MDS_ATTR_CTIME_SET;
285 if (ia_valid & ATTR_FROM_OPEN)
286 sa_valid |= MDS_ATTR_FROM_OPEN;
287 if (ia_valid & ATTR_BLOCKS)
288 sa_valid |= MDS_ATTR_BLOCKS;
289 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
290 /* NFSD hack (see bug 5781) */
291 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
295 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
296 struct md_op_data *op_data)
298 rec->sa_opcode = REINT_SETATTR;
299 rec->sa_fsuid = current->fsuid;
300 rec->sa_fsgid = current->fsgid;
301 rec->sa_cap = current->cap_effective;
302 rec->sa_suppgid = -1;
304 rec->sa_fid = op_data->op_fid1;
305 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
306 rec->sa_mode = op_data->op_attr.ia_mode;
307 rec->sa_uid = op_data->op_attr.ia_uid;
308 rec->sa_gid = op_data->op_attr.ia_gid;
309 rec->sa_size = op_data->op_attr.ia_size;
310 rec->sa_blocks = op_data->op_attr_blocks;
311 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
312 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
313 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
314 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
315 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
316 in_group_p(op_data->op_attr.ia_gid))
317 rec->sa_suppgid = op_data->op_attr.ia_gid;
319 rec->sa_suppgid = op_data->op_suppgids[0];
322 static void mdc_epoch_pack(struct mdt_epoch *epoch, struct md_op_data *op_data)
324 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
325 epoch->ioepoch = op_data->op_ioepoch;
326 epoch->flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
329 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
330 void *ea, int ealen, void *ea2, int ea2len)
332 struct mdt_rec_setattr *rec;
333 struct mdt_epoch *epoch;
335 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
336 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
337 mdc_setattr_pack_rec(rec, op_data);
339 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
341 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
342 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
343 mdc_epoch_pack(epoch, op_data);
349 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_EADATA), ea, ealen);
354 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
358 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
360 struct mdt_rec_unlink *rec;
363 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
364 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
365 LASSERT (rec != NULL);
367 rec->ul_opcode = REINT_UNLINK;
368 rec->ul_fsuid = op_data->op_fsuid;
369 rec->ul_fsgid = op_data->op_fsgid;
370 rec->ul_cap = op_data->op_cap;
371 rec->ul_mode = op_data->op_mode;
372 rec->ul_suppgid1= op_data->op_suppgids[0];
373 rec->ul_suppgid2= -1;
374 rec->ul_fid1 = op_data->op_fid1;
375 rec->ul_fid2 = op_data->op_fid2;
376 rec->ul_time = op_data->op_mod_time;
377 rec->ul_bias = op_data->op_bias;
379 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
381 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
382 LASSERT(tmp != NULL);
383 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
386 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
388 struct mdt_rec_link *rec;
391 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
392 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
393 LASSERT (rec != NULL);
395 rec->lk_opcode = REINT_LINK;
396 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
397 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
398 rec->lk_cap = op_data->op_cap;//current->cap_effective;
399 rec->lk_suppgid1 = op_data->op_suppgids[0];
400 rec->lk_suppgid2 = op_data->op_suppgids[1];
401 rec->lk_fid1 = op_data->op_fid1;
402 rec->lk_fid2 = op_data->op_fid2;
403 rec->lk_time = op_data->op_mod_time;
404 rec->lk_bias = op_data->op_bias;
406 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
407 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
409 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
410 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
413 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
414 const char *old, int oldlen, const char *new, int newlen)
416 struct mdt_rec_rename *rec;
419 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
420 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
422 /* XXX do something about time, uid, gid */
423 rec->rn_opcode = REINT_RENAME;
424 rec->rn_fsuid = op_data->op_fsuid;
425 rec->rn_fsgid = op_data->op_fsgid;
426 rec->rn_cap = op_data->op_cap;
427 rec->rn_suppgid1 = op_data->op_suppgids[0];
428 rec->rn_suppgid2 = op_data->op_suppgids[1];
429 rec->rn_fid1 = op_data->op_fid1;
430 rec->rn_fid2 = op_data->op_fid2;
431 rec->rn_time = op_data->op_mod_time;
432 rec->rn_mode = op_data->op_mode;
433 rec->rn_bias = op_data->op_bias;
435 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
436 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
438 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
439 LOGL0(old, oldlen, tmp);
442 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
443 LOGL0(new, newlen, tmp);
447 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
448 struct md_op_data *op_data)
450 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
453 b->fsuid = current->fsuid;
454 b->fsgid = current->fsgid;
455 b->capability = current->cap_effective;
457 if (op_data->op_bias & MDS_CHECK_SPLIT)
458 b->valid |= OBD_MD_FLCKSPLIT;
459 if (op_data->op_bias & MDS_CROSS_REF)
460 b->valid |= OBD_MD_FLCROSSREF;
461 b->flags = flags | MDS_BFLAG_EXT_FLAGS;
462 b->suppgid = op_data->op_suppgids[0];
464 b->fid1 = op_data->op_fid1;
465 b->fid2 = op_data->op_fid2;
466 b->valid |= OBD_MD_FLID;
468 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
470 if (op_data->op_name) {
471 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
472 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
477 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
479 struct mdt_epoch *epoch;
480 struct mdt_rec_setattr *rec;
482 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
483 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
485 mdc_setattr_pack_rec(rec, op_data);
486 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
487 mdc_epoch_pack(epoch, op_data);
490 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
494 spin_lock(&cli->cl_loi_list_lock);
495 rc = list_empty(&mcw->mcw_entry);
496 spin_unlock(&cli->cl_loi_list_lock);
500 /* We record requests in flight in cli->cl_r_in_flight here.
501 * There is only one write rpc possible in mdc anyway. If this to change
502 * in the future - the code may need to be revisited. */
503 void mdc_enter_request(struct client_obd *cli)
505 struct mdc_cache_waiter mcw;
506 struct l_wait_info lwi = { 0 };
508 spin_lock(&cli->cl_loi_list_lock);
509 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
510 list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
511 cfs_waitq_init(&mcw.mcw_waitq);
512 spin_unlock(&cli->cl_loi_list_lock);
513 l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
515 cli->cl_r_in_flight++;
516 spin_unlock(&cli->cl_loi_list_lock);
520 void mdc_exit_request(struct client_obd *cli)
522 struct list_head *l, *tmp;
523 struct mdc_cache_waiter *mcw;
525 spin_lock(&cli->cl_loi_list_lock);
526 cli->cl_r_in_flight--;
527 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
529 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
530 /* No free request slots anymore */
534 mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
535 list_del_init(&mcw->mcw_entry);
536 cli->cl_r_in_flight++;
537 cfs_waitq_signal(&mcw->mcw_waitq);
539 /* Empty waiting list? Decrease reqs in-flight number */
541 spin_unlock(&cli->cl_loi_list_lock);