1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2003 Cluster File Systems, Inc.
6 * This file is part of the Lustre file system, http://www.lustre.org
7 * Lustre is a trademark of Cluster File Systems, Inc.
9 * You may have signed or agreed to another license before downloading
10 * this software. If so, you are bound by the terms and conditions
11 * of that agreement, and the following does not apply to you. See the
12 * LICENSE file included with this distribution for more information.
14 * If you did not agree to a different license, then this copy of Lustre
15 * is open source software; you can redistribute it and/or modify it
16 * under the terms of version 2 of the GNU General Public License as
17 * published by the Free Software Foundation.
19 * In either case, Lustre is distributed in the hope that it will be
20 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
21 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * license text for more details.
25 #define DEBUG_SUBSYSTEM S_MDC
28 # include <liblustre.h>
30 #include <lustre/lustre_idl.h>
31 #include <lustre_net.h>
32 #include "mdc_internal.h"
35 /* some liblustre hackings here */
41 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
46 b->uid = current->uid;
47 b->gid = current->gid;
48 b->fsuid = current->fsuid;
49 b->fsgid = current->fsgid;
50 b->capability = current->cap_effective;
53 void mdc_pack_capa(struct ptlrpc_request *req, const struct req_msg_field *field,
56 struct req_capsule *pill = &req->rq_pill;
57 struct lustre_capa *c;
60 LASSERT(req_capsule_get_size(pill, field, RCL_CLIENT) == 0);
64 c = req_capsule_client_get(pill, field);
67 DEBUG_CAPA(D_SEC, c, "pack");
70 void mdc_is_subdir_pack(struct ptlrpc_request *req, const struct lu_fid *pfid,
71 const struct lu_fid *cfid, int flags)
73 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
78 b->valid = OBD_MD_FLID;
85 void mdc_pack_body(struct ptlrpc_request *req,
86 const struct lu_fid *fid, struct obd_capa *oc,
87 __u64 valid, int ea_size, __u32 suppgid, int flags)
89 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
93 b->eadatasize = ea_size;
95 __mdc_pack_body(b, suppgid);
98 b->valid |= OBD_MD_FLID;
99 mdc_pack_capa(req, &RMF_CAPA1, oc);
103 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff,
104 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
106 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
109 b->valid |= OBD_MD_FLID;
110 b->size = pgoff; /* !! */
111 b->nlink = size; /* !! */
112 __mdc_pack_body(b, -1);
113 mdc_pack_capa(req, &RMF_CAPA1, oc);
116 /* packing of MDS records */
117 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
118 const void *data, int datalen, __u32 mode,
119 __u32 uid, __u32 gid, __u32 cap_effective, __u64 rdev)
121 struct mdt_rec_create *rec;
124 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
125 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
128 rec->cr_opcode = REINT_CREATE;
131 rec->cr_cap = cap_effective;
132 rec->cr_fid1 = op_data->op_fid1;
133 rec->cr_fid2 = op_data->op_fid2;
136 rec->cr_time = op_data->op_mod_time;
137 rec->cr_suppgid1 = op_data->op_suppgids[0];
138 rec->cr_suppgid2 = op_data->op_suppgids[1];
139 rec->cr_flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
140 rec->cr_bias = op_data->op_bias;
142 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
144 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
145 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
148 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
149 memcpy(tmp, data, datalen);
153 static __u32 mds_pack_open_flags(__u32 flags)
155 __u32 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
156 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
157 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK));
159 cr_flags |= MDS_OPEN_CREAT;
161 cr_flags |= MDS_OPEN_EXCL;
163 cr_flags |= MDS_OPEN_TRUNC;
164 if (flags & O_APPEND)
165 cr_flags |= MDS_OPEN_APPEND;
167 cr_flags |= MDS_OPEN_SYNC;
168 if (flags & O_DIRECTORY)
169 cr_flags |= MDS_OPEN_DIRECTORY;
170 if (flags & O_JOIN_FILE)
171 cr_flags |= MDS_OPEN_JOIN_FILE;
173 if (flags & FMODE_EXEC)
174 cr_flags |= MDS_FMODE_EXEC;
176 if (flags & O_LOV_DELAY_CREATE)
177 cr_flags |= MDS_OPEN_DELAY_CREATE;
182 /* packing of MDS records */
183 void mdc_join_pack(struct ptlrpc_request *req,
184 struct md_op_data *op_data,
187 struct mdt_rec_join *rec;
189 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_JOINFILE);
190 LASSERT(rec != NULL);
191 rec->jr_fid = op_data->op_fid2;
192 rec->jr_headsize = head_size;
195 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
196 __u32 mode, __u64 rdev, __u32 flags, const void *lmm,
199 struct mdt_rec_create *rec;
202 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
203 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
205 /* XXX do something about time, uid, gid */
206 rec->cr_opcode = REINT_OPEN;
207 rec->cr_fsuid = current->fsuid;
208 rec->cr_fsgid = current->fsgid;
209 rec->cr_cap = current->cap_effective;
210 if (op_data != NULL) {
211 rec->cr_fid1 = op_data->op_fid1;
212 rec->cr_fid2 = op_data->op_fid2;
215 rec->cr_flags = mds_pack_open_flags(flags);
217 rec->cr_time = op_data->op_mod_time;
218 rec->cr_suppgid1 = op_data->op_suppgids[0];
219 rec->cr_suppgid2 = op_data->op_suppgids[1];
220 rec->cr_bias = op_data->op_bias;
222 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
223 /* the next buffer is child capa, which is used for replay,
224 * will be packed from the data in reply message. */
226 if (op_data->op_name) {
227 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
228 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
232 rec->cr_flags |= MDS_OPEN_HAS_EA;
234 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
235 rec->cr_fid2 = op_data->op_fid2;
237 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
238 memcpy (tmp, lmm, lmmlen);
242 static inline __u64 attr_pack(unsigned int ia_valid) {
245 if (ia_valid & ATTR_MODE)
246 sa_valid |= MDS_ATTR_MODE;
247 if (ia_valid & ATTR_UID)
248 sa_valid |= MDS_ATTR_UID;
249 if (ia_valid & ATTR_GID)
250 sa_valid |= MDS_ATTR_GID;
251 if (ia_valid & ATTR_SIZE)
252 sa_valid |= MDS_ATTR_SIZE;
253 if (ia_valid & ATTR_ATIME)
254 sa_valid |= MDS_ATTR_ATIME;
255 if (ia_valid & ATTR_MTIME)
256 sa_valid |= MDS_ATTR_MTIME;
257 if (ia_valid & ATTR_CTIME)
258 sa_valid |= MDS_ATTR_CTIME;
259 if (ia_valid & ATTR_ATIME_SET)
260 sa_valid |= MDS_ATTR_ATIME_SET;
261 if (ia_valid & ATTR_MTIME_SET)
262 sa_valid |= MDS_ATTR_MTIME_SET;
263 if (ia_valid & ATTR_FORCE)
264 sa_valid |= MDS_ATTR_FORCE;
265 if (ia_valid & ATTR_ATTR_FLAG)
266 sa_valid |= MDS_ATTR_ATTR_FLAG;
267 if (ia_valid & ATTR_KILL_SUID)
268 sa_valid |= MDS_ATTR_KILL_SUID;
269 if (ia_valid & ATTR_KILL_SGID)
270 sa_valid |= MDS_ATTR_KILL_SGID;
271 if (ia_valid & ATTR_CTIME_SET)
272 sa_valid |= MDS_ATTR_CTIME_SET;
273 if (ia_valid & ATTR_FROM_OPEN)
274 sa_valid |= MDS_ATTR_FROM_OPEN;
275 if (ia_valid & ATTR_BLOCKS)
276 sa_valid |= MDS_ATTR_BLOCKS;
277 if (ia_valid & ATTR_TRUNC)
278 sa_valid |= MDS_ATTR_TRUNC;
279 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
280 /* NFSD hack (see bug 5781) */
281 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
285 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
286 struct md_op_data *op_data)
288 rec->sa_opcode = REINT_SETATTR;
289 rec->sa_fsuid = current->fsuid;
290 rec->sa_fsgid = current->fsgid;
291 rec->sa_cap = current->cap_effective;
292 rec->sa_suppgid = -1;
294 rec->sa_fid = op_data->op_fid1;
295 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
296 rec->sa_mode = op_data->op_attr.ia_mode;
297 rec->sa_uid = op_data->op_attr.ia_uid;
298 rec->sa_gid = op_data->op_attr.ia_gid;
299 rec->sa_size = op_data->op_attr.ia_size;
300 rec->sa_blocks = op_data->op_attr_blocks;
301 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
302 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
303 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
304 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
305 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
306 in_group_p(op_data->op_attr.ia_gid))
307 rec->sa_suppgid = op_data->op_attr.ia_gid;
309 rec->sa_suppgid = op_data->op_suppgids[0];
312 static void mdc_epoch_pack(struct mdt_epoch *epoch, struct md_op_data *op_data)
314 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
315 epoch->ioepoch = op_data->op_ioepoch;
316 epoch->flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
319 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
320 void *ea, int ealen, void *ea2, int ea2len)
322 struct mdt_rec_setattr *rec;
323 struct mdt_epoch *epoch;
325 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
326 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
327 mdc_setattr_pack_rec(rec, op_data);
329 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
331 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
332 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
333 mdc_epoch_pack(epoch, op_data);
339 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_EADATA), ea, ealen);
344 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
348 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
350 struct mdt_rec_unlink *rec;
353 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
354 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
355 LASSERT (rec != NULL);
357 rec->ul_opcode = REINT_UNLINK;
358 rec->ul_fsuid = op_data->op_fsuid;
359 rec->ul_fsgid = op_data->op_fsgid;
360 rec->ul_cap = op_data->op_cap;
361 rec->ul_mode = op_data->op_mode;
362 rec->ul_suppgid1= op_data->op_suppgids[0];
363 rec->ul_suppgid2= -1;
364 rec->ul_fid1 = op_data->op_fid1;
365 rec->ul_fid2 = op_data->op_fid2;
366 rec->ul_time = op_data->op_mod_time;
367 rec->ul_bias = op_data->op_bias;
369 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
371 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
372 LASSERT(tmp != NULL);
373 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
376 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
378 struct mdt_rec_link *rec;
381 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
382 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
383 LASSERT (rec != NULL);
385 rec->lk_opcode = REINT_LINK;
386 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
387 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
388 rec->lk_cap = op_data->op_cap;//current->cap_effective;
389 rec->lk_suppgid1 = op_data->op_suppgids[0];
390 rec->lk_suppgid2 = op_data->op_suppgids[1];
391 rec->lk_fid1 = op_data->op_fid1;
392 rec->lk_fid2 = op_data->op_fid2;
393 rec->lk_time = op_data->op_mod_time;
394 rec->lk_bias = op_data->op_bias;
396 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
397 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
399 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
400 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
403 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
404 const char *old, int oldlen, const char *new, int newlen)
406 struct mdt_rec_rename *rec;
409 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
410 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
412 /* XXX do something about time, uid, gid */
413 rec->rn_opcode = REINT_RENAME;
414 rec->rn_fsuid = op_data->op_fsuid;
415 rec->rn_fsgid = op_data->op_fsgid;
416 rec->rn_cap = op_data->op_cap;
417 rec->rn_suppgid1 = op_data->op_suppgids[0];
418 rec->rn_suppgid2 = op_data->op_suppgids[1];
419 rec->rn_fid1 = op_data->op_fid1;
420 rec->rn_fid2 = op_data->op_fid2;
421 rec->rn_time = op_data->op_mod_time;
422 rec->rn_mode = op_data->op_mode;
423 rec->rn_bias = op_data->op_bias;
425 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
426 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
428 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
429 LOGL0(old, oldlen, tmp);
432 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
433 LOGL0(new, newlen, tmp);
437 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
438 struct md_op_data *op_data)
440 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
443 b->fsuid = current->fsuid;
444 b->fsgid = current->fsgid;
445 b->capability = current->cap_effective;
447 if (op_data->op_bias & MDS_CHECK_SPLIT)
448 b->valid |= OBD_MD_FLCKSPLIT;
449 if (op_data->op_bias & MDS_CROSS_REF)
450 b->valid |= OBD_MD_FLCROSSREF;
451 b->flags = flags | MDS_BFLAG_EXT_FLAGS;
452 b->suppgid = op_data->op_suppgids[0];
454 b->fid1 = op_data->op_fid1;
455 b->fid2 = op_data->op_fid2;
456 b->valid |= OBD_MD_FLID;
458 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
460 if (op_data->op_name) {
461 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
462 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
467 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
469 struct mdt_epoch *epoch;
470 struct mdt_rec_setattr *rec;
472 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
473 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
475 mdc_setattr_pack_rec(rec, op_data);
476 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
477 mdc_epoch_pack(epoch, op_data);
480 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
484 spin_lock(&cli->cl_loi_list_lock);
485 rc = list_empty(&mcw->mcw_entry);
486 spin_unlock(&cli->cl_loi_list_lock);
490 /* We record requests in flight in cli->cl_r_in_flight here.
491 * There is only one write rpc possible in mdc anyway. If this to change
492 * in the future - the code may need to be revisited. */
493 void mdc_enter_request(struct client_obd *cli)
495 struct mdc_cache_waiter mcw;
496 struct l_wait_info lwi = { 0 };
498 spin_lock(&cli->cl_loi_list_lock);
499 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
500 list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
501 cfs_waitq_init(&mcw.mcw_waitq);
502 spin_unlock(&cli->cl_loi_list_lock);
503 l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
505 cli->cl_r_in_flight++;
506 spin_unlock(&cli->cl_loi_list_lock);
510 void mdc_exit_request(struct client_obd *cli)
512 struct list_head *l, *tmp;
513 struct mdc_cache_waiter *mcw;
515 spin_lock(&cli->cl_loi_list_lock);
516 cli->cl_r_in_flight--;
517 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
519 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
520 /* No free request slots anymore */
524 mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
525 list_del_init(&mcw->mcw_entry);
526 cli->cl_r_in_flight++;
527 cfs_waitq_signal(&mcw->mcw_waitq);
529 /* Empty waiting list? Decrease reqs in-flight number */
531 spin_unlock(&cli->cl_loi_list_lock);