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, Whamcloud, Inc.
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_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 set_mrc_cr_flags(rec, 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 __u64 mds_pack_open_flags(__u32 flags, __u32 mode)
169 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
170 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
171 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK |
174 cr_flags |= MDS_OPEN_CREAT;
176 cr_flags |= MDS_OPEN_EXCL;
178 cr_flags |= MDS_OPEN_TRUNC;
179 if (flags & O_APPEND)
180 cr_flags |= MDS_OPEN_APPEND;
182 cr_flags |= MDS_OPEN_SYNC;
183 if (flags & O_DIRECTORY)
184 cr_flags |= MDS_OPEN_DIRECTORY;
186 if (flags & FMODE_EXEC)
187 cr_flags |= MDS_FMODE_EXEC;
189 if (flags & O_LOV_DELAY_CREATE)
190 cr_flags |= MDS_OPEN_DELAY_CREATE;
192 if ((flags & O_NOACCESS) || (flags & O_NONBLOCK))
193 cr_flags |= MDS_OPEN_NORESTORE;
198 /* packing of MDS records */
199 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
200 __u32 mode, __u64 rdev, __u32 flags, const void *lmm,
203 struct mdt_rec_create *rec;
207 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
208 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
210 /* XXX do something about time, uid, gid */
211 rec->cr_opcode = REINT_OPEN;
212 rec->cr_fsuid = cfs_curproc_fsuid();
213 rec->cr_fsgid = cfs_curproc_fsgid();
214 rec->cr_cap = cfs_curproc_cap_pack();
215 if (op_data != NULL) {
216 rec->cr_fid1 = op_data->op_fid1;
217 rec->cr_fid2 = op_data->op_fid2;
220 cr_flags = mds_pack_open_flags(flags, mode);
222 rec->cr_time = op_data->op_mod_time;
223 rec->cr_suppgid1 = op_data->op_suppgids[0];
224 rec->cr_suppgid2 = op_data->op_suppgids[1];
225 rec->cr_bias = op_data->op_bias;
227 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
228 /* the next buffer is child capa, which is used for replay,
229 * will be packed from the data in reply message. */
231 if (op_data->op_name) {
232 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
233 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
237 cr_flags |= MDS_OPEN_HAS_EA;
239 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
240 rec->cr_fid2 = op_data->op_fid2;
242 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
243 memcpy (tmp, lmm, lmmlen);
245 set_mrc_cr_flags(rec, cr_flags);
248 static inline __u64 attr_pack(unsigned int ia_valid) {
251 if (ia_valid & ATTR_MODE)
252 sa_valid |= MDS_ATTR_MODE;
253 if (ia_valid & ATTR_UID)
254 sa_valid |= MDS_ATTR_UID;
255 if (ia_valid & ATTR_GID)
256 sa_valid |= MDS_ATTR_GID;
257 if (ia_valid & ATTR_SIZE)
258 sa_valid |= MDS_ATTR_SIZE;
259 if (ia_valid & ATTR_ATIME)
260 sa_valid |= MDS_ATTR_ATIME;
261 if (ia_valid & ATTR_MTIME)
262 sa_valid |= MDS_ATTR_MTIME;
263 if (ia_valid & ATTR_CTIME)
264 sa_valid |= MDS_ATTR_CTIME;
265 if (ia_valid & ATTR_ATIME_SET)
266 sa_valid |= MDS_ATTR_ATIME_SET;
267 if (ia_valid & ATTR_MTIME_SET)
268 sa_valid |= MDS_ATTR_MTIME_SET;
269 if (ia_valid & ATTR_FORCE)
270 sa_valid |= MDS_ATTR_FORCE;
271 if (ia_valid & ATTR_ATTR_FLAG)
272 sa_valid |= MDS_ATTR_ATTR_FLAG;
273 if (ia_valid & ATTR_KILL_SUID)
274 sa_valid |= MDS_ATTR_KILL_SUID;
275 if (ia_valid & ATTR_KILL_SGID)
276 sa_valid |= MDS_ATTR_KILL_SGID;
277 if (ia_valid & ATTR_CTIME_SET)
278 sa_valid |= MDS_ATTR_CTIME_SET;
279 if (ia_valid & ATTR_FROM_OPEN)
280 sa_valid |= MDS_ATTR_FROM_OPEN;
281 if (ia_valid & ATTR_BLOCKS)
282 sa_valid |= MDS_ATTR_BLOCKS;
283 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
284 /* NFSD hack (see bug 5781) */
285 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
289 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
290 struct md_op_data *op_data)
292 rec->sa_opcode = REINT_SETATTR;
293 rec->sa_fsuid = cfs_curproc_fsuid();
294 rec->sa_fsgid = cfs_curproc_fsgid();
295 rec->sa_cap = cfs_curproc_cap_pack();
296 rec->sa_suppgid = -1;
298 rec->sa_fid = op_data->op_fid1;
299 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
300 rec->sa_mode = op_data->op_attr.ia_mode;
301 rec->sa_uid = op_data->op_attr.ia_uid;
302 rec->sa_gid = op_data->op_attr.ia_gid;
303 rec->sa_size = op_data->op_attr.ia_size;
304 rec->sa_blocks = op_data->op_attr_blocks;
305 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
306 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
307 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
308 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
309 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
310 cfs_curproc_is_in_groups(op_data->op_attr.ia_gid))
311 rec->sa_suppgid = op_data->op_attr.ia_gid;
313 rec->sa_suppgid = op_data->op_suppgids[0];
316 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
317 struct md_op_data *op_data)
319 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
320 epoch->ioepoch = op_data->op_ioepoch;
321 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
324 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
325 void *ea, int ealen, void *ea2, int ea2len)
327 struct mdt_rec_setattr *rec;
328 struct mdt_ioepoch *epoch;
329 struct lov_user_md *lum = NULL;
331 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
332 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
333 mdc_setattr_pack_rec(rec, op_data);
335 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
337 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
338 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
339 mdc_ioepoch_pack(epoch, op_data);
345 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
346 if (ea == NULL) { /* Remove LOV EA */
347 lum->lmm_magic = LOV_USER_MAGIC_V1;
348 lum->lmm_stripe_size = 0;
349 lum->lmm_stripe_count = 0;
350 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
352 memcpy(lum, ea, ealen);
358 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
362 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
364 struct mdt_rec_unlink *rec;
367 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
368 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
369 LASSERT (rec != NULL);
371 rec->ul_opcode = REINT_UNLINK;
372 rec->ul_fsuid = op_data->op_fsuid;
373 rec->ul_fsgid = op_data->op_fsgid;
374 rec->ul_cap = op_data->op_cap;
375 rec->ul_mode = op_data->op_mode;
376 rec->ul_suppgid1= op_data->op_suppgids[0];
377 rec->ul_suppgid2= -1;
378 rec->ul_fid1 = op_data->op_fid1;
379 rec->ul_fid2 = op_data->op_fid2;
380 rec->ul_time = op_data->op_mod_time;
381 rec->ul_bias = op_data->op_bias;
383 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
385 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
386 LASSERT(tmp != NULL);
387 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
390 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
392 struct mdt_rec_link *rec;
395 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
396 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
397 LASSERT (rec != NULL);
399 rec->lk_opcode = REINT_LINK;
400 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
401 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
402 rec->lk_cap = op_data->op_cap;//current->cap_effective;
403 rec->lk_suppgid1 = op_data->op_suppgids[0];
404 rec->lk_suppgid2 = op_data->op_suppgids[1];
405 rec->lk_fid1 = op_data->op_fid1;
406 rec->lk_fid2 = op_data->op_fid2;
407 rec->lk_time = op_data->op_mod_time;
408 rec->lk_bias = op_data->op_bias;
410 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
411 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
413 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
414 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
417 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
418 const char *old, int oldlen, const char *new, int newlen)
420 struct mdt_rec_rename *rec;
423 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
424 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
426 /* XXX do something about time, uid, gid */
427 rec->rn_opcode = REINT_RENAME;
428 rec->rn_fsuid = op_data->op_fsuid;
429 rec->rn_fsgid = op_data->op_fsgid;
430 rec->rn_cap = op_data->op_cap;
431 rec->rn_suppgid1 = op_data->op_suppgids[0];
432 rec->rn_suppgid2 = op_data->op_suppgids[1];
433 rec->rn_fid1 = op_data->op_fid1;
434 rec->rn_fid2 = op_data->op_fid2;
435 rec->rn_time = op_data->op_mod_time;
436 rec->rn_mode = op_data->op_mode;
437 rec->rn_bias = op_data->op_bias;
439 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
440 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
442 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
443 LOGL0(old, oldlen, tmp);
446 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
447 LOGL0(new, newlen, tmp);
451 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
452 struct md_op_data *op_data, int ea_size)
454 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
458 if (op_data->op_bias & MDS_CHECK_SPLIT)
459 b->valid |= OBD_MD_FLCKSPLIT;
460 if (op_data->op_bias & MDS_CROSS_REF)
461 b->valid |= OBD_MD_FLCROSSREF;
462 b->eadatasize = ea_size;
464 __mdc_pack_body(b, 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_ioepoch *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_ioepoch_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 = cfs_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 int mdc_enter_request(struct client_obd *cli)
508 struct mdc_cache_waiter mcw;
509 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
511 client_obd_list_lock(&cli->cl_loi_list_lock);
512 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
513 cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
514 cfs_waitq_init(&mcw.mcw_waitq);
515 client_obd_list_unlock(&cli->cl_loi_list_lock);
516 rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
518 client_obd_list_lock(&cli->cl_loi_list_lock);
519 if (cfs_list_empty(&mcw.mcw_entry))
520 cli->cl_r_in_flight--;
521 cfs_list_del_init(&mcw.mcw_entry);
522 client_obd_list_unlock(&cli->cl_loi_list_lock);
525 cli->cl_r_in_flight++;
526 client_obd_list_unlock(&cli->cl_loi_list_lock);
531 void mdc_exit_request(struct client_obd *cli)
534 struct mdc_cache_waiter *mcw;
536 client_obd_list_lock(&cli->cl_loi_list_lock);
537 cli->cl_r_in_flight--;
538 cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
539 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
540 /* No free request slots anymore */
544 mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
545 cfs_list_del_init(&mcw->mcw_entry);
546 cli->cl_r_in_flight++;
547 cfs_waitq_signal(&mcw->mcw_waitq);
549 /* Empty waiting list? Decrease reqs in-flight number */
551 client_obd_list_unlock(&cli->cl_loi_list_lock);