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_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;
139 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
140 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
143 rec->cr_opcode = REINT_CREATE;
146 rec->cr_cap = cap_effective;
147 rec->cr_fid1 = op_data->op_fid1;
148 rec->cr_fid2 = op_data->op_fid2;
151 rec->cr_time = op_data->op_mod_time;
152 rec->cr_suppgid1 = op_data->op_suppgids[0];
153 rec->cr_suppgid2 = op_data->op_suppgids[1];
154 flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
155 if (op_data->op_bias & MDS_CREATE_VOLATILE)
156 flags |= MDS_OPEN_VOLATILE;
157 set_mrc_cr_flags(rec, flags);
158 rec->cr_bias = op_data->op_bias;
159 rec->cr_umask = cfs_curproc_umask();
161 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
163 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
164 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
167 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
168 memcpy(tmp, data, datalen);
172 static __u64 mds_pack_open_flags(__u32 flags, __u32 mode)
174 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
175 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
176 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK |
179 cr_flags |= MDS_OPEN_CREAT;
181 cr_flags |= MDS_OPEN_EXCL;
183 cr_flags |= MDS_OPEN_TRUNC;
184 if (flags & O_APPEND)
185 cr_flags |= MDS_OPEN_APPEND;
187 cr_flags |= MDS_OPEN_SYNC;
188 if (flags & O_DIRECTORY)
189 cr_flags |= MDS_OPEN_DIRECTORY;
191 if (flags & FMODE_EXEC)
192 cr_flags |= MDS_FMODE_EXEC;
194 if (flags & O_LOV_DELAY_CREATE)
195 cr_flags |= MDS_OPEN_DELAY_CREATE;
197 if ((flags & O_NOACCESS) || (flags & O_NONBLOCK))
198 cr_flags |= MDS_OPEN_NORESTORE;
203 /* packing of MDS records */
204 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
205 __u32 mode, __u64 rdev, __u32 flags, const void *lmm,
208 struct mdt_rec_create *rec;
212 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
213 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
215 /* XXX do something about time, uid, gid */
216 rec->cr_opcode = REINT_OPEN;
217 rec->cr_fsuid = cfs_curproc_fsuid();
218 rec->cr_fsgid = cfs_curproc_fsgid();
219 rec->cr_cap = cfs_curproc_cap_pack();
220 if (op_data != NULL) {
221 rec->cr_fid1 = op_data->op_fid1;
222 rec->cr_fid2 = op_data->op_fid2;
225 cr_flags = mds_pack_open_flags(flags, mode);
227 rec->cr_time = op_data->op_mod_time;
228 rec->cr_suppgid1 = op_data->op_suppgids[0];
229 rec->cr_suppgid2 = op_data->op_suppgids[1];
230 rec->cr_bias = op_data->op_bias;
231 rec->cr_umask = cfs_curproc_umask();
233 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
234 /* the next buffer is child capa, which is used for replay,
235 * will be packed from the data in reply message. */
237 if (op_data->op_name) {
238 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
239 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
240 if (op_data->op_bias & MDS_CREATE_VOLATILE)
241 cr_flags |= MDS_OPEN_VOLATILE;
245 cr_flags |= MDS_OPEN_HAS_EA;
247 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
248 rec->cr_fid2 = op_data->op_fid2;
250 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
251 memcpy(tmp, lmm, lmmlen);
253 set_mrc_cr_flags(rec, cr_flags);
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 = cfs_curproc_fsuid();
302 rec->sa_fsgid = cfs_curproc_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 cfs_curproc_is_in_groups(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];
323 rec->sa_bias = op_data->op_bias;
326 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
327 struct md_op_data *op_data)
329 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
330 epoch->ioepoch = op_data->op_ioepoch;
331 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
334 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
335 void *ea, int ealen, void *ea2, int ea2len)
337 struct mdt_rec_setattr *rec;
338 struct mdt_ioepoch *epoch;
339 struct lov_user_md *lum = NULL;
341 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
342 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
343 mdc_setattr_pack_rec(rec, op_data);
345 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
347 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
348 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
349 mdc_ioepoch_pack(epoch, op_data);
355 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
356 if (ea == NULL) { /* Remove LOV EA */
357 lum->lmm_magic = LOV_USER_MAGIC_V1;
358 lum->lmm_stripe_size = 0;
359 lum->lmm_stripe_count = 0;
360 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
362 memcpy(lum, ea, ealen);
368 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
372 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
374 struct mdt_rec_unlink *rec;
377 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
378 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
379 LASSERT (rec != NULL);
381 rec->ul_opcode = REINT_UNLINK;
382 rec->ul_fsuid = op_data->op_fsuid;
383 rec->ul_fsgid = op_data->op_fsgid;
384 rec->ul_cap = op_data->op_cap;
385 rec->ul_mode = op_data->op_mode;
386 rec->ul_suppgid1= op_data->op_suppgids[0];
387 rec->ul_suppgid2= -1;
388 rec->ul_fid1 = op_data->op_fid1;
389 rec->ul_fid2 = op_data->op_fid2;
390 rec->ul_time = op_data->op_mod_time;
391 rec->ul_bias = op_data->op_bias;
393 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
395 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
396 LASSERT(tmp != NULL);
397 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
400 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
402 struct mdt_rec_link *rec;
405 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
406 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
407 LASSERT (rec != NULL);
409 rec->lk_opcode = REINT_LINK;
410 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
411 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
412 rec->lk_cap = op_data->op_cap;//current->cap_effective;
413 rec->lk_suppgid1 = op_data->op_suppgids[0];
414 rec->lk_suppgid2 = op_data->op_suppgids[1];
415 rec->lk_fid1 = op_data->op_fid1;
416 rec->lk_fid2 = op_data->op_fid2;
417 rec->lk_time = op_data->op_mod_time;
418 rec->lk_bias = op_data->op_bias;
420 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
421 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
423 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
424 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
427 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
428 const char *old, int oldlen, const char *new, int newlen)
430 struct mdt_rec_rename *rec;
433 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
434 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
436 /* XXX do something about time, uid, gid */
437 rec->rn_opcode = REINT_RENAME;
438 rec->rn_fsuid = op_data->op_fsuid;
439 rec->rn_fsgid = op_data->op_fsgid;
440 rec->rn_cap = op_data->op_cap;
441 rec->rn_suppgid1 = op_data->op_suppgids[0];
442 rec->rn_suppgid2 = op_data->op_suppgids[1];
443 rec->rn_fid1 = op_data->op_fid1;
444 rec->rn_fid2 = op_data->op_fid2;
445 rec->rn_time = op_data->op_mod_time;
446 rec->rn_mode = op_data->op_mode;
447 rec->rn_bias = op_data->op_bias;
449 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
450 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
452 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
453 LOGL0(old, oldlen, tmp);
456 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
457 LOGL0(new, newlen, tmp);
461 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
462 struct md_op_data *op_data, int ea_size)
464 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
468 if (op_data->op_bias & MDS_CHECK_SPLIT)
469 b->valid |= OBD_MD_FLCKSPLIT;
470 if (op_data->op_bias & MDS_CROSS_REF)
471 b->valid |= OBD_MD_FLCROSSREF;
472 b->eadatasize = ea_size;
474 __mdc_pack_body(b, op_data->op_suppgids[0]);
476 b->fid1 = op_data->op_fid1;
477 b->fid2 = op_data->op_fid2;
478 b->valid |= OBD_MD_FLID;
480 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
482 if (op_data->op_name) {
483 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
484 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
489 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
491 struct mdt_ioepoch *epoch;
492 struct mdt_rec_setattr *rec;
494 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
495 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
497 mdc_setattr_pack_rec(rec, op_data);
498 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
499 mdc_ioepoch_pack(epoch, op_data);
502 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
506 client_obd_list_lock(&cli->cl_loi_list_lock);
507 rc = cfs_list_empty(&mcw->mcw_entry);
508 client_obd_list_unlock(&cli->cl_loi_list_lock);
512 /* We record requests in flight in cli->cl_r_in_flight here.
513 * There is only one write rpc possible in mdc anyway. If this to change
514 * in the future - the code may need to be revisited. */
515 int mdc_enter_request(struct client_obd *cli)
518 struct mdc_cache_waiter mcw;
519 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
521 client_obd_list_lock(&cli->cl_loi_list_lock);
522 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
523 cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
524 cfs_waitq_init(&mcw.mcw_waitq);
525 client_obd_list_unlock(&cli->cl_loi_list_lock);
526 rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
528 client_obd_list_lock(&cli->cl_loi_list_lock);
529 if (cfs_list_empty(&mcw.mcw_entry))
530 cli->cl_r_in_flight--;
531 cfs_list_del_init(&mcw.mcw_entry);
532 client_obd_list_unlock(&cli->cl_loi_list_lock);
535 cli->cl_r_in_flight++;
536 client_obd_list_unlock(&cli->cl_loi_list_lock);
541 void mdc_exit_request(struct client_obd *cli)
544 struct mdc_cache_waiter *mcw;
546 client_obd_list_lock(&cli->cl_loi_list_lock);
547 cli->cl_r_in_flight--;
548 cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
549 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
550 /* No free request slots anymore */
554 mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
555 cfs_list_del_init(&mcw->mcw_entry);
556 cli->cl_r_in_flight++;
557 cfs_waitq_signal(&mcw->mcw_waitq);
559 /* Empty waiting list? Decrease reqs in-flight number */
561 client_obd_list_unlock(&cli->cl_loi_list_lock);