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 (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
39 #define DEBUG_SUBSYSTEM S_MDC
42 # include <liblustre.h>
44 #include <lustre_net.h>
45 #include <lustre/lustre_idl.h>
46 #include "mdc_internal.h"
49 /* some liblustre hackings here */
55 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
60 b->uid = cfs_curproc_uid();
61 b->gid = cfs_curproc_gid();
62 b->fsuid = cfs_curproc_fsuid();
63 b->fsgid = cfs_curproc_fsgid();
64 b->capability = cfs_curproc_cap_pack();
67 void mdc_pack_capa(struct ptlrpc_request *req, const struct req_msg_field *field,
70 struct req_capsule *pill = &req->rq_pill;
71 struct lustre_capa *c;
74 LASSERT(req_capsule_get_size(pill, field, RCL_CLIENT) == 0);
78 c = req_capsule_client_get(pill, field);
81 DEBUG_CAPA(D_SEC, c, "pack");
84 void mdc_is_subdir_pack(struct ptlrpc_request *req, const struct lu_fid *pfid,
85 const struct lu_fid *cfid, int flags)
87 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
92 b->valid = OBD_MD_FLID;
99 void mdc_pack_body(struct ptlrpc_request *req,
100 const struct lu_fid *fid, struct obd_capa *oc,
101 __u64 valid, int ea_size, __u32 suppgid, int flags)
103 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
107 b->eadatasize = ea_size;
109 __mdc_pack_body(b, suppgid);
112 b->valid |= OBD_MD_FLID;
113 mdc_pack_capa(req, &RMF_CAPA1, oc);
117 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff,
118 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
120 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
123 b->valid |= OBD_MD_FLID;
124 b->size = pgoff; /* !! */
125 b->nlink = size; /* !! */
126 __mdc_pack_body(b, -1);
127 b->mode = LUDA_FID | LUDA_TYPE;
129 mdc_pack_capa(req, &RMF_CAPA1, oc);
132 /* packing of MDS records */
133 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
134 const void *data, int datalen, __u32 mode,
135 __u32 uid, __u32 gid, cfs_cap_t cap_effective, __u64 rdev)
137 struct mdt_rec_create *rec;
140 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
141 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
144 rec->cr_opcode = REINT_CREATE;
147 rec->cr_cap = cap_effective;
148 rec->cr_fid1 = op_data->op_fid1;
149 rec->cr_fid2 = op_data->op_fid2;
152 rec->cr_time = op_data->op_mod_time;
153 rec->cr_suppgid1 = op_data->op_suppgids[0];
154 rec->cr_suppgid2 = op_data->op_suppgids[1];
155 set_mrc_cr_flags(rec, op_data->op_flags & MF_SOM_LOCAL_FLAGS);
156 rec->cr_bias = op_data->op_bias;
158 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
160 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
161 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
164 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
165 memcpy(tmp, data, datalen);
169 static __u64 mds_pack_open_flags(__u32 flags, __u32 mode)
171 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
172 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
173 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK));
175 cr_flags |= MDS_OPEN_CREAT;
177 cr_flags |= MDS_OPEN_EXCL;
179 cr_flags |= MDS_OPEN_TRUNC;
180 if (flags & O_APPEND)
181 cr_flags |= MDS_OPEN_APPEND;
183 cr_flags |= MDS_OPEN_SYNC;
184 if (flags & O_DIRECTORY)
185 cr_flags |= MDS_OPEN_DIRECTORY;
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;
193 if ((flags & O_NOACCESS) || (flags & O_NONBLOCK))
194 cr_flags |= MDS_OPEN_NORESTORE;
199 /* packing of MDS records */
200 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
201 __u32 mode, __u64 rdev, __u32 flags, const void *lmm,
204 struct mdt_rec_create *rec;
208 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
209 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
211 /* XXX do something about time, uid, gid */
212 rec->cr_opcode = REINT_OPEN;
213 rec->cr_fsuid = cfs_curproc_fsuid();
214 rec->cr_fsgid = cfs_curproc_fsgid();
215 rec->cr_cap = cfs_curproc_cap_pack();
216 if (op_data != NULL) {
217 rec->cr_fid1 = op_data->op_fid1;
218 rec->cr_fid2 = op_data->op_fid2;
221 cr_flags = mds_pack_open_flags(flags, mode);
223 rec->cr_time = op_data->op_mod_time;
224 rec->cr_suppgid1 = op_data->op_suppgids[0];
225 rec->cr_suppgid2 = op_data->op_suppgids[1];
226 rec->cr_bias = op_data->op_bias;
228 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
229 /* the next buffer is child capa, which is used for replay,
230 * will be packed from the data in reply message. */
232 if (op_data->op_name) {
233 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
234 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
238 cr_flags |= MDS_OPEN_HAS_EA;
240 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
241 rec->cr_fid2 = op_data->op_fid2;
243 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
244 memcpy (tmp, lmm, lmmlen);
246 set_mrc_cr_flags(rec, cr_flags);
249 static inline __u64 attr_pack(unsigned int ia_valid) {
252 if (ia_valid & ATTR_MODE)
253 sa_valid |= MDS_ATTR_MODE;
254 if (ia_valid & ATTR_UID)
255 sa_valid |= MDS_ATTR_UID;
256 if (ia_valid & ATTR_GID)
257 sa_valid |= MDS_ATTR_GID;
258 if (ia_valid & ATTR_SIZE)
259 sa_valid |= MDS_ATTR_SIZE;
260 if (ia_valid & ATTR_ATIME)
261 sa_valid |= MDS_ATTR_ATIME;
262 if (ia_valid & ATTR_MTIME)
263 sa_valid |= MDS_ATTR_MTIME;
264 if (ia_valid & ATTR_CTIME)
265 sa_valid |= MDS_ATTR_CTIME;
266 if (ia_valid & ATTR_ATIME_SET)
267 sa_valid |= MDS_ATTR_ATIME_SET;
268 if (ia_valid & ATTR_MTIME_SET)
269 sa_valid |= MDS_ATTR_MTIME_SET;
270 if (ia_valid & ATTR_FORCE)
271 sa_valid |= MDS_ATTR_FORCE;
272 if (ia_valid & ATTR_ATTR_FLAG)
273 sa_valid |= MDS_ATTR_ATTR_FLAG;
274 if (ia_valid & ATTR_KILL_SUID)
275 sa_valid |= MDS_ATTR_KILL_SUID;
276 if (ia_valid & ATTR_KILL_SGID)
277 sa_valid |= MDS_ATTR_KILL_SGID;
278 if (ia_valid & ATTR_CTIME_SET)
279 sa_valid |= MDS_ATTR_CTIME_SET;
280 if (ia_valid & ATTR_FROM_OPEN)
281 sa_valid |= MDS_ATTR_FROM_OPEN;
282 if (ia_valid & ATTR_BLOCKS)
283 sa_valid |= MDS_ATTR_BLOCKS;
284 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
285 /* NFSD hack (see bug 5781) */
286 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
290 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
291 struct md_op_data *op_data)
293 rec->sa_opcode = REINT_SETATTR;
294 rec->sa_fsuid = cfs_curproc_fsuid();
295 rec->sa_fsgid = cfs_curproc_fsgid();
296 rec->sa_cap = cfs_curproc_cap_pack();
297 rec->sa_suppgid = -1;
299 rec->sa_fid = op_data->op_fid1;
300 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
301 rec->sa_mode = op_data->op_attr.ia_mode;
302 rec->sa_uid = op_data->op_attr.ia_uid;
303 rec->sa_gid = op_data->op_attr.ia_gid;
304 rec->sa_size = op_data->op_attr.ia_size;
305 rec->sa_blocks = op_data->op_attr_blocks;
306 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
307 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
308 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
309 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
310 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
311 cfs_curproc_is_in_groups(op_data->op_attr.ia_gid))
312 rec->sa_suppgid = op_data->op_attr.ia_gid;
314 rec->sa_suppgid = op_data->op_suppgids[0];
317 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
318 struct md_op_data *op_data)
320 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
321 epoch->ioepoch = op_data->op_ioepoch;
322 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
325 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
326 void *ea, int ealen, void *ea2, int ea2len)
328 struct mdt_rec_setattr *rec;
329 struct mdt_ioepoch *epoch;
330 struct lov_user_md *lum = NULL;
332 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
333 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
334 mdc_setattr_pack_rec(rec, op_data);
336 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
338 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
339 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
340 mdc_ioepoch_pack(epoch, op_data);
346 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
347 if (ea == NULL) { /* Remove LOV EA */
348 lum->lmm_magic = LOV_USER_MAGIC_V1;
349 lum->lmm_stripe_size = 0;
350 lum->lmm_stripe_count = 0;
351 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
353 memcpy(lum, ea, ealen);
359 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
363 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
365 struct mdt_rec_unlink *rec;
368 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
369 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
370 LASSERT (rec != NULL);
372 rec->ul_opcode = REINT_UNLINK;
373 rec->ul_fsuid = op_data->op_fsuid;
374 rec->ul_fsgid = op_data->op_fsgid;
375 rec->ul_cap = op_data->op_cap;
376 rec->ul_mode = op_data->op_mode;
377 rec->ul_suppgid1= op_data->op_suppgids[0];
378 rec->ul_suppgid2= -1;
379 rec->ul_fid1 = op_data->op_fid1;
380 rec->ul_fid2 = op_data->op_fid2;
381 rec->ul_time = op_data->op_mod_time;
382 rec->ul_bias = op_data->op_bias;
384 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
386 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
387 LASSERT(tmp != NULL);
388 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
391 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
393 struct mdt_rec_link *rec;
396 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
397 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
398 LASSERT (rec != NULL);
400 rec->lk_opcode = REINT_LINK;
401 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
402 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
403 rec->lk_cap = op_data->op_cap;//current->cap_effective;
404 rec->lk_suppgid1 = op_data->op_suppgids[0];
405 rec->lk_suppgid2 = op_data->op_suppgids[1];
406 rec->lk_fid1 = op_data->op_fid1;
407 rec->lk_fid2 = op_data->op_fid2;
408 rec->lk_time = op_data->op_mod_time;
409 rec->lk_bias = op_data->op_bias;
411 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
412 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
414 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
415 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
418 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
419 const char *old, int oldlen, const char *new, int newlen)
421 struct mdt_rec_rename *rec;
424 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
425 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
427 /* XXX do something about time, uid, gid */
428 rec->rn_opcode = REINT_RENAME;
429 rec->rn_fsuid = op_data->op_fsuid;
430 rec->rn_fsgid = op_data->op_fsgid;
431 rec->rn_cap = op_data->op_cap;
432 rec->rn_suppgid1 = op_data->op_suppgids[0];
433 rec->rn_suppgid2 = op_data->op_suppgids[1];
434 rec->rn_fid1 = op_data->op_fid1;
435 rec->rn_fid2 = op_data->op_fid2;
436 rec->rn_time = op_data->op_mod_time;
437 rec->rn_mode = op_data->op_mode;
438 rec->rn_bias = op_data->op_bias;
440 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
441 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
443 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
444 LOGL0(old, oldlen, tmp);
447 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
448 LOGL0(new, newlen, tmp);
452 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
453 struct md_op_data *op_data, int ea_size)
455 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
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->eadatasize = ea_size;
465 __mdc_pack_body(b, op_data->op_suppgids[0]);
467 b->fid1 = op_data->op_fid1;
468 b->fid2 = op_data->op_fid2;
469 b->valid |= OBD_MD_FLID;
471 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
473 if (op_data->op_name) {
474 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
475 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
480 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
482 struct mdt_ioepoch *epoch;
483 struct mdt_rec_setattr *rec;
485 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
486 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
488 mdc_setattr_pack_rec(rec, op_data);
489 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
490 mdc_ioepoch_pack(epoch, op_data);
493 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
497 client_obd_list_lock(&cli->cl_loi_list_lock);
498 rc = cfs_list_empty(&mcw->mcw_entry);
499 client_obd_list_unlock(&cli->cl_loi_list_lock);
503 /* We record requests in flight in cli->cl_r_in_flight here.
504 * There is only one write rpc possible in mdc anyway. If this to change
505 * in the future - the code may need to be revisited. */
506 int mdc_enter_request(struct client_obd *cli)
509 struct mdc_cache_waiter mcw;
510 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
512 client_obd_list_lock(&cli->cl_loi_list_lock);
513 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
514 cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
515 cfs_waitq_init(&mcw.mcw_waitq);
516 client_obd_list_unlock(&cli->cl_loi_list_lock);
517 rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
519 client_obd_list_lock(&cli->cl_loi_list_lock);
520 if (cfs_list_empty(&mcw.mcw_entry))
521 cli->cl_r_in_flight--;
522 cfs_list_del_init(&mcw.mcw_entry);
523 client_obd_list_unlock(&cli->cl_loi_list_lock);
526 cli->cl_r_in_flight++;
527 client_obd_list_unlock(&cli->cl_loi_list_lock);
532 void mdc_exit_request(struct client_obd *cli)
535 struct mdc_cache_waiter *mcw;
537 client_obd_list_lock(&cli->cl_loi_list_lock);
538 cli->cl_r_in_flight--;
539 cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
540 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
541 /* No free request slots anymore */
545 mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
546 cfs_list_del_init(&mcw->mcw_entry);
547 cli->cl_r_in_flight++;
548 cfs_waitq_signal(&mcw->mcw_waitq);
550 /* Empty waiting list? Decrease reqs in-flight number */
552 client_obd_list_unlock(&cli->cl_loi_list_lock);