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.
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 rec->cr_flags = 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 __u32 mds_pack_open_flags(__u32 flags, __u32 mode)
169 __u32 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
170 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
171 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK));
173 cr_flags |= MDS_OPEN_CREAT;
175 cr_flags |= MDS_OPEN_EXCL;
177 cr_flags |= MDS_OPEN_TRUNC;
178 if (flags & O_APPEND)
179 cr_flags |= MDS_OPEN_APPEND;
181 cr_flags |= MDS_OPEN_SYNC;
182 if (flags & O_DIRECTORY)
183 cr_flags |= MDS_OPEN_DIRECTORY;
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_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 = cfs_curproc_fsuid();
208 rec->cr_fsgid = cfs_curproc_fsgid();
209 rec->cr_cap = cfs_curproc_cap_pack();
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, mode);
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 & MDS_OPEN_OWNEROVERRIDE)
278 /* NFSD hack (see bug 5781) */
279 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
283 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
284 struct md_op_data *op_data)
286 rec->sa_opcode = REINT_SETATTR;
287 rec->sa_fsuid = cfs_curproc_fsuid();
288 rec->sa_fsgid = cfs_curproc_fsgid();
289 rec->sa_cap = cfs_curproc_cap_pack();
290 rec->sa_suppgid = -1;
292 rec->sa_fid = op_data->op_fid1;
293 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
294 rec->sa_mode = op_data->op_attr.ia_mode;
295 rec->sa_uid = op_data->op_attr.ia_uid;
296 rec->sa_gid = op_data->op_attr.ia_gid;
297 rec->sa_size = op_data->op_attr.ia_size;
298 rec->sa_blocks = op_data->op_attr_blocks;
299 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
300 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
301 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
302 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
303 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
304 cfs_curproc_is_in_groups(op_data->op_attr.ia_gid))
305 rec->sa_suppgid = op_data->op_attr.ia_gid;
307 rec->sa_suppgid = op_data->op_suppgids[0];
310 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
311 struct md_op_data *op_data)
313 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
314 epoch->ioepoch = op_data->op_ioepoch;
315 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
318 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
319 void *ea, int ealen, void *ea2, int ea2len)
321 struct mdt_rec_setattr *rec;
322 struct mdt_ioepoch *epoch;
323 struct lov_user_md *lum = NULL;
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_ioepoch_pack(epoch, op_data);
339 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
340 if (ea == NULL) { /* Remove LOV EA */
341 lum->lmm_magic = LOV_USER_MAGIC_V1;
342 lum->lmm_stripe_size = 0;
343 lum->lmm_stripe_count = 0;
344 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
346 memcpy(lum, ea, ealen);
352 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
356 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
358 struct mdt_rec_unlink *rec;
361 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
362 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
363 LASSERT (rec != NULL);
365 rec->ul_opcode = REINT_UNLINK;
366 rec->ul_fsuid = op_data->op_fsuid;
367 rec->ul_fsgid = op_data->op_fsgid;
368 rec->ul_cap = op_data->op_cap;
369 rec->ul_mode = op_data->op_mode;
370 rec->ul_suppgid1= op_data->op_suppgids[0];
371 rec->ul_suppgid2= -1;
372 rec->ul_fid1 = op_data->op_fid1;
373 rec->ul_fid2 = op_data->op_fid2;
374 rec->ul_time = op_data->op_mod_time;
375 rec->ul_bias = op_data->op_bias;
377 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
379 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
380 LASSERT(tmp != NULL);
381 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
384 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
386 struct mdt_rec_link *rec;
389 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
390 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
391 LASSERT (rec != NULL);
393 rec->lk_opcode = REINT_LINK;
394 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
395 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
396 rec->lk_cap = op_data->op_cap;//current->cap_effective;
397 rec->lk_suppgid1 = op_data->op_suppgids[0];
398 rec->lk_suppgid2 = op_data->op_suppgids[1];
399 rec->lk_fid1 = op_data->op_fid1;
400 rec->lk_fid2 = op_data->op_fid2;
401 rec->lk_time = op_data->op_mod_time;
402 rec->lk_bias = op_data->op_bias;
404 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
405 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
407 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
408 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
411 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
412 const char *old, int oldlen, const char *new, int newlen)
414 struct mdt_rec_rename *rec;
417 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
418 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
420 /* XXX do something about time, uid, gid */
421 rec->rn_opcode = REINT_RENAME;
422 rec->rn_fsuid = op_data->op_fsuid;
423 rec->rn_fsgid = op_data->op_fsgid;
424 rec->rn_cap = op_data->op_cap;
425 rec->rn_suppgid1 = op_data->op_suppgids[0];
426 rec->rn_suppgid2 = op_data->op_suppgids[1];
427 rec->rn_fid1 = op_data->op_fid1;
428 rec->rn_fid2 = op_data->op_fid2;
429 rec->rn_time = op_data->op_mod_time;
430 rec->rn_mode = op_data->op_mode;
431 rec->rn_bias = op_data->op_bias;
433 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
434 mdc_pack_capa(req, &RMF_CAPA2, op_data->op_capa2);
436 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
437 LOGL0(old, oldlen, tmp);
440 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
441 LOGL0(new, newlen, tmp);
445 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
446 struct md_op_data *op_data)
448 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
451 b->fsuid = cfs_curproc_fsuid();
452 b->fsgid = cfs_curproc_fsgid();
453 b->capability = cfs_curproc_cap_pack();
455 if (op_data->op_bias & MDS_CHECK_SPLIT)
456 b->valid |= OBD_MD_FLCKSPLIT;
457 if (op_data->op_bias & MDS_CROSS_REF)
458 b->valid |= OBD_MD_FLCROSSREF;
460 b->suppgid = op_data->op_suppgids[0];
462 b->fid1 = op_data->op_fid1;
463 b->fid2 = op_data->op_fid2;
464 b->valid |= OBD_MD_FLID;
466 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
468 if (op_data->op_name) {
469 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
470 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
475 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
477 struct mdt_ioepoch *epoch;
478 struct mdt_rec_setattr *rec;
480 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
481 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
483 mdc_setattr_pack_rec(rec, op_data);
484 mdc_pack_capa(req, &RMF_CAPA1, op_data->op_capa1);
485 mdc_ioepoch_pack(epoch, op_data);
488 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
492 client_obd_list_lock(&cli->cl_loi_list_lock);
493 rc = cfs_list_empty(&mcw->mcw_entry);
494 client_obd_list_unlock(&cli->cl_loi_list_lock);
498 /* We record requests in flight in cli->cl_r_in_flight here.
499 * There is only one write rpc possible in mdc anyway. If this to change
500 * in the future - the code may need to be revisited. */
501 void mdc_enter_request(struct client_obd *cli)
503 struct mdc_cache_waiter mcw;
504 struct l_wait_info lwi = { 0 };
506 client_obd_list_lock(&cli->cl_loi_list_lock);
507 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
508 cfs_list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
509 cfs_waitq_init(&mcw.mcw_waitq);
510 client_obd_list_unlock(&cli->cl_loi_list_lock);
511 l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
513 cli->cl_r_in_flight++;
514 client_obd_list_unlock(&cli->cl_loi_list_lock);
518 void mdc_exit_request(struct client_obd *cli)
521 struct mdc_cache_waiter *mcw;
523 client_obd_list_lock(&cli->cl_loi_list_lock);
524 cli->cl_r_in_flight--;
525 cfs_list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
527 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
528 /* No free request slots anymore */
532 mcw = cfs_list_entry(l, struct mdc_cache_waiter, mcw_entry);
533 cfs_list_del_init(&mcw->mcw_entry);
534 cli->cl_r_in_flight++;
535 cfs_waitq_signal(&mcw->mcw_waitq);
537 /* Empty waiting list? Decrease reqs in-flight number */
539 client_obd_list_unlock(&cli->cl_loi_list_lock);