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)
45 b->fsuid = current->fsuid;
46 b->fsgid = current->fsgid;
47 b->capability = current->cap_effective;
50 void mdc_pack_capa(struct ptlrpc_request *req, int offset, struct obd_capa *oc)
52 struct lustre_capa *c;
55 LASSERT(lustre_msg_buflen(req->rq_reqmsg, offset) == 0);
59 c = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*c));
62 DEBUG_CAPA(D_SEC, c, "pack");
65 void mdc_is_subdir_pack(struct ptlrpc_request *req, int offset,
66 const struct lu_fid *pfid,
67 const struct lu_fid *cfid, int flags)
69 struct mdt_body *b = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*b));
75 b->valid = OBD_MD_FLID;
79 void mdc_pack_req_body(struct ptlrpc_request *req, int offset,
80 __u64 valid, const struct lu_fid *fid,
81 struct obd_capa *oc, int ea_size, int flags)
83 struct mdt_body *b = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*b));
86 b->eadatasize = ea_size;
91 mdc_pack_capa(req, offset + 1, oc);
95 void mdc_readdir_pack(struct ptlrpc_request *req, int offset, __u64 pgoff,
96 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
100 b = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*b));
102 b->size = pgoff; /* !! */
104 b->nlink = size; /* !! */
106 mdc_pack_capa(req, offset + 1, oc);
109 /* packing of MDS records */
110 void mdc_create_pack(struct ptlrpc_request *req, int offset,
111 struct md_op_data *op_data, const void *data, int datalen,
112 __u32 mode, __u32 uid, __u32 gid, __u32 cap_effective,
115 struct mdt_rec_create *rec;
118 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*rec));
120 rec->cr_opcode = REINT_CREATE;
123 rec->cr_cap = cap_effective;
124 rec->cr_fid1 = op_data->op_fid1;
125 rec->cr_fid2 = op_data->op_fid2;
128 rec->cr_time = op_data->op_mod_time;
129 rec->cr_suppgid1 = op_data->op_suppgids[0];
130 rec->cr_suppgid2 = op_data->op_suppgids[1];
131 rec->cr_flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
132 rec->cr_bias = op_data->op_bias;
134 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
136 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 2, op_data->op_namelen + 1);
137 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
140 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 3, datalen);
141 memcpy(tmp, data, datalen);
145 static __u32 mds_pack_open_flags(__u32 flags)
147 __u32 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
148 MDS_OPEN_DELAY_CREATE | MDS_OPEN_HAS_EA |
149 MDS_OPEN_HAS_OBJS | MDS_OPEN_OWNEROVERRIDE |
152 cr_flags |= MDS_OPEN_CREAT;
154 cr_flags |= MDS_OPEN_EXCL;
156 cr_flags |= MDS_OPEN_TRUNC;
157 if (flags & O_APPEND)
158 cr_flags |= MDS_OPEN_APPEND;
160 cr_flags |= MDS_OPEN_SYNC;
161 if (flags & O_DIRECTORY)
162 cr_flags |= MDS_OPEN_DIRECTORY;
163 if (flags & O_JOIN_FILE)
164 cr_flags |= MDS_OPEN_JOIN_FILE;
166 if (flags & FMODE_EXEC)
167 cr_flags |= MDS_FMODE_EXEC;
172 /* packing of MDS records */
173 void mdc_join_pack(struct ptlrpc_request *req, int offset,
174 struct md_op_data *op_data, __u64 head_size)
176 struct mdt_rec_join *rec;
178 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*rec));
179 LASSERT(rec != NULL);
180 rec->jr_fid = op_data->op_fid2;
181 rec->jr_headsize = head_size;
184 void mdc_open_pack(struct ptlrpc_request *req, int offset,
185 struct md_op_data *op_data, __u32 mode, __u64 rdev,
186 __u32 flags, const void *lmm, int lmmlen)
188 struct mdt_rec_create *rec;
190 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
192 /* XXX do something about time, uid, gid */
193 rec->cr_opcode = REINT_OPEN;
194 rec->cr_fsuid = current->fsuid;
195 rec->cr_fsgid = current->fsgid;
196 rec->cr_cap = current->cap_effective;
197 if (op_data != NULL) {
198 rec->cr_fid1 = op_data->op_fid1;
199 rec->cr_fid2 = op_data->op_fid2;
202 rec->cr_flags = mds_pack_open_flags(flags);
203 rec->cr_time = op_data->op_mod_time;
205 rec->cr_suppgid1 = op_data->op_suppgids[0];
206 rec->cr_suppgid2 = op_data->op_suppgids[1];
207 rec->cr_bias = op_data->op_bias;
209 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
210 /* the next buffer is child capa, which is used for replay,
211 * will be packed from the data in reply message. */
213 if (op_data->op_name) {
214 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 3,
215 op_data->op_namelen + 1);
216 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
220 rec->cr_flags |= MDS_OPEN_HAS_EA;
222 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
223 rec->cr_fid2 = op_data->op_fid2;
225 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 4, lmmlen);
226 memcpy (tmp, lmm, lmmlen);
230 static inline __u64 attr_pack(unsigned int ia_valid) {
233 if (ia_valid & ATTR_MODE)
234 sa_valid |= MDS_ATTR_MODE;
235 if (ia_valid & ATTR_UID)
236 sa_valid |= MDS_ATTR_UID;
237 if (ia_valid & ATTR_GID)
238 sa_valid |= MDS_ATTR_GID;
239 if (ia_valid & ATTR_SIZE)
240 sa_valid |= MDS_ATTR_SIZE;
241 if (ia_valid & ATTR_ATIME)
242 sa_valid |= MDS_ATTR_ATIME;
243 if (ia_valid & ATTR_MTIME)
244 sa_valid |= MDS_ATTR_MTIME;
245 if (ia_valid & ATTR_CTIME)
246 sa_valid |= MDS_ATTR_CTIME;
247 if (ia_valid & ATTR_ATIME_SET)
248 sa_valid |= MDS_ATTR_ATIME_SET;
249 if (ia_valid & ATTR_MTIME_SET)
250 sa_valid |= MDS_ATTR_MTIME_SET;
251 if (ia_valid & ATTR_FORCE)
252 sa_valid |= MDS_ATTR_FORCE;
253 if (ia_valid & ATTR_ATTR_FLAG)
254 sa_valid |= MDS_ATTR_ATTR_FLAG;
255 if (ia_valid & ATTR_KILL_SUID)
256 sa_valid |= MDS_ATTR_KILL_SUID;
257 if (ia_valid & ATTR_KILL_SGID)
258 sa_valid |= MDS_ATTR_KILL_SGID;
259 if (ia_valid & ATTR_CTIME_SET)
260 sa_valid |= MDS_ATTR_CTIME_SET;
261 if (ia_valid & ATTR_FROM_OPEN)
262 sa_valid |= MDS_ATTR_FROM_OPEN;
263 if (ia_valid & ATTR_BLOCKS)
264 sa_valid |= MDS_ATTR_BLOCKS;
265 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
266 /* NFSD hack (see bug 5781) */
267 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
271 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
272 struct md_op_data *op_data)
274 rec->sa_opcode = REINT_SETATTR;
275 rec->sa_fsuid = current->fsuid;
276 rec->sa_fsgid = current->fsgid;
277 rec->sa_cap = current->cap_effective;
278 rec->sa_suppgid = -1;
280 rec->sa_fid = op_data->op_fid1;
281 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
282 rec->sa_mode = op_data->op_attr.ia_mode;
283 rec->sa_uid = op_data->op_attr.ia_uid;
284 rec->sa_gid = op_data->op_attr.ia_gid;
285 rec->sa_size = op_data->op_attr.ia_size;
286 rec->sa_blocks = op_data->op_attr_blocks;
287 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
288 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
289 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
290 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
291 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
292 in_group_p(op_data->op_attr.ia_gid))
293 rec->sa_suppgid = op_data->op_attr.ia_gid;
295 rec->sa_suppgid = op_data->op_suppgids[0];
298 static void mdc_epoch_pack(struct mdt_epoch *epoch, struct md_op_data *op_data)
300 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
301 epoch->ioepoch = op_data->op_ioepoch;
302 epoch->flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
305 void mdc_setattr_pack(struct ptlrpc_request *req, int offset,
306 struct md_op_data *op_data, void *ea,
307 int ealen, void *ea2, int ea2len)
309 struct mdt_rec_setattr *rec;
310 struct mdt_epoch *epoch;
312 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
313 mdc_setattr_pack_rec(rec, op_data);
315 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
317 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
318 epoch = lustre_msg_buf(req->rq_reqmsg, offset + 2,
320 mdc_epoch_pack(epoch, op_data);
326 memcpy(lustre_msg_buf(req->rq_reqmsg, offset + 3, ealen), ea, ealen);
331 memcpy(lustre_msg_buf(req->rq_reqmsg, offset + 4, ea2len), ea2, ea2len);
334 void mdc_unlink_pack(struct ptlrpc_request *req, int offset,
335 struct md_op_data *op_data)
337 struct mdt_rec_unlink *rec;
340 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
341 LASSERT (rec != NULL);
343 rec->ul_opcode = REINT_UNLINK;
344 rec->ul_fsuid = op_data->op_fsuid;//current->fsuid;
345 rec->ul_fsgid = op_data->op_fsgid;//current->fsgid;
346 rec->ul_cap = op_data->op_cap;//current->cap_effective;
347 rec->ul_mode = op_data->op_mode;
348 rec->ul_suppgid = op_data->op_suppgids[0];
349 rec->ul_fid1 = op_data->op_fid1;
350 rec->ul_fid2 = op_data->op_fid2;
351 rec->ul_time = op_data->op_mod_time;
352 rec->ul_bias = op_data->op_bias;
354 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
356 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 2, op_data->op_namelen + 1);
357 LASSERT(tmp != NULL);
358 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
361 void mdc_link_pack(struct ptlrpc_request *req, int offset,
362 struct md_op_data *op_data)
364 struct mdt_rec_link *rec;
367 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
369 rec->lk_opcode = REINT_LINK;
370 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
371 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
372 rec->lk_cap = op_data->op_cap;//current->cap_effective;
373 rec->lk_suppgid1 = op_data->op_suppgids[0];
374 rec->lk_suppgid2 = op_data->op_suppgids[1];
375 rec->lk_fid1 = op_data->op_fid1;
376 rec->lk_fid2 = op_data->op_fid2;
377 rec->lk_time = op_data->op_mod_time;
378 rec->lk_bias = op_data->op_bias;
380 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
381 mdc_pack_capa(req, offset + 2, op_data->op_capa2);
383 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 3, op_data->op_namelen + 1);
384 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
387 void mdc_rename_pack(struct ptlrpc_request *req, int offset,
388 struct md_op_data *op_data,
389 const char *old, int oldlen, const char *new, int newlen)
391 struct mdt_rec_rename *rec;
394 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
396 /* XXX do something about time, uid, gid */
397 rec->rn_opcode = REINT_RENAME;
398 rec->rn_fsuid = op_data->op_fsuid;//current->fsuid;
399 rec->rn_fsgid = op_data->op_fsgid;//current->fsgid;
400 rec->rn_cap = op_data->op_cap;//current->cap_effective;
401 rec->rn_suppgid1 = op_data->op_suppgids[0];
402 rec->rn_suppgid2 = op_data->op_suppgids[1];
403 rec->rn_fid1 = op_data->op_fid1;
404 rec->rn_fid2 = op_data->op_fid2;
405 rec->rn_time = op_data->op_mod_time;
406 rec->rn_mode = op_data->op_mode;
407 rec->rn_bias = op_data->op_bias;
409 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
410 mdc_pack_capa(req, offset + 2, op_data->op_capa2);
412 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 3, oldlen + 1);
413 LOGL0(old, oldlen, tmp);
416 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 4, newlen + 1);
417 LOGL0(new, newlen, tmp);
421 void mdc_getattr_pack(struct ptlrpc_request *req, int offset, __u64 valid,
422 int flags, struct md_op_data *op_data)
425 b = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*b));
427 b->fsuid = current->fsuid;
428 b->fsgid = current->fsgid;
429 b->capability = current->cap_effective;
431 if (op_data->op_bias & MDS_CHECK_SPLIT)
432 b->valid |= OBD_MD_FLCKSPLIT;
433 if (op_data->op_bias & MDS_CROSS_REF)
434 b->valid |= OBD_MD_FLCROSSREF;
435 b->flags = flags | MDS_BFLAG_EXT_FLAGS;
436 b->suppgid = op_data->op_suppgids[0];
438 b->fid1 = op_data->op_fid1;
439 b->fid2 = op_data->op_fid2;
441 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
443 if (op_data->op_name) {
445 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 2,
446 op_data->op_namelen + 1);
447 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
451 void mdc_close_pack(struct ptlrpc_request *req, int offset,
452 struct md_op_data *op_data)
454 struct mdt_epoch *epoch;
455 struct mdt_rec_setattr *rec;
457 epoch = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*epoch));
458 rec = lustre_msg_buf(req->rq_reqmsg, offset + 1, sizeof(*rec));
460 mdc_setattr_pack_rec(rec, op_data);
461 mdc_pack_capa(req, offset + 2, op_data->op_capa1);
462 mdc_epoch_pack(epoch, op_data);
465 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
469 spin_lock(&cli->cl_loi_list_lock);
470 rc = list_empty(&mcw->mcw_entry);
471 spin_unlock(&cli->cl_loi_list_lock);
475 /* We record requests in flight in cli->cl_r_in_flight here.
476 * There is only one write rpc possible in mdc anyway. If this to change
477 * in the future - the code may need to be revisited. */
478 void mdc_enter_request(struct client_obd *cli)
480 struct mdc_cache_waiter mcw;
481 struct l_wait_info lwi = { 0 };
483 spin_lock(&cli->cl_loi_list_lock);
484 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
485 list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
486 init_waitqueue_head(&mcw.mcw_waitq);
487 spin_unlock(&cli->cl_loi_list_lock);
488 l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
490 cli->cl_r_in_flight++;
491 spin_unlock(&cli->cl_loi_list_lock);
495 void mdc_exit_request(struct client_obd *cli)
497 struct list_head *l, *tmp;
498 struct mdc_cache_waiter *mcw;
500 spin_lock(&cli->cl_loi_list_lock);
501 cli->cl_r_in_flight--;
502 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
504 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
505 /* No free request slots anymore */
509 mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
510 list_del_init(&mcw->mcw_entry);
511 cli->cl_r_in_flight++;
512 wake_up(&mcw->mcw_waitq);
514 /* Empty waiting list? Decrease reqs in-flight number */
516 spin_unlock(&cli->cl_loi_list_lock);