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, __u32 suppgid)
46 b->uid = current->uid;
47 b->gid = current->gid;
48 b->fsuid = current->fsuid;
49 b->fsgid = current->fsgid;
50 b->capability = current->cap_effective;
53 void mdc_pack_capa(struct ptlrpc_request *req, int offset, struct obd_capa *oc)
55 struct lustre_capa *c;
58 LASSERT(lustre_msg_buflen(req->rq_reqmsg, offset) == 0);
62 c = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*c));
65 DEBUG_CAPA(D_SEC, c, "pack");
68 void mdc_is_subdir_pack(struct ptlrpc_request *req, int offset,
69 const struct lu_fid *pfid,
70 const struct lu_fid *cfid, int flags)
72 struct mdt_body *b = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*b));
78 b->valid = OBD_MD_FLID;
82 void mdc_pack_req_body(struct ptlrpc_request *req, int offset,
83 __u64 valid, const struct lu_fid *fid,
84 struct obd_capa *oc, int ea_size, __u32 suppgid,
87 struct mdt_body *b = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*b));
90 b->eadatasize = ea_size;
92 mdc_pack_body(b, suppgid);
95 mdc_pack_capa(req, offset + 1, oc);
99 void mdc_readdir_pack(struct ptlrpc_request *req, int offset, __u64 pgoff,
100 __u32 size, const struct lu_fid *fid, struct obd_capa *oc)
104 b = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*b));
106 b->size = pgoff; /* !! */
107 b->nlink = size; /* !! */
108 mdc_pack_body(b, -1);
109 mdc_pack_capa(req, offset + 1, oc);
112 /* packing of MDS records */
113 void mdc_create_pack(struct ptlrpc_request *req, int offset,
114 struct md_op_data *op_data, const void *data, int datalen,
115 __u32 mode, __u32 uid, __u32 gid, __u32 cap_effective,
118 struct mdt_rec_create *rec;
121 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*rec));
123 rec->cr_opcode = REINT_CREATE;
126 rec->cr_cap = cap_effective;
127 rec->cr_fid1 = op_data->op_fid1;
128 rec->cr_fid2 = op_data->op_fid2;
131 rec->cr_time = op_data->op_mod_time;
132 rec->cr_suppgid1 = op_data->op_suppgids[0];
133 rec->cr_suppgid2 = op_data->op_suppgids[1];
134 rec->cr_flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
135 rec->cr_bias = op_data->op_bias;
137 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
139 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 2, op_data->op_namelen + 1);
140 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
143 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 3, datalen);
144 memcpy(tmp, data, datalen);
148 static __u32 mds_pack_open_flags(__u32 flags)
150 __u32 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
151 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
152 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK));
154 cr_flags |= MDS_OPEN_CREAT;
156 cr_flags |= MDS_OPEN_EXCL;
158 cr_flags |= MDS_OPEN_TRUNC;
159 if (flags & O_APPEND)
160 cr_flags |= MDS_OPEN_APPEND;
162 cr_flags |= MDS_OPEN_SYNC;
163 if (flags & O_DIRECTORY)
164 cr_flags |= MDS_OPEN_DIRECTORY;
165 if (flags & O_JOIN_FILE)
166 cr_flags |= MDS_OPEN_JOIN_FILE;
168 if (flags & FMODE_EXEC)
169 cr_flags |= MDS_FMODE_EXEC;
171 if (flags & O_LOV_DELAY_CREATE)
172 cr_flags |= MDS_OPEN_DELAY_CREATE;
177 /* packing of MDS records */
178 void mdc_join_pack(struct ptlrpc_request *req, int offset,
179 struct md_op_data *op_data, __u64 head_size)
181 struct mdt_rec_join *rec;
183 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*rec));
184 LASSERT(rec != NULL);
185 rec->jr_fid = op_data->op_fid2;
186 rec->jr_headsize = head_size;
189 void mdc_open_pack(struct ptlrpc_request *req, int offset,
190 struct md_op_data *op_data, __u32 mode, __u64 rdev,
191 __u32 flags, const void *lmm, int lmmlen)
193 struct mdt_rec_create *rec;
195 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
197 /* XXX do something about time, uid, gid */
198 rec->cr_opcode = REINT_OPEN;
199 rec->cr_fsuid = current->fsuid;
200 rec->cr_fsgid = current->fsgid;
201 rec->cr_cap = current->cap_effective;
202 if (op_data != NULL) {
203 rec->cr_fid1 = op_data->op_fid1;
204 rec->cr_fid2 = op_data->op_fid2;
207 rec->cr_flags = mds_pack_open_flags(flags);
208 rec->cr_time = op_data->op_mod_time;
210 rec->cr_suppgid1 = op_data->op_suppgids[0];
211 rec->cr_suppgid2 = op_data->op_suppgids[1];
212 rec->cr_bias = op_data->op_bias;
214 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
215 /* the next buffer is child capa, which is used for replay,
216 * will be packed from the data in reply message. */
218 if (op_data->op_name) {
219 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 3,
220 op_data->op_namelen + 1);
221 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
225 rec->cr_flags |= MDS_OPEN_HAS_EA;
227 /*XXX a hack for liblustre to set EA (LL_IOC_LOV_SETSTRIPE) */
228 rec->cr_fid2 = op_data->op_fid2;
230 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 4, lmmlen);
231 memcpy (tmp, lmm, lmmlen);
235 static inline __u64 attr_pack(unsigned int ia_valid) {
238 if (ia_valid & ATTR_MODE)
239 sa_valid |= MDS_ATTR_MODE;
240 if (ia_valid & ATTR_UID)
241 sa_valid |= MDS_ATTR_UID;
242 if (ia_valid & ATTR_GID)
243 sa_valid |= MDS_ATTR_GID;
244 if (ia_valid & ATTR_SIZE)
245 sa_valid |= MDS_ATTR_SIZE;
246 if (ia_valid & ATTR_ATIME)
247 sa_valid |= MDS_ATTR_ATIME;
248 if (ia_valid & ATTR_MTIME)
249 sa_valid |= MDS_ATTR_MTIME;
250 if (ia_valid & ATTR_CTIME)
251 sa_valid |= MDS_ATTR_CTIME;
252 if (ia_valid & ATTR_ATIME_SET)
253 sa_valid |= MDS_ATTR_ATIME_SET;
254 if (ia_valid & ATTR_MTIME_SET)
255 sa_valid |= MDS_ATTR_MTIME_SET;
256 if (ia_valid & ATTR_FORCE)
257 sa_valid |= MDS_ATTR_FORCE;
258 if (ia_valid & ATTR_ATTR_FLAG)
259 sa_valid |= MDS_ATTR_ATTR_FLAG;
260 if (ia_valid & ATTR_KILL_SUID)
261 sa_valid |= MDS_ATTR_KILL_SUID;
262 if (ia_valid & ATTR_KILL_SGID)
263 sa_valid |= MDS_ATTR_KILL_SGID;
264 if (ia_valid & ATTR_CTIME_SET)
265 sa_valid |= MDS_ATTR_CTIME_SET;
266 if (ia_valid & ATTR_FROM_OPEN)
267 sa_valid |= MDS_ATTR_FROM_OPEN;
268 if (ia_valid & ATTR_BLOCKS)
269 sa_valid |= MDS_ATTR_BLOCKS;
270 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
271 /* NFSD hack (see bug 5781) */
272 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
276 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
277 struct md_op_data *op_data)
279 rec->sa_opcode = REINT_SETATTR;
280 rec->sa_fsuid = current->fsuid;
281 rec->sa_fsgid = current->fsgid;
282 rec->sa_cap = current->cap_effective;
283 rec->sa_suppgid = -1;
285 rec->sa_fid = op_data->op_fid1;
286 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
287 rec->sa_mode = op_data->op_attr.ia_mode;
288 rec->sa_uid = op_data->op_attr.ia_uid;
289 rec->sa_gid = op_data->op_attr.ia_gid;
290 rec->sa_size = op_data->op_attr.ia_size;
291 rec->sa_blocks = op_data->op_attr_blocks;
292 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
293 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
294 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
295 rec->sa_attr_flags = ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
296 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
297 in_group_p(op_data->op_attr.ia_gid))
298 rec->sa_suppgid = op_data->op_attr.ia_gid;
300 rec->sa_suppgid = op_data->op_suppgids[0];
303 static void mdc_epoch_pack(struct mdt_epoch *epoch, struct md_op_data *op_data)
305 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
306 epoch->ioepoch = op_data->op_ioepoch;
307 epoch->flags = op_data->op_flags & ~MF_SOM_LOCAL_FLAGS;
310 void mdc_setattr_pack(struct ptlrpc_request *req, int offset,
311 struct md_op_data *op_data, void *ea,
312 int ealen, void *ea2, int ea2len)
314 struct mdt_rec_setattr *rec;
315 struct mdt_epoch *epoch;
317 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
318 mdc_setattr_pack_rec(rec, op_data);
320 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
322 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
323 epoch = lustre_msg_buf(req->rq_reqmsg, offset + 2,
325 mdc_epoch_pack(epoch, op_data);
331 memcpy(lustre_msg_buf(req->rq_reqmsg, offset + 3, ealen), ea, ealen);
336 memcpy(lustre_msg_buf(req->rq_reqmsg, offset + 4, ea2len), ea2, ea2len);
339 void mdc_unlink_pack(struct ptlrpc_request *req, int offset,
340 struct md_op_data *op_data)
342 struct mdt_rec_unlink *rec;
345 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
346 LASSERT (rec != NULL);
348 rec->ul_opcode = REINT_UNLINK;
349 rec->ul_fsuid = op_data->op_fsuid;//current->fsuid;
350 rec->ul_fsgid = op_data->op_fsgid;//current->fsgid;
351 rec->ul_cap = op_data->op_cap;//current->cap_effective;
352 rec->ul_mode = op_data->op_mode;
353 rec->ul_suppgid = op_data->op_suppgids[0];
354 rec->ul_fid1 = op_data->op_fid1;
355 rec->ul_fid2 = op_data->op_fid2;
356 rec->ul_time = op_data->op_mod_time;
357 rec->ul_bias = op_data->op_bias;
359 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
361 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 2, op_data->op_namelen + 1);
362 LASSERT(tmp != NULL);
363 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
366 void mdc_link_pack(struct ptlrpc_request *req, int offset,
367 struct md_op_data *op_data)
369 struct mdt_rec_link *rec;
372 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
374 rec->lk_opcode = REINT_LINK;
375 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
376 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
377 rec->lk_cap = op_data->op_cap;//current->cap_effective;
378 rec->lk_suppgid1 = op_data->op_suppgids[0];
379 rec->lk_suppgid2 = op_data->op_suppgids[1];
380 rec->lk_fid1 = op_data->op_fid1;
381 rec->lk_fid2 = op_data->op_fid2;
382 rec->lk_time = op_data->op_mod_time;
383 rec->lk_bias = op_data->op_bias;
385 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
386 mdc_pack_capa(req, offset + 2, op_data->op_capa2);
388 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 3, op_data->op_namelen + 1);
389 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
392 void mdc_rename_pack(struct ptlrpc_request *req, int offset,
393 struct md_op_data *op_data,
394 const char *old, int oldlen, const char *new, int newlen)
396 struct mdt_rec_rename *rec;
399 rec = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*rec));
401 /* XXX do something about time, uid, gid */
402 rec->rn_opcode = REINT_RENAME;
403 rec->rn_fsuid = op_data->op_fsuid;//current->fsuid;
404 rec->rn_fsgid = op_data->op_fsgid;//current->fsgid;
405 rec->rn_cap = op_data->op_cap;//current->cap_effective;
406 rec->rn_suppgid1 = op_data->op_suppgids[0];
407 rec->rn_suppgid2 = op_data->op_suppgids[1];
408 rec->rn_fid1 = op_data->op_fid1;
409 rec->rn_fid2 = op_data->op_fid2;
410 rec->rn_time = op_data->op_mod_time;
411 rec->rn_mode = op_data->op_mode;
412 rec->rn_bias = op_data->op_bias;
414 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
415 mdc_pack_capa(req, offset + 2, op_data->op_capa2);
417 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 3, oldlen + 1);
418 LOGL0(old, oldlen, tmp);
421 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 4, newlen + 1);
422 LOGL0(new, newlen, tmp);
426 void mdc_getattr_pack(struct ptlrpc_request *req, int offset, __u64 valid,
427 int flags, struct md_op_data *op_data)
430 b = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*b));
432 b->fsuid = current->fsuid;
433 b->fsgid = current->fsgid;
434 b->capability = current->cap_effective;
436 if (op_data->op_bias & MDS_CHECK_SPLIT)
437 b->valid |= OBD_MD_FLCKSPLIT;
438 if (op_data->op_bias & MDS_CROSS_REF)
439 b->valid |= OBD_MD_FLCROSSREF;
440 b->flags = flags | MDS_BFLAG_EXT_FLAGS;
441 b->suppgid = op_data->op_suppgids[0];
443 b->fid1 = op_data->op_fid1;
444 b->fid2 = op_data->op_fid2;
446 mdc_pack_capa(req, offset + 1, op_data->op_capa1);
448 if (op_data->op_name) {
450 tmp = lustre_msg_buf(req->rq_reqmsg, offset + 2,
451 op_data->op_namelen + 1);
452 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
456 void mdc_close_pack(struct ptlrpc_request *req, int offset,
457 struct md_op_data *op_data)
459 struct mdt_epoch *epoch;
460 struct mdt_rec_setattr *rec;
462 epoch = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*epoch));
463 rec = lustre_msg_buf(req->rq_reqmsg, offset + 1, sizeof(*rec));
465 mdc_setattr_pack_rec(rec, op_data);
466 mdc_pack_capa(req, offset + 2, op_data->op_capa1);
467 mdc_epoch_pack(epoch, op_data);
470 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
474 spin_lock(&cli->cl_loi_list_lock);
475 rc = list_empty(&mcw->mcw_entry);
476 spin_unlock(&cli->cl_loi_list_lock);
480 /* We record requests in flight in cli->cl_r_in_flight here.
481 * There is only one write rpc possible in mdc anyway. If this to change
482 * in the future - the code may need to be revisited. */
483 void mdc_enter_request(struct client_obd *cli)
485 struct mdc_cache_waiter mcw;
486 struct l_wait_info lwi = { 0 };
488 spin_lock(&cli->cl_loi_list_lock);
489 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
490 list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
491 init_waitqueue_head(&mcw.mcw_waitq);
492 spin_unlock(&cli->cl_loi_list_lock);
493 l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw), &lwi);
495 cli->cl_r_in_flight++;
496 spin_unlock(&cli->cl_loi_list_lock);
500 void mdc_exit_request(struct client_obd *cli)
502 struct list_head *l, *tmp;
503 struct mdc_cache_waiter *mcw;
505 spin_lock(&cli->cl_loi_list_lock);
506 cli->cl_r_in_flight--;
507 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
509 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
510 /* No free request slots anymore */
514 mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
515 list_del_init(&mcw->mcw_entry);
516 cli->cl_r_in_flight++;
517 wake_up(&mcw->mcw_waitq);
519 /* Empty waiting list? Decrease reqs in-flight number */
521 spin_unlock(&cli->cl_loi_list_lock);