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, 2015, 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
38 #include "../include/lustre_net.h"
39 #include "../include/lustre/lustre_idl.h"
40 #include "mdc_internal.h"
42 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
45 b->uid = from_kuid(&init_user_ns, current_uid());
46 b->gid = from_kgid(&init_user_ns, current_gid());
47 b->fsuid = from_kuid(&init_user_ns, current_fsuid());
48 b->fsgid = from_kgid(&init_user_ns, current_fsgid());
49 b->capability = cfs_curproc_cap_pack();
52 void mdc_is_subdir_pack(struct ptlrpc_request *req, const struct lu_fid *pfid,
53 const struct lu_fid *cfid, int flags)
55 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
60 b->valid = OBD_MD_FLID;
67 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
68 struct md_op_data *op_data)
70 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
73 __mdc_pack_body(b, op_data->op_suppgids[0]);
74 b->fid1 = op_data->op_fid1;
75 b->fid2 = op_data->op_fid2;
76 b->valid |= OBD_MD_FLID;
79 void mdc_pack_body(struct ptlrpc_request *req, const struct lu_fid *fid,
80 __u64 valid, int ea_size, __u32 suppgid, int flags)
82 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
85 b->eadatasize = ea_size;
87 __mdc_pack_body(b, suppgid);
90 b->valid |= OBD_MD_FLID;
94 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff,
95 __u32 size, const struct lu_fid *fid)
97 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
100 b->valid |= OBD_MD_FLID;
101 b->size = pgoff; /* !! */
102 b->nlink = size; /* !! */
103 __mdc_pack_body(b, -1);
104 b->mode = LUDA_FID | LUDA_TYPE;
107 /* packing of MDS records */
108 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
109 const void *data, int datalen, __u32 mode,
110 __u32 uid, __u32 gid, cfs_cap_t cap_effective, __u64 rdev)
112 struct mdt_rec_create *rec;
116 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
117 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
119 rec->cr_opcode = REINT_CREATE;
122 rec->cr_cap = cap_effective;
123 rec->cr_fid1 = op_data->op_fid1;
124 rec->cr_fid2 = op_data->op_fid2;
127 rec->cr_time = op_data->op_mod_time;
128 rec->cr_suppgid1 = op_data->op_suppgids[0];
129 rec->cr_suppgid2 = op_data->op_suppgids[1];
130 flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
131 if (op_data->op_bias & MDS_CREATE_VOLATILE)
132 flags |= MDS_OPEN_VOLATILE;
133 set_mrc_cr_flags(rec, flags);
134 rec->cr_bias = op_data->op_bias;
135 rec->cr_umask = current_umask();
137 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
138 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
141 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
142 memcpy(tmp, data, datalen);
146 static __u64 mds_pack_open_flags(__u64 flags, __u32 mode)
148 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
149 MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS |
150 MDS_OPEN_OWNEROVERRIDE | MDS_OPEN_LOCK |
151 MDS_OPEN_BY_FID | MDS_OPEN_LEASE |
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 & __FMODE_EXEC)
166 cr_flags |= MDS_FMODE_EXEC;
167 if (cl_is_lov_delay_create(flags))
168 cr_flags |= MDS_OPEN_DELAY_CREATE;
170 if (flags & O_NONBLOCK)
171 cr_flags |= MDS_OPEN_NORESTORE;
176 /* packing of MDS records */
177 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
178 __u32 mode, __u64 rdev, __u64 flags, const void *lmm,
181 struct mdt_rec_create *rec;
185 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
186 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
188 /* XXX do something about time, uid, gid */
189 rec->cr_opcode = REINT_OPEN;
190 rec->cr_fsuid = from_kuid(&init_user_ns, current_fsuid());
191 rec->cr_fsgid = from_kgid(&init_user_ns, current_fsgid());
192 rec->cr_cap = cfs_curproc_cap_pack();
193 rec->cr_fid1 = op_data->op_fid1;
194 rec->cr_fid2 = op_data->op_fid2;
197 cr_flags = mds_pack_open_flags(flags, mode);
199 rec->cr_time = op_data->op_mod_time;
200 rec->cr_suppgid1 = op_data->op_suppgids[0];
201 rec->cr_suppgid2 = op_data->op_suppgids[1];
202 rec->cr_bias = op_data->op_bias;
203 rec->cr_umask = current_umask();
204 rec->cr_old_handle = op_data->op_handle;
206 if (op_data->op_name) {
207 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
208 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
209 if (op_data->op_bias & MDS_CREATE_VOLATILE)
210 cr_flags |= MDS_OPEN_VOLATILE;
214 cr_flags |= MDS_OPEN_HAS_EA;
215 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
216 memcpy(tmp, lmm, lmmlen);
218 set_mrc_cr_flags(rec, cr_flags);
221 static inline __u64 attr_pack(unsigned int ia_valid)
225 if (ia_valid & ATTR_MODE)
226 sa_valid |= MDS_ATTR_MODE;
227 if (ia_valid & ATTR_UID)
228 sa_valid |= MDS_ATTR_UID;
229 if (ia_valid & ATTR_GID)
230 sa_valid |= MDS_ATTR_GID;
231 if (ia_valid & ATTR_SIZE)
232 sa_valid |= MDS_ATTR_SIZE;
233 if (ia_valid & ATTR_ATIME)
234 sa_valid |= MDS_ATTR_ATIME;
235 if (ia_valid & ATTR_MTIME)
236 sa_valid |= MDS_ATTR_MTIME;
237 if (ia_valid & ATTR_CTIME)
238 sa_valid |= MDS_ATTR_CTIME;
239 if (ia_valid & ATTR_ATIME_SET)
240 sa_valid |= MDS_ATTR_ATIME_SET;
241 if (ia_valid & ATTR_MTIME_SET)
242 sa_valid |= MDS_ATTR_MTIME_SET;
243 if (ia_valid & ATTR_FORCE)
244 sa_valid |= MDS_ATTR_FORCE;
245 if (ia_valid & ATTR_ATTR_FLAG)
246 sa_valid |= MDS_ATTR_ATTR_FLAG;
247 if (ia_valid & ATTR_KILL_SUID)
248 sa_valid |= MDS_ATTR_KILL_SUID;
249 if (ia_valid & ATTR_KILL_SGID)
250 sa_valid |= MDS_ATTR_KILL_SGID;
251 if (ia_valid & ATTR_CTIME_SET)
252 sa_valid |= MDS_ATTR_CTIME_SET;
253 if (ia_valid & ATTR_OPEN)
254 sa_valid |= MDS_ATTR_FROM_OPEN;
255 if (ia_valid & ATTR_BLOCKS)
256 sa_valid |= MDS_ATTR_BLOCKS;
257 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
258 /* NFSD hack (see bug 5781) */
259 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
263 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
264 struct md_op_data *op_data)
266 rec->sa_opcode = REINT_SETATTR;
267 rec->sa_fsuid = from_kuid(&init_user_ns, current_fsuid());
268 rec->sa_fsgid = from_kgid(&init_user_ns, current_fsgid());
269 rec->sa_cap = cfs_curproc_cap_pack();
270 rec->sa_suppgid = -1;
272 rec->sa_fid = op_data->op_fid1;
273 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
274 rec->sa_mode = op_data->op_attr.ia_mode;
275 rec->sa_uid = from_kuid(&init_user_ns, op_data->op_attr.ia_uid);
276 rec->sa_gid = from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
277 rec->sa_size = op_data->op_attr.ia_size;
278 rec->sa_blocks = op_data->op_attr_blocks;
279 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
280 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
281 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
283 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags;
284 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
285 in_group_p(op_data->op_attr.ia_gid))
287 from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
289 rec->sa_suppgid = op_data->op_suppgids[0];
291 rec->sa_bias = op_data->op_bias;
294 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
295 struct md_op_data *op_data)
297 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
298 epoch->ioepoch = op_data->op_ioepoch;
299 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
302 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
303 void *ea, int ealen, void *ea2, int ea2len)
305 struct mdt_rec_setattr *rec;
306 struct mdt_ioepoch *epoch;
307 struct lov_user_md *lum = NULL;
309 CLASSERT(sizeof(struct mdt_rec_reint) ==
310 sizeof(struct mdt_rec_setattr));
311 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
312 mdc_setattr_pack_rec(rec, op_data);
314 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
315 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
316 mdc_ioepoch_pack(epoch, op_data);
322 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
323 if (!ea) { /* Remove LOV EA */
324 lum->lmm_magic = LOV_USER_MAGIC_V1;
325 lum->lmm_stripe_size = 0;
326 lum->lmm_stripe_count = 0;
327 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
329 memcpy(lum, ea, ealen);
335 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
339 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
341 struct mdt_rec_unlink *rec;
344 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
345 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
347 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
348 REINT_RMENTRY : REINT_UNLINK;
349 rec->ul_fsuid = op_data->op_fsuid;
350 rec->ul_fsgid = op_data->op_fsgid;
351 rec->ul_cap = op_data->op_cap;
352 rec->ul_mode = op_data->op_mode;
353 rec->ul_suppgid1 = op_data->op_suppgids[0];
354 rec->ul_suppgid2 = -1;
355 rec->ul_fid1 = op_data->op_fid1;
356 rec->ul_fid2 = op_data->op_fid2;
357 rec->ul_time = op_data->op_mod_time;
358 rec->ul_bias = op_data->op_bias;
360 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
362 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
365 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
367 struct mdt_rec_link *rec;
370 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
371 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
373 rec->lk_opcode = REINT_LINK;
374 rec->lk_fsuid = op_data->op_fsuid; /* current->fsuid; */
375 rec->lk_fsgid = op_data->op_fsgid; /* current->fsgid; */
376 rec->lk_cap = op_data->op_cap; /* current->cap_effective; */
377 rec->lk_suppgid1 = op_data->op_suppgids[0];
378 rec->lk_suppgid2 = op_data->op_suppgids[1];
379 rec->lk_fid1 = op_data->op_fid1;
380 rec->lk_fid2 = op_data->op_fid2;
381 rec->lk_time = op_data->op_mod_time;
382 rec->lk_bias = op_data->op_bias;
384 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
385 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
388 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
389 const char *old, int oldlen, const char *new, int newlen)
391 struct mdt_rec_rename *rec;
394 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
395 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
397 /* XXX do something about time, uid, gid */
398 rec->rn_opcode = REINT_RENAME;
399 rec->rn_fsuid = op_data->op_fsuid;
400 rec->rn_fsgid = op_data->op_fsgid;
401 rec->rn_cap = op_data->op_cap;
402 rec->rn_suppgid1 = op_data->op_suppgids[0];
403 rec->rn_suppgid2 = op_data->op_suppgids[1];
404 rec->rn_fid1 = op_data->op_fid1;
405 rec->rn_fid2 = op_data->op_fid2;
406 rec->rn_time = op_data->op_mod_time;
407 rec->rn_mode = op_data->op_mode;
408 rec->rn_bias = op_data->op_bias;
410 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
411 LOGL0(old, oldlen, tmp);
414 tmp = req_capsule_client_get(&req->rq_pill, &RMF_SYMTGT);
415 LOGL0(new, newlen, tmp);
419 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, int flags,
420 struct md_op_data *op_data, int ea_size)
422 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
426 if (op_data->op_bias & MDS_CHECK_SPLIT)
427 b->valid |= OBD_MD_FLCKSPLIT;
428 if (op_data->op_bias & MDS_CROSS_REF)
429 b->valid |= OBD_MD_FLCROSSREF;
430 b->eadatasize = ea_size;
432 __mdc_pack_body(b, op_data->op_suppgids[0]);
434 b->fid1 = op_data->op_fid1;
435 b->fid2 = op_data->op_fid2;
436 b->valid |= OBD_MD_FLID;
438 if (op_data->op_name) {
439 char *tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
441 LOGL0(op_data->op_name, op_data->op_namelen, tmp);
446 static void mdc_hsm_release_pack(struct ptlrpc_request *req,
447 struct md_op_data *op_data)
449 if (op_data->op_bias & MDS_HSM_RELEASE) {
450 struct close_data *data;
451 struct ldlm_lock *lock;
453 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
455 lock = ldlm_handle2lock(&op_data->op_lease_handle);
457 data->cd_handle = lock->l_remote_handle;
460 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
462 data->cd_data_version = op_data->op_data_version;
463 data->cd_fid = op_data->op_fid2;
467 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
469 struct mdt_ioepoch *epoch;
470 struct mdt_rec_setattr *rec;
472 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
473 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
475 mdc_setattr_pack_rec(rec, op_data);
476 mdc_ioepoch_pack(epoch, op_data);
477 mdc_hsm_release_pack(req, op_data);
480 static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
484 client_obd_list_lock(&cli->cl_loi_list_lock);
485 rc = list_empty(&mcw->mcw_entry);
486 client_obd_list_unlock(&cli->cl_loi_list_lock);
490 /* We record requests in flight in cli->cl_r_in_flight here.
491 * There is only one write rpc possible in mdc anyway. If this to change
492 * in the future - the code may need to be revisited.
494 int mdc_enter_request(struct client_obd *cli)
497 struct mdc_cache_waiter mcw;
498 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
500 client_obd_list_lock(&cli->cl_loi_list_lock);
501 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
502 list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
503 init_waitqueue_head(&mcw.mcw_waitq);
504 client_obd_list_unlock(&cli->cl_loi_list_lock);
505 rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw),
508 client_obd_list_lock(&cli->cl_loi_list_lock);
509 if (list_empty(&mcw.mcw_entry))
510 cli->cl_r_in_flight--;
511 list_del_init(&mcw.mcw_entry);
512 client_obd_list_unlock(&cli->cl_loi_list_lock);
515 cli->cl_r_in_flight++;
516 client_obd_list_unlock(&cli->cl_loi_list_lock);
521 void mdc_exit_request(struct client_obd *cli)
523 struct list_head *l, *tmp;
524 struct mdc_cache_waiter *mcw;
526 client_obd_list_lock(&cli->cl_loi_list_lock);
527 cli->cl_r_in_flight--;
528 list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
529 if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
530 /* No free request slots anymore */
534 mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
535 list_del_init(&mcw->mcw_entry);
536 cli->cl_r_in_flight++;
537 wake_up(&mcw->mcw_waitq);
539 /* Empty waiting list? Decrease reqs in-flight number */
541 client_obd_list_unlock(&cli->cl_loi_list_lock);