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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_MDC
34 #include "../include/lustre_net.h"
35 #include "../include/lustre/lustre_idl.h"
36 #include "mdc_internal.h"
38 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
40 b->mbo_suppgid = suppgid;
41 b->mbo_uid = from_kuid(&init_user_ns, current_uid());
42 b->mbo_gid = from_kgid(&init_user_ns, current_gid());
43 b->mbo_fsuid = from_kuid(&init_user_ns, current_fsuid());
44 b->mbo_fsgid = from_kgid(&init_user_ns, current_fsgid());
45 b->mbo_capability = cfs_curproc_cap_pack();
48 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
49 struct md_op_data *op_data)
51 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
54 __mdc_pack_body(b, op_data->op_suppgids[0]);
55 b->mbo_fid1 = op_data->op_fid1;
56 b->mbo_fid2 = op_data->op_fid2;
57 b->mbo_valid |= OBD_MD_FLID;
60 void mdc_pack_body(struct ptlrpc_request *req, const struct lu_fid *fid,
61 __u64 valid, size_t ea_size, __u32 suppgid, u32 flags)
63 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
66 b->mbo_eadatasize = ea_size;
68 __mdc_pack_body(b, suppgid);
71 b->mbo_valid |= OBD_MD_FLID;
76 * Pack a name (path component) into a request
78 * \param[in] req request
79 * \param[in] field request field (usually RMF_NAME)
80 * \param[in] name path component
81 * \param[in] name_len length of path component
83 * \a field must be present in \a req and of size \a name_len + 1.
85 * \a name must be '\0' terminated of length \a name_len and represent
86 * a single path component (not contain '/').
88 static void mdc_pack_name(struct ptlrpc_request *req,
89 const struct req_msg_field *field,
90 const char *name, size_t name_len)
96 buf = req_capsule_client_get(&req->rq_pill, field);
97 buf_size = req_capsule_get_size(&req->rq_pill, field, RCL_CLIENT);
99 LASSERT(name && name_len && buf && buf_size == name_len + 1);
101 cpy_len = strlcpy(buf, name, buf_size);
103 LASSERT(cpy_len == name_len && lu_name_is_valid_2(buf, cpy_len));
106 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff, size_t size,
107 const struct lu_fid *fid)
109 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
112 b->mbo_valid |= OBD_MD_FLID;
113 b->mbo_size = pgoff; /* !! */
114 b->mbo_nlink = size; /* !! */
115 __mdc_pack_body(b, -1);
116 b->mbo_mode = LUDA_FID | LUDA_TYPE;
119 /* packing of MDS records */
120 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
121 const void *data, size_t datalen, umode_t mode,
122 uid_t uid, gid_t gid, cfs_cap_t cap_effective, __u64 rdev)
124 struct mdt_rec_create *rec;
128 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
129 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
131 rec->cr_opcode = REINT_CREATE;
134 rec->cr_cap = cap_effective;
135 rec->cr_fid1 = op_data->op_fid1;
136 rec->cr_fid2 = op_data->op_fid2;
139 rec->cr_time = op_data->op_mod_time;
140 rec->cr_suppgid1 = op_data->op_suppgids[0];
141 rec->cr_suppgid2 = op_data->op_suppgids[1];
142 flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
143 if (op_data->op_bias & MDS_CREATE_VOLATILE)
144 flags |= MDS_OPEN_VOLATILE;
145 set_mrc_cr_flags(rec, flags);
146 rec->cr_bias = op_data->op_bias;
147 rec->cr_umask = current_umask();
149 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
151 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
152 memcpy(tmp, data, datalen);
156 static inline __u64 mds_pack_open_flags(__u64 flags)
158 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
159 MDS_OPEN_FL_INTERNAL));
161 cr_flags |= MDS_OPEN_CREAT;
163 cr_flags |= MDS_OPEN_EXCL;
165 cr_flags |= MDS_OPEN_TRUNC;
166 if (flags & O_APPEND)
167 cr_flags |= MDS_OPEN_APPEND;
169 cr_flags |= MDS_OPEN_SYNC;
170 if (flags & O_DIRECTORY)
171 cr_flags |= MDS_OPEN_DIRECTORY;
172 if (flags & __FMODE_EXEC)
173 cr_flags |= MDS_FMODE_EXEC;
174 if (cl_is_lov_delay_create(flags))
175 cr_flags |= MDS_OPEN_DELAY_CREATE;
177 if (flags & O_NONBLOCK)
178 cr_flags |= MDS_OPEN_NORESTORE;
183 /* packing of MDS records */
184 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
185 umode_t mode, __u64 rdev, __u64 flags, const void *lmm,
188 struct mdt_rec_create *rec;
192 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
193 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
195 /* XXX do something about time, uid, gid */
196 rec->cr_opcode = REINT_OPEN;
197 rec->cr_fsuid = from_kuid(&init_user_ns, current_fsuid());
198 rec->cr_fsgid = from_kgid(&init_user_ns, current_fsgid());
199 rec->cr_cap = cfs_curproc_cap_pack();
200 rec->cr_fid1 = op_data->op_fid1;
201 rec->cr_fid2 = op_data->op_fid2;
204 cr_flags = mds_pack_open_flags(flags);
206 rec->cr_time = op_data->op_mod_time;
207 rec->cr_suppgid1 = op_data->op_suppgids[0];
208 rec->cr_suppgid2 = op_data->op_suppgids[1];
209 rec->cr_bias = op_data->op_bias;
210 rec->cr_umask = current_umask();
211 rec->cr_old_handle = op_data->op_handle;
213 if (op_data->op_name) {
214 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
215 op_data->op_namelen);
217 if (op_data->op_bias & MDS_CREATE_VOLATILE)
218 cr_flags |= MDS_OPEN_VOLATILE;
222 cr_flags |= MDS_OPEN_HAS_EA;
223 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
224 memcpy(tmp, lmm, lmmlen);
226 set_mrc_cr_flags(rec, cr_flags);
229 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_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 = from_kuid(&init_user_ns, current_fsuid());
276 rec->sa_fsgid = from_kgid(&init_user_ns, current_fsgid());
277 rec->sa_cap = cfs_curproc_cap_pack();
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 = from_kuid(&init_user_ns, op_data->op_attr.ia_uid);
284 rec->sa_gid = from_kgid(&init_user_ns, 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 = op_data->op_attr_flags;
291 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
292 in_group_p(op_data->op_attr.ia_gid))
294 from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
296 rec->sa_suppgid = op_data->op_suppgids[0];
298 rec->sa_bias = op_data->op_bias;
301 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
302 struct md_op_data *op_data)
304 memcpy(&epoch->handle, &op_data->op_handle, sizeof(epoch->handle));
305 epoch->ioepoch = op_data->op_ioepoch;
306 epoch->flags = op_data->op_flags & MF_SOM_LOCAL_FLAGS;
309 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
310 void *ea, size_t ealen, void *ea2, size_t ea2len)
312 struct mdt_rec_setattr *rec;
313 struct mdt_ioepoch *epoch;
314 struct lov_user_md *lum = NULL;
316 CLASSERT(sizeof(struct mdt_rec_reint) ==
317 sizeof(struct mdt_rec_setattr));
318 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
319 mdc_setattr_pack_rec(rec, op_data);
321 if (op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) {
322 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
323 mdc_ioepoch_pack(epoch, op_data);
329 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
330 if (!ea) { /* Remove LOV EA */
331 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
332 lum->lmm_stripe_size = 0;
333 lum->lmm_stripe_count = 0;
334 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
336 memcpy(lum, ea, ealen);
342 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_LOGCOOKIES), ea2,
346 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
348 struct mdt_rec_unlink *rec;
350 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
351 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
353 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
354 REINT_RMENTRY : REINT_UNLINK;
355 rec->ul_fsuid = op_data->op_fsuid;
356 rec->ul_fsgid = op_data->op_fsgid;
357 rec->ul_cap = op_data->op_cap;
358 rec->ul_mode = op_data->op_mode;
359 rec->ul_suppgid1 = op_data->op_suppgids[0];
360 rec->ul_suppgid2 = -1;
361 rec->ul_fid1 = op_data->op_fid1;
362 rec->ul_fid2 = op_data->op_fid2;
363 rec->ul_time = op_data->op_mod_time;
364 rec->ul_bias = op_data->op_bias;
366 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
369 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
371 struct mdt_rec_link *rec;
373 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
374 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
376 rec->lk_opcode = REINT_LINK;
377 rec->lk_fsuid = op_data->op_fsuid; /* current->fsuid; */
378 rec->lk_fsgid = op_data->op_fsgid; /* current->fsgid; */
379 rec->lk_cap = op_data->op_cap; /* current->cap_effective; */
380 rec->lk_suppgid1 = op_data->op_suppgids[0];
381 rec->lk_suppgid2 = op_data->op_suppgids[1];
382 rec->lk_fid1 = op_data->op_fid1;
383 rec->lk_fid2 = op_data->op_fid2;
384 rec->lk_time = op_data->op_mod_time;
385 rec->lk_bias = op_data->op_bias;
387 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
390 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
391 const char *old, size_t oldlen,
392 const char *new, size_t newlen)
394 struct mdt_rec_rename *rec;
396 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
397 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
399 /* XXX do something about time, uid, gid */
400 rec->rn_opcode = op_data->op_cli_flags & CLI_MIGRATE ?
401 REINT_MIGRATE : REINT_RENAME;
402 rec->rn_opcode = REINT_RENAME;
403 rec->rn_fsuid = op_data->op_fsuid;
404 rec->rn_fsgid = op_data->op_fsgid;
405 rec->rn_cap = op_data->op_cap;
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_name(req, &RMF_NAME, old, oldlen);
417 mdc_pack_name(req, &RMF_SYMTGT, new, newlen);
420 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, u32 flags,
421 struct md_op_data *op_data, size_t ea_size)
423 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
426 b->mbo_valid = valid;
427 if (op_data->op_bias & MDS_CHECK_SPLIT)
428 b->mbo_valid |= OBD_MD_FLCKSPLIT;
429 if (op_data->op_bias & MDS_CROSS_REF)
430 b->mbo_valid |= OBD_MD_FLCROSSREF;
431 b->mbo_eadatasize = ea_size;
432 b->mbo_flags = flags;
433 __mdc_pack_body(b, op_data->op_suppgids[0]);
435 b->mbo_fid1 = op_data->op_fid1;
436 b->mbo_fid2 = op_data->op_fid2;
437 b->mbo_valid |= OBD_MD_FLID;
439 if (op_data->op_name)
440 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
441 op_data->op_namelen);
444 static void mdc_hsm_release_pack(struct ptlrpc_request *req,
445 struct md_op_data *op_data)
447 if (op_data->op_bias & MDS_HSM_RELEASE) {
448 struct close_data *data;
449 struct ldlm_lock *lock;
451 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
453 lock = ldlm_handle2lock(&op_data->op_lease_handle);
455 data->cd_handle = lock->l_remote_handle;
458 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
460 data->cd_data_version = op_data->op_data_version;
461 data->cd_fid = op_data->op_fid2;
465 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
467 struct mdt_ioepoch *epoch;
468 struct mdt_rec_setattr *rec;
470 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
471 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
473 mdc_setattr_pack_rec(rec, op_data);
475 * The client will zero out local timestamps when losing the IBITS lock
476 * so any new RPC timestamps will update the client inode's timestamps.
477 * There was a defect on the server side which allowed the atime to be
478 * overwritten by a zeroed-out atime packed into the close RPC.
480 * Proactively clear the MDS_ATTR_ATIME flag in the RPC in this case
481 * to avoid zeroing the atime on old unpatched servers. See LU-8041.
483 if (rec->sa_atime == 0)
484 rec->sa_valid &= ~MDS_ATTR_ATIME;
486 mdc_ioepoch_pack(epoch, op_data);
487 mdc_hsm_release_pack(req, op_data);