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 <linux/user_namespace.h>
39 #ifdef HAVE_UIDGID_HEADER
40 # include <linux/uidgid.h>
42 #include <lustre_net.h>
43 #include <lustre/lustre_idl.h>
44 #include <obd_class.h>
46 #include <cl_object.h>
47 #include "mdc_internal.h"
50 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
54 b->mbo_suppgid = suppgid;
55 b->mbo_uid = from_kuid(&init_user_ns, current_uid());
56 b->mbo_gid = from_kgid(&init_user_ns, current_gid());
57 b->mbo_fsuid = from_kuid(&init_user_ns, current_fsuid());
58 b->mbo_fsgid = from_kgid(&init_user_ns, current_fsgid());
59 b->mbo_capability = cfs_curproc_cap_pack();
62 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
63 struct md_op_data *op_data)
65 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
68 __mdc_pack_body(b, op_data->op_suppgids[0]);
69 b->mbo_fid1 = op_data->op_fid1;
70 b->mbo_fid2 = op_data->op_fid2;
71 b->mbo_valid |= OBD_MD_FLID;
74 void mdc_pack_body(struct ptlrpc_request *req, const struct lu_fid *fid,
75 u64 valid, size_t ea_size, u32 suppgid, u32 flags)
77 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
81 b->mbo_eadatasize = ea_size;
83 __mdc_pack_body(b, suppgid);
86 b->mbo_valid |= OBD_MD_FLID;
91 * Pack a name (path component) into a request
93 * \param[in] req request
94 * \param[in] field request field (usually RMF_NAME)
95 * \param[in] name path component
96 * \param[in] name_len length of path component
98 * \a field must be present in \a req and of size \a name_len + 1.
100 * \a name must be '\0' terminated of length \a name_len and represent
101 * a single path component (not contain '/').
103 static void mdc_pack_name(struct ptlrpc_request *req,
104 const struct req_msg_field *field,
105 const char *name, size_t name_len)
111 buf = req_capsule_client_get(&req->rq_pill, field);
112 buf_size = req_capsule_get_size(&req->rq_pill, field, RCL_CLIENT);
114 LASSERT(name != NULL && name_len != 0 &&
115 buf != NULL && buf_size == name_len + 1);
117 cpy_len = strlcpy(buf, name, buf_size);
119 LASSERT(cpy_len == name_len && lu_name_is_valid_2(buf, cpy_len));
122 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff, size_t size,
123 const struct lu_fid *fid)
125 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
128 b->mbo_valid |= OBD_MD_FLID;
129 b->mbo_size = pgoff; /* !! */
130 b->mbo_nlink = size; /* !! */
131 __mdc_pack_body(b, -1);
132 b->mbo_mode = LUDA_FID | LUDA_TYPE;
135 /* packing of MDS records */
136 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
137 const void *data, size_t datalen, umode_t mode,
138 uid_t uid, gid_t gid, cfs_cap_t cap_effective, __u64 rdev)
140 struct mdt_rec_create *rec;
144 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
145 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
148 rec->cr_opcode = REINT_CREATE;
151 rec->cr_cap = cap_effective;
152 rec->cr_fid1 = op_data->op_fid1;
153 rec->cr_fid2 = op_data->op_fid2;
156 rec->cr_time = op_data->op_mod_time;
157 rec->cr_suppgid1 = op_data->op_suppgids[0];
158 rec->cr_suppgid2 = op_data->op_suppgids[1];
160 if (op_data->op_bias & MDS_CREATE_VOLATILE)
161 flags |= MDS_OPEN_VOLATILE;
162 set_mrc_cr_flags(rec, flags);
163 rec->cr_bias = op_data->op_bias;
164 rec->cr_umask = current_umask();
166 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
168 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
169 memcpy(tmp, data, datalen);
173 static inline __u64 mds_pack_open_flags(__u64 flags)
175 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
176 MDS_OPEN_FL_INTERNAL));
179 cr_flags |= MDS_OPEN_CREAT;
181 cr_flags |= MDS_OPEN_EXCL;
183 cr_flags |= MDS_OPEN_TRUNC;
184 if (flags & O_APPEND)
185 cr_flags |= MDS_OPEN_APPEND;
187 cr_flags |= MDS_OPEN_SYNC;
188 if (flags & O_DIRECTORY)
189 cr_flags |= MDS_OPEN_DIRECTORY;
191 if (flags & FMODE_EXEC)
192 cr_flags |= MDS_FMODE_EXEC;
194 if (cl_is_lov_delay_create(flags))
195 cr_flags |= MDS_OPEN_DELAY_CREATE;
197 if (flags & O_NONBLOCK)
198 cr_flags |= MDS_OPEN_NORESTORE;
203 /* packing of MDS records */
204 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
205 umode_t mode, __u64 rdev, __u64 flags, const void *lmm,
208 struct mdt_rec_create *rec;
212 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
213 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
215 /* XXX do something about time, uid, gid */
216 rec->cr_opcode = REINT_OPEN;
217 rec->cr_fsuid = from_kuid(&init_user_ns, current_fsuid());
218 rec->cr_fsgid = from_kgid(&init_user_ns, current_fsgid());
219 rec->cr_cap = cfs_curproc_cap_pack();
221 cr_flags = mds_pack_open_flags(flags);
223 rec->cr_umask = current_umask();
224 if (op_data != NULL) {
225 rec->cr_fid1 = op_data->op_fid1;
226 rec->cr_fid2 = op_data->op_fid2;
227 rec->cr_time = op_data->op_mod_time;
228 rec->cr_suppgid1 = op_data->op_suppgids[0];
229 rec->cr_suppgid2 = op_data->op_suppgids[1];
230 rec->cr_bias = op_data->op_bias;
231 rec->cr_old_handle = op_data->op_handle;
233 if (op_data->op_name) {
234 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
235 op_data->op_namelen);
237 if (op_data->op_bias & MDS_CREATE_VOLATILE)
238 cr_flags |= MDS_OPEN_VOLATILE;
243 cr_flags |= MDS_OPEN_HAS_EA;
244 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
245 memcpy(tmp, lmm, lmmlen);
247 set_mrc_cr_flags(rec, cr_flags);
250 static inline __u64 attr_pack(unsigned int ia_valid) {
253 if (ia_valid & ATTR_MODE)
254 sa_valid |= MDS_ATTR_MODE;
255 if (ia_valid & ATTR_UID)
256 sa_valid |= MDS_ATTR_UID;
257 if (ia_valid & ATTR_GID)
258 sa_valid |= MDS_ATTR_GID;
259 if (ia_valid & ATTR_SIZE)
260 sa_valid |= MDS_ATTR_SIZE;
261 if (ia_valid & ATTR_ATIME)
262 sa_valid |= MDS_ATTR_ATIME;
263 if (ia_valid & ATTR_MTIME)
264 sa_valid |= MDS_ATTR_MTIME;
265 if (ia_valid & ATTR_CTIME)
266 sa_valid |= MDS_ATTR_CTIME;
267 if (ia_valid & ATTR_ATIME_SET)
268 sa_valid |= MDS_ATTR_ATIME_SET;
269 if (ia_valid & ATTR_MTIME_SET)
270 sa_valid |= MDS_ATTR_MTIME_SET;
271 if (ia_valid & ATTR_FORCE)
272 sa_valid |= MDS_ATTR_FORCE;
273 if (ia_valid & ATTR_ATTR_FLAG)
274 sa_valid |= MDS_ATTR_ATTR_FLAG;
275 if (ia_valid & ATTR_KILL_SUID)
276 sa_valid |= MDS_ATTR_KILL_SUID;
277 if (ia_valid & ATTR_KILL_SGID)
278 sa_valid |= MDS_ATTR_KILL_SGID;
279 if (ia_valid & ATTR_CTIME_SET)
280 sa_valid |= MDS_ATTR_CTIME_SET;
281 if (ia_valid & ATTR_FROM_OPEN)
282 sa_valid |= MDS_ATTR_FROM_OPEN;
283 if (ia_valid & ATTR_BLOCKS)
284 sa_valid |= MDS_ATTR_BLOCKS;
285 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
286 /* NFSD hack (see bug 5781) */
287 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
291 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
292 struct md_op_data *op_data)
294 rec->sa_opcode = REINT_SETATTR;
295 rec->sa_fsuid = from_kuid(&init_user_ns, current_fsuid());
296 rec->sa_fsgid = from_kgid(&init_user_ns, current_fsgid());
297 rec->sa_cap = cfs_curproc_cap_pack();
298 rec->sa_suppgid = -1;
300 rec->sa_fid = op_data->op_fid1;
301 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
302 rec->sa_mode = op_data->op_attr.ia_mode;
303 rec->sa_uid = from_kuid(&init_user_ns, op_data->op_attr.ia_uid);
304 rec->sa_gid = from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
305 rec->sa_size = op_data->op_attr.ia_size;
306 rec->sa_blocks = op_data->op_attr_blocks;
307 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
308 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
309 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
310 rec->sa_attr_flags = op_data->op_attr_flags;
311 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
312 in_group_p(op_data->op_attr.ia_gid))
314 from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
316 rec->sa_suppgid = op_data->op_suppgids[0];
318 rec->sa_bias = op_data->op_bias;
321 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
322 struct md_op_data *op_data)
324 epoch->mio_handle = op_data->op_handle;
325 epoch->mio_unused1 = 0;
326 epoch->mio_unused2 = 0;
327 epoch->mio_padding = 0;
330 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
331 void *ea, size_t ealen)
333 struct mdt_rec_setattr *rec;
334 struct lov_user_md *lum = NULL;
336 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
337 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
338 mdc_setattr_pack_rec(rec, op_data);
343 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
344 if (ea == NULL) { /* Remove LOV EA */
345 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
346 lum->lmm_stripe_size = 0;
347 lum->lmm_stripe_count = 0;
348 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
350 memcpy(lum, ea, ealen);
354 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
356 struct mdt_rec_unlink *rec;
358 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
359 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
360 LASSERT(rec != NULL);
362 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
363 REINT_RMENTRY : REINT_UNLINK;
364 rec->ul_fsuid = op_data->op_fsuid;
365 rec->ul_fsgid = op_data->op_fsgid;
366 rec->ul_cap = op_data->op_cap;
367 rec->ul_mode = op_data->op_mode;
368 rec->ul_suppgid1= op_data->op_suppgids[0];
369 rec->ul_suppgid2= -1;
370 rec->ul_fid1 = op_data->op_fid1;
371 rec->ul_fid2 = op_data->op_fid2;
372 rec->ul_time = op_data->op_mod_time;
373 rec->ul_bias = op_data->op_bias;
375 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
378 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
380 struct mdt_rec_link *rec;
382 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
383 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
384 LASSERT (rec != NULL);
386 rec->lk_opcode = REINT_LINK;
387 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
388 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
389 rec->lk_cap = op_data->op_cap;//current->cap_effective;
390 rec->lk_suppgid1 = op_data->op_suppgids[0];
391 rec->lk_suppgid2 = op_data->op_suppgids[1];
392 rec->lk_fid1 = op_data->op_fid1;
393 rec->lk_fid2 = op_data->op_fid2;
394 rec->lk_time = op_data->op_mod_time;
395 rec->lk_bias = op_data->op_bias;
397 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
400 static void mdc_intent_close_pack(struct ptlrpc_request *req,
401 struct md_op_data *op_data)
403 struct close_data *data;
404 struct ldlm_lock *lock;
405 enum mds_op_bias bias = op_data->op_bias;
407 if (!(bias & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP |
408 MDS_RENAME_MIGRATE)))
411 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
412 LASSERT(data != NULL);
414 lock = ldlm_handle2lock(&op_data->op_lease_handle);
416 data->cd_handle = lock->l_remote_handle;
419 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
421 data->cd_data_version = op_data->op_data_version;
422 data->cd_fid = op_data->op_fid2;
425 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
426 const char *old, size_t oldlen,
427 const char *new, size_t newlen)
429 struct mdt_rec_rename *rec;
431 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
432 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
434 /* XXX do something about time, uid, gid */
435 rec->rn_opcode = op_data->op_cli_flags & CLI_MIGRATE ?
436 REINT_MIGRATE : REINT_RENAME;
437 rec->rn_fsuid = op_data->op_fsuid;
438 rec->rn_fsgid = op_data->op_fsgid;
439 rec->rn_cap = op_data->op_cap;
440 rec->rn_suppgid1 = op_data->op_suppgids[0];
441 rec->rn_suppgid2 = op_data->op_suppgids[1];
442 rec->rn_fid1 = op_data->op_fid1;
443 rec->rn_fid2 = op_data->op_fid2;
444 rec->rn_time = op_data->op_mod_time;
445 rec->rn_mode = op_data->op_mode;
446 rec->rn_bias = op_data->op_bias;
448 mdc_pack_name(req, &RMF_NAME, old, oldlen);
451 mdc_pack_name(req, &RMF_SYMTGT, new, newlen);
453 if (op_data->op_cli_flags & CLI_MIGRATE &&
454 op_data->op_bias & MDS_RENAME_MIGRATE) {
455 struct mdt_ioepoch *epoch;
457 mdc_intent_close_pack(req, op_data);
458 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
459 mdc_ioepoch_pack(epoch, op_data);
463 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, __u32 flags,
464 struct md_op_data *op_data, size_t ea_size)
466 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
469 b->mbo_valid = valid;
470 if (op_data->op_bias & MDS_CHECK_SPLIT)
471 b->mbo_valid |= OBD_MD_FLCKSPLIT;
472 if (op_data->op_bias & MDS_CROSS_REF)
473 b->mbo_valid |= OBD_MD_FLCROSSREF;
474 b->mbo_eadatasize = ea_size;
475 b->mbo_flags = flags;
476 __mdc_pack_body(b, op_data->op_suppgids[0]);
478 b->mbo_fid1 = op_data->op_fid1;
479 b->mbo_fid2 = op_data->op_fid2;
480 b->mbo_valid |= OBD_MD_FLID;
482 if (op_data->op_name != NULL)
483 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
484 op_data->op_namelen);
487 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
489 struct mdt_ioepoch *epoch;
490 struct mdt_rec_setattr *rec;
492 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
493 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
495 mdc_setattr_pack_rec(rec, op_data);
497 * The client will zero out local timestamps when losing the IBITS lock
498 * so any new RPC timestamps will update the client inode's timestamps.
499 * There was a defect on the server side which allowed the atime to be
500 * overwritten by a zeroed-out atime packed into the close RPC.
502 * Proactively clear the MDS_ATTR_ATIME flag in the RPC in this case
503 * to avoid zeroing the atime on old unpatched servers. See LU-8041.
505 if (rec->sa_atime == 0)
506 rec->sa_valid &= ~MDS_ATTR_ATIME;
508 mdc_ioepoch_pack(epoch, op_data);
509 mdc_intent_close_pack(req, op_data);