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_file_secctx_pack(struct ptlrpc_request *req, const char *secctx_name,
123 const void *secctx, size_t secctx_size)
128 if (secctx_name == NULL)
131 buf = req_capsule_client_get(&req->rq_pill, &RMF_FILE_SECCTX_NAME);
132 buf_size = req_capsule_get_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
135 LASSERT(buf_size == strlen(secctx_name) + 1);
136 memcpy(buf, secctx_name, buf_size);
138 buf = req_capsule_client_get(&req->rq_pill, &RMF_FILE_SECCTX);
139 buf_size = req_capsule_get_size(&req->rq_pill, &RMF_FILE_SECCTX,
142 LASSERT(buf_size == secctx_size);
143 memcpy(buf, secctx, buf_size);
146 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff, size_t size,
147 const struct lu_fid *fid)
149 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
152 b->mbo_valid |= OBD_MD_FLID;
153 b->mbo_size = pgoff; /* !! */
154 b->mbo_nlink = size; /* !! */
155 __mdc_pack_body(b, -1);
156 b->mbo_mode = LUDA_FID | LUDA_TYPE;
159 /* packing of MDS records */
160 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
161 const void *data, size_t datalen, umode_t mode,
162 uid_t uid, gid_t gid, cfs_cap_t cap_effective, __u64 rdev)
164 struct mdt_rec_create *rec;
168 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
169 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
172 rec->cr_opcode = REINT_CREATE;
175 rec->cr_cap = cap_effective;
176 rec->cr_fid1 = op_data->op_fid1;
177 rec->cr_fid2 = op_data->op_fid2;
180 rec->cr_time = op_data->op_mod_time;
181 rec->cr_suppgid1 = op_data->op_suppgids[0];
182 rec->cr_suppgid2 = op_data->op_suppgids[1];
184 if (op_data->op_bias & MDS_CREATE_VOLATILE)
185 flags |= MDS_OPEN_VOLATILE;
186 set_mrc_cr_flags(rec, flags);
187 rec->cr_bias = op_data->op_bias;
188 rec->cr_umask = current_umask();
190 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
192 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
193 memcpy(tmp, data, datalen);
196 mdc_file_secctx_pack(req, op_data->op_file_secctx_name,
197 op_data->op_file_secctx,
198 op_data->op_file_secctx_size);
201 static inline __u64 mds_pack_open_flags(__u64 flags)
203 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
204 MDS_OPEN_FL_INTERNAL));
207 cr_flags |= MDS_OPEN_CREAT;
209 cr_flags |= MDS_OPEN_EXCL;
211 cr_flags |= MDS_OPEN_TRUNC;
212 if (flags & O_APPEND)
213 cr_flags |= MDS_OPEN_APPEND;
215 cr_flags |= MDS_OPEN_SYNC;
216 if (flags & O_DIRECTORY)
217 cr_flags |= MDS_OPEN_DIRECTORY;
219 if (flags & FMODE_EXEC)
220 cr_flags |= MDS_FMODE_EXEC;
222 if (cl_is_lov_delay_create(flags))
223 cr_flags |= MDS_OPEN_DELAY_CREATE;
225 if (flags & O_NONBLOCK)
226 cr_flags |= MDS_OPEN_NORESTORE;
231 /* packing of MDS records */
232 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
233 umode_t mode, __u64 rdev, __u64 flags, const void *lmm,
236 struct mdt_rec_create *rec;
240 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
241 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
243 /* XXX do something about time, uid, gid */
244 rec->cr_opcode = REINT_OPEN;
245 rec->cr_fsuid = from_kuid(&init_user_ns, current_fsuid());
246 rec->cr_fsgid = from_kgid(&init_user_ns, current_fsgid());
247 rec->cr_cap = cfs_curproc_cap_pack();
249 cr_flags = mds_pack_open_flags(flags);
251 rec->cr_umask = current_umask();
252 if (op_data != NULL) {
253 rec->cr_fid1 = op_data->op_fid1;
254 rec->cr_fid2 = op_data->op_fid2;
255 rec->cr_time = op_data->op_mod_time;
256 rec->cr_suppgid1 = op_data->op_suppgids[0];
257 rec->cr_suppgid2 = op_data->op_suppgids[1];
258 rec->cr_bias = op_data->op_bias;
259 rec->cr_old_handle = op_data->op_handle;
261 if (op_data->op_name) {
262 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
263 op_data->op_namelen);
265 if (op_data->op_bias & MDS_CREATE_VOLATILE)
266 cr_flags |= MDS_OPEN_VOLATILE;
269 mdc_file_secctx_pack(req, op_data->op_file_secctx_name,
270 op_data->op_file_secctx,
271 op_data->op_file_secctx_size);
275 cr_flags |= MDS_OPEN_HAS_EA;
276 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
277 memcpy(tmp, lmm, lmmlen);
279 set_mrc_cr_flags(rec, cr_flags);
282 static inline __u64 attr_pack(unsigned int ia_valid) {
285 if (ia_valid & ATTR_MODE)
286 sa_valid |= MDS_ATTR_MODE;
287 if (ia_valid & ATTR_UID)
288 sa_valid |= MDS_ATTR_UID;
289 if (ia_valid & ATTR_GID)
290 sa_valid |= MDS_ATTR_GID;
291 if (ia_valid & ATTR_SIZE)
292 sa_valid |= MDS_ATTR_SIZE;
293 if (ia_valid & ATTR_ATIME)
294 sa_valid |= MDS_ATTR_ATIME;
295 if (ia_valid & ATTR_MTIME)
296 sa_valid |= MDS_ATTR_MTIME;
297 if (ia_valid & ATTR_CTIME)
298 sa_valid |= MDS_ATTR_CTIME;
299 if (ia_valid & ATTR_ATIME_SET)
300 sa_valid |= MDS_ATTR_ATIME_SET;
301 if (ia_valid & ATTR_MTIME_SET)
302 sa_valid |= MDS_ATTR_MTIME_SET;
303 if (ia_valid & ATTR_FORCE)
304 sa_valid |= MDS_ATTR_FORCE;
305 if (ia_valid & ATTR_ATTR_FLAG)
306 sa_valid |= MDS_ATTR_ATTR_FLAG;
307 if (ia_valid & ATTR_KILL_SUID)
308 sa_valid |= MDS_ATTR_KILL_SUID;
309 if (ia_valid & ATTR_KILL_SGID)
310 sa_valid |= MDS_ATTR_KILL_SGID;
311 if (ia_valid & ATTR_CTIME_SET)
312 sa_valid |= MDS_ATTR_CTIME_SET;
313 if (ia_valid & ATTR_FROM_OPEN)
314 sa_valid |= MDS_ATTR_FROM_OPEN;
315 if (ia_valid & ATTR_BLOCKS)
316 sa_valid |= MDS_ATTR_BLOCKS;
317 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
318 /* NFSD hack (see bug 5781) */
319 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
323 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
324 struct md_op_data *op_data)
326 rec->sa_opcode = REINT_SETATTR;
327 rec->sa_fsuid = from_kuid(&init_user_ns, current_fsuid());
328 rec->sa_fsgid = from_kgid(&init_user_ns, current_fsgid());
329 rec->sa_cap = cfs_curproc_cap_pack();
330 rec->sa_suppgid = -1;
332 rec->sa_fid = op_data->op_fid1;
333 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
334 rec->sa_mode = op_data->op_attr.ia_mode;
335 rec->sa_uid = from_kuid(&init_user_ns, op_data->op_attr.ia_uid);
336 rec->sa_gid = from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
337 rec->sa_size = op_data->op_attr.ia_size;
338 rec->sa_blocks = op_data->op_attr_blocks;
339 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
340 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
341 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
342 rec->sa_attr_flags = op_data->op_attr_flags;
343 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
344 in_group_p(op_data->op_attr.ia_gid))
346 from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
348 rec->sa_suppgid = op_data->op_suppgids[0];
350 rec->sa_bias = op_data->op_bias;
353 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
354 struct md_op_data *op_data)
356 epoch->mio_handle = op_data->op_handle;
357 epoch->mio_unused1 = 0;
358 epoch->mio_unused2 = 0;
359 epoch->mio_padding = 0;
362 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
363 void *ea, size_t ealen)
365 struct mdt_rec_setattr *rec;
366 struct lov_user_md *lum = NULL;
368 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
369 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
370 mdc_setattr_pack_rec(rec, op_data);
375 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
376 if (ea == NULL) { /* Remove LOV EA */
377 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
378 lum->lmm_stripe_size = 0;
379 lum->lmm_stripe_count = 0;
380 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
382 memcpy(lum, ea, ealen);
386 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
388 struct mdt_rec_unlink *rec;
390 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
391 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
392 LASSERT(rec != NULL);
394 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
395 REINT_RMENTRY : REINT_UNLINK;
396 rec->ul_fsuid = op_data->op_fsuid;
397 rec->ul_fsgid = op_data->op_fsgid;
398 rec->ul_cap = op_data->op_cap;
399 rec->ul_mode = op_data->op_mode;
400 rec->ul_suppgid1= op_data->op_suppgids[0];
401 rec->ul_suppgid2= -1;
402 rec->ul_fid1 = op_data->op_fid1;
403 rec->ul_fid2 = op_data->op_fid2;
404 rec->ul_time = op_data->op_mod_time;
405 rec->ul_bias = op_data->op_bias;
407 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
410 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
412 struct mdt_rec_link *rec;
414 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
415 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
416 LASSERT (rec != NULL);
418 rec->lk_opcode = REINT_LINK;
419 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
420 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
421 rec->lk_cap = op_data->op_cap;//current->cap_effective;
422 rec->lk_suppgid1 = op_data->op_suppgids[0];
423 rec->lk_suppgid2 = op_data->op_suppgids[1];
424 rec->lk_fid1 = op_data->op_fid1;
425 rec->lk_fid2 = op_data->op_fid2;
426 rec->lk_time = op_data->op_mod_time;
427 rec->lk_bias = op_data->op_bias;
429 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
432 static void mdc_intent_close_pack(struct ptlrpc_request *req,
433 struct md_op_data *op_data)
435 struct close_data *data;
436 struct ldlm_lock *lock;
437 enum mds_op_bias bias = op_data->op_bias;
439 if (!(bias & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP |
440 MDS_RENAME_MIGRATE)))
443 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
444 LASSERT(data != NULL);
446 lock = ldlm_handle2lock(&op_data->op_lease_handle);
448 data->cd_handle = lock->l_remote_handle;
451 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
453 data->cd_data_version = op_data->op_data_version;
454 data->cd_fid = op_data->op_fid2;
457 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
458 const char *old, size_t oldlen,
459 const char *new, size_t newlen)
461 struct mdt_rec_rename *rec;
463 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
464 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
466 /* XXX do something about time, uid, gid */
467 rec->rn_opcode = op_data->op_cli_flags & CLI_MIGRATE ?
468 REINT_MIGRATE : REINT_RENAME;
469 rec->rn_fsuid = op_data->op_fsuid;
470 rec->rn_fsgid = op_data->op_fsgid;
471 rec->rn_cap = op_data->op_cap;
472 rec->rn_suppgid1 = op_data->op_suppgids[0];
473 rec->rn_suppgid2 = op_data->op_suppgids[1];
474 rec->rn_fid1 = op_data->op_fid1;
475 rec->rn_fid2 = op_data->op_fid2;
476 rec->rn_time = op_data->op_mod_time;
477 rec->rn_mode = op_data->op_mode;
478 rec->rn_bias = op_data->op_bias;
480 mdc_pack_name(req, &RMF_NAME, old, oldlen);
483 mdc_pack_name(req, &RMF_SYMTGT, new, newlen);
485 if (op_data->op_cli_flags & CLI_MIGRATE &&
486 op_data->op_bias & MDS_RENAME_MIGRATE) {
487 struct mdt_ioepoch *epoch;
489 mdc_intent_close_pack(req, op_data);
490 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
491 mdc_ioepoch_pack(epoch, op_data);
495 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, __u32 flags,
496 struct md_op_data *op_data, size_t ea_size)
498 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
501 b->mbo_valid = valid;
502 if (op_data->op_bias & MDS_CHECK_SPLIT)
503 b->mbo_valid |= OBD_MD_FLCKSPLIT;
504 if (op_data->op_bias & MDS_CROSS_REF)
505 b->mbo_valid |= OBD_MD_FLCROSSREF;
506 b->mbo_eadatasize = ea_size;
507 b->mbo_flags = flags;
508 __mdc_pack_body(b, op_data->op_suppgids[0]);
510 b->mbo_fid1 = op_data->op_fid1;
511 b->mbo_fid2 = op_data->op_fid2;
512 b->mbo_valid |= OBD_MD_FLID;
514 if (op_data->op_name != NULL)
515 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
516 op_data->op_namelen);
519 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
521 struct mdt_ioepoch *epoch;
522 struct mdt_rec_setattr *rec;
524 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
525 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
527 mdc_setattr_pack_rec(rec, op_data);
529 * The client will zero out local timestamps when losing the IBITS lock
530 * so any new RPC timestamps will update the client inode's timestamps.
531 * There was a defect on the server side which allowed the atime to be
532 * overwritten by a zeroed-out atime packed into the close RPC.
534 * Proactively clear the MDS_ATTR_ATIME flag in the RPC in this case
535 * to avoid zeroing the atime on old unpatched servers. See LU-8041.
537 if (rec->sa_atime == 0)
538 rec->sa_valid &= ~MDS_ATTR_ATIME;
540 mdc_ioepoch_pack(epoch, op_data);
541 mdc_intent_close_pack(req, op_data);