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, 2017, 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 <linux/user_namespace.h>
35 #ifdef HAVE_UIDGID_HEADER
36 # include <linux/uidgid.h>
38 #include <lustre_net.h>
39 #include <obd_class.h>
41 #include <cl_object.h>
42 #include "mdc_internal.h"
44 static void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
46 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
47 mrc->cr_flags_h = (__u32)(flags >> 32);
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(lu_name_is_valid_2(buf, cpy_len));
120 if (cpy_len != name_len)
121 CDEBUG(D_DENTRY, "%s: %s len %zd != %zd, concurrent rename?\n",
122 req->rq_export->exp_obd->obd_name, buf, name_len,
126 void mdc_file_secctx_pack(struct ptlrpc_request *req, const char *secctx_name,
127 const void *secctx, size_t secctx_size)
132 if (secctx_name == NULL)
135 buf = req_capsule_client_get(&req->rq_pill, &RMF_FILE_SECCTX_NAME);
136 buf_size = req_capsule_get_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
139 LASSERT(buf_size == strlen(secctx_name) + 1);
140 memcpy(buf, secctx_name, buf_size);
142 buf = req_capsule_client_get(&req->rq_pill, &RMF_FILE_SECCTX);
143 buf_size = req_capsule_get_size(&req->rq_pill, &RMF_FILE_SECCTX,
146 LASSERT(buf_size == secctx_size);
147 memcpy(buf, secctx, buf_size);
150 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff, size_t size,
151 const struct lu_fid *fid)
153 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
156 b->mbo_valid |= OBD_MD_FLID;
157 b->mbo_size = pgoff; /* !! */
158 b->mbo_nlink = size; /* !! */
159 __mdc_pack_body(b, -1);
160 b->mbo_mode = LUDA_FID | LUDA_TYPE;
163 /* packing of MDS records */
164 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
165 const void *data, size_t datalen, umode_t mode,
166 uid_t uid, gid_t gid, cfs_cap_t cap_effective, __u64 rdev)
168 struct mdt_rec_create *rec;
172 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
173 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
176 rec->cr_opcode = REINT_CREATE;
179 rec->cr_cap = cap_effective;
180 rec->cr_fid1 = op_data->op_fid1;
181 rec->cr_fid2 = op_data->op_fid2;
184 rec->cr_time = op_data->op_mod_time;
185 rec->cr_suppgid1 = op_data->op_suppgids[0];
186 rec->cr_suppgid2 = op_data->op_suppgids[1];
188 if (op_data->op_bias & MDS_CREATE_VOLATILE)
189 flags |= MDS_OPEN_VOLATILE;
190 set_mrc_cr_flags(rec, flags);
191 rec->cr_bias = op_data->op_bias;
192 rec->cr_umask = current_umask();
194 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
196 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
197 memcpy(tmp, data, datalen);
200 mdc_file_secctx_pack(req, op_data->op_file_secctx_name,
201 op_data->op_file_secctx,
202 op_data->op_file_secctx_size);
205 static inline __u64 mds_pack_open_flags(__u64 flags)
207 __u64 cr_flags = (flags & MDS_OPEN_FL_INTERNAL);
209 if (flags & FMODE_READ)
210 cr_flags |= MDS_FMODE_READ;
211 if (flags & FMODE_WRITE)
212 cr_flags |= MDS_FMODE_WRITE;
214 cr_flags |= MDS_OPEN_CREAT;
216 cr_flags |= MDS_OPEN_EXCL;
218 cr_flags |= MDS_OPEN_TRUNC;
219 if (flags & O_APPEND)
220 cr_flags |= MDS_OPEN_APPEND;
222 cr_flags |= MDS_OPEN_SYNC;
223 if (flags & O_DIRECTORY)
224 cr_flags |= MDS_OPEN_DIRECTORY;
226 if (flags & FMODE_EXEC)
227 cr_flags |= MDS_FMODE_EXEC;
229 if (cl_is_lov_delay_create(flags))
230 cr_flags |= MDS_OPEN_DELAY_CREATE;
232 if (flags & O_NONBLOCK)
233 cr_flags |= MDS_OPEN_NORESTORE;
238 /* packing of MDS records */
239 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
240 umode_t mode, __u64 rdev, __u64 flags, const void *lmm,
243 struct mdt_rec_create *rec;
247 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
248 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
250 /* XXX do something about time, uid, gid */
251 rec->cr_opcode = REINT_OPEN;
252 rec->cr_fsuid = from_kuid(&init_user_ns, current_fsuid());
253 rec->cr_fsgid = from_kgid(&init_user_ns, current_fsgid());
254 rec->cr_cap = cfs_curproc_cap_pack();
256 cr_flags = mds_pack_open_flags(flags);
258 rec->cr_umask = current_umask();
259 if (op_data != NULL) {
260 rec->cr_fid1 = op_data->op_fid1;
261 rec->cr_fid2 = op_data->op_fid2;
262 rec->cr_time = op_data->op_mod_time;
263 rec->cr_suppgid1 = op_data->op_suppgids[0];
264 rec->cr_suppgid2 = op_data->op_suppgids[1];
265 rec->cr_bias = op_data->op_bias;
266 rec->cr_open_handle_old = op_data->op_open_handle;
268 if (op_data->op_name) {
269 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
270 op_data->op_namelen);
272 if (op_data->op_bias & MDS_CREATE_VOLATILE)
273 cr_flags |= MDS_OPEN_VOLATILE;
276 mdc_file_secctx_pack(req, op_data->op_file_secctx_name,
277 op_data->op_file_secctx,
278 op_data->op_file_secctx_size);
282 cr_flags |= MDS_OPEN_HAS_EA;
283 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
284 memcpy(tmp, lmm, lmmlen);
286 set_mrc_cr_flags(rec, cr_flags);
289 static inline u64 attr_pack(unsigned int ia_valid, enum op_xvalid ia_xvalid)
293 if (ia_valid & ATTR_MODE)
294 sa_valid |= MDS_ATTR_MODE;
295 if (ia_valid & ATTR_UID)
296 sa_valid |= MDS_ATTR_UID;
297 if (ia_valid & ATTR_GID)
298 sa_valid |= MDS_ATTR_GID;
299 if (ia_valid & ATTR_SIZE)
300 sa_valid |= MDS_ATTR_SIZE;
301 if (ia_valid & ATTR_ATIME)
302 sa_valid |= MDS_ATTR_ATIME;
303 if (ia_valid & ATTR_MTIME)
304 sa_valid |= MDS_ATTR_MTIME;
305 if (ia_valid & ATTR_CTIME)
306 sa_valid |= MDS_ATTR_CTIME;
307 if (ia_valid & ATTR_ATIME_SET)
308 sa_valid |= MDS_ATTR_ATIME_SET;
309 if (ia_valid & ATTR_MTIME_SET)
310 sa_valid |= MDS_ATTR_MTIME_SET;
311 if (ia_valid & ATTR_FORCE)
312 sa_valid |= MDS_ATTR_FORCE;
313 if (ia_xvalid & OP_XVALID_FLAGS)
314 sa_valid |= MDS_ATTR_ATTR_FLAG;
315 if (ia_valid & ATTR_KILL_SUID)
316 sa_valid |= MDS_ATTR_KILL_SUID;
317 if (ia_valid & ATTR_KILL_SGID)
318 sa_valid |= MDS_ATTR_KILL_SGID;
319 if (ia_xvalid & OP_XVALID_CTIME_SET)
320 sa_valid |= MDS_ATTR_CTIME_SET;
321 if (ia_valid & ATTR_OPEN)
322 sa_valid |= MDS_ATTR_FROM_OPEN;
323 if (ia_xvalid & OP_XVALID_BLOCKS)
324 sa_valid |= MDS_ATTR_BLOCKS;
325 if (ia_xvalid & OP_XVALID_OWNEROVERRIDE)
326 /* NFSD hack (see bug 5781) */
327 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
328 if (ia_xvalid & OP_XVALID_PROJID)
329 sa_valid |= MDS_ATTR_PROJID;
330 if (ia_xvalid & OP_XVALID_LAZYSIZE)
331 sa_valid |= MDS_ATTR_LSIZE;
332 if (ia_xvalid & OP_XVALID_LAZYBLOCKS)
333 sa_valid |= MDS_ATTR_LBLOCKS;
337 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
338 struct md_op_data *op_data)
340 rec->sa_opcode = REINT_SETATTR;
341 rec->sa_fsuid = from_kuid(&init_user_ns, current_fsuid());
342 rec->sa_fsgid = from_kgid(&init_user_ns, current_fsgid());
343 rec->sa_cap = cfs_curproc_cap_pack();
344 rec->sa_suppgid = -1;
346 rec->sa_fid = op_data->op_fid1;
347 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid,
349 rec->sa_mode = op_data->op_attr.ia_mode;
350 rec->sa_uid = from_kuid(&init_user_ns, op_data->op_attr.ia_uid);
351 rec->sa_gid = from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
352 rec->sa_projid = op_data->op_projid;
353 rec->sa_size = op_data->op_attr.ia_size;
354 rec->sa_blocks = op_data->op_attr_blocks;
355 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
356 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
357 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
358 rec->sa_attr_flags = op_data->op_attr_flags;
359 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
360 in_group_p(op_data->op_attr.ia_gid))
362 from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
364 rec->sa_suppgid = op_data->op_suppgids[0];
366 rec->sa_bias = op_data->op_bias;
369 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
370 struct md_op_data *op_data)
372 epoch->mio_open_handle = op_data->op_open_handle;
373 epoch->mio_unused1 = 0;
374 epoch->mio_unused2 = 0;
375 epoch->mio_padding = 0;
378 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
379 void *ea, size_t ealen)
381 struct mdt_rec_setattr *rec;
382 struct lov_user_md *lum = NULL;
384 CLASSERT(sizeof(struct mdt_rec_reint) ==
385 sizeof(struct mdt_rec_setattr));
386 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
387 mdc_setattr_pack_rec(rec, op_data);
392 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
393 if (ea == NULL) { /* Remove LOV EA */
394 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
395 lum->lmm_stripe_size = 0;
396 lum->lmm_stripe_count = 0;
397 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
399 memcpy(lum, ea, ealen);
403 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
405 struct mdt_rec_unlink *rec;
407 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
408 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
409 LASSERT(rec != NULL);
411 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
412 REINT_RMENTRY : REINT_UNLINK;
413 rec->ul_fsuid = op_data->op_fsuid;
414 rec->ul_fsgid = op_data->op_fsgid;
415 rec->ul_cap = op_data->op_cap;
416 rec->ul_mode = op_data->op_mode;
417 rec->ul_suppgid1= op_data->op_suppgids[0];
418 rec->ul_suppgid2= -1;
419 rec->ul_fid1 = op_data->op_fid1;
420 rec->ul_fid2 = op_data->op_fid2;
421 rec->ul_time = op_data->op_mod_time;
422 rec->ul_bias = op_data->op_bias;
424 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
427 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
429 struct mdt_rec_link *rec;
431 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
432 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
433 LASSERT (rec != NULL);
435 rec->lk_opcode = REINT_LINK;
436 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
437 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
438 rec->lk_cap = op_data->op_cap;//current->cap_effective;
439 rec->lk_suppgid1 = op_data->op_suppgids[0];
440 rec->lk_suppgid2 = op_data->op_suppgids[1];
441 rec->lk_fid1 = op_data->op_fid1;
442 rec->lk_fid2 = op_data->op_fid2;
443 rec->lk_time = op_data->op_mod_time;
444 rec->lk_bias = op_data->op_bias;
446 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
449 static void mdc_close_intent_pack(struct ptlrpc_request *req,
450 struct md_op_data *op_data)
452 struct close_data *data;
453 struct ldlm_lock *lock;
454 enum mds_op_bias bias = op_data->op_bias;
456 if (!(bias & (MDS_CLOSE_INTENT | MDS_CLOSE_MIGRATE)))
459 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
460 LASSERT(data != NULL);
462 lock = ldlm_handle2lock(&op_data->op_lease_handle);
464 data->cd_handle = lock->l_remote_handle;
467 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
469 data->cd_data_version = op_data->op_data_version;
470 data->cd_fid = op_data->op_fid2;
472 if (bias & MDS_CLOSE_LAYOUT_SPLIT) {
473 data->cd_mirror_id = op_data->op_mirror_id;
474 } else if (bias & MDS_CLOSE_RESYNC_DONE) {
475 struct close_data_resync_done *sync = &data->cd_resync;
477 CLASSERT(sizeof(data->cd_resync) <= sizeof(data->cd_reserved));
478 sync->resync_count = op_data->op_data_size / sizeof(__u32);
479 if (sync->resync_count <= INLINE_RESYNC_ARRAY_SIZE) {
480 memcpy(sync->resync_ids_inline, op_data->op_data,
481 op_data->op_data_size);
483 size_t count = sync->resync_count;
485 memcpy(req_capsule_client_get(&req->rq_pill, &RMF_U32),
486 op_data->op_data, count * sizeof(__u32));
491 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
492 const char *old, size_t oldlen,
493 const char *new, size_t newlen)
495 struct mdt_rec_rename *rec;
497 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
498 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
500 /* XXX do something about time, uid, gid */
501 rec->rn_opcode = REINT_RENAME;
502 rec->rn_fsuid = op_data->op_fsuid;
503 rec->rn_fsgid = op_data->op_fsgid;
504 rec->rn_cap = op_data->op_cap;
505 rec->rn_suppgid1 = op_data->op_suppgids[0];
506 rec->rn_suppgid2 = op_data->op_suppgids[1];
507 rec->rn_fid1 = op_data->op_fid1;
508 rec->rn_fid2 = op_data->op_fid2;
509 rec->rn_time = op_data->op_mod_time;
510 rec->rn_mode = op_data->op_mode;
511 rec->rn_bias = op_data->op_bias;
513 mdc_pack_name(req, &RMF_NAME, old, oldlen);
516 mdc_pack_name(req, &RMF_SYMTGT, new, newlen);
519 void mdc_migrate_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
520 const char *name, size_t namelen)
522 struct mdt_rec_rename *rec;
525 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
526 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
528 rec->rn_opcode = REINT_MIGRATE;
529 rec->rn_fsuid = op_data->op_fsuid;
530 rec->rn_fsgid = op_data->op_fsgid;
531 rec->rn_cap = op_data->op_cap;
532 rec->rn_suppgid1 = op_data->op_suppgids[0];
533 rec->rn_suppgid2 = op_data->op_suppgids[1];
534 rec->rn_fid1 = op_data->op_fid1;
535 rec->rn_fid2 = op_data->op_fid4;
536 rec->rn_time = op_data->op_mod_time;
537 rec->rn_mode = op_data->op_mode;
538 rec->rn_bias = op_data->op_bias;
540 mdc_pack_name(req, &RMF_NAME, name, namelen);
542 if (op_data->op_bias & MDS_CLOSE_MIGRATE) {
543 struct mdt_ioepoch *epoch;
545 mdc_close_intent_pack(req, op_data);
546 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
547 mdc_ioepoch_pack(epoch, op_data);
550 ea = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
551 memcpy(ea, op_data->op_data, op_data->op_data_size);
554 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, __u32 flags,
555 struct md_op_data *op_data, size_t ea_size)
557 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
560 b->mbo_valid = valid;
561 if (op_data->op_bias & MDS_CROSS_REF)
562 b->mbo_valid |= OBD_MD_FLCROSSREF;
563 b->mbo_eadatasize = ea_size;
564 b->mbo_flags = flags;
565 __mdc_pack_body(b, op_data->op_suppgids[0]);
567 b->mbo_fid1 = op_data->op_fid1;
568 b->mbo_fid2 = op_data->op_fid2;
569 b->mbo_valid |= OBD_MD_FLID;
571 if (op_data->op_name != NULL)
572 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
573 op_data->op_namelen);
576 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
578 struct mdt_ioepoch *epoch;
579 struct mdt_rec_setattr *rec;
581 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
582 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
584 mdc_setattr_pack_rec(rec, op_data);
586 * The client will zero out local timestamps when losing the IBITS lock
587 * so any new RPC timestamps will update the client inode's timestamps.
588 * There was a defect on the server side which allowed the atime to be
589 * overwritten by a zeroed-out atime packed into the close RPC.
591 * Proactively clear the MDS_ATTR_ATIME flag in the RPC in this case
592 * to avoid zeroing the atime on old unpatched servers. See LU-8041.
594 if (rec->sa_atime == 0)
595 rec->sa_valid &= ~MDS_ATTR_ATIME;
597 mdc_ioepoch_pack(epoch, op_data);
598 mdc_close_intent_pack(req, op_data);