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, 2016, 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 <lustre/lustre_idl.h>
40 #include <obd_class.h>
42 #include <cl_object.h>
43 #include "mdc_internal.h"
45 static void set_mrc_cr_flags(struct mdt_rec_create *mrc, __u64 flags)
47 mrc->cr_flags_l = (__u32)(flags & 0xFFFFFFFFUll);
48 mrc->cr_flags_h = (__u32)(flags >> 32);
51 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
55 b->mbo_suppgid = suppgid;
56 b->mbo_uid = from_kuid(&init_user_ns, current_uid());
57 b->mbo_gid = from_kgid(&init_user_ns, current_gid());
58 b->mbo_fsuid = from_kuid(&init_user_ns, current_fsuid());
59 b->mbo_fsgid = from_kgid(&init_user_ns, current_fsgid());
60 b->mbo_capability = cfs_curproc_cap_pack();
63 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
64 struct md_op_data *op_data)
66 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
69 __mdc_pack_body(b, op_data->op_suppgids[0]);
70 b->mbo_fid1 = op_data->op_fid1;
71 b->mbo_fid2 = op_data->op_fid2;
72 b->mbo_valid |= OBD_MD_FLID;
75 void mdc_pack_body(struct ptlrpc_request *req, const struct lu_fid *fid,
76 u64 valid, size_t ea_size, u32 suppgid, u32 flags)
78 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
82 b->mbo_eadatasize = ea_size;
84 __mdc_pack_body(b, suppgid);
87 b->mbo_valid |= OBD_MD_FLID;
92 * Pack a name (path component) into a request
94 * \param[in] req request
95 * \param[in] field request field (usually RMF_NAME)
96 * \param[in] name path component
97 * \param[in] name_len length of path component
99 * \a field must be present in \a req and of size \a name_len + 1.
101 * \a name must be '\0' terminated of length \a name_len and represent
102 * a single path component (not contain '/').
104 static void mdc_pack_name(struct ptlrpc_request *req,
105 const struct req_msg_field *field,
106 const char *name, size_t name_len)
112 buf = req_capsule_client_get(&req->rq_pill, field);
113 buf_size = req_capsule_get_size(&req->rq_pill, field, RCL_CLIENT);
115 LASSERT(name != NULL && name_len != 0 &&
116 buf != NULL && buf_size == name_len + 1);
118 cpy_len = strlcpy(buf, name, buf_size);
120 LASSERT(cpy_len == name_len && lu_name_is_valid_2(buf, cpy_len));
123 void mdc_file_secctx_pack(struct ptlrpc_request *req, const char *secctx_name,
124 const void *secctx, size_t secctx_size)
129 if (secctx_name == NULL)
132 buf = req_capsule_client_get(&req->rq_pill, &RMF_FILE_SECCTX_NAME);
133 buf_size = req_capsule_get_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
136 LASSERT(buf_size == strlen(secctx_name) + 1);
137 memcpy(buf, secctx_name, buf_size);
139 buf = req_capsule_client_get(&req->rq_pill, &RMF_FILE_SECCTX);
140 buf_size = req_capsule_get_size(&req->rq_pill, &RMF_FILE_SECCTX,
143 LASSERT(buf_size == secctx_size);
144 memcpy(buf, secctx, buf_size);
147 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff, size_t size,
148 const struct lu_fid *fid)
150 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
153 b->mbo_valid |= OBD_MD_FLID;
154 b->mbo_size = pgoff; /* !! */
155 b->mbo_nlink = size; /* !! */
156 __mdc_pack_body(b, -1);
157 b->mbo_mode = LUDA_FID | LUDA_TYPE;
160 /* packing of MDS records */
161 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
162 const void *data, size_t datalen, umode_t mode,
163 uid_t uid, gid_t gid, cfs_cap_t cap_effective, __u64 rdev)
165 struct mdt_rec_create *rec;
169 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
170 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
173 rec->cr_opcode = REINT_CREATE;
176 rec->cr_cap = cap_effective;
177 rec->cr_fid1 = op_data->op_fid1;
178 rec->cr_fid2 = op_data->op_fid2;
181 rec->cr_time = op_data->op_mod_time;
182 rec->cr_suppgid1 = op_data->op_suppgids[0];
183 rec->cr_suppgid2 = op_data->op_suppgids[1];
185 if (op_data->op_bias & MDS_CREATE_VOLATILE)
186 flags |= MDS_OPEN_VOLATILE;
187 set_mrc_cr_flags(rec, flags);
188 rec->cr_bias = op_data->op_bias;
189 rec->cr_umask = current_umask();
191 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
193 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
194 memcpy(tmp, data, datalen);
197 mdc_file_secctx_pack(req, op_data->op_file_secctx_name,
198 op_data->op_file_secctx,
199 op_data->op_file_secctx_size);
202 static inline __u64 mds_pack_open_flags(__u64 flags)
204 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
205 MDS_OPEN_FL_INTERNAL));
208 cr_flags |= MDS_OPEN_CREAT;
210 cr_flags |= MDS_OPEN_EXCL;
212 cr_flags |= MDS_OPEN_TRUNC;
213 if (flags & O_APPEND)
214 cr_flags |= MDS_OPEN_APPEND;
216 cr_flags |= MDS_OPEN_SYNC;
217 if (flags & O_DIRECTORY)
218 cr_flags |= MDS_OPEN_DIRECTORY;
220 if (flags & FMODE_EXEC)
221 cr_flags |= MDS_FMODE_EXEC;
223 if (cl_is_lov_delay_create(flags))
224 cr_flags |= MDS_OPEN_DELAY_CREATE;
226 if (flags & O_NONBLOCK)
227 cr_flags |= MDS_OPEN_NORESTORE;
232 /* packing of MDS records */
233 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
234 umode_t mode, __u64 rdev, __u64 flags, const void *lmm,
237 struct mdt_rec_create *rec;
241 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_create));
242 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
244 /* XXX do something about time, uid, gid */
245 rec->cr_opcode = REINT_OPEN;
246 rec->cr_fsuid = from_kuid(&init_user_ns, current_fsuid());
247 rec->cr_fsgid = from_kgid(&init_user_ns, current_fsgid());
248 rec->cr_cap = cfs_curproc_cap_pack();
250 cr_flags = mds_pack_open_flags(flags);
252 rec->cr_umask = current_umask();
253 if (op_data != NULL) {
254 rec->cr_fid1 = op_data->op_fid1;
255 rec->cr_fid2 = op_data->op_fid2;
256 rec->cr_time = op_data->op_mod_time;
257 rec->cr_suppgid1 = op_data->op_suppgids[0];
258 rec->cr_suppgid2 = op_data->op_suppgids[1];
259 rec->cr_bias = op_data->op_bias;
260 rec->cr_old_handle = op_data->op_handle;
262 if (op_data->op_name) {
263 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
264 op_data->op_namelen);
266 if (op_data->op_bias & MDS_CREATE_VOLATILE)
267 cr_flags |= MDS_OPEN_VOLATILE;
270 mdc_file_secctx_pack(req, op_data->op_file_secctx_name,
271 op_data->op_file_secctx,
272 op_data->op_file_secctx_size);
276 cr_flags |= MDS_OPEN_HAS_EA;
277 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
278 memcpy(tmp, lmm, lmmlen);
280 set_mrc_cr_flags(rec, cr_flags);
283 static inline __u64 attr_pack(unsigned int ia_valid) {
286 if (ia_valid & ATTR_MODE)
287 sa_valid |= MDS_ATTR_MODE;
288 if (ia_valid & ATTR_UID)
289 sa_valid |= MDS_ATTR_UID;
290 if (ia_valid & ATTR_GID)
291 sa_valid |= MDS_ATTR_GID;
292 if (ia_valid & ATTR_SIZE)
293 sa_valid |= MDS_ATTR_SIZE;
294 if (ia_valid & ATTR_ATIME)
295 sa_valid |= MDS_ATTR_ATIME;
296 if (ia_valid & ATTR_MTIME)
297 sa_valid |= MDS_ATTR_MTIME;
298 if (ia_valid & ATTR_CTIME)
299 sa_valid |= MDS_ATTR_CTIME;
300 if (ia_valid & ATTR_ATIME_SET)
301 sa_valid |= MDS_ATTR_ATIME_SET;
302 if (ia_valid & ATTR_MTIME_SET)
303 sa_valid |= MDS_ATTR_MTIME_SET;
304 if (ia_valid & ATTR_FORCE)
305 sa_valid |= MDS_ATTR_FORCE;
306 if (ia_valid & ATTR_ATTR_FLAG)
307 sa_valid |= MDS_ATTR_ATTR_FLAG;
308 if (ia_valid & ATTR_KILL_SUID)
309 sa_valid |= MDS_ATTR_KILL_SUID;
310 if (ia_valid & ATTR_KILL_SGID)
311 sa_valid |= MDS_ATTR_KILL_SGID;
312 if (ia_valid & ATTR_CTIME_SET)
313 sa_valid |= MDS_ATTR_CTIME_SET;
314 if (ia_valid & ATTR_FROM_OPEN)
315 sa_valid |= MDS_ATTR_FROM_OPEN;
316 if (ia_valid & ATTR_BLOCKS)
317 sa_valid |= MDS_ATTR_BLOCKS;
318 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
319 /* NFSD hack (see bug 5781) */
320 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
321 if (ia_valid & MDS_ATTR_PROJID)
322 sa_valid |= MDS_ATTR_PROJID;
326 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
327 struct md_op_data *op_data)
329 rec->sa_opcode = REINT_SETATTR;
330 rec->sa_fsuid = from_kuid(&init_user_ns, current_fsuid());
331 rec->sa_fsgid = from_kgid(&init_user_ns, current_fsgid());
332 rec->sa_cap = cfs_curproc_cap_pack();
333 rec->sa_suppgid = -1;
335 rec->sa_fid = op_data->op_fid1;
336 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
337 rec->sa_mode = op_data->op_attr.ia_mode;
338 rec->sa_uid = from_kuid(&init_user_ns, op_data->op_attr.ia_uid);
339 rec->sa_gid = from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
340 rec->sa_projid = op_data->op_projid;
341 rec->sa_size = op_data->op_attr.ia_size;
342 rec->sa_blocks = op_data->op_attr_blocks;
343 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
344 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
345 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
346 rec->sa_attr_flags = op_data->op_attr_flags;
347 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
348 in_group_p(op_data->op_attr.ia_gid))
350 from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
352 rec->sa_suppgid = op_data->op_suppgids[0];
354 rec->sa_bias = op_data->op_bias;
357 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
358 struct md_op_data *op_data)
360 epoch->mio_handle = op_data->op_handle;
361 epoch->mio_unused1 = 0;
362 epoch->mio_unused2 = 0;
363 epoch->mio_padding = 0;
366 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
367 void *ea, size_t ealen)
369 struct mdt_rec_setattr *rec;
370 struct lov_user_md *lum = NULL;
372 CLASSERT(sizeof(struct mdt_rec_reint) ==sizeof(struct mdt_rec_setattr));
373 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
374 mdc_setattr_pack_rec(rec, op_data);
379 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
380 if (ea == NULL) { /* Remove LOV EA */
381 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
382 lum->lmm_stripe_size = 0;
383 lum->lmm_stripe_count = 0;
384 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
386 memcpy(lum, ea, ealen);
390 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
392 struct mdt_rec_unlink *rec;
394 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_unlink));
395 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
396 LASSERT(rec != NULL);
398 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
399 REINT_RMENTRY : REINT_UNLINK;
400 rec->ul_fsuid = op_data->op_fsuid;
401 rec->ul_fsgid = op_data->op_fsgid;
402 rec->ul_cap = op_data->op_cap;
403 rec->ul_mode = op_data->op_mode;
404 rec->ul_suppgid1= op_data->op_suppgids[0];
405 rec->ul_suppgid2= -1;
406 rec->ul_fid1 = op_data->op_fid1;
407 rec->ul_fid2 = op_data->op_fid2;
408 rec->ul_time = op_data->op_mod_time;
409 rec->ul_bias = op_data->op_bias;
411 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
414 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
416 struct mdt_rec_link *rec;
418 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_link));
419 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
420 LASSERT (rec != NULL);
422 rec->lk_opcode = REINT_LINK;
423 rec->lk_fsuid = op_data->op_fsuid;//current->fsuid;
424 rec->lk_fsgid = op_data->op_fsgid;//current->fsgid;
425 rec->lk_cap = op_data->op_cap;//current->cap_effective;
426 rec->lk_suppgid1 = op_data->op_suppgids[0];
427 rec->lk_suppgid2 = op_data->op_suppgids[1];
428 rec->lk_fid1 = op_data->op_fid1;
429 rec->lk_fid2 = op_data->op_fid2;
430 rec->lk_time = op_data->op_mod_time;
431 rec->lk_bias = op_data->op_bias;
433 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
436 static void mdc_intent_close_pack(struct ptlrpc_request *req,
437 struct md_op_data *op_data)
439 struct close_data *data;
440 struct ldlm_lock *lock;
441 enum mds_op_bias bias = op_data->op_bias;
443 if (!(bias & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP |
444 MDS_RENAME_MIGRATE)))
447 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
448 LASSERT(data != NULL);
450 lock = ldlm_handle2lock(&op_data->op_lease_handle);
452 data->cd_handle = lock->l_remote_handle;
455 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
457 data->cd_data_version = op_data->op_data_version;
458 data->cd_fid = op_data->op_fid2;
461 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
462 const char *old, size_t oldlen,
463 const char *new, size_t newlen)
465 struct mdt_rec_rename *rec;
467 CLASSERT(sizeof(struct mdt_rec_reint) == sizeof(struct mdt_rec_rename));
468 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
470 /* XXX do something about time, uid, gid */
471 rec->rn_opcode = op_data->op_cli_flags & CLI_MIGRATE ?
472 REINT_MIGRATE : REINT_RENAME;
473 rec->rn_fsuid = op_data->op_fsuid;
474 rec->rn_fsgid = op_data->op_fsgid;
475 rec->rn_cap = op_data->op_cap;
476 rec->rn_suppgid1 = op_data->op_suppgids[0];
477 rec->rn_suppgid2 = op_data->op_suppgids[1];
478 rec->rn_fid1 = op_data->op_fid1;
479 rec->rn_fid2 = op_data->op_fid2;
480 rec->rn_time = op_data->op_mod_time;
481 rec->rn_mode = op_data->op_mode;
482 rec->rn_bias = op_data->op_bias;
484 mdc_pack_name(req, &RMF_NAME, old, oldlen);
487 mdc_pack_name(req, &RMF_SYMTGT, new, newlen);
489 if (op_data->op_cli_flags & CLI_MIGRATE &&
490 op_data->op_bias & MDS_RENAME_MIGRATE) {
491 struct mdt_ioepoch *epoch;
493 mdc_intent_close_pack(req, op_data);
494 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
495 mdc_ioepoch_pack(epoch, op_data);
499 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, __u32 flags,
500 struct md_op_data *op_data, size_t ea_size)
502 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
505 b->mbo_valid = valid;
506 if (op_data->op_bias & MDS_CHECK_SPLIT)
507 b->mbo_valid |= OBD_MD_FLCKSPLIT;
508 if (op_data->op_bias & MDS_CROSS_REF)
509 b->mbo_valid |= OBD_MD_FLCROSSREF;
510 b->mbo_eadatasize = ea_size;
511 b->mbo_flags = flags;
512 __mdc_pack_body(b, op_data->op_suppgids[0]);
514 b->mbo_fid1 = op_data->op_fid1;
515 b->mbo_fid2 = op_data->op_fid2;
516 b->mbo_valid |= OBD_MD_FLID;
518 if (op_data->op_name != NULL)
519 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
520 op_data->op_namelen);
523 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
525 struct mdt_ioepoch *epoch;
526 struct mdt_rec_setattr *rec;
528 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
529 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
531 mdc_setattr_pack_rec(rec, op_data);
533 * The client will zero out local timestamps when losing the IBITS lock
534 * so any new RPC timestamps will update the client inode's timestamps.
535 * There was a defect on the server side which allowed the atime to be
536 * overwritten by a zeroed-out atime packed into the close RPC.
538 * Proactively clear the MDS_ATTR_ATIME flag in the RPC in this case
539 * to avoid zeroing the atime on old unpatched servers. See LU-8041.
541 if (rec->sa_atime == 0)
542 rec->sa_valid &= ~MDS_ATTR_ATIME;
544 mdc_ioepoch_pack(epoch, op_data);
545 mdc_intent_close_pack(req, op_data);