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 <lustre_net.h>
35 #include <uapi/linux/lustre/lustre_idl.h>
36 #include "mdc_internal.h"
38 static void set_mrc_cr_flags(struct mdt_rec_create *mrc, u64 flags)
40 mrc->cr_flags_l = (u32)(flags & 0xFFFFFFFFUll);
41 mrc->cr_flags_h = (u32)(flags >> 32);
44 static void __mdc_pack_body(struct mdt_body *b, __u32 suppgid)
46 b->mbo_suppgid = suppgid;
47 b->mbo_uid = from_kuid(&init_user_ns, current_uid());
48 b->mbo_gid = from_kgid(&init_user_ns, current_gid());
49 b->mbo_fsuid = from_kuid(&init_user_ns, current_fsuid());
50 b->mbo_fsgid = from_kgid(&init_user_ns, current_fsgid());
51 b->mbo_capability = cfs_curproc_cap_pack();
54 void mdc_swap_layouts_pack(struct ptlrpc_request *req,
55 struct md_op_data *op_data)
57 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
60 __mdc_pack_body(b, op_data->op_suppgids[0]);
61 b->mbo_fid1 = op_data->op_fid1;
62 b->mbo_fid2 = op_data->op_fid2;
63 b->mbo_valid |= OBD_MD_FLID;
66 void mdc_pack_body(struct ptlrpc_request *req, const struct lu_fid *fid,
67 __u64 valid, size_t ea_size, __u32 suppgid, u32 flags)
69 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
72 b->mbo_eadatasize = ea_size;
74 __mdc_pack_body(b, suppgid);
77 b->mbo_valid |= OBD_MD_FLID;
82 * Pack a name (path component) into a request
84 * \param[in] req request
85 * \param[in] field request field (usually RMF_NAME)
86 * \param[in] name path component
87 * \param[in] name_len length of path component
89 * \a field must be present in \a req and of size \a name_len + 1.
91 * \a name must be '\0' terminated of length \a name_len and represent
92 * a single path component (not contain '/').
94 static void mdc_pack_name(struct ptlrpc_request *req,
95 const struct req_msg_field *field,
96 const char *name, size_t name_len)
102 buf = req_capsule_client_get(&req->rq_pill, field);
103 buf_size = req_capsule_get_size(&req->rq_pill, field, RCL_CLIENT);
105 LASSERT(name && name_len && buf && buf_size == name_len + 1);
107 cpy_len = strlcpy(buf, name, buf_size);
109 LASSERT(cpy_len == name_len && lu_name_is_valid_2(buf, cpy_len));
112 void mdc_readdir_pack(struct ptlrpc_request *req, __u64 pgoff, size_t size,
113 const struct lu_fid *fid)
115 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
118 b->mbo_valid |= OBD_MD_FLID;
119 b->mbo_size = pgoff; /* !! */
120 b->mbo_nlink = size; /* !! */
121 __mdc_pack_body(b, -1);
122 b->mbo_mode = LUDA_FID | LUDA_TYPE;
125 /* packing of MDS records */
126 void mdc_create_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
127 const void *data, size_t datalen, umode_t mode,
128 uid_t uid, gid_t gid, cfs_cap_t cap_effective, __u64 rdev)
130 struct mdt_rec_create *rec;
134 BUILD_BUG_ON(sizeof(struct mdt_rec_reint) != sizeof(struct mdt_rec_create));
135 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
137 rec->cr_opcode = REINT_CREATE;
140 rec->cr_cap = cap_effective;
141 rec->cr_fid1 = op_data->op_fid1;
142 rec->cr_fid2 = op_data->op_fid2;
145 rec->cr_time = op_data->op_mod_time;
146 rec->cr_suppgid1 = op_data->op_suppgids[0];
147 rec->cr_suppgid2 = op_data->op_suppgids[1];
149 if (op_data->op_bias & MDS_CREATE_VOLATILE)
150 flags |= MDS_OPEN_VOLATILE;
151 set_mrc_cr_flags(rec, flags);
152 rec->cr_bias = op_data->op_bias;
153 rec->cr_umask = current_umask();
155 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
157 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
158 memcpy(tmp, data, datalen);
162 static inline __u64 mds_pack_open_flags(__u64 flags)
164 __u64 cr_flags = (flags & (FMODE_READ | FMODE_WRITE |
165 MDS_OPEN_FL_INTERNAL));
167 cr_flags |= MDS_OPEN_CREAT;
169 cr_flags |= MDS_OPEN_EXCL;
171 cr_flags |= MDS_OPEN_TRUNC;
172 if (flags & O_APPEND)
173 cr_flags |= MDS_OPEN_APPEND;
175 cr_flags |= MDS_OPEN_SYNC;
176 if (flags & O_DIRECTORY)
177 cr_flags |= MDS_OPEN_DIRECTORY;
178 if (flags & __FMODE_EXEC)
179 cr_flags |= MDS_FMODE_EXEC;
180 if (cl_is_lov_delay_create(flags))
181 cr_flags |= MDS_OPEN_DELAY_CREATE;
183 if (flags & O_NONBLOCK)
184 cr_flags |= MDS_OPEN_NORESTORE;
189 /* packing of MDS records */
190 void mdc_open_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
191 umode_t mode, __u64 rdev, __u64 flags, const void *lmm,
194 struct mdt_rec_create *rec;
198 BUILD_BUG_ON(sizeof(struct mdt_rec_reint) != sizeof(struct mdt_rec_create));
199 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
201 /* XXX do something about time, uid, gid */
202 rec->cr_opcode = REINT_OPEN;
203 rec->cr_fsuid = from_kuid(&init_user_ns, current_fsuid());
204 rec->cr_fsgid = from_kgid(&init_user_ns, current_fsgid());
205 rec->cr_cap = cfs_curproc_cap_pack();
206 rec->cr_fid1 = op_data->op_fid1;
207 rec->cr_fid2 = op_data->op_fid2;
210 cr_flags = mds_pack_open_flags(flags);
212 rec->cr_time = op_data->op_mod_time;
213 rec->cr_suppgid1 = op_data->op_suppgids[0];
214 rec->cr_suppgid2 = op_data->op_suppgids[1];
215 rec->cr_bias = op_data->op_bias;
216 rec->cr_umask = current_umask();
217 rec->cr_old_handle = op_data->op_handle;
219 if (op_data->op_name) {
220 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
221 op_data->op_namelen);
223 if (op_data->op_bias & MDS_CREATE_VOLATILE)
224 cr_flags |= MDS_OPEN_VOLATILE;
228 cr_flags |= MDS_OPEN_HAS_EA;
229 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
230 memcpy(tmp, lmm, lmmlen);
232 set_mrc_cr_flags(rec, cr_flags);
235 static inline __u64 attr_pack(unsigned int ia_valid)
239 if (ia_valid & ATTR_MODE)
240 sa_valid |= MDS_ATTR_MODE;
241 if (ia_valid & ATTR_UID)
242 sa_valid |= MDS_ATTR_UID;
243 if (ia_valid & ATTR_GID)
244 sa_valid |= MDS_ATTR_GID;
245 if (ia_valid & ATTR_SIZE)
246 sa_valid |= MDS_ATTR_SIZE;
247 if (ia_valid & ATTR_ATIME)
248 sa_valid |= MDS_ATTR_ATIME;
249 if (ia_valid & ATTR_MTIME)
250 sa_valid |= MDS_ATTR_MTIME;
251 if (ia_valid & ATTR_CTIME)
252 sa_valid |= MDS_ATTR_CTIME;
253 if (ia_valid & ATTR_ATIME_SET)
254 sa_valid |= MDS_ATTR_ATIME_SET;
255 if (ia_valid & ATTR_MTIME_SET)
256 sa_valid |= MDS_ATTR_MTIME_SET;
257 if (ia_valid & ATTR_FORCE)
258 sa_valid |= MDS_ATTR_FORCE;
259 if (ia_valid & ATTR_ATTR_FLAG)
260 sa_valid |= MDS_ATTR_ATTR_FLAG;
261 if (ia_valid & ATTR_KILL_SUID)
262 sa_valid |= MDS_ATTR_KILL_SUID;
263 if (ia_valid & ATTR_KILL_SGID)
264 sa_valid |= MDS_ATTR_KILL_SGID;
265 if (ia_valid & ATTR_CTIME_SET)
266 sa_valid |= MDS_ATTR_CTIME_SET;
267 if (ia_valid & ATTR_OPEN)
268 sa_valid |= MDS_ATTR_FROM_OPEN;
269 if (ia_valid & ATTR_BLOCKS)
270 sa_valid |= MDS_ATTR_BLOCKS;
271 if (ia_valid & MDS_OPEN_OWNEROVERRIDE)
272 /* NFSD hack (see bug 5781) */
273 sa_valid |= MDS_OPEN_OWNEROVERRIDE;
277 static void mdc_setattr_pack_rec(struct mdt_rec_setattr *rec,
278 struct md_op_data *op_data)
280 rec->sa_opcode = REINT_SETATTR;
281 rec->sa_fsuid = from_kuid(&init_user_ns, current_fsuid());
282 rec->sa_fsgid = from_kgid(&init_user_ns, current_fsgid());
283 rec->sa_cap = cfs_curproc_cap_pack();
284 rec->sa_suppgid = -1;
286 rec->sa_fid = op_data->op_fid1;
287 rec->sa_valid = attr_pack(op_data->op_attr.ia_valid);
288 rec->sa_mode = op_data->op_attr.ia_mode;
289 rec->sa_uid = from_kuid(&init_user_ns, op_data->op_attr.ia_uid);
290 rec->sa_gid = from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
291 rec->sa_size = op_data->op_attr.ia_size;
292 rec->sa_blocks = op_data->op_attr_blocks;
293 rec->sa_atime = LTIME_S(op_data->op_attr.ia_atime);
294 rec->sa_mtime = LTIME_S(op_data->op_attr.ia_mtime);
295 rec->sa_ctime = LTIME_S(op_data->op_attr.ia_ctime);
296 rec->sa_attr_flags = op_data->op_attr_flags;
297 if ((op_data->op_attr.ia_valid & ATTR_GID) &&
298 in_group_p(op_data->op_attr.ia_gid))
300 from_kgid(&init_user_ns, op_data->op_attr.ia_gid);
302 rec->sa_suppgid = op_data->op_suppgids[0];
304 rec->sa_bias = op_data->op_bias;
307 static void mdc_ioepoch_pack(struct mdt_ioepoch *epoch,
308 struct md_op_data *op_data)
310 epoch->mio_handle = op_data->op_handle;
311 epoch->mio_unused1 = 0;
312 epoch->mio_unused2 = 0;
313 epoch->mio_padding = 0;
316 void mdc_setattr_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
317 void *ea, size_t ealen)
319 struct mdt_rec_setattr *rec;
320 struct lov_user_md *lum = NULL;
322 BUILD_BUG_ON(sizeof(struct mdt_rec_reint) !=
323 sizeof(struct mdt_rec_setattr));
324 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
325 mdc_setattr_pack_rec(rec, op_data);
330 lum = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
331 if (!ea) { /* Remove LOV EA */
332 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
333 lum->lmm_stripe_size = 0;
334 lum->lmm_stripe_count = 0;
335 lum->lmm_stripe_offset = (typeof(lum->lmm_stripe_offset))(-1);
337 memcpy(lum, ea, ealen);
341 void mdc_unlink_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
343 struct mdt_rec_unlink *rec;
345 BUILD_BUG_ON(sizeof(struct mdt_rec_reint) != sizeof(struct mdt_rec_unlink));
346 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
348 rec->ul_opcode = op_data->op_cli_flags & CLI_RM_ENTRY ?
349 REINT_RMENTRY : REINT_UNLINK;
350 rec->ul_fsuid = op_data->op_fsuid;
351 rec->ul_fsgid = op_data->op_fsgid;
352 rec->ul_cap = op_data->op_cap;
353 rec->ul_mode = op_data->op_mode;
354 rec->ul_suppgid1 = op_data->op_suppgids[0];
355 rec->ul_suppgid2 = -1;
356 rec->ul_fid1 = op_data->op_fid1;
357 rec->ul_fid2 = op_data->op_fid2;
358 rec->ul_time = op_data->op_mod_time;
359 rec->ul_bias = op_data->op_bias;
361 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
364 void mdc_link_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
366 struct mdt_rec_link *rec;
368 BUILD_BUG_ON(sizeof(struct mdt_rec_reint) != sizeof(struct mdt_rec_link));
369 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
371 rec->lk_opcode = REINT_LINK;
372 rec->lk_fsuid = op_data->op_fsuid; /* current->fsuid; */
373 rec->lk_fsgid = op_data->op_fsgid; /* current->fsgid; */
374 rec->lk_cap = op_data->op_cap; /* current->cap_effective; */
375 rec->lk_suppgid1 = op_data->op_suppgids[0];
376 rec->lk_suppgid2 = op_data->op_suppgids[1];
377 rec->lk_fid1 = op_data->op_fid1;
378 rec->lk_fid2 = op_data->op_fid2;
379 rec->lk_time = op_data->op_mod_time;
380 rec->lk_bias = op_data->op_bias;
382 mdc_pack_name(req, &RMF_NAME, op_data->op_name, op_data->op_namelen);
385 static void mdc_intent_close_pack(struct ptlrpc_request *req,
386 struct md_op_data *op_data)
388 enum mds_op_bias bias = op_data->op_bias;
389 struct close_data *data;
390 struct ldlm_lock *lock;
392 if (!(bias & (MDS_HSM_RELEASE | MDS_CLOSE_LAYOUT_SWAP |
393 MDS_RENAME_MIGRATE)))
396 data = req_capsule_client_get(&req->rq_pill, &RMF_CLOSE_DATA);
399 lock = ldlm_handle2lock(&op_data->op_lease_handle);
401 data->cd_handle = lock->l_remote_handle;
404 ldlm_cli_cancel(&op_data->op_lease_handle, LCF_LOCAL);
406 data->cd_data_version = op_data->op_data_version;
407 data->cd_fid = op_data->op_fid2;
410 void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
411 const char *old, size_t oldlen,
412 const char *new, size_t newlen)
414 struct mdt_rec_rename *rec;
416 BUILD_BUG_ON(sizeof(struct mdt_rec_reint) != sizeof(struct mdt_rec_rename));
417 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
419 /* XXX do something about time, uid, gid */
420 rec->rn_opcode = op_data->op_cli_flags & CLI_MIGRATE ?
421 REINT_MIGRATE : REINT_RENAME;
422 rec->rn_opcode = REINT_RENAME;
423 rec->rn_fsuid = op_data->op_fsuid;
424 rec->rn_fsgid = op_data->op_fsgid;
425 rec->rn_cap = op_data->op_cap;
426 rec->rn_suppgid1 = op_data->op_suppgids[0];
427 rec->rn_suppgid2 = op_data->op_suppgids[1];
428 rec->rn_fid1 = op_data->op_fid1;
429 rec->rn_fid2 = op_data->op_fid2;
430 rec->rn_time = op_data->op_mod_time;
431 rec->rn_mode = op_data->op_mode;
432 rec->rn_bias = op_data->op_bias;
434 mdc_pack_name(req, &RMF_NAME, old, oldlen);
437 mdc_pack_name(req, &RMF_SYMTGT, new, newlen);
439 if (op_data->op_cli_flags & CLI_MIGRATE &&
440 op_data->op_bias & MDS_RENAME_MIGRATE) {
441 struct mdt_ioepoch *epoch;
443 mdc_intent_close_pack(req, op_data);
444 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
445 mdc_ioepoch_pack(epoch, op_data);
449 void mdc_getattr_pack(struct ptlrpc_request *req, __u64 valid, u32 flags,
450 struct md_op_data *op_data, size_t ea_size)
452 struct mdt_body *b = req_capsule_client_get(&req->rq_pill,
455 b->mbo_valid = valid;
456 if (op_data->op_bias & MDS_CHECK_SPLIT)
457 b->mbo_valid |= OBD_MD_FLCKSPLIT;
458 if (op_data->op_bias & MDS_CROSS_REF)
459 b->mbo_valid |= OBD_MD_FLCROSSREF;
460 b->mbo_eadatasize = ea_size;
461 b->mbo_flags = flags;
462 __mdc_pack_body(b, op_data->op_suppgids[0]);
464 b->mbo_fid1 = op_data->op_fid1;
465 b->mbo_fid2 = op_data->op_fid2;
466 b->mbo_valid |= OBD_MD_FLID;
468 if (op_data->op_name)
469 mdc_pack_name(req, &RMF_NAME, op_data->op_name,
470 op_data->op_namelen);
473 void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data)
475 struct mdt_ioepoch *epoch;
476 struct mdt_rec_setattr *rec;
478 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
479 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
481 mdc_setattr_pack_rec(rec, op_data);
483 * The client will zero out local timestamps when losing the IBITS lock
484 * so any new RPC timestamps will update the client inode's timestamps.
485 * There was a defect on the server side which allowed the atime to be
486 * overwritten by a zeroed-out atime packed into the close RPC.
488 * Proactively clear the MDS_ATTR_ATIME flag in the RPC in this case
489 * to avoid zeroing the atime on old unpatched servers. See LU-8041.
491 if (rec->sa_atime == 0)
492 rec->sa_valid &= ~MDS_ATTR_ATIME;
494 mdc_ioepoch_pack(epoch, op_data);
495 mdc_intent_close_pack(req, op_data);