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) 2004, 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_LMV
34 #include <linux/slab.h>
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/slab.h>
38 #include <linux/pagemap.h>
39 #include <linux/math64.h>
40 #include <linux/seq_file.h>
41 #include <linux/namei.h>
42 #include <lustre_intent.h>
44 #include <obd_support.h>
45 #include <lustre_lib.h>
46 #include <lustre_net.h>
47 #include <lustre_dlm.h>
48 #include <lustre_mdc.h>
49 #include <obd_class.h>
50 #include <lprocfs_status.h>
51 #include "lmv_internal.h"
53 static int lmv_intent_remote(struct obd_export *exp, struct lookup_intent *it,
54 const struct lu_fid *parent_fid,
55 struct ptlrpc_request **reqp,
56 ldlm_blocking_callback cb_blocking,
57 __u64 extra_lock_flags,
58 const char *secctx_name, __u32 secctx_name_size)
60 struct obd_device *obd = exp->exp_obd;
61 struct lmv_obd *lmv = &obd->u.lmv;
62 struct ptlrpc_request *req = NULL;
63 struct lustre_handle plock;
64 struct md_op_data *op_data;
65 struct lmv_tgt_desc *tgt;
66 struct mdt_body *body;
71 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
75 LASSERT((body->mbo_valid & OBD_MD_MDS));
78 * We got LOOKUP lock, but we really need attrs.
80 pmode = it->it_lock_mode;
82 plock.cookie = it->it_lock_handle;
84 it->it_request = NULL;
87 LASSERT(fid_is_sane(&body->mbo_fid1));
89 tgt = lmv_fid2tgt(lmv, &body->mbo_fid1);
91 GOTO(out, rc = PTR_ERR(tgt));
93 OBD_ALLOC_PTR(op_data);
95 GOTO(out, rc = -ENOMEM);
97 op_data->op_fid1 = body->mbo_fid1;
98 /* Sent the parent FID to the remote MDT */
99 if (parent_fid != NULL) {
100 /* The parent fid is only for remote open to
101 * check whether the open is from OBF,
102 * see mdt_cross_open */
103 LASSERT(it->it_op & IT_OPEN);
104 op_data->op_fid2 = *parent_fid;
107 op_data->op_bias = MDS_CROSS_REF;
108 CDEBUG(D_INODE, "REMOTE_INTENT with fid="DFID" -> mds #%u\n",
109 PFID(&body->mbo_fid1), tgt->ltd_index);
111 /* ask for security context upon intent */
112 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN) &&
113 secctx_name_size != 0 && secctx_name != NULL) {
114 op_data->op_file_secctx_name = secctx_name;
115 op_data->op_file_secctx_name_size = secctx_name_size;
116 CDEBUG(D_SEC, "'%.*s' is security xattr to fetch for "
118 secctx_name_size, secctx_name, PFID(&body->mbo_fid1));
121 rc = md_intent_lock(tgt->ltd_exp, op_data, it, &req, cb_blocking,
124 GOTO(out_free_op_data, rc);
127 * LLite needs LOOKUP lock to track dentry revocation in order to
128 * maintain dcache consistency. Thus drop UPDATE|PERM lock here
129 * and put LOOKUP in request.
131 if (it->it_lock_mode != 0) {
132 it->it_remote_lock_handle =
134 it->it_remote_lock_mode = it->it_lock_mode;
138 it->it_lock_handle = plock.cookie;
139 it->it_lock_mode = pmode;
144 OBD_FREE_PTR(op_data);
147 ldlm_lock_decref(&plock, pmode);
149 ptlrpc_req_finished(*reqp);
154 int lmv_revalidate_slaves(struct obd_export *exp,
155 const struct lu_fid *pfid,
156 const struct lmv_stripe_md *lsm,
157 ldlm_blocking_callback cb_blocking,
158 int extra_lock_flags)
160 struct obd_device *obd = exp->exp_obd;
161 struct lmv_obd *lmv = &obd->u.lmv;
162 struct ptlrpc_request *req = NULL;
163 struct mdt_body *body;
164 struct md_op_data *op_data;
166 int valid_stripe_count = 0;
172 * revalidate slaves has some problems, temporarily return,
173 * we may not need that
175 OBD_ALLOC_PTR(op_data);
180 * Loop over the stripe information, check validity and update them
181 * from MDS if needed.
183 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
185 struct lookup_intent it = { .it_op = IT_GETATTR };
186 struct lustre_handle *lockh = NULL;
187 struct lmv_tgt_desc *tgt = NULL;
190 fid = lsm->lsm_md_oinfo[i].lmo_fid;
191 inode = lsm->lsm_md_oinfo[i].lmo_root;
197 * Prepare op_data for revalidating. Note that @fid2 shluld be
198 * defined otherwise it will go to server and take new lock
199 * which is not needed here.
201 memset(op_data, 0, sizeof(*op_data));
202 if (exp_connect_flags2(exp) & OBD_CONNECT2_GETATTR_PFID)
203 op_data->op_fid1 = *pfid;
205 op_data->op_fid1 = fid;
206 op_data->op_fid2 = fid;
208 tgt = lmv_tgt(lmv, lsm->lsm_md_oinfo[i].lmo_mds);
210 GOTO(cleanup, rc = -ENODEV);
212 CDEBUG(D_INODE, "Revalidate slave "DFID" -> mds #%u\n",
213 PFID(&fid), tgt->ltd_index);
216 ptlrpc_req_finished(req);
220 rc = md_intent_lock(tgt->ltd_exp, op_data, &it, &req,
221 cb_blocking, extra_lock_flags);
223 /* skip stripe is not exists */
231 lockh = (struct lustre_handle *)&it.it_lock_handle;
232 if (rc > 0 && req == NULL) {
233 /* slave inode is still valid */
234 CDEBUG(D_INODE, "slave "DFID" is still valid.\n",
238 /* refresh slave from server */
239 body = req_capsule_server_get(&req->rq_pill,
242 if (it.it_lock_mode && lockh) {
243 ldlm_lock_decref(lockh,
247 GOTO(cleanup, rc = -ENOENT);
250 i_size_write(inode, body->mbo_size);
251 inode->i_blocks = body->mbo_blocks;
252 set_nlink(inode, body->mbo_nlink);
253 inode->i_atime.tv_sec = body->mbo_atime;
254 inode->i_ctime.tv_sec = body->mbo_ctime;
255 inode->i_mtime.tv_sec = body->mbo_mtime;
258 md_set_lock_data(tgt->ltd_exp, lockh, inode, NULL);
259 if (it.it_lock_mode != 0 && lockh != NULL) {
260 ldlm_lock_decref(lockh, it.it_lock_mode);
264 valid_stripe_count++;
269 ptlrpc_req_finished(req);
271 /* if all stripes are invalid, return -ENOENT to notify user */
272 if (!rc && !valid_stripe_count)
275 OBD_FREE_PTR(op_data);
280 * IT_OPEN is intended to open (and create, possible) an object. Parent (pid)
283 static int lmv_intent_open(struct obd_export *exp, struct md_op_data *op_data,
284 struct lookup_intent *it,
285 struct ptlrpc_request **reqp,
286 ldlm_blocking_callback cb_blocking,
287 __u64 extra_lock_flags)
289 struct obd_device *obd = exp->exp_obd;
290 struct lmv_obd *lmv = &obd->u.lmv;
291 struct lmv_tgt_desc *tgt;
292 struct mdt_body *body;
293 __u64 flags = it->it_flags;
298 /* do not allow file creation in foreign dir */
299 if ((it->it_op & IT_CREAT) && lmv_dir_foreign(op_data->op_mea1))
302 if ((it->it_op & IT_CREAT) && !(flags & MDS_OPEN_BY_FID)) {
303 /* don't allow create under dir with bad hash */
304 if (lmv_dir_bad_hash(op_data->op_mea1))
307 if (lmv_dir_layout_changing(op_data->op_mea1)) {
308 if (flags & O_EXCL) {
310 * open(O_CREAT | O_EXCL) needs to check
311 * existing name, which should be done on both
312 * old and new layout, check old layout on
315 rc = lmv_old_layout_lookup(lmv, op_data);
319 op_data->op_new_layout = true;
322 * open(O_CREAT) will be sent to MDT in old
323 * layout first, to avoid creating new file
324 * under old layout, clear O_CREAT.
326 it->it_flags &= ~O_CREAT;
332 if (it->it_flags & MDS_OPEN_BY_FID) {
333 LASSERT(fid_is_sane(&op_data->op_fid2));
335 /* for striped directory, we can't know parent stripe fid
336 * without name, but we can set it to child fid, and MDT
337 * will obtain it from linkea in open in such case. */
338 if (lmv_dir_striped(op_data->op_mea1))
339 op_data->op_fid1 = op_data->op_fid2;
341 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
343 RETURN(PTR_ERR(tgt));
345 op_data->op_mds = tgt->ltd_index;
347 LASSERT(fid_is_sane(&op_data->op_fid1));
348 LASSERT(it->it_flags & MDS_OPEN_PCC ||
349 fid_is_zero(&op_data->op_fid2));
350 LASSERT(op_data->op_name != NULL);
352 tgt = lmv_locate_tgt(lmv, op_data);
354 RETURN(PTR_ERR(tgt));
357 /* If it is ready to open the file by FID, do not need
358 * allocate FID at all, otherwise it will confuse MDT */
359 if ((it->it_op & IT_CREAT) && !(it->it_flags & MDS_OPEN_BY_FID ||
360 it->it_flags & MDS_OPEN_PCC)) {
362 * For lookup(IT_CREATE) cases allocate new fid and setup FLD
365 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
370 CDEBUG(D_INODE, "OPEN_INTENT with fid1="DFID", fid2="DFID","
371 " name='%s' -> mds #%u\n", PFID(&op_data->op_fid1),
372 PFID(&op_data->op_fid2), op_data->op_name, tgt->ltd_index);
374 rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp, cb_blocking,
379 * Nothing is found, do not access body->fid1 as it is zero and thus
382 if ((it->it_disposition & DISP_LOOKUP_NEG) &&
383 !(it->it_disposition & DISP_OPEN_CREATE) &&
384 !(it->it_disposition & DISP_OPEN_OPEN)) {
385 if (!(it->it_flags & MDS_OPEN_BY_FID) &&
386 lmv_dir_retry_check_update(op_data)) {
387 ptlrpc_req_finished(*reqp);
388 it->it_request = NULL;
389 it->it_disposition = 0;
392 it->it_flags = flags;
393 fid_zero(&op_data->op_fid2);
400 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
404 /* Not cross-ref case, just get out of here. */
405 if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
406 rc = lmv_intent_remote(exp, it, &op_data->op_fid1, reqp,
407 cb_blocking, extra_lock_flags,
408 op_data->op_file_secctx_name,
409 op_data->op_file_secctx_name_size);
413 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
422 * Handler for: getattr, lookup and revalidate cases.
425 lmv_intent_lookup(struct obd_export *exp, struct md_op_data *op_data,
426 struct lookup_intent *it, struct ptlrpc_request **reqp,
427 ldlm_blocking_callback cb_blocking,
428 __u64 extra_lock_flags)
430 struct obd_device *obd = exp->exp_obd;
431 struct lmv_obd *lmv = &obd->u.lmv;
432 struct lmv_tgt_desc *tgt = NULL;
433 struct mdt_body *body;
437 /* foreign dir is not striped */
438 if (lmv_dir_foreign(op_data->op_mea1)) {
439 /* only allow getattr/lookup for itself */
440 if (op_data->op_name != NULL)
446 if (op_data->op_name) {
447 tgt = lmv_locate_tgt(lmv, op_data);
448 if (!fid_is_sane(&op_data->op_fid2))
449 fid_zero(&op_data->op_fid2);
450 } else if (fid_is_sane(&op_data->op_fid2)) {
451 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
453 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
456 RETURN(PTR_ERR(tgt));
458 CDEBUG(D_INODE, "LOOKUP_INTENT with fid1="DFID", fid2="DFID
459 ", name='%s' -> mds #%u\n",
460 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
461 op_data->op_name ? op_data->op_name : "<NULL>",
464 op_data->op_bias &= ~MDS_CROSS_REF;
466 rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp, cb_blocking,
472 /* If RPC happens, lsm information will be revalidated
473 * during update_inode process (see ll_update_lsm_md) */
474 if (lmv_dir_striped(op_data->op_mea2)) {
475 rc = lmv_revalidate_slaves(exp, &op_data->op_fid2,
483 } else if (it_disposition(it, DISP_LOOKUP_NEG) &&
484 lmv_dir_retry_check_update(op_data)) {
485 ptlrpc_req_finished(*reqp);
486 it->it_request = NULL;
487 it->it_disposition = 0;
493 if (!it_has_reply_body(it))
497 * MDS has returned success. Probably name has been resolved in
498 * remote inode. Let's check this.
500 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
504 /* Not cross-ref case, just get out of here. */
505 if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
506 rc = lmv_intent_remote(exp, it, NULL, reqp, cb_blocking,
508 op_data->op_file_secctx_name,
509 op_data->op_file_secctx_name_size);
512 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
520 int lmv_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
521 struct lookup_intent *it, struct ptlrpc_request **reqp,
522 ldlm_blocking_callback cb_blocking,
523 __u64 extra_lock_flags)
529 LASSERT(fid_is_sane(&op_data->op_fid1));
531 CDEBUG(D_INODE, "INTENT LOCK '%s' for "DFID" '%.*s' on "DFID"\n",
532 LL_IT2STR(it), PFID(&op_data->op_fid2),
533 (int)op_data->op_namelen, op_data->op_name,
534 PFID(&op_data->op_fid1));
536 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_LAYOUT | IT_GETXATTR))
537 rc = lmv_intent_lookup(exp, op_data, it, reqp, cb_blocking,
539 else if (it->it_op & IT_OPEN)
540 rc = lmv_intent_open(exp, op_data, it, reqp, cb_blocking,
546 struct lustre_handle lock_handle;
548 if (it->it_lock_mode != 0) {
549 lock_handle.cookie = it->it_lock_handle;
550 ldlm_lock_decref_and_cancel(&lock_handle,
554 it->it_lock_handle = 0;
555 it->it_lock_mode = 0;
557 if (it->it_remote_lock_mode != 0) {
558 lock_handle.cookie = it->it_remote_lock_handle;
559 ldlm_lock_decref_and_cancel(&lock_handle,
560 it->it_remote_lock_mode);
563 it->it_remote_lock_handle = 0;
564 it->it_remote_lock_mode = 0;