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)
59 struct obd_device *obd = exp->exp_obd;
60 struct lmv_obd *lmv = &obd->u.lmv;
61 struct ptlrpc_request *req = NULL;
62 struct lustre_handle plock;
63 struct md_op_data *op_data;
64 struct lmv_tgt_desc *tgt;
65 struct mdt_body *body;
70 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
74 LASSERT((body->mbo_valid & OBD_MD_MDS));
77 * We got LOOKUP lock, but we really need attrs.
79 pmode = it->it_lock_mode;
81 plock.cookie = it->it_lock_handle;
83 it->it_request = NULL;
86 LASSERT(fid_is_sane(&body->mbo_fid1));
88 tgt = lmv_find_target(lmv, &body->mbo_fid1);
90 GOTO(out, rc = PTR_ERR(tgt));
92 OBD_ALLOC_PTR(op_data);
94 GOTO(out, rc = -ENOMEM);
96 op_data->op_fid1 = body->mbo_fid1;
97 /* Sent the parent FID to the remote MDT */
98 if (parent_fid != NULL) {
99 /* The parent fid is only for remote open to
100 * check whether the open is from OBF,
101 * see mdt_cross_open */
102 LASSERT(it->it_op & IT_OPEN);
103 op_data->op_fid2 = *parent_fid;
106 op_data->op_bias = MDS_CROSS_REF;
107 CDEBUG(D_INODE, "REMOTE_INTENT with fid="DFID" -> mds #%u\n",
108 PFID(&body->mbo_fid1), tgt->ltd_idx);
110 rc = md_intent_lock(tgt->ltd_exp, op_data, it, &req, cb_blocking,
113 GOTO(out_free_op_data, rc);
116 * LLite needs LOOKUP lock to track dentry revocation in order to
117 * maintain dcache consistency. Thus drop UPDATE|PERM lock here
118 * and put LOOKUP in request.
120 if (it->it_lock_mode != 0) {
121 it->it_remote_lock_handle =
123 it->it_remote_lock_mode = it->it_lock_mode;
127 it->it_lock_handle = plock.cookie;
128 it->it_lock_mode = pmode;
133 OBD_FREE_PTR(op_data);
136 ldlm_lock_decref(&plock, pmode);
138 ptlrpc_req_finished(*reqp);
143 int lmv_revalidate_slaves(struct obd_export *exp,
144 const struct lmv_stripe_md *lsm,
145 ldlm_blocking_callback cb_blocking,
146 int extra_lock_flags)
148 struct obd_device *obd = exp->exp_obd;
149 struct lmv_obd *lmv = &obd->u.lmv;
150 struct ptlrpc_request *req = NULL;
151 struct mdt_body *body;
152 struct md_op_data *op_data;
154 int valid_stripe_count = 0;
160 * revalidate slaves has some problems, temporarily return,
161 * we may not need that
163 OBD_ALLOC_PTR(op_data);
168 * Loop over the stripe information, check validity and update them
169 * from MDS if needed.
171 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
173 struct lookup_intent it = { .it_op = IT_GETATTR };
174 struct lustre_handle *lockh = NULL;
175 struct lmv_tgt_desc *tgt = NULL;
178 fid = lsm->lsm_md_oinfo[i].lmo_fid;
179 inode = lsm->lsm_md_oinfo[i].lmo_root;
185 * Prepare op_data for revalidating. Note that @fid2 shluld be
186 * defined otherwise it will go to server and take new lock
187 * which is not needed here.
189 memset(op_data, 0, sizeof(*op_data));
190 op_data->op_fid1 = fid;
191 op_data->op_fid2 = fid;
192 /* shard revalidate only needs to fetch attributes and UPDATE
193 * lock, which is similar to the bottom half of remote object
194 * getattr, set this flag so that MDT skips checking whether
195 * it's remote object.
197 op_data->op_bias = MDS_CROSS_REF;
199 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
201 GOTO(cleanup, rc = PTR_ERR(tgt));
203 CDEBUG(D_INODE, "Revalidate slave "DFID" -> mds #%u\n",
204 PFID(&fid), tgt->ltd_idx);
207 ptlrpc_req_finished(req);
211 rc = md_intent_lock(tgt->ltd_exp, op_data, &it, &req,
212 cb_blocking, extra_lock_flags);
214 /* skip stripe is not exists */
222 lockh = (struct lustre_handle *)&it.it_lock_handle;
223 if (rc > 0 && req == NULL) {
224 /* slave inode is still valid */
225 CDEBUG(D_INODE, "slave "DFID" is still valid.\n",
229 /* refresh slave from server */
230 body = req_capsule_server_get(&req->rq_pill,
233 if (it.it_lock_mode && lockh) {
234 ldlm_lock_decref(lockh,
238 GOTO(cleanup, rc = -ENOENT);
241 i_size_write(inode, body->mbo_size);
242 inode->i_blocks = body->mbo_blocks;
243 set_nlink(inode, body->mbo_nlink);
244 inode->i_atime.tv_sec = body->mbo_atime;
245 inode->i_ctime.tv_sec = body->mbo_ctime;
246 inode->i_mtime.tv_sec = body->mbo_mtime;
249 md_set_lock_data(tgt->ltd_exp, lockh, inode, NULL);
250 if (it.it_lock_mode != 0 && lockh != NULL) {
251 ldlm_lock_decref(lockh, it.it_lock_mode);
255 valid_stripe_count++;
260 ptlrpc_req_finished(req);
262 /* if all stripes are invalid, return -ENOENT to notify user */
263 if (!rc && !valid_stripe_count)
266 OBD_FREE_PTR(op_data);
271 * IT_OPEN is intended to open (and create, possible) an object. Parent (pid)
274 static int lmv_intent_open(struct obd_export *exp, struct md_op_data *op_data,
275 struct lookup_intent *it,
276 struct ptlrpc_request **reqp,
277 ldlm_blocking_callback cb_blocking,
278 __u64 extra_lock_flags)
280 struct obd_device *obd = exp->exp_obd;
281 struct lmv_obd *lmv = &obd->u.lmv;
282 struct lmv_tgt_desc *tgt;
283 struct mdt_body *body;
284 __u64 flags = it->it_flags;
289 if ((it->it_op & IT_CREAT) && !(flags & MDS_OPEN_BY_FID)) {
290 /* don't allow create under dir with bad hash */
291 if (lmv_is_dir_bad_hash(op_data->op_mea1))
294 if (lmv_is_dir_migrating(op_data->op_mea1)) {
295 if (flags & O_EXCL) {
297 * open(O_CREAT | O_EXCL) needs to check
298 * existing name, which should be done on both
299 * old and new layout, to avoid creating new
300 * file under old layout, check old layout on
303 tgt = lmv_locate_tgt(lmv, op_data,
306 RETURN(PTR_ERR(tgt));
308 rc = md_getattr_name(tgt->ltd_exp, op_data,
311 ptlrpc_req_finished(*reqp);
319 op_data->op_post_migrate = 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 (op_data->op_mea1 != NULL)
339 op_data->op_fid1 = op_data->op_fid2;
341 tgt = lmv_find_target(lmv, &op_data->op_fid2);
343 RETURN(PTR_ERR(tgt));
345 op_data->op_mds = tgt->ltd_idx;
347 LASSERT(fid_is_sane(&op_data->op_fid1));
348 LASSERT(fid_is_zero(&op_data->op_fid2));
349 LASSERT(op_data->op_name != NULL);
351 tgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
353 RETURN(PTR_ERR(tgt));
356 /* If it is ready to open the file by FID, do not need
357 * allocate FID at all, otherwise it will confuse MDT */
358 if ((it->it_op & IT_CREAT) && !(it->it_flags & MDS_OPEN_BY_FID)) {
360 * For lookup(IT_CREATE) cases allocate new fid and setup FLD
363 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
368 CDEBUG(D_INODE, "OPEN_INTENT with fid1="DFID", fid2="DFID","
369 " name='%s' -> mds #%u\n", PFID(&op_data->op_fid1),
370 PFID(&op_data->op_fid2), op_data->op_name, tgt->ltd_idx);
372 rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp, cb_blocking,
377 * Nothing is found, do not access body->fid1 as it is zero and thus
380 if ((it->it_disposition & DISP_LOOKUP_NEG) &&
381 !(it->it_disposition & DISP_OPEN_CREATE) &&
382 !(it->it_disposition & DISP_OPEN_OPEN)) {
383 if (!(it->it_flags & MDS_OPEN_BY_FID) &&
384 lmv_dir_retry_check_update(op_data)) {
385 ptlrpc_req_finished(*reqp);
386 it->it_request = NULL;
387 it->it_disposition = 0;
390 it->it_flags = flags;
391 fid_zero(&op_data->op_fid2);
398 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
402 /* Not cross-ref case, just get out of here. */
403 if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
404 rc = lmv_intent_remote(exp, it, &op_data->op_fid1, reqp,
405 cb_blocking, extra_lock_flags);
409 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
418 * Handler for: getattr, lookup and revalidate cases.
421 lmv_intent_lookup(struct obd_export *exp, struct md_op_data *op_data,
422 struct lookup_intent *it, struct ptlrpc_request **reqp,
423 ldlm_blocking_callback cb_blocking,
424 __u64 extra_lock_flags)
426 struct obd_device *obd = exp->exp_obd;
427 struct lmv_obd *lmv = &obd->u.lmv;
428 struct lmv_tgt_desc *tgt = NULL;
429 struct mdt_body *body;
434 if (op_data->op_flags & MF_GETATTR_BY_FID) {
435 /* getattr by FID, replace fid1 with stripe FID */
436 LASSERT(op_data->op_name);
437 tgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
439 RETURN(PTR_ERR(tgt));
441 /* name is used to locate stripe target, clear it here
442 * to avoid packing name in request, so that MDS knows
443 * it's getattr by FID.
445 op_data->op_name = NULL;
446 op_data->op_namelen = 0;
448 /* getattr request is sent to MDT where fid2 inode is */
449 tgt = lmv_find_target(lmv, &op_data->op_fid2);
450 } else if (op_data->op_name) {
451 /* getattr by name */
452 tgt = lmv_locate_tgt(lmv, op_data, &op_data->op_fid1);
453 if (!fid_is_sane(&op_data->op_fid2))
454 fid_zero(&op_data->op_fid2);
456 /* old way to getattr by FID, parent FID not packed */
457 tgt = lmv_find_target(lmv, &op_data->op_fid1);
460 RETURN(PTR_ERR(tgt));
462 CDEBUG(D_INODE, "LOOKUP_INTENT with fid1="DFID", fid2="DFID
463 ", name='%s' -> mds #%u\n",
464 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
465 op_data->op_name ? op_data->op_name : "<NULL>",
468 op_data->op_bias &= ~MDS_CROSS_REF;
470 rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp, cb_blocking,
476 /* If RPC happens, lsm information will be revalidated
477 * during update_inode process (see ll_update_lsm_md) */
478 if (op_data->op_mea2 != NULL) {
479 rc = lmv_revalidate_slaves(exp, op_data->op_mea2,
486 } else if (it_disposition(it, DISP_LOOKUP_NEG) &&
487 lmv_dir_retry_check_update(op_data)) {
488 ptlrpc_req_finished(*reqp);
489 it->it_request = NULL;
490 it->it_disposition = 0;
496 if (!it_has_reply_body(it))
500 * MDS has returned success. Probably name has been resolved in
501 * remote inode. Let's check this.
503 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
507 /* Not cross-ref case, just get out of here. */
508 if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
509 rc = lmv_intent_remote(exp, it, NULL, reqp, cb_blocking,
513 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
521 int lmv_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
522 struct lookup_intent *it, struct ptlrpc_request **reqp,
523 ldlm_blocking_callback cb_blocking,
524 __u64 extra_lock_flags)
530 LASSERT(fid_is_sane(&op_data->op_fid1));
532 CDEBUG(D_INODE, "INTENT LOCK '%s' for "DFID" '%.*s' on "DFID"\n",
533 LL_IT2STR(it), PFID(&op_data->op_fid2),
534 (int)op_data->op_namelen, op_data->op_name,
535 PFID(&op_data->op_fid1));
537 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_LAYOUT | IT_GETXATTR))
538 rc = lmv_intent_lookup(exp, op_data, it, reqp, cb_blocking,
540 else if (it->it_op & IT_OPEN)
541 rc = lmv_intent_open(exp, op_data, it, reqp, cb_blocking,
547 struct lustre_handle lock_handle;
549 if (it->it_lock_mode != 0) {
550 lock_handle.cookie = it->it_lock_handle;
551 ldlm_lock_decref_and_cancel(&lock_handle,
555 it->it_lock_handle = 0;
556 it->it_lock_mode = 0;
558 if (it->it_remote_lock_mode != 0) {
559 lock_handle.cookie = it->it_remote_lock_handle;
560 ldlm_lock_decref_and_cancel(&lock_handle,
561 it->it_remote_lock_mode);
564 it->it_remote_lock_handle = 0;
565 it->it_remote_lock_mode = 0;