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 lmv_stripe_md *lsm,
156 ldlm_blocking_callback cb_blocking,
157 int extra_lock_flags)
159 struct obd_device *obd = exp->exp_obd;
160 struct lmv_obd *lmv = &obd->u.lmv;
161 struct ptlrpc_request *req = NULL;
162 struct mdt_body *body;
163 struct md_op_data *op_data;
165 int valid_stripe_count = 0;
171 * revalidate slaves has some problems, temporarily return,
172 * we may not need that
174 OBD_ALLOC_PTR(op_data);
179 * Loop over the stripe information, check validity and update them
180 * from MDS if needed.
182 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
184 struct lookup_intent it = { .it_op = IT_GETATTR };
185 struct lustre_handle *lockh = NULL;
186 struct lmv_tgt_desc *tgt = NULL;
189 fid = lsm->lsm_md_oinfo[i].lmo_fid;
190 inode = lsm->lsm_md_oinfo[i].lmo_root;
196 * Prepare op_data for revalidating. Note that @fid2 shluld be
197 * defined otherwise it will go to server and take new lock
198 * which is not needed here.
200 memset(op_data, 0, sizeof(*op_data));
201 op_data->op_fid1 = fid;
202 op_data->op_fid2 = fid;
203 /* shard revalidate only needs to fetch attributes and UPDATE
204 * lock, which is similar to the bottom half of remote object
205 * getattr, set this flag so that MDT skips checking whether
206 * it's remote object.
208 op_data->op_bias = MDS_CROSS_REF;
210 tgt = lmv_tgt(lmv, lsm->lsm_md_oinfo[i].lmo_mds);
212 GOTO(cleanup, rc = -ENODEV);
214 CDEBUG(D_INODE, "Revalidate slave "DFID" -> mds #%u\n",
215 PFID(&fid), tgt->ltd_index);
218 ptlrpc_req_finished(req);
222 rc = md_intent_lock(tgt->ltd_exp, op_data, &it, &req,
223 cb_blocking, extra_lock_flags);
225 /* skip stripe is not exists */
233 lockh = (struct lustre_handle *)&it.it_lock_handle;
234 if (rc > 0 && req == NULL) {
235 /* slave inode is still valid */
236 CDEBUG(D_INODE, "slave "DFID" is still valid.\n",
240 /* refresh slave from server */
241 body = req_capsule_server_get(&req->rq_pill,
244 if (it.it_lock_mode && lockh) {
245 ldlm_lock_decref(lockh,
249 GOTO(cleanup, rc = -ENOENT);
252 i_size_write(inode, body->mbo_size);
253 inode->i_blocks = body->mbo_blocks;
254 set_nlink(inode, body->mbo_nlink);
255 inode->i_atime.tv_sec = body->mbo_atime;
256 inode->i_ctime.tv_sec = body->mbo_ctime;
257 inode->i_mtime.tv_sec = body->mbo_mtime;
260 md_set_lock_data(tgt->ltd_exp, lockh, inode, NULL);
261 if (it.it_lock_mode != 0 && lockh != NULL) {
262 ldlm_lock_decref(lockh, it.it_lock_mode);
266 valid_stripe_count++;
271 ptlrpc_req_finished(req);
273 /* if all stripes are invalid, return -ENOENT to notify user */
274 if (!rc && !valid_stripe_count)
277 OBD_FREE_PTR(op_data);
282 * IT_OPEN is intended to open (and create, possible) an object. Parent (pid)
285 static int lmv_intent_open(struct obd_export *exp, struct md_op_data *op_data,
286 struct lookup_intent *it,
287 struct ptlrpc_request **reqp,
288 ldlm_blocking_callback cb_blocking,
289 __u64 extra_lock_flags)
291 struct obd_device *obd = exp->exp_obd;
292 struct lmv_obd *lmv = &obd->u.lmv;
293 struct lmv_tgt_desc *tgt;
294 struct mdt_body *body;
295 __u64 flags = it->it_flags;
300 /* do not allow file creation in foreign dir */
301 if ((it->it_op & IT_CREAT) && lmv_dir_foreign(op_data->op_mea1))
304 if ((it->it_op & IT_CREAT) && !(flags & MDS_OPEN_BY_FID)) {
305 /* don't allow create under dir with bad hash */
306 if (lmv_dir_bad_hash(op_data->op_mea1))
309 if (lmv_dir_layout_changing(op_data->op_mea1)) {
310 if (flags & O_EXCL) {
312 * open(O_CREAT | O_EXCL) needs to check
313 * existing name, which should be done on both
314 * old and new layout, check old layout on
317 rc = lmv_old_layout_lookup(lmv, op_data);
321 op_data->op_new_layout = true;
324 * open(O_CREAT) will be sent to MDT in old
325 * layout first, to avoid creating new file
326 * under old layout, clear O_CREAT.
328 it->it_flags &= ~O_CREAT;
334 if (it->it_flags & MDS_OPEN_BY_FID) {
335 LASSERT(fid_is_sane(&op_data->op_fid2));
337 /* for striped directory, we can't know parent stripe fid
338 * without name, but we can set it to child fid, and MDT
339 * will obtain it from linkea in open in such case. */
340 if (lmv_dir_striped(op_data->op_mea1))
341 op_data->op_fid1 = op_data->op_fid2;
343 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
345 RETURN(PTR_ERR(tgt));
347 op_data->op_mds = tgt->ltd_index;
349 LASSERT(fid_is_sane(&op_data->op_fid1));
350 LASSERT(it->it_flags & MDS_OPEN_PCC ||
351 fid_is_zero(&op_data->op_fid2));
352 LASSERT(op_data->op_name != NULL);
354 tgt = lmv_locate_tgt(lmv, op_data);
356 RETURN(PTR_ERR(tgt));
359 /* If it is ready to open the file by FID, do not need
360 * allocate FID at all, otherwise it will confuse MDT */
361 if ((it->it_op & IT_CREAT) && !(it->it_flags & MDS_OPEN_BY_FID ||
362 it->it_flags & MDS_OPEN_PCC)) {
364 * For lookup(IT_CREATE) cases allocate new fid and setup FLD
367 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
372 CDEBUG(D_INODE, "OPEN_INTENT with fid1="DFID", fid2="DFID","
373 " name='%s' -> mds #%u\n", PFID(&op_data->op_fid1),
374 PFID(&op_data->op_fid2), op_data->op_name, tgt->ltd_index);
376 rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp, cb_blocking,
381 * Nothing is found, do not access body->fid1 as it is zero and thus
384 if ((it->it_disposition & DISP_LOOKUP_NEG) &&
385 !(it->it_disposition & DISP_OPEN_CREATE) &&
386 !(it->it_disposition & DISP_OPEN_OPEN)) {
387 if (!(it->it_flags & MDS_OPEN_BY_FID) &&
388 lmv_dir_retry_check_update(op_data)) {
389 ptlrpc_req_finished(*reqp);
390 it->it_request = NULL;
391 it->it_disposition = 0;
394 it->it_flags = flags;
395 fid_zero(&op_data->op_fid2);
402 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
406 /* Not cross-ref case, just get out of here. */
407 if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
408 rc = lmv_intent_remote(exp, it, &op_data->op_fid1, reqp,
409 cb_blocking, extra_lock_flags,
410 op_data->op_file_secctx_name,
411 op_data->op_file_secctx_name_size);
415 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
424 * Handler for: getattr, lookup and revalidate cases.
427 lmv_intent_lookup(struct obd_export *exp, struct md_op_data *op_data,
428 struct lookup_intent *it, struct ptlrpc_request **reqp,
429 ldlm_blocking_callback cb_blocking,
430 __u64 extra_lock_flags)
432 struct obd_device *obd = exp->exp_obd;
433 struct lmv_obd *lmv = &obd->u.lmv;
434 struct lmv_tgt_desc *tgt = NULL;
435 struct mdt_body *body;
439 /* foreign dir is not striped */
440 if (lmv_dir_foreign(op_data->op_mea1)) {
441 /* only allow getattr/lookup for itself */
442 if (op_data->op_name != NULL)
448 if (op_data->op_flags & MF_GETATTR_BY_FID) {
449 /* getattr by FID, replace fid1 with stripe FID */
450 LASSERT(op_data->op_name);
451 tgt = lmv_locate_tgt(lmv, op_data);
453 RETURN(PTR_ERR(tgt));
455 /* name is used to locate stripe target, clear it here
456 * to avoid packing name in request, so that MDS knows
457 * it's getattr by FID.
459 op_data->op_name = NULL;
460 op_data->op_namelen = 0;
462 /* getattr request is sent to MDT where fid2 inode is */
463 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
464 } else if (op_data->op_name) {
465 /* getattr by name */
466 tgt = lmv_locate_tgt(lmv, op_data);
467 if (!fid_is_sane(&op_data->op_fid2))
468 fid_zero(&op_data->op_fid2);
470 /* old way to getattr by FID, parent FID not packed */
471 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
474 RETURN(PTR_ERR(tgt));
476 CDEBUG(D_INODE, "LOOKUP_INTENT with fid1="DFID", fid2="DFID
477 ", name='%s' -> mds #%u\n",
478 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
479 op_data->op_name ? op_data->op_name : "<NULL>",
482 op_data->op_bias &= ~MDS_CROSS_REF;
484 rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp, cb_blocking,
490 /* If RPC happens, lsm information will be revalidated
491 * during update_inode process (see ll_update_lsm_md) */
492 if (lmv_dir_striped(op_data->op_mea2)) {
493 rc = lmv_revalidate_slaves(exp, op_data->op_mea2,
500 } else if (it_disposition(it, DISP_LOOKUP_NEG) &&
501 lmv_dir_retry_check_update(op_data)) {
502 ptlrpc_req_finished(*reqp);
503 it->it_request = NULL;
504 it->it_disposition = 0;
510 if (!it_has_reply_body(it))
514 * MDS has returned success. Probably name has been resolved in
515 * remote inode. Let's check this.
517 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
521 /* Not cross-ref case, just get out of here. */
522 if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
523 rc = lmv_intent_remote(exp, it, NULL, reqp, cb_blocking,
525 op_data->op_file_secctx_name,
526 op_data->op_file_secctx_name_size);
529 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
537 int lmv_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
538 struct lookup_intent *it, struct ptlrpc_request **reqp,
539 ldlm_blocking_callback cb_blocking,
540 __u64 extra_lock_flags)
546 LASSERT(fid_is_sane(&op_data->op_fid1));
548 CDEBUG(D_INODE, "INTENT LOCK '%s' for "DFID" '%.*s' on "DFID"\n",
549 LL_IT2STR(it), PFID(&op_data->op_fid2),
550 (int)op_data->op_namelen, op_data->op_name,
551 PFID(&op_data->op_fid1));
553 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_LAYOUT | IT_GETXATTR))
554 rc = lmv_intent_lookup(exp, op_data, it, reqp, cb_blocking,
556 else if (it->it_op & IT_OPEN)
557 rc = lmv_intent_open(exp, op_data, it, reqp, cb_blocking,
563 struct lustre_handle lock_handle;
565 if (it->it_lock_mode != 0) {
566 lock_handle.cookie = it->it_lock_handle;
567 ldlm_lock_decref_and_cancel(&lock_handle,
571 it->it_lock_handle = 0;
572 it->it_lock_mode = 0;
574 if (it->it_remote_lock_mode != 0) {
575 lock_handle.cookie = it->it_remote_lock_handle;
576 ldlm_lock_decref_and_cancel(&lock_handle,
577 it->it_remote_lock_mode);
580 it->it_remote_lock_handle = 0;
581 it->it_remote_lock_mode = 0;