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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LMV
39 #include <linux/slab.h>
40 #include <linux/module.h>
41 #include <linux/init.h>
42 #include <linux/slab.h>
43 #include <linux/pagemap.h>
44 #include <linux/math64.h>
45 #include <linux/seq_file.h>
46 #include <linux/namei.h>
47 #include <linux/lustre_intent.h>
49 #include <liblustre.h>
52 #include <obd_support.h>
53 #include <lustre/lustre_idl.h>
54 #include <lustre_lib.h>
55 #include <lustre_net.h>
56 #include <lustre_dlm.h>
57 #include <lustre_mdc.h>
58 #include <obd_class.h>
59 #include <lprocfs_status.h>
60 #include "lmv_internal.h"
62 static int lmv_intent_remote(struct obd_export *exp, struct lookup_intent *it,
63 const struct lu_fid *parent_fid,
64 struct ptlrpc_request **reqp,
65 ldlm_blocking_callback cb_blocking,
66 __u64 extra_lock_flags)
68 struct obd_device *obd = exp->exp_obd;
69 struct lmv_obd *lmv = &obd->u.lmv;
70 struct ptlrpc_request *req = NULL;
71 struct lustre_handle plock;
72 struct md_op_data *op_data;
73 struct lmv_tgt_desc *tgt;
74 struct mdt_body *body;
79 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
83 LASSERT((body->mbo_valid & OBD_MD_MDS));
86 * Unfortunately, we have to lie to MDC/MDS to retrieve
87 * attributes llite needs and provideproper locking.
89 if (it->it_op & IT_LOOKUP)
90 it->it_op = IT_GETATTR;
93 * We got LOOKUP lock, but we really need attrs.
95 pmode = it->d.lustre.it_lock_mode;
97 plock.cookie = it->d.lustre.it_lock_handle;
98 it->d.lustre.it_lock_mode = 0;
99 it->d.lustre.it_data = NULL;
102 LASSERT(fid_is_sane(&body->mbo_fid1));
104 tgt = lmv_find_target(lmv, &body->mbo_fid1);
106 GOTO(out, rc = PTR_ERR(tgt));
108 OBD_ALLOC_PTR(op_data);
110 GOTO(out, rc = -ENOMEM);
112 op_data->op_fid1 = body->mbo_fid1;
113 /* Sent the parent FID to the remote MDT */
114 if (parent_fid != NULL) {
115 /* The parent fid is only for remote open to
116 * check whether the open is from OBF,
117 * see mdt_cross_open */
118 LASSERT(it->it_op & IT_OPEN);
119 op_data->op_fid2 = *parent_fid;
122 op_data->op_bias = MDS_CROSS_REF;
123 CDEBUG(D_INODE, "REMOTE_INTENT with fid="DFID" -> mds #%d\n",
124 PFID(&body->mbo_fid1), tgt->ltd_idx);
126 rc = md_intent_lock(tgt->ltd_exp, op_data, it, &req, cb_blocking,
129 GOTO(out_free_op_data, rc);
132 * LLite needs LOOKUP lock to track dentry revocation in order to
133 * maintain dcache consistency. Thus drop UPDATE|PERM lock here
134 * and put LOOKUP in request.
136 if (it->d.lustre.it_lock_mode != 0) {
137 it->d.lustre.it_remote_lock_handle =
138 it->d.lustre.it_lock_handle;
139 it->d.lustre.it_remote_lock_mode = it->d.lustre.it_lock_mode;
143 it->d.lustre.it_lock_handle = plock.cookie;
144 it->d.lustre.it_lock_mode = pmode;
149 OBD_FREE_PTR(op_data);
152 ldlm_lock_decref(&plock, pmode);
154 ptlrpc_req_finished(*reqp);
160 int lmv_revalidate_slaves(struct obd_export *exp, struct mdt_body *mbody,
161 struct lmv_stripe_md *lsm,
162 ldlm_blocking_callback cb_blocking,
163 int extra_lock_flags)
165 struct obd_device *obd = exp->exp_obd;
166 struct lmv_obd *lmv = &obd->u.lmv;
167 struct mdt_body *body;
168 struct md_op_data *op_data;
169 unsigned long size = 0;
170 unsigned long nlink = 0;
180 * revalidate slaves has some problems, temporarily return,
181 * we may not need that
183 OBD_ALLOC_PTR(op_data);
188 * Loop over the stripe information, check validity and update them
189 * from MDS if needed.
191 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
193 struct lookup_intent it = { .it_op = IT_GETATTR };
194 struct ptlrpc_request *req = NULL;
195 struct lustre_handle *lockh = NULL;
196 struct lmv_tgt_desc *tgt = NULL;
199 fid = lsm->lsm_md_oinfo[i].lmo_fid;
200 inode = lsm->lsm_md_oinfo[i].lmo_root;
203 * Prepare op_data for revalidating. Note that @fid2 shluld be
204 * defined otherwise it will go to server and take new lock
205 * which is not needed here.
207 memset(op_data, 0, sizeof(*op_data));
208 op_data->op_fid1 = fid;
209 op_data->op_fid2 = fid;
211 tgt = lmv_locate_mds(lmv, op_data, &fid);
213 GOTO(cleanup, rc = PTR_ERR(tgt));
215 CDEBUG(D_INODE, "Revalidate slave "DFID" -> mds #%d\n",
216 PFID(&fid), tgt->ltd_idx);
218 rc = md_intent_lock(tgt->ltd_exp, op_data, &it, &req,
219 cb_blocking, extra_lock_flags);
223 lockh = (struct lustre_handle *)&it.d.lustre.it_lock_handle;
224 if (rc > 0 && req == NULL) {
225 /* slave inode is still valid */
226 CDEBUG(D_INODE, "slave "DFID" is still valid.\n",
230 /* refresh slave from server */
231 body = req_capsule_server_get(&req->rq_pill,
233 LASSERT(body != NULL);
234 if (unlikely(body->mbo_nlink < 2)) {
235 CERROR("%s: nlink %d < 2 corrupt stripe %d "DFID
237 obd->obd_name, body->mbo_nlink, i,
238 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
239 PFID(&lsm->lsm_md_oinfo[0].lmo_fid));
242 ptlrpc_req_finished(req);
244 if (it.d.lustre.it_lock_mode && lockh) {
245 ldlm_lock_decref(lockh,
246 it.d.lustre.it_lock_mode);
247 it.d.lustre.it_lock_mode = 0;
250 GOTO(cleanup, rc = -EIO);
254 i_size_write(inode, body->mbo_size);
255 set_nlink(inode, body->mbo_nlink);
256 LTIME_S(inode->i_atime) = body->mbo_atime;
257 LTIME_S(inode->i_ctime) = body->mbo_ctime;
258 LTIME_S(inode->i_mtime) = body->mbo_mtime;
261 ptlrpc_req_finished(req);
264 md_set_lock_data(tgt->ltd_exp, &lockh->cookie, inode, NULL);
266 size += i_size_read(inode);
269 nlink += inode->i_nlink - 2;
271 nlink += inode->i_nlink;
273 atime = LTIME_S(inode->i_atime) > atime ?
274 LTIME_S(inode->i_atime) : atime;
275 ctime = LTIME_S(inode->i_ctime) > ctime ?
276 LTIME_S(inode->i_ctime) : ctime;
277 mtime = LTIME_S(inode->i_mtime) > mtime ?
278 LTIME_S(inode->i_mtime) : mtime;
280 if (it.d.lustre.it_lock_mode != 0 && lockh != NULL) {
281 ldlm_lock_decref(lockh, it.d.lustre.it_lock_mode);
282 it.d.lustre.it_lock_mode = 0;
285 CDEBUG(D_INODE, "i %d "DFID" size %llu, nlink %u, atime "
286 "%lu, mtime %lu, ctime %lu.\n", i, PFID(&fid),
287 i_size_read(inode), inode->i_nlink,
288 LTIME_S(inode->i_atime), LTIME_S(inode->i_mtime),
289 LTIME_S(inode->i_ctime));
293 * update attr of master request.
295 CDEBUG(D_INODE, "Return refreshed attrs: size = %lu nlink %lu atime "
296 LPU64 "ctime "LPU64" mtime "LPU64" for " DFID"\n", size, nlink,
297 atime, ctime, mtime, PFID(&lsm->lsm_md_oinfo[0].lmo_fid));
300 mbody->mbo_atime = atime;
301 mbody->mbo_ctime = ctime;
302 mbody->mbo_mtime = mtime;
305 OBD_FREE_PTR(op_data);
311 int lmv_revalidate_slaves(struct obd_export *exp, struct mdt_body *mbody,
312 struct lmv_stripe_md *lsm,
313 ldlm_blocking_callback cb_blocking,
314 int extra_lock_flags)
322 * IT_OPEN is intended to open (and create, possible) an object. Parent (pid)
325 int lmv_intent_open(struct obd_export *exp, struct md_op_data *op_data,
326 struct lookup_intent *it, struct ptlrpc_request **reqp,
327 ldlm_blocking_callback cb_blocking, __u64 extra_lock_flags)
329 struct obd_device *obd = exp->exp_obd;
330 struct lmv_obd *lmv = &obd->u.lmv;
331 struct lmv_tgt_desc *tgt;
332 struct mdt_body *body;
336 if (it->it_flags & MDS_OPEN_BY_FID) {
337 LASSERT(fid_is_sane(&op_data->op_fid2));
339 /* for striped directory, we can't know parent stripe fid
340 * without name, but we can set it to child fid, and MDT
341 * will obtain it from linkea in open in such case. */
342 if (op_data->op_mea1 != NULL)
343 op_data->op_fid1 = op_data->op_fid2;
345 tgt = lmv_find_target(lmv, &op_data->op_fid2);
347 RETURN(PTR_ERR(tgt));
349 op_data->op_mds = tgt->ltd_idx;
351 LASSERT(fid_is_sane(&op_data->op_fid1));
352 LASSERT(fid_is_zero(&op_data->op_fid2));
353 LASSERT(op_data->op_name != NULL);
355 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
357 RETURN(PTR_ERR(tgt));
360 /* If it is ready to open the file by FID, do not need
361 * allocate FID at all, otherwise it will confuse MDT */
362 if ((it->it_op & IT_CREAT) && !(it->it_flags & MDS_OPEN_BY_FID)) {
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 #%d\n", PFID(&op_data->op_fid1),
374 PFID(&op_data->op_fid2), op_data->op_name, tgt->ltd_idx);
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->d.lustre.it_disposition & DISP_LOOKUP_NEG) &&
385 !(it->d.lustre.it_disposition & DISP_OPEN_CREATE) &&
386 !(it->d.lustre.it_disposition & DISP_OPEN_OPEN))
389 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
393 /* Not cross-ref case, just get out of here. */
394 if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
395 rc = lmv_intent_remote(exp, it, &op_data->op_fid1, reqp,
396 cb_blocking, extra_lock_flags);
400 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
409 * Handler for: getattr, lookup and revalidate cases.
412 lmv_intent_lookup(struct obd_export *exp, struct md_op_data *op_data,
413 struct lookup_intent *it, struct ptlrpc_request **reqp,
414 ldlm_blocking_callback cb_blocking,
415 __u64 extra_lock_flags)
417 struct obd_device *obd = exp->exp_obd;
418 struct lmv_obd *lmv = &obd->u.lmv;
419 struct lmv_tgt_desc *tgt = NULL;
420 struct mdt_body *body;
421 struct lmv_stripe_md *lsm = op_data->op_mea1;
425 /* If it returns ERR_PTR(-EBADFD) then it is an unknown hash type
426 * it will try all stripes to locate the object */
427 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
428 if (IS_ERR(tgt) && (PTR_ERR(tgt) != -EBADFD))
429 RETURN(PTR_ERR(tgt));
431 /* Both migrating dir and unknown hash dir need to try
432 * all of sub-stripes */
433 if (lsm != NULL && !lmv_is_known_hash_type(lsm)) {
434 struct lmv_oinfo *oinfo;
436 oinfo = &lsm->lsm_md_oinfo[0];
438 op_data->op_fid1 = oinfo->lmo_fid;
439 op_data->op_mds = oinfo->lmo_mds;
440 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
442 RETURN(PTR_ERR(tgt));
445 if (!fid_is_sane(&op_data->op_fid2))
446 fid_zero(&op_data->op_fid2);
448 CDEBUG(D_INODE, "LOOKUP_INTENT with fid1="DFID", fid2="DFID
449 ", name='%s' -> mds #%d lsm=%p lsm_magic=%x\n",
450 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
451 op_data->op_name ? op_data->op_name : "<NULL>",
452 tgt->ltd_idx, lsm, lsm == NULL ? -1 : lsm->lsm_md_magic);
454 op_data->op_bias &= ~MDS_CROSS_REF;
456 rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp, cb_blocking,
462 /* If RPC happens, lsm information will be revalidated
463 * during update_inode process (see ll_update_lsm_md) */
464 if (op_data->op_mea2 != NULL) {
465 rc = lmv_revalidate_slaves(exp, NULL, op_data->op_mea2,
472 } else if (it_disposition(it, DISP_LOOKUP_NEG) && lsm != NULL &&
473 lmv_need_try_all_stripes(lsm)) {
474 /* For migrating and unknown hash type directory, it will
475 * try to target the entry on other stripes */
478 for (stripe_index = 1;
479 stripe_index < lsm->lsm_md_stripe_count &&
480 it_disposition(it, DISP_LOOKUP_NEG); stripe_index++) {
481 struct lmv_oinfo *oinfo;
483 /* release the previous request */
484 ptlrpc_req_finished(*reqp);
485 it->d.lustre.it_data = NULL;
488 oinfo = &lsm->lsm_md_oinfo[stripe_index];
489 tgt = lmv_find_target(lmv, &oinfo->lmo_fid);
491 RETURN(PTR_ERR(tgt));
493 CDEBUG(D_INODE, "Try other stripes " DFID"\n",
494 PFID(&oinfo->lmo_fid));
496 op_data->op_fid1 = oinfo->lmo_fid;
497 it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
498 rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp,
499 cb_blocking, extra_lock_flags);
506 * MDS has returned success. Probably name has been resolved in
507 * remote inode. Let's check this.
509 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
513 /* Not cross-ref case, just get out of here. */
514 if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
515 rc = lmv_intent_remote(exp, it, NULL, reqp, cb_blocking,
519 body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
527 int lmv_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
528 struct lookup_intent *it, struct ptlrpc_request **reqp,
529 ldlm_blocking_callback cb_blocking,
530 __u64 extra_lock_flags)
532 struct obd_device *obd = exp->exp_obd;
537 LASSERT(fid_is_sane(&op_data->op_fid1));
539 CDEBUG(D_INODE, "INTENT LOCK '%s' for "DFID" '%.*s' on "DFID"\n",
540 LL_IT2STR(it), PFID(&op_data->op_fid2), op_data->op_namelen,
541 op_data->op_name, PFID(&op_data->op_fid1));
543 rc = lmv_check_connect(obd);
547 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_LAYOUT))
548 rc = lmv_intent_lookup(exp, op_data, it, reqp, cb_blocking,
550 else if (it->it_op & IT_OPEN)
551 rc = lmv_intent_open(exp, op_data, it, reqp, cb_blocking,