X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Fdcache.c;h=029e62120e2d0fd0d430779e62df13a3614419b4;hp=4e6c5a91777323b44084c1e71d11cf087a7af607;hb=ea766bd959cc2347b223d597b3f8a31e010bed72;hpb=5900da4464d90468519964571c441382cd512232 diff --git a/lustre/llite/dcache.c b/lustre/llite/dcache.c index 4e6c5a9..029e621 100644 --- a/lustre/llite/dcache.c +++ b/lustre/llite/dcache.c @@ -1,27 +1,41 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: +/* + * GPL HEADER START * - * Copyright (c) 2001-2003 Cluster File Systems, Inc. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * This file is part of Lustre, http://www.lustre.org. + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * Lustre is free software; you can redistribute it and/or - * modify it under the terms of version 2 of the GNU General Public - * License as published by the Free Software Foundation. + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * Lustre is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * - * You should have received a copy of the GNU General Public License - * along with Lustre; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + * GPL HEADER END + */ +/* + * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. + * Use is subject to license terms. + * + * Copyright (c) 2011, 2013, Intel Corporation. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. */ #include #include -#include #include #define DEBUG_SUBSYSTEM S_LLITE @@ -30,12 +44,17 @@ #include #include #include -#include -//#include -//#include #include "llite_internal.h" +static void free_dentry_data(struct rcu_head *head) +{ + struct ll_dentry_data *lld; + + lld = container_of(head, struct ll_dentry_data, lld_rcu_head); + OBD_FREE_PTR(lld); +} + /* should NOT be called with the dcache lock, see fs/dcache.c */ static void ll_release(struct dentry *de) { @@ -43,106 +62,226 @@ static void ll_release(struct dentry *de) ENTRY; LASSERT(de != NULL); lld = ll_d2d(de); - if (lld == NULL) { /* NFS copies the de->d_op methods (bug 4655) */ - EXIT; - return; - } -#ifndef LUSTRE_KERNEL_VERSION + if (lld == NULL) /* NFS copies the de->d_op methods (bug 4655) */ + RETURN_EXIT; + if (lld->lld_it) { ll_intent_release(lld->lld_it); OBD_FREE(lld->lld_it, sizeof(*lld->lld_it)); } -#endif - LASSERT(lld->lld_cwd_count == 0); - LASSERT(lld->lld_mnt_count == 0); - OBD_FREE(de->d_fsdata, sizeof(*lld)); + LASSERT(lld->lld_cwd_count == 0); + LASSERT(lld->lld_mnt_count == 0); + de->d_fsdata = NULL; + call_rcu(&lld->lld_rcu_head, free_dentry_data); - EXIT; + EXIT; } -#ifdef LUSTRE_KERNEL_VERSION /* Compare if two dentries are the same. Don't match if the existing dentry - * is marked DCACHE_LUSTRE_INVALID. Returns 1 if different, 0 if the same. + * is marked invalid. Returns 1 if different, 0 if the same. * * This avoids a race where ll_lookup_it() instantiates a dentry, but we get * an AST before calling d_revalidate_it(). The dentry still exists (marked * INVALID) so d_lookup() matches it, but we have no lock on it (so * lock_match() fails) and we spin around real_lookup(). */ +#ifdef HAVE_D_COMPARE_7ARGS +int ll_dcompare(const struct dentry *parent, const struct inode *pinode, + const struct dentry *dentry, const struct inode *inode, + unsigned int len, const char *str, const struct qstr *name) +#else int ll_dcompare(struct dentry *parent, struct qstr *d_name, struct qstr *name) +#endif { - struct dentry *dchild; - ENTRY; +#ifdef HAVE_D_COMPARE_7ARGS + ENTRY; - if (d_name->len != name->len) - RETURN(1); + if (len != name->len) + RETURN(1); - if (memcmp(d_name->name, name->name, name->len)) - RETURN(1); + if (memcmp(str, name->name, len)) + RETURN(1); +#else + struct dentry *dentry; + ENTRY; + + if (d_name->len != name->len) + RETURN(1); + + if (memcmp(d_name->name, name->name, name->len)) + RETURN(1); - /* XXX: d_name must be in-dentry structure */ - dchild = container_of(d_name, struct dentry, d_name); /* ugh */ - if (dchild->d_flags & DCACHE_LUSTRE_INVALID) { - CDEBUG(D_DENTRY,"INVALID dentry %p not matched, was bug 3784\n", - dchild); + /* XXX: d_name must be in-dentry structure */ + dentry = container_of(d_name, struct dentry, d_name); /* ugh */ +#endif + + CDEBUG(D_DENTRY, "found name %.*s(%p) flags %#x refc %d\n", + name->len, name->name, dentry, dentry->d_flags, + d_refcount(dentry)); + + /* mountpoint is always valid */ + if (d_mountpoint((struct dentry *)dentry)) + RETURN(0); + + if (d_lustre_invalid(dentry)) RETURN(1); - } RETURN(0); } -#endif -/* should NOT be called with the dcache lock, see fs/dcache.c */ -static int ll_ddelete(struct dentry *de) +static inline int return_if_equal(struct ldlm_lock *lock, void *data) +{ + if ((lock->l_flags & + (LDLM_FL_CANCELING | LDLM_FL_DISCARD_DATA)) == + (LDLM_FL_CANCELING | LDLM_FL_DISCARD_DATA)) + return LDLM_ITER_CONTINUE; + return LDLM_ITER_STOP; +} + +/* find any ldlm lock of the inode in mdc and lov + * return 0 not find + * 1 find one + * < 0 error */ +static int find_cbdata(struct inode *inode) { + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct lov_stripe_md *lsm; + int rc = 0; ENTRY; - LASSERT(de); -#ifndef DCACHE_LUSTRE_INVALID -#define DCACHE_LUSTRE_INVALID 0 + + LASSERT(inode); + rc = md_find_cbdata(sbi->ll_md_exp, ll_inode2fid(inode), + return_if_equal, NULL); + if (rc != 0) + RETURN(rc); + + lsm = ccc_inode_lsm_get(inode); + if (lsm == NULL) + RETURN(rc); + + rc = obd_find_cbdata(sbi->ll_dt_exp, lsm, return_if_equal, NULL); + ccc_inode_lsm_put(inode, lsm); + + RETURN(rc); +} + +/** + * Called when last reference to a dentry is dropped and dcache wants to know + * whether or not it should cache it: + * - return 1 to delete the dentry immediately + * - return 0 to cache the dentry + * Should NOT be called with the dcache lock, see fs/dcache.c + */ +static int ll_ddelete(HAVE_D_DELETE_CONST struct dentry *de) +{ + ENTRY; + LASSERT(de); + + CDEBUG(D_DENTRY, "%s dentry %.*s (%p, parent %p, inode %p) %s%s\n", + d_lustre_invalid((struct dentry *)de) ? "deleting" : "keeping", + de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode, + d_unhashed((struct dentry *)de) ? "" : "hashed,", + list_empty(&de->d_subdirs) ? "" : "subdirs"); + +#ifdef HAVE_DCACHE_LOCK + LASSERT(d_refcount(de) == 0); +#else + /* kernel >= 2.6.38 last refcount is decreased after this function. */ + LASSERT(d_refcount(de) == 1); #endif - CDEBUG(D_DENTRY, "%s dentry %.*s (%p, parent %p, inode %p) %s%s\n", - (de->d_flags & DCACHE_LUSTRE_INVALID ? "deleting" : "keeping"), - de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode, - d_unhashed(de) ? "" : "hashed,", - list_empty(&de->d_subdirs) ? "" : "subdirs"); -#if DCACHE_LUSTRE_INVALID == 0 -#undef DCACHE_LUSTRE_INVALID + /* Disable this piece of code temproarily because this is called + * inside dcache_lock so it's not appropriate to do lots of work + * here. ATTENTION: Before this piece of code enabling, LU-2487 must be + * resolved. */ +#if 0 + /* if not ldlm lock for this inode, set i_nlink to 0 so that + * this inode can be recycled later b=20433 */ + if (de->d_inode && !find_cbdata(de->d_inode)) + clear_nlink(de->d_inode); #endif - RETURN(0); + if (d_lustre_invalid((struct dentry *)de)) + RETURN(1); + RETURN(0); } -void ll_set_dd(struct dentry *de) +static int ll_set_dd(struct dentry *de) { - ENTRY; - LASSERT(de != NULL); + ENTRY; + LASSERT(de != NULL); + + CDEBUG(D_DENTRY, "ldd on dentry %.*s (%p) parent %p inode %p refc %d\n", + de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode, + d_refcount(de)); + + if (de->d_fsdata == NULL) { + struct ll_dentry_data *lld; + + OBD_ALLOC_PTR(lld); + if (likely(lld != NULL)) { + spin_lock(&de->d_lock); + if (likely(de->d_fsdata == NULL)) + de->d_fsdata = lld; + else + OBD_FREE_PTR(lld); + spin_unlock(&de->d_lock); + } else { + RETURN(-ENOMEM); + } + } + + RETURN(0); +} + +int ll_dops_init(struct dentry *de, int block, int init_sa) +{ + struct ll_dentry_data *lld = ll_d2d(de); + int rc = 0; - CDEBUG(D_DENTRY, "ldd on dentry %.*s (%p) parent %p inode %p refc %d\n", - de->d_name.len, de->d_name.name, de, de->d_parent, de->d_inode, - atomic_read(&de->d_count)); - lock_kernel(); - if (de->d_fsdata == NULL) { - OBD_ALLOC(de->d_fsdata, sizeof(struct ll_dentry_data)); + if (lld == NULL && block != 0) { + rc = ll_set_dd(de); + if (rc) + return rc; + + lld = ll_d2d(de); } - unlock_kernel(); - EXIT; + if (lld != NULL && init_sa != 0) + lld->lld_sa_generation = 0; + +#ifdef HAVE_DCACHE_LOCK + de->d_op = &ll_d_ops; +#else + /* kernel >= 2.6.38 d_op is set in d_alloc() */ + LASSERT(de->d_op == &ll_d_ops); +#endif + return rc; } void ll_intent_drop_lock(struct lookup_intent *it) { - struct lustre_handle *handle; - if (it->it_op && it->d.lustre.it_lock_mode) { - handle = (struct lustre_handle *)&it->d.lustre.it_lock_handle; + struct lustre_handle handle; + + handle.cookie = it->d.lustre.it_lock_handle; + CDEBUG(D_DLMTRACE, "releasing lock with cookie "LPX64 - " from it %p\n", handle->cookie, it); - ldlm_lock_decref(handle, it->d.lustre.it_lock_mode); + " from it %p\n", handle.cookie, it); + ldlm_lock_decref(&handle, it->d.lustre.it_lock_mode); /* bug 494: intent_release may be called multiple times, from * this thread and we don't want to double-decref this lock */ it->d.lustre.it_lock_mode = 0; - } + if (it->d.lustre.it_remote_lock_mode != 0) { + handle.cookie = it->d.lustre.it_remote_lock_handle; + + CDEBUG(D_DLMTRACE, "releasing remote lock with cookie" + LPX64" from it %p\n", handle.cookie, it); + ldlm_lock_decref(&handle, + it->d.lustre.it_remote_lock_mode); + it->d.lustre.it_remote_lock_mode = 0; + } + } } void ll_intent_release(struct lookup_intent *it) @@ -151,13 +290,9 @@ void ll_intent_release(struct lookup_intent *it) CDEBUG(D_INFO, "intent %p released\n", it); ll_intent_drop_lock(it); -#ifdef LUSTRE_KERNEL_VERSION - it->it_magic = 0; - it->it_op_release = 0; -#endif /* We are still holding extra reference on a request, need to free it */ - if (it_disposition(it, DISP_ENQ_OPEN_REF)) /* open req for llfile_open*/ - ptlrpc_req_finished(it->d.lustre.it_data); + if (it_disposition(it, DISP_ENQ_OPEN_REF)) + ptlrpc_req_finished(it->d.lustre.it_data); /* ll_file_open */ if (it_disposition(it, DISP_ENQ_CREATE_REF)) /* create rec */ ptlrpc_req_finished(it->d.lustre.it_data); if (it_disposition(it, DISP_ENQ_COMPLETE)) /* saved req from revalidate @@ -169,107 +304,40 @@ void ll_intent_release(struct lookup_intent *it) EXIT; } -/* Drop dentry if it is not used already, unhash otherwise. - Should be called with dcache lock held! - Returns: 1 if dentry was dropped, 0 if unhashed. */ -int ll_drop_dentry(struct dentry *dentry) +void ll_invalidate_aliases(struct inode *inode) { - lock_dentry(dentry); - if (atomic_read(&dentry->d_count) == 0) { - CDEBUG(D_DENTRY, "deleting dentry %.*s (%p) parent %p " - "inode %p\n", dentry->d_name.len, - dentry->d_name.name, dentry, dentry->d_parent, - dentry->d_inode); - dget_locked(dentry); - __d_drop(dentry); - unlock_dentry(dentry); - spin_unlock(&dcache_lock); - dput(dentry); - spin_lock(&dcache_lock); - return 1; - } - /* disconected dentry can not be find without lookup, because we - * not need his to unhash or mark invalid. */ - if (dentry->d_flags & DCACHE_DISCONNECTED) { - unlock_dentry(dentry); - RETURN (0); - } + struct dentry *dentry; + struct ll_d_hlist_node *p; + ENTRY; -#ifdef LUSTRE_KERNEL_VERSION - if (!(dentry->d_flags & DCACHE_LUSTRE_INVALID)) { -#else - if (!d_unhashed(dentry)) { -#endif - CDEBUG(D_DENTRY, "unhashing dentry %.*s (%p) parent %p " - "inode %p refc %d\n", dentry->d_name.len, - dentry->d_name.name, dentry, dentry->d_parent, - dentry->d_inode, atomic_read(&dentry->d_count)); - /* actually we don't unhash the dentry, rather just - * mark it inaccessible for to __d_lookup(). otherwise - * sys_getcwd() could return -ENOENT -bzzz */ -#ifdef LUSTRE_KERNEL_VERSION - dentry->d_flags |= DCACHE_LUSTRE_INVALID; -#endif - if (!dentry->d_inode || !S_ISDIR(dentry->d_inode->i_mode)) - __d_drop(dentry); + LASSERT(inode != NULL); - } - unlock_dentry(dentry); - return 0; -} + CDEBUG(D_INODE, "marking dentries for ino %lu/%u(%p) invalid\n", + inode->i_ino, inode->i_generation, inode); -void ll_unhash_aliases(struct inode *inode) -{ - struct list_head *tmp, *head; - ENTRY; - - if (inode == NULL) { - CERROR("unexpected NULL inode, tell phil\n"); - return; - } - - CDEBUG(D_INODE, "marking dentries for ino %lu/%u(%p) invalid\n", - inode->i_ino, inode->i_generation, inode); - - head = &inode->i_dentry; - spin_lock(&dcache_lock); -restart: - tmp = head; - while ((tmp = tmp->next) != head) { - struct dentry *dentry = list_entry(tmp, struct dentry, d_alias); - - CDEBUG(D_DENTRY, "dentry in drop %.*s (%p) parent %p " - "inode %p flags %d\n", dentry->d_name.len, - dentry->d_name.name, dentry, dentry->d_parent, - dentry->d_inode, dentry->d_flags); + ll_lock_dcache(inode); + ll_d_hlist_for_each_entry(dentry, p, &inode->i_dentry, d_alias) { + CDEBUG(D_DENTRY, "dentry in drop %.*s (%p) parent %p " + "inode %p flags %d\n", dentry->d_name.len, + dentry->d_name.name, dentry, dentry->d_parent, + dentry->d_inode, dentry->d_flags); if (dentry->d_name.len == 1 && dentry->d_name.name[0] == '/') { CERROR("called on root (?) dentry=%p, inode=%p " "ino=%lu\n", dentry, inode, inode->i_ino); lustre_dump_dentry(dentry, 1); libcfs_debug_dumpstack(NULL); - } else if (d_mountpoint(dentry)) { - /* For mountpoints we skip removal of the dentry - which happens solely because we have a lock on it - obtained when this dentry was not a mountpoint yet */ - CDEBUG(D_DENTRY, "Skippind mountpoint dentry removal " - "%.*s (%p) parent %p\n", - dentry->d_name.len, - dentry->d_name.name, - dentry, dentry->d_parent); - - continue; } - if (ll_drop_dentry(dentry)) - goto restart; - } - spin_unlock(&dcache_lock); + d_lustre_invalidate(dentry, 0); + } + ll_unlock_dcache(inode); + EXIT; } int ll_revalidate_it_finish(struct ptlrpc_request *request, - int offset, struct lookup_intent *it, + struct lookup_intent *it, struct dentry *de) { int rc = 0; @@ -278,11 +346,10 @@ int ll_revalidate_it_finish(struct ptlrpc_request *request, if (!request) RETURN(0); - if (it_disposition(it, DISP_LOOKUP_NEG)) + if (it_disposition(it, DISP_LOOKUP_NEG)) RETURN(-ENOENT); - rc = ll_prep_inode(&de->d_inode, - request, offset, NULL); + rc = ll_prep_inode(&de->d_inode, request, NULL, it); RETURN(rc); } @@ -298,8 +365,7 @@ void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry) CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n", inode, inode->i_ino, inode->i_generation); - md_set_lock_data(sbi->ll_md_exp, &it->d.lustre.it_lock_handle, - inode); + ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL); } /* drop lookup or getattr locks immediately */ @@ -314,37 +380,29 @@ void ll_lookup_finish_locks(struct lookup_intent *it, struct dentry *dentry) void ll_frob_intent(struct lookup_intent **itp, struct lookup_intent *deft) { struct lookup_intent *it = *itp; -#if defined(LUSTRE_KERNEL_VERSION) - if (it) { - LASSERTF(it->it_magic == INTENT_MAGIC, - "%p has bad intent magic: %x\n", - it, it->it_magic); - } -#endif if (!it || it->it_op == IT_GETXATTR) it = *itp = deft; -#ifdef LUSTRE_KERNEL_VERSION - it->it_op_release = ll_intent_release; -#endif } int ll_revalidate_it(struct dentry *de, int lookup_flags, struct lookup_intent *it) { - int rc; struct md_op_data *op_data; struct ptlrpc_request *req = NULL; struct lookup_intent lookup_it = { .it_op = IT_LOOKUP }; struct obd_export *exp; - struct inode *parent; + struct inode *parent = de->d_parent->d_inode; + int rc; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:name=%s,intent=%s\n", de->d_name.name, LL_IT2STR(it)); if (de->d_inode == NULL) { + __u64 ibits; + /* We can only use negative dentries if this is stat or lookup, for opens and stuff we do need to query server. */ /* If there is IT_CREAT in intent op set, then we must throw @@ -353,54 +411,43 @@ int ll_revalidate_it(struct dentry *de, int lookup_flags, if (it && (it->it_op & IT_CREAT)) RETURN(0); -#ifdef LUSTRE_KERNEL_VERSION - if (de->d_flags & DCACHE_LUSTRE_INVALID) + if (d_lustre_invalid(de)) RETURN(0); -#endif - rc = ll_have_md_lock(de->d_parent->d_inode, - MDS_INODELOCK_UPDATE); - - RETURN(rc); + ibits = MDS_INODELOCK_UPDATE; + rc = ll_have_md_lock(parent, &ibits, LCK_MINMODE); + GOTO(out_sa, rc); } - exp = ll_i2mdexp(de->d_inode); - /* Never execute intents for mount points. * Attributes will be fixed up in ll_inode_revalidate_it */ if (d_mountpoint(de)) - RETURN(1); + GOTO(out_sa, rc = 1); + + /* need to get attributes in case root got changed from other client */ + if (de == de->d_sb->s_root) { + rc = __ll_inode_revalidate_it(de, it, MDS_INODELOCK_LOOKUP); + if (rc == 0) + rc = 1; + GOTO(out_sa, rc); + } - /* Root of the lustre tree. Always valid. - * Attributes will be fixed up in ll_inode_revalidate_it */ - if (de == de->d_sb->s_root) - RETURN(1); + exp = ll_i2mdexp(de->d_inode); OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_REVALIDATE_PAUSE, 5); ll_frob_intent(&it, &lookup_it); LASSERT(it); - parent = de->d_parent->d_inode; - - if (it->it_op & IT_CREAT) { - op_data = ll_prep_md_op_data(NULL, parent, NULL, - de->d_name.name, de->d_name.len, - 0, LUSTRE_OPC_CREATE, NULL); - } else { - op_data = ll_prep_md_op_data(NULL, parent, de->d_inode, - de->d_name.name, de->d_name.len, - 0, LUSTRE_OPC_ANY, NULL); - } - if (IS_ERR(op_data)) - RETURN(PTR_ERR(op_data)); + if (it->it_op == IT_LOOKUP && !d_lustre_invalid(de)) + RETURN(1); + if (it->it_op == IT_OPEN) { + struct inode *inode = de->d_inode; + struct ll_inode_info *lli = ll_i2info(inode); + struct obd_client_handle **och_p; + __u64 *och_usecount; + __u64 ibits; - if ((it->it_op == IT_OPEN) && de->d_inode) { - struct inode *inode = de->d_inode; - struct ll_inode_info *lli = ll_i2info(inode); - struct obd_client_handle **och_p; - __u64 *och_usecount; - /* * We used to check for MDS_INODELOCK_OPEN here, but in fact * just having LOOKUP lock is enough to justify inode is the @@ -423,12 +470,14 @@ int ll_revalidate_it(struct dentry *de, int lookup_flags, och_usecount = &lli->lli_open_fd_read_count; } /* Check for the proper lock. */ - if (!ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP)) + ibits = MDS_INODELOCK_LOOKUP; + if (!ll_have_md_lock(inode, &ibits, LCK_MINMODE)) goto do_lock; - down(&lli->lli_och_sem); + mutex_lock(&lli->lli_och_mutex); if (*och_p) { /* Everything is open already, do nothing */ /*(*och_usecount)++; Do not let them steal our open handle from under us */ + SET_BUT_UNUSED(och_usecount); /* XXX The code above was my original idea, but in case we have the handle, but we cannot use it due to later checks (e.g. O_CREAT|O_EXCL flags set), nobody @@ -436,22 +485,37 @@ int ll_revalidate_it(struct dentry *de, int lookup_flags, hope the lock won't be invalidated in between. But if it would be, we'll reopen the open request to MDS later during file open path */ - up(&lli->lli_och_sem); - ll_finish_md_op_data(op_data); + mutex_unlock(&lli->lli_och_mutex); RETURN(1); } else { - up(&lli->lli_och_sem); + mutex_unlock(&lli->lli_och_mutex); } } + if (it->it_op == IT_GETATTR) { + rc = ll_statahead_enter(parent, &de, 0); + if (rc == 1) + goto mark; + else if (rc != -EAGAIN && rc != 0) + GOTO(out, rc = 0); + } + do_lock: - it->it_create_mode &= ~current->fs->umask; - it->it_flags |= O_CHECK_STALE; + op_data = ll_prep_md_op_data(NULL, parent, de->d_inode, + de->d_name.name, de->d_name.len, + 0, LUSTRE_OPC_ANY, NULL); + if (IS_ERR(op_data)) + RETURN(PTR_ERR(op_data)); + + if (!IS_POSIXACL(parent) || !exp_connect_umask(exp)) + it->it_create_mode &= ~cfs_curproc_umask(); + it->it_create_mode |= M_CHECK_STALE; rc = md_intent_lock(exp, op_data, NULL, 0, it, lookup_flags, &req, ll_md_blocking_ast, 0); - it->it_flags &= ~O_CHECK_STALE; + it->it_create_mode &= ~M_CHECK_STALE; ll_finish_md_op_data(op_data); + /* If req is NULL, then md_intent_lock only tried to do a lock match; * if all was well, it will return 1 if it found locks, 0 otherwise. */ if (req == NULL && rc >= 0) { @@ -469,29 +533,20 @@ do_lock: } revalidate_finish: - rc = ll_revalidate_it_finish(req, DLM_REPLY_REC_OFF, it, de); + rc = ll_revalidate_it_finish(req, it, de); if (rc != 0) { if (rc != -ESTALE && rc != -ENOENT) ll_intent_release(it); GOTO(out, rc = 0); } - if ((it->it_op & IT_OPEN) && de->d_inode && - !S_ISREG(de->d_inode->i_mode) && + if ((it->it_op & IT_OPEN) && de->d_inode && + !S_ISREG(de->d_inode->i_mode) && !S_ISDIR(de->d_inode->i_mode)) { ll_release_openhandle(de, it); } rc = 1; - /* unfortunately ll_intent_lock may cause a callback and revoke our - * dentry */ - spin_lock(&dcache_lock); - lock_dentry(de); - __d_drop(de); - unlock_dentry(de); - __d_rehash(de, 0); - spin_unlock(&dcache_lock); - out: /* We do not free request as it may be reused during following lookup * (see comment in mdc/mdc_locks.c::mdc_intent_lock()), request will @@ -500,28 +555,45 @@ out: if (req != NULL && !it_disposition(it, DISP_ENQ_COMPLETE)) ptlrpc_req_finished(req); if (rc == 0) { -#ifdef LUSTRE_KERNEL_VERSION - ll_unhash_aliases(de->d_inode); - /* done in ll_unhash_aliases() - dentry->d_flags |= DCACHE_LUSTRE_INVALID; */ -#else - /* We do not want d_invalidate to kill all child dentries too */ - d_drop(de); -#endif - } else { - CDEBUG(D_DENTRY, "revalidated dentry %.*s (%p) parent %p " - "inode %p refc %d\n", de->d_name.len, - de->d_name.name, de, de->d_parent, de->d_inode, - atomic_read(&de->d_count)); - ll_lookup_finish_locks(it, de); -#ifdef LUSTRE_KERNEL_VERSION - lock_dentry(de); - de->d_flags &= ~DCACHE_LUSTRE_INVALID; - unlock_dentry(de); -#endif - } + /* mdt may grant layout lock for the newly created file, so + * release the lock to avoid leaking */ + ll_intent_drop_lock(it); + ll_invalidate_aliases(de->d_inode); + } else { + __u64 bits = 0; + __u64 matched_bits = 0; + + CDEBUG(D_DENTRY, "revalidated dentry %.*s (%p) parent %p " + "inode %p refc %d\n", de->d_name.len, + de->d_name.name, de, de->d_parent, de->d_inode, + d_refcount(de)); + + ll_set_lock_data(exp, de->d_inode, it, &bits); + + /* Note: We have to match both LOOKUP and PERM lock + * here to make sure the dentry is valid and no one + * changing the permission. + * But if the client connects < 2.4 server, which will + * only grant LOOKUP lock, so we can only Match LOOKUP + * lock for old server */ + if (exp_connect_flags(ll_i2mdexp(de->d_inode)) && + OBD_CONNECT_LVB_TYPE) + matched_bits = + MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM; + else + matched_bits = MDS_INODELOCK_LOOKUP; + + if (((bits & matched_bits) == matched_bits) && + d_lustre_invalid(de)) + d_lustre_revalidate(de); + ll_lookup_finish_locks(it, de); + } + +mark: + if (it != NULL && it->it_op == IT_GETATTR && rc > 0) + ll_statahead_mark(parent, de); RETURN(rc); - + /* * This part is here to combat evil-evil race in real_lookup on 2.6 * kernels. The race details are: We enter do_lookup() looking for some @@ -545,7 +617,7 @@ do_lookup: ll_lookup_finish_locks(it, de); it = &lookup_it; } - + /* Do real lookup here. */ op_data = ll_prep_md_op_data(NULL, parent, NULL, de->d_name.name, de->d_name.len, 0, (it->it_op & IT_CREAT ? @@ -557,11 +629,10 @@ do_lookup: rc = md_intent_lock(exp, op_data, NULL, 0, it, 0, &req, ll_md_blocking_ast, 0); if (rc >= 0) { - struct mdt_body *mdt_body = lustre_msg_buf(req->rq_repmsg, - DLM_REPLY_REC_OFF, - sizeof(*mdt_body)); + struct mdt_body *mdt_body; struct lu_fid fid = {.f_seq = 0, .f_oid = 0, .f_ver = 0}; - + mdt_body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); + if (de->d_inode) fid = *ll_inode2fid(de->d_inode); @@ -575,129 +646,74 @@ do_lookup: } ll_finish_md_op_data(op_data); GOTO(out, rc = 0); -} -/*static*/ void ll_pin(struct dentry *de, struct vfsmount *mnt, int flag) -{ - struct inode *inode= de->d_inode; - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ll_dentry_data *ldd = ll_d2d(de); - struct obd_client_handle *handle; - struct obd_capa *oc; - int rc = 0; - ENTRY; - LASSERT(ldd); - - lock_kernel(); - /* Strictly speaking this introduces an additional race: the - * increments should wait until the rpc has returned. - * However, given that at present the function is void, this - * issue is moot. */ - if (flag == 1 && (++ldd->lld_mnt_count) > 1) { - unlock_kernel(); - EXIT; - return; - } - - if (flag == 0 && (++ldd->lld_cwd_count) > 1) { - unlock_kernel(); - EXIT; - return; - } - unlock_kernel(); - - handle = (flag) ? &ldd->lld_mnt_och : &ldd->lld_cwd_och; - oc = ll_mdscapa_get(inode); - rc = obd_pin(sbi->ll_md_exp, ll_inode2fid(inode), oc, handle, flag); - capa_put(oc); - if (rc) { - lock_kernel(); - memset(handle, 0, sizeof(*handle)); - if (flag == 0) - ldd->lld_cwd_count--; - else - ldd->lld_mnt_count--; - unlock_kernel(); - } - - EXIT; - return; +out_sa: + /* + * For rc == 1 case, should not return directly to prevent losing + * statahead windows; for rc == 0 case, the "lookup" will be done later. + */ + if (it != NULL && it->it_op == IT_GETATTR && rc == 1) + ll_statahead_enter(parent, &de, 1); + goto mark; } -/*static*/ void ll_unpin(struct dentry *de, struct vfsmount *mnt, int flag) +#ifdef HAVE_IOP_ATOMIC_OPEN +/* + * Always trust cached dentries. Update statahead window if necessary. + */ +int ll_revalidate_nd(struct dentry *dentry, unsigned int flags) { - struct ll_sb_info *sbi = ll_i2sbi(de->d_inode); - struct ll_dentry_data *ldd = ll_d2d(de); - struct obd_client_handle handle; - int count, rc = 0; - ENTRY; - LASSERT(ldd); - - lock_kernel(); - /* Strictly speaking this introduces an additional race: the - * increments should wait until the rpc has returned. - * However, given that at present the function is void, this - * issue is moot. */ - handle = (flag) ? ldd->lld_mnt_och : ldd->lld_cwd_och; - if (handle.och_magic != OBD_CLIENT_HANDLE_MAGIC) { - /* the "pin" failed */ - unlock_kernel(); - EXIT; - return; - } + struct inode *parent = dentry->d_parent->d_inode; + int unplug = 0; - if (flag) - count = --ldd->lld_mnt_count; - else - count = --ldd->lld_cwd_count; - unlock_kernel(); + ENTRY; + CDEBUG(D_VFSTRACE, "VFS Op:name=%s,flags=%u\n", + dentry->d_name.name, flags); - if (count != 0) { - EXIT; - return; - } - - rc = obd_unpin(sbi->ll_md_exp, &handle, flag); - EXIT; - return; -} - -#ifdef LUSTRE_KERNEL_VERSION -static int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd) -{ - int rc; - ENTRY; + if (!(flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE)) && + ll_need_statahead(parent, dentry) > 0) { + if (flags & LOOKUP_RCU) + RETURN(-ECHILD); - if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST)) - rc = ll_revalidate_it(dentry, nd->flags, &nd->intent); - else - rc = ll_revalidate_it(dentry, 0, NULL); + if (dentry->d_inode == NULL) + unplug = 1; + do_statahead_enter(parent, &dentry, unplug); + ll_statahead_mark(parent, dentry); + } - RETURN(rc); + RETURN(1); } -#else + +#else /* !HAVE_IOP_ATOMIC_OPEN */ int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd) { int rc; ENTRY; +#ifndef HAVE_DCACHE_LOCK + /* kernel >= 2.6.38 supports rcu-walk, but lustre doesn't. */ + if (nd->flags & LOOKUP_RCU) + return -ECHILD; +#endif + if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) { struct lookup_intent *it; + it = ll_convert_intent(&nd->intent.open, nd->flags); if (IS_ERR(it)) RETURN(0); - if (it->it_op == (IT_OPEN|IT_CREAT)) - if (nd->intent.open.flags & O_EXCL) { - CDEBUG(D_VFSTRACE, "create O_EXCL, returning 0\n"); - rc = 0; - goto out_it; - } + + if (it->it_op == (IT_OPEN|IT_CREAT) && + nd->intent.open.flags & O_EXCL) { + CDEBUG(D_VFSTRACE, "create O_EXCL, returning 0\n"); + rc = 0; + goto out_it; + } rc = ll_revalidate_it(dentry, nd->flags, it); if (rc && (nd->flags & LOOKUP_OPEN) && it_disposition(it, DISP_OPEN_OPEN)) {/*Open*/ -#ifdef HAVE_FILE_IN_STRUCT_INTENT // XXX Code duplication with ll_lookup_nd if (S_ISFIFO(dentry->d_inode->i_mode)) { // We cannot call open here as it would @@ -706,21 +722,13 @@ int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd) (struct ptlrpc_request *) it->d.lustre.it_data); } else { - struct file *filp; - - nd->intent.open.file->private_data = it; - filp = lookup_instantiate_filp(nd, dentry,NULL); -#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)) -/* 2.6.1[456] have a bug in open_namei() that forgets to check - * nd->intent.open.file for error, so we need to return it as lookup's result - * instead */ - if (IS_ERR(filp)) - rc = 0; -#endif + struct file *filp; + + nd->intent.open.file->private_data = it; + filp = lookup_instantiate_filp(nd, dentry,NULL); + if (IS_ERR(filp)) + rc = PTR_ERR(filp); } -#else - ll_release_openhandle(dentry, it); -#endif /* HAVE_FILE_IN_STRUCT_INTENT */ } if (!rc && (nd->flags & LOOKUP_CREATE) && it_disposition(it, DISP_OPEN_CREATE)) { @@ -732,7 +740,7 @@ int ll_revalidate_nd(struct dentry *dentry, struct nameidata *nd) ll_d2d(dentry)->lld_it = it; it = NULL; /* avoid freeing */ } - + out_it: if (it) { ll_intent_release(it); @@ -744,17 +752,20 @@ out_it: RETURN(rc); } -#endif +#endif /* HAVE_IOP_ATOMIC_OPEN */ + +void ll_d_iput(struct dentry *de, struct inode *inode) +{ + LASSERT(inode); + if (!find_cbdata(inode)) + clear_nlink(inode); + iput(inode); +} struct dentry_operations ll_d_ops = { .d_revalidate = ll_revalidate_nd, .d_release = ll_release, - .d_delete = ll_ddelete, -#ifdef LUSTRE_KERNEL_VERSION + .d_delete = ll_ddelete, + .d_iput = ll_d_iput, .d_compare = ll_dcompare, -#endif -#if 0 - .d_pin = ll_pin, - .d_unpin = ll_unpin, -#endif };