X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_object.c;h=d4eff75567bc10433cb07423867a5c0a6ad33b86;hp=f057b2e875c0383b740a9fa62b044f96db6baafd;hb=60a9fd37185edf57fc42bf6a5ad9b9c61fc10e14;hpb=6c8ed1d8edb0f6fe1d79172f7fa4a0fcf5053dfe diff --git a/lustre/mdd/mdd_object.c b/lustre/mdd/mdd_object.c index f057b2e..d4eff75 100644 --- a/lustre/mdd/mdd_object.c +++ b/lustre/mdd/mdd_object.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -40,33 +40,50 @@ * Author: Wang Di */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS #include -#include #include #include #include #include /* fid_be_cpu(), fid_cpu_to_be(). */ #include +#include -#include +#include #include #include #include "mdd_internal.h" -static struct lu_object_operations mdd_lu_obj_ops; +static const struct lu_object_operations mdd_lu_obj_ops; +extern cfs_mem_cache_t *mdd_object_kmem; + +static int mdd_xattr_get(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf, + const char *name); + +int mdd_data_get(const struct lu_env *env, struct mdd_object *obj, + void **data) +{ + if (mdd_object_exists(obj) == 0) { + CERROR("%s: object "DFID" not found: rc = -2\n", + mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj))); + return -ENOENT; + } + mdo_data_get(env, obj, data); + return 0; +} int mdd_la_get(const struct lu_env *env, struct mdd_object *obj, struct lu_attr *la, struct lustre_capa *capa) { - LASSERTF(mdd_object_exists(obj), "FID is "DFID"\n", - PFID(mdd_object_fid(obj))); + if (mdd_object_exists(obj) == 0) { + CERROR("%s: object "DFID" not found: rc = -2\n", + mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj))); + return -ENOENT; + } return mdo_attr_get(env, obj, la, capa); } @@ -81,6 +98,15 @@ static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags) obj->mod_flags |= IMMUTE_OBJ; } +struct mdd_thread_info *mdd_env_info(const struct lu_env *env) +{ + struct mdd_thread_info *info; + + info = lu_context_key_get(&env->le_ctx, &mdd_thread_key); + LASSERT(info != NULL); + return info; +} + struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len) { struct lu_buf *buf; @@ -91,46 +117,12 @@ struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len) return buf; } -struct llog_cookie *mdd_max_cookie_get(const struct lu_env *env, - struct mdd_device *mdd) -{ - struct mdd_thread_info *mti = mdd_env_info(env); - int max_cookie_size; - - max_cookie_size = mdd_lov_cookiesize(env, mdd); - if (unlikely(mti->mti_max_cookie_size < max_cookie_size)) { - if (mti->mti_max_cookie) - OBD_FREE(mti->mti_max_cookie, mti->mti_max_cookie_size); - mti->mti_max_cookie = NULL; - mti->mti_max_cookie_size = 0; - } - if (unlikely(mti->mti_max_cookie == NULL)) { - OBD_ALLOC(mti->mti_max_cookie, max_cookie_size); - if (unlikely(mti->mti_max_cookie != NULL)) - mti->mti_max_cookie_size = max_cookie_size; - } - return mti->mti_max_cookie; -} - -struct lov_mds_md *mdd_max_lmm_get(const struct lu_env *env, - struct mdd_device *mdd) +void mdd_buf_put(struct lu_buf *buf) { - struct mdd_thread_info *mti = mdd_env_info(env); - int max_lmm_size; - - max_lmm_size = mdd_lov_mdsize(env, mdd); - if (unlikely(mti->mti_max_lmm_size < max_lmm_size)) { - if (mti->mti_max_lmm) - OBD_FREE(mti->mti_max_lmm, mti->mti_max_lmm_size); - mti->mti_max_lmm = NULL; - mti->mti_max_lmm_size = 0; - } - if (unlikely(mti->mti_max_lmm == NULL)) { - OBD_ALLOC(mti->mti_max_lmm, max_lmm_size); - if (unlikely(mti->mti_max_lmm != NULL)) - mti->mti_max_lmm_size = max_lmm_size; - } - return mti->mti_max_lmm; + if (buf == NULL || buf->lb_buf == NULL) + return; + OBD_FREE_LARGE(buf->lb_buf, buf->lb_len); + *buf = LU_BUF_NULL; } const struct lu_buf *mdd_buf_get_const(const struct lu_env *env, @@ -144,13 +136,47 @@ const struct lu_buf *mdd_buf_get_const(const struct lu_env *env, return buf; } -struct mdd_thread_info *mdd_env_info(const struct lu_env *env) +struct lu_buf *mdd_buf_alloc(const struct lu_env *env, ssize_t len) { - struct mdd_thread_info *info; + struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf; + + if ((len > buf->lb_len) && (buf->lb_buf != NULL)) { + OBD_FREE_LARGE(buf->lb_buf, buf->lb_len); + *buf = LU_BUF_NULL; + } + if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) { + buf->lb_len = len; + OBD_ALLOC_LARGE(buf->lb_buf, buf->lb_len); + if (buf->lb_buf == NULL) + *buf = LU_BUF_NULL; + } + return buf; +} - info = lu_context_key_get(&env->le_ctx, &mdd_thread_key); - LASSERT(info != NULL); - return info; +/** Increase the size of the \a mti_big_buf. + * preserves old data in buffer + * old buffer remains unchanged on error + * \retval 0 or -ENOMEM + */ +int mdd_buf_grow(const struct lu_env *env, ssize_t len) +{ + struct lu_buf *oldbuf = &mdd_env_info(env)->mti_big_buf; + struct lu_buf buf; + + LASSERT(len >= oldbuf->lb_len); + OBD_ALLOC_LARGE(buf.lb_buf, len); + + if (buf.lb_buf == NULL) + return -ENOMEM; + + buf.lb_len = len; + memcpy(buf.lb_buf, oldbuf->lb_buf, oldbuf->lb_len); + + OBD_FREE_LARGE(oldbuf->lb_buf, oldbuf->lb_len); + + memcpy(oldbuf, &buf, sizeof(buf)); + + return 0; } struct lu_object *mdd_object_alloc(const struct lu_env *env, @@ -159,7 +185,7 @@ struct lu_object *mdd_object_alloc(const struct lu_env *env, { struct mdd_object *mdd_obj; - OBD_ALLOC_PTR(mdd_obj); + OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, CFS_ALLOC_IO); if (mdd_obj != NULL) { struct lu_object *o; @@ -175,20 +201,24 @@ struct lu_object *mdd_object_alloc(const struct lu_env *env, } } -static int mdd_object_init(const struct lu_env *env, struct lu_object *o) +static int mdd_object_init(const struct lu_env *env, struct lu_object *o, + const struct lu_object_conf *unused) { - struct mdd_device *d = lu2mdd_dev(o->lo_dev); - struct lu_object *below; + struct mdd_device *d = lu2mdd_dev(o->lo_dev); + struct mdd_object *mdd_obj = lu2mdd_obj(o); + struct lu_object *below; struct lu_device *under; ENTRY; - under = &d->mdd_child->dd_lu_dev; - below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under); - mdd_pdlock_init(lu2mdd_obj(o)); + mdd_obj->mod_cltime = 0; + under = &d->mdd_child->dd_lu_dev; + below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under); + mdd_pdlock_init(mdd_obj); if (below == NULL) - RETURN(-ENOMEM); + RETURN(-ENOMEM); lu_object_add(o, below); + RETURN(0); } @@ -203,240 +233,299 @@ static int mdd_object_start(const struct lu_env *env, struct lu_object *o) static void mdd_object_free(const struct lu_env *env, struct lu_object *o) { struct mdd_object *mdd = lu2mdd_obj(o); - + lu_object_fini(o); - OBD_FREE_PTR(mdd); + OBD_SLAB_FREE_PTR(mdd, mdd_object_kmem); } static int mdd_object_print(const struct lu_env *env, void *cookie, lu_printer_t p, const struct lu_object *o) { - return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p", o); + struct mdd_object *mdd = lu2mdd_obj((struct lu_object *)o); + return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p(open_count=%d, " + "valid=%x, cltime="LPU64", flags=%lx)", + mdd, mdd->mod_count, mdd->mod_valid, + mdd->mod_cltime, mdd->mod_flags); } -/* orphan handling is here */ -static void mdd_object_delete(const struct lu_env *env, - struct lu_object *o) -{ - struct mdd_object *mdd_obj = lu2mdd_obj(o); - struct thandle *handle = NULL; - ENTRY; - - if (lu2mdd_dev(o->lo_dev)->mdd_orphans == NULL) - return; - - if (mdd_obj->mod_flags & ORPHAN_OBJ) { - mdd_txn_param_build(env, lu2mdd_dev(o->lo_dev), - MDD_TXN_INDEX_DELETE_OP); - handle = mdd_trans_start(env, lu2mdd_dev(o->lo_dev)); - if (IS_ERR(handle)) - CERROR("Cannot get thandle\n"); - else { - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - /* let's remove obj from the orphan list */ - __mdd_orphan_del(env, mdd_obj, handle); - mdd_write_unlock(env, mdd_obj); - mdd_trans_stop(env, lu2mdd_dev(o->lo_dev), - 0, handle); - } - } -} - -static struct lu_object_operations mdd_lu_obj_ops = { - .loo_object_init = mdd_object_init, - .loo_object_start = mdd_object_start, - .loo_object_free = mdd_object_free, - .loo_object_print = mdd_object_print, - .loo_object_delete = mdd_object_delete +static const struct lu_object_operations mdd_lu_obj_ops = { + .loo_object_init = mdd_object_init, + .loo_object_start = mdd_object_start, + .loo_object_free = mdd_object_free, + .loo_object_print = mdd_object_print, }; struct mdd_object *mdd_object_find(const struct lu_env *env, struct mdd_device *d, const struct lu_fid *f) { - struct lu_object *o, *lo; - struct mdd_object *m; - ENTRY; - - o = lu_object_find(env, mdd2lu_dev(d)->ld_site, f); - if (IS_ERR(o)) - m = (struct mdd_object *)o; - else { - lo = lu_object_locate(o->lo_header, mdd2lu_dev(d)->ld_type); - /* remote object can't be located and should be put then */ - if (lo == NULL) - lu_object_put(env, o); - m = lu2mdd_obj(lo); - } - RETURN(m); -} - -int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj) -{ - struct lu_attr *la = &mdd_env_info(env)->mti_la; - int rc; - - ENTRY; - rc = mdd_la_get(env, obj, la, BYPASS_CAPA); - if (rc == 0) { - mdd_flags_xlate(obj, la->la_flags); - if (S_ISDIR(la->la_mode) && la->la_nlink == 1) - obj->mod_flags |= MNLINK_OBJ; - } - RETURN(rc); + return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f)); } -/* get only inode attributes */ -int mdd_iattr_get(const struct lu_env *env, struct mdd_object *mdd_obj, - struct md_attr *ma) +static int mdd_path2fid(const struct lu_env *env, struct mdd_device *mdd, + const char *path, struct lu_fid *fid) { + struct lu_buf *buf; + struct lu_fid *f = &mdd_env_info(env)->mti_fid; + struct mdd_object *obj; + struct lu_name *lname = &mdd_env_info(env)->mti_name; + char *name; int rc = 0; ENTRY; - if (ma->ma_valid & MA_INODE) - RETURN(0); - - rc = mdd_la_get(env, mdd_obj, &ma->ma_attr, - mdd_object_capa(env, mdd_obj)); - if (rc == 0) - ma->ma_valid |= MA_INODE; - RETURN(rc); -} + /* temp buffer for path element */ + buf = mdd_buf_alloc(env, PATH_MAX); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); -static int mdd_get_default_md(struct mdd_object *mdd_obj, - struct lov_mds_md *lmm, int *size) -{ - struct lov_desc *ldesc; - struct mdd_device *mdd = mdo2mdd(&mdd_obj->mod_obj); - ENTRY; + lname->ln_name = name = buf->lb_buf; + lname->ln_namelen = 0; + *f = mdd->mdd_root_fid; - ldesc = &mdd->mdd_obd_dev->u.mds.mds_lov_desc; - LASSERT(ldesc != NULL); + while(1) { + while (*path == '/') + path++; + if (*path == '\0') + break; + while (*path != '/' && *path != '\0') { + *name = *path; + path++; + name++; + lname->ln_namelen++; + } - if (!lmm) - RETURN(0); + *name = '\0'; + /* find obj corresponding to fid */ + obj = mdd_object_find(env, mdd, f); + if (obj == NULL) + GOTO(out, rc = -EREMOTE); + if (IS_ERR(obj)) + GOTO(out, rc = PTR_ERR(obj)); + /* get child fid from parent and name */ + rc = mdd_lookup(env, &obj->mod_obj, lname, f, NULL); + mdd_object_put(env, obj); + if (rc) + break; - lmm->lmm_magic = LOV_MAGIC_V1; - lmm->lmm_object_gr = LOV_OBJECT_GROUP_DEFAULT; - lmm->lmm_pattern = ldesc->ld_pattern; - lmm->lmm_stripe_size = ldesc->ld_default_stripe_size; - lmm->lmm_stripe_count = ldesc->ld_default_stripe_count; - *size = sizeof(struct lov_mds_md); + name = buf->lb_buf; + lname->ln_namelen = 0; + } - RETURN(sizeof(struct lov_mds_md)); + if (!rc) + *fid = *f; +out: + RETURN(rc); } -/* get lov EA only */ -static int __mdd_lmm_get(const struct lu_env *env, - struct mdd_object *mdd_obj, struct md_attr *ma) +/** The maximum depth that fid2path() will search. + * This is limited only because we want to store the fids for + * historical path lookup purposes. + */ +#define MAX_PATH_DEPTH 100 + +/** mdd_path() lookup structure. */ +struct path_lookup_info { + __u64 pli_recno; /**< history point */ + __u64 pli_currec; /**< current record */ + struct lu_fid pli_fid; + struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */ + struct mdd_object *pli_mdd_obj; + char *pli_path; /**< full path */ + int pli_pathlen; + int pli_linkno; /**< which hardlink to follow */ + int pli_fidcount; /**< number of \a pli_fids */ +}; + +static int mdd_path_current(const struct lu_env *env, + struct path_lookup_info *pli) { + struct mdd_device *mdd = mdo2mdd(&pli->pli_mdd_obj->mod_obj); + struct mdd_object *mdd_obj; + struct lu_buf *buf = NULL; + struct link_ea_header *leh; + struct link_ea_entry *lee; + struct lu_name *tmpname = &mdd_env_info(env)->mti_name; + struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid; + char *ptr; + int reclen; int rc; ENTRY; - if (ma->ma_valid & MA_LOV) - RETURN(0); + ptr = pli->pli_path + pli->pli_pathlen - 1; + *ptr = 0; + --ptr; + pli->pli_fidcount = 0; + pli->pli_fids[0] = *(struct lu_fid *)mdd_object_fid(pli->pli_mdd_obj); + + while (!mdd_is_root(mdd, &pli->pli_fids[pli->pli_fidcount])) { + mdd_obj = mdd_object_find(env, mdd, + &pli->pli_fids[pli->pli_fidcount]); + if (mdd_obj == NULL) + GOTO(out, rc = -EREMOTE); + if (IS_ERR(mdd_obj)) + GOTO(out, rc = PTR_ERR(mdd_obj)); + rc = lu_object_exists(&mdd_obj->mod_obj.mo_lu); + if (rc <= 0) { + mdd_object_put(env, mdd_obj); + if (rc == -1) + rc = -EREMOTE; + else if (rc == 0) + /* Do I need to error out here? */ + rc = -ENOENT; + GOTO(out, rc); + } - rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &ma->ma_lmm_size, - MDS_LOV_MD_NAME); + /* Get parent fid and object name */ + mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); + buf = mdd_links_get(env, mdd_obj); + mdd_read_unlock(env, mdd_obj); + mdd_object_put(env, mdd_obj); + if (IS_ERR(buf)) + GOTO(out, rc = PTR_ERR(buf)); + + leh = buf->lb_buf; + lee = (struct link_ea_entry *)(leh + 1); /* link #0 */ + mdd_lee_unpack(lee, &reclen, tmpname, tmpfid); + + /* If set, use link #linkno for path lookup, otherwise use + link #0. Only do this for the final path element. */ + if ((pli->pli_fidcount == 0) && + (pli->pli_linkno < leh->leh_reccount)) { + int count; + for (count = 0; count < pli->pli_linkno; count++) { + lee = (struct link_ea_entry *) + ((char *)lee + reclen); + mdd_lee_unpack(lee, &reclen, tmpname, tmpfid); + } + if (pli->pli_linkno < leh->leh_reccount - 1) + /* indicate to user there are more links */ + pli->pli_linkno++; + } - if (rc == 0 && (ma->ma_need & MA_LOV_DEF)) { - rc = mdd_get_default_md(mdd_obj, ma->ma_lmm, - &ma->ma_lmm_size); + /* Pack the name in the end of the buffer */ + ptr -= tmpname->ln_namelen; + if (ptr - 1 <= pli->pli_path) + GOTO(out, rc = -EOVERFLOW); + strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen); + *(--ptr) = '/'; + + /* Store the parent fid for historic lookup */ + if (++pli->pli_fidcount >= MAX_PATH_DEPTH) + GOTO(out, rc = -EOVERFLOW); + pli->pli_fids[pli->pli_fidcount] = *tmpfid; } - if (rc > 0) { - ma->ma_valid |= MA_LOV; - rc = 0; + /* Verify that our path hasn't changed since we started the lookup. + Record the current index, and verify the path resolves to the + same fid. If it does, then the path is correct as of this index. */ + spin_lock(&mdd->mdd_cl.mc_lock); + pli->pli_currec = mdd->mdd_cl.mc_index; + spin_unlock(&mdd->mdd_cl.mc_lock); + rc = mdd_path2fid(env, mdd, ptr, &pli->pli_fid); + if (rc) { + CDEBUG(D_INFO, "mdd_path2fid(%s) failed %d\n", ptr, rc); + GOTO (out, rc = -EAGAIN); } - RETURN(rc); -} + if (!lu_fid_eq(&pli->pli_fids[0], &pli->pli_fid)) { + CDEBUG(D_INFO, "mdd_path2fid(%s) found another FID o="DFID + " n="DFID"\n", ptr, PFID(&pli->pli_fids[0]), + PFID(&pli->pli_fid)); + GOTO(out, rc = -EAGAIN); + } + ptr++; /* skip leading / */ + memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr); -int mdd_lmm_get_locked(const struct lu_env *env, struct mdd_object *mdd_obj, - struct md_attr *ma) -{ - int rc; - ENTRY; + EXIT; +out: + if (buf && !IS_ERR(buf) && buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + mdd_buf_put(buf); - mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = __mdd_lmm_get(env, mdd_obj, ma); - mdd_read_unlock(env, mdd_obj); - RETURN(rc); + return rc; } -/* get lmv EA only*/ -static int __mdd_lmv_get(const struct lu_env *env, - struct mdd_object *mdd_obj, struct md_attr *ma) +static int mdd_path_historic(const struct lu_env *env, + struct path_lookup_info *pli) { - int rc; - ENTRY; - - if (ma->ma_valid & MA_LMV) - RETURN(0); - - rc = mdd_get_md(env, mdd_obj, ma->ma_lmv, &ma->ma_lmv_size, - MDS_LMV_MD_NAME); - if (rc > 0) { - ma->ma_valid |= MA_LMV; - rc = 0; - } - RETURN(rc); + return 0; } -static int mdd_attr_get_internal(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct md_attr *ma) +/* Returns the full path to this fid, as of changelog record recno. */ +static int mdd_path(const struct lu_env *env, struct md_object *obj, + char *path, int pathlen, __u64 *recno, int *linkno) { - int rc = 0; + struct path_lookup_info *pli; + int tries = 3; + int rc = -EAGAIN; ENTRY; - if (ma->ma_need & MA_INODE) - rc = mdd_iattr_get(env, mdd_obj, ma); + if (pathlen < 3) + RETURN(-EOVERFLOW); - if (rc == 0 && ma->ma_need & MA_LOV) { - if (S_ISREG(mdd_object_type(mdd_obj)) || - S_ISDIR(mdd_object_type(mdd_obj))) - rc = __mdd_lmm_get(env, mdd_obj, ma); - } - if (rc == 0 && ma->ma_need & MA_LMV) { - if (S_ISDIR(mdd_object_type(mdd_obj))) - rc = __mdd_lmv_get(env, mdd_obj, ma); + if (mdd_is_root(mdo2mdd(obj), mdd_object_fid(md2mdd_obj(obj)))) { + path[0] = '\0'; + RETURN(0); } -#ifdef CONFIG_FS_POSIX_ACL - if (rc == 0 && ma->ma_need & MA_ACL_DEF) { - if (S_ISDIR(mdd_object_type(mdd_obj))) - rc = mdd_def_acl_get(env, mdd_obj, ma); + + OBD_ALLOC_PTR(pli); + if (pli == NULL) + RETURN(-ENOMEM); + + pli->pli_mdd_obj = md2mdd_obj(obj); + pli->pli_recno = *recno; + pli->pli_path = path; + pli->pli_pathlen = pathlen; + pli->pli_linkno = *linkno; + + /* Retry multiple times in case file is being moved */ + while (tries-- && rc == -EAGAIN) + rc = mdd_path_current(env, pli); + + /* For historical path lookup, the current links may not have existed + * at "recno" time. We must switch over to earlier links/parents + * by using the changelog records. If the earlier parent doesn't + * exist, we must search back through the changelog to reconstruct + * its parents, then check if it exists, etc. + * We may ignore this problem for the initial implementation and + * state that an "original" hardlink must still exist for us to find + * historic path name. */ + if (pli->pli_recno != -1) { + rc = mdd_path_historic(env, pli); + } else { + *recno = pli->pli_currec; + /* Return next link index to caller */ + *linkno = pli->pli_linkno; } -#endif - CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64"\n", - rc, ma->ma_valid); - RETURN(rc); + + OBD_FREE_PTR(pli); + + RETURN (rc); } -int mdd_attr_get_internal_locked(const struct lu_env *env, - struct mdd_object *mdd_obj, struct md_attr *ma) +int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj) { + struct lu_attr *la = &mdd_env_info(env)->mti_la; int rc; - int needlock = ma->ma_need & (MA_LOV | MA_LMV | MA_ACL_DEF); - if (needlock) - mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = mdd_attr_get_internal(env, mdd_obj, ma); - if (needlock) - mdd_read_unlock(env, mdd_obj); - return rc; + ENTRY; + rc = mdd_la_get(env, obj, la, BYPASS_CAPA); + if (rc == 0) { + mdd_flags_xlate(obj, la->la_flags); + } + RETURN(rc); } /* * No permission check is needed. */ -static int mdd_attr_get(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) +int mdd_attr_get(const struct lu_env *env, struct md_object *obj, + struct md_attr *ma) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); - int rc; - + int rc; ENTRY; - rc = mdd_attr_get_internal_locked(env, mdd_obj, ma); + + return mdd_la_get(env, md2mdd_obj(obj), &ma->ma_attr, + mdd_object_capa(env, md2mdd_obj(obj))); RETURN(rc); } @@ -452,7 +541,11 @@ static int mdd_xattr_get(const struct lu_env *env, ENTRY; - LASSERT(mdd_object_exists(mdd_obj)); + if (mdd_object_exists(mdd_obj) == 0) { + CERROR("%s: object "DFID" not found: rc = -2\n", + mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj))); + return -ENOENT; + } mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); rc = mdo_xattr_get(env, mdd_obj, buf, name, @@ -475,7 +568,11 @@ static int mdd_readlink(const struct lu_env *env, struct md_object *obj, int rc; ENTRY; - LASSERT(mdd_object_exists(mdd_obj)); + if (mdd_object_exists(mdd_obj) == 0) { + CERROR("%s: object "DFID" not found: rc = -2\n", + mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj))); + return -ENOENT; + } next = mdd_object_child(mdd_obj); mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); @@ -503,29 +600,57 @@ static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj, RETURN(rc); } +int mdd_declare_object_create_internal(const struct lu_env *env, + struct mdd_object *p, + struct mdd_object *c, + struct lu_attr *attr, + struct thandle *handle, + const struct md_op_spec *spec) +{ + struct dt_object_format *dof = &mdd_env_info(env)->mti_dof; + const struct dt_index_features *feat = spec->sp_feat; + int rc; + ENTRY; + + if (feat != &dt_directory_features && feat != NULL) { + dof->dof_type = DFT_INDEX; + dof->u.dof_idx.di_feat = feat; + + } else { + dof->dof_type = dt_mode_to_dft(attr->la_mode); + if (dof->dof_type == DFT_REGULAR) { + dof->u.dof_reg.striped = + md_should_create(spec->sp_cr_flags); + if (spec->sp_cr_flags & MDS_OPEN_HAS_EA) + dof->u.dof_reg.striped = 0; + /* is this replay? */ + if (spec->no_create) + dof->u.dof_reg.striped = 0; + } + } + + rc = mdo_declare_create_obj(env, c, attr, NULL, dof, handle); + + RETURN(rc); +} + int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p, - struct mdd_object *c, struct md_attr *ma, - struct thandle *handle) + struct mdd_object *c, struct lu_attr *attr, + struct thandle *handle, + const struct md_op_spec *spec) { - struct lu_attr *attr = &ma->ma_attr; struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint; + struct dt_object_format *dof = &mdd_env_info(env)->mti_dof; int rc; ENTRY; - if (!mdd_object_exists(c)) { - struct dt_object *next = mdd_object_child(c); - LASSERT(next); + LASSERT(!mdd_object_exists(c)); - /* @hint will be initialized by underlying device. */ - next->do_ops->do_ah_init(env, hint, - p ? mdd_object_child(p) : NULL, - attr->la_mode & S_IFMT); - rc = mdo_create_obj(env, c, attr, hint, handle); - LASSERT(ergo(rc == 0, mdd_object_exists(c))); - } else - rc = -EEXIST; + rc = mdo_create_obj(env, c, attr, hint, dof, handle); - RETURN(rc); + LASSERT(ergo(rc == 0, mdd_object_exists(c))); + + RETURN(rc); } /** @@ -553,11 +678,9 @@ static inline int mdd_attr_check(const struct lu_env *env, RETURN(0); } -int mdd_attr_set_internal(const struct lu_env *env, - struct mdd_object *obj, - struct lu_attr *attr, - struct thandle *handle, - int needacl) +int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj, + struct lu_attr *attr, struct thandle *handle, + int needacl) { int rc; ENTRY; @@ -571,10 +694,8 @@ int mdd_attr_set_internal(const struct lu_env *env, } int mdd_attr_check_set_internal(const struct lu_env *env, - struct mdd_object *obj, - struct lu_attr *attr, - struct thandle *handle, - int needacl) + struct mdd_object *obj, struct lu_attr *attr, + struct thandle *handle, int needacl) { int rc; ENTRY; @@ -588,58 +709,6 @@ int mdd_attr_check_set_internal(const struct lu_env *env, RETURN(rc); } -static int mdd_attr_set_internal_locked(const struct lu_env *env, - struct mdd_object *obj, - struct lu_attr *attr, - struct thandle *handle, - int needacl) -{ - int rc; - ENTRY; - - needacl = needacl && (attr->la_valid & LA_MODE); - if (needacl) - mdd_write_lock(env, obj, MOR_TGT_CHILD); - rc = mdd_attr_set_internal(env, obj, attr, handle, needacl); - if (needacl) - mdd_write_unlock(env, obj); - RETURN(rc); -} - -int mdd_attr_check_set_internal_locked(const struct lu_env *env, - struct mdd_object *obj, - struct lu_attr *attr, - struct thandle *handle, - int needacl) -{ - int rc; - ENTRY; - - needacl = needacl && (attr->la_valid & LA_MODE); - if (needacl) - mdd_write_lock(env, obj, MOR_TGT_CHILD); - rc = mdd_attr_check_set_internal(env, obj, attr, handle, needacl); - if (needacl) - mdd_write_unlock(env, obj); - RETURN(rc); -} - -static int __mdd_xattr_set(const struct lu_env *env, struct mdd_object *obj, - const struct lu_buf *buf, const char *name, - int fl, struct thandle *handle) -{ - struct lustre_capa *capa = mdd_object_capa(env, obj); - int rc = -EINVAL; - ENTRY; - - if (buf->lb_buf && buf->lb_len > 0) - rc = mdo_xattr_set(env, obj, buf, name, 0, handle, capa); - else if (buf->lb_buf == NULL && buf->lb_len == 0) - rc = mdo_xattr_del(env, obj, name, handle, capa); - - RETURN(rc); -} - /* * This gives the same functionality as the code between * sys_chmod and inode_setattr @@ -648,10 +717,10 @@ static int __mdd_xattr_set(const struct lu_env *env, struct mdd_object *obj, * This API is ported from mds_fix_attr but remove some unnecesssary stuff. */ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, - struct lu_attr *la, const struct md_attr *ma) + struct lu_attr *la, const unsigned long flags) { struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; - struct md_ucred *uc = md_ucred(env); + struct lu_ucred *uc; int rc; ENTRY; @@ -666,16 +735,21 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE)) RETURN(-EPERM); + /* export destroy does not have ->le_ses, but we may want + * to drop LUSTRE_SOM_FL. */ + uc = lu_ucred_check(env); + if (uc == NULL) + RETURN(0); + rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA); if (rc) RETURN(rc); if (la->la_valid == LA_CTIME) { - if (!(ma->ma_attr_flags & MDS_PERM_BYPASS)) + if (!(flags & MDS_PERM_BYPASS)) /* This is only for set ctime when rename's source is * on remote MDS. */ - rc = mdd_may_delete(env, NULL, obj, - (struct md_attr *)ma, 1, 0); + rc = mdd_may_delete(env, NULL, obj, tmp_la, NULL, 1, 0); if (rc == 0 && la->la_ctime <= tmp_la->la_ctime) la->la_valid &= ~LA_CTIME; RETURN(rc); @@ -683,28 +757,29 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, if (la->la_valid == LA_ATIME) { /* This is atime only set for read atime update on close. */ - if (la->la_atime <= tmp_la->la_atime + - mdd_obj2mdd_dev(obj)->mdd_atime_diff) + if (la->la_atime >= tmp_la->la_atime && + la->la_atime < (tmp_la->la_atime + + mdd_obj2mdd_dev(obj)->mdd_atime_diff)) la->la_valid &= ~LA_ATIME; RETURN(0); } - + /* Check if flags change. */ if (la->la_valid & LA_FLAGS) { unsigned int oldflags = 0; unsigned int newflags = la->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL); - if ((uc->mu_fsuid != tmp_la->la_uid) && - !mdd_capable(uc, CFS_CAP_FOWNER)) - RETURN(-EPERM); + if ((uc->uc_fsuid != tmp_la->la_uid) && + !mdd_capable(uc, CFS_CAP_FOWNER)) + RETURN(-EPERM); /* XXX: the IMMUTABLE and APPEND_ONLY flags can * only be changed by the relevant capability. */ if (mdd_is_immutable(obj)) oldflags |= LUSTRE_IMMUTABLE_FL; if (mdd_is_append(obj)) - oldflags |= LUSTRE_APPEND_FL; + oldflags |= LUSTRE_APPEND_FL; if ((oldflags ^ newflags) && !mdd_capable(uc, CFS_CAP_LINUX_IMMUTABLE)) RETURN(-EPERM); @@ -715,33 +790,50 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, if ((mdd_is_immutable(obj) || mdd_is_append(obj)) && (la->la_valid & ~LA_FLAGS) && - !(ma->ma_attr_flags & MDS_PERM_BYPASS)) + !(flags & MDS_PERM_BYPASS)) RETURN(-EPERM); - /* Check for setting the obj time. */ - if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) && - !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) { - if ((uc->mu_fsuid != tmp_la->la_uid) && - !mdd_capable(uc, CFS_CAP_FOWNER)) { - rc = mdd_permission_internal_locked(env, obj, tmp_la, - MAY_WRITE, - MOR_TGT_CHILD); - if (rc) - RETURN(rc); + /* Check for setting the obj time. */ + if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) && + !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) { + if ((uc->uc_fsuid != tmp_la->la_uid) && + !mdd_capable(uc, CFS_CAP_FOWNER)) { + rc = mdd_permission_internal(env, obj, tmp_la, + MAY_WRITE); + if (rc) + RETURN(rc); + } + } + + if (la->la_valid & LA_KILL_SUID) { + la->la_valid &= ~LA_KILL_SUID; + if ((tmp_la->la_mode & S_ISUID) && + !(la->la_valid & LA_MODE)) { + la->la_mode = tmp_la->la_mode; + la->la_valid |= LA_MODE; } + la->la_mode &= ~S_ISUID; + } + + if (la->la_valid & LA_KILL_SGID) { + la->la_valid &= ~LA_KILL_SGID; + if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) == + (S_ISGID | S_IXGRP)) && + !(la->la_valid & LA_MODE)) { + la->la_mode = tmp_la->la_mode; + la->la_valid |= LA_MODE; + } + la->la_mode &= ~S_ISGID; } /* Make sure a caller can chmod. */ if (la->la_valid & LA_MODE) { - /* Bypass la_vaild == LA_MODE, - * this is for changing file with SUID or SGID. */ - if ((la->la_valid & ~LA_MODE) && - !(ma->ma_attr_flags & MDS_PERM_BYPASS) && - (uc->mu_fsuid != tmp_la->la_uid) && - !mdd_capable(uc, CFS_CAP_FOWNER)) - RETURN(-EPERM); + if (!(flags & MDS_PERM_BYPASS) && + (uc->uc_fsuid != tmp_la->la_uid) && + !mdd_capable(uc, CFS_CAP_FOWNER)) + RETURN(-EPERM); - if (la->la_mode == (umode_t) -1) + if (la->la_mode == (cfs_umode_t) -1) la->la_mode = tmp_la->la_mode; else la->la_mode = (la->la_mode & S_IALLUGO) | @@ -760,10 +852,10 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, if (la->la_valid & LA_UID) { if (la->la_uid == (uid_t) -1) la->la_uid = tmp_la->la_uid; - if (((uc->mu_fsuid != tmp_la->la_uid) || - (la->la_uid != tmp_la->la_uid)) && - !mdd_capable(uc, CFS_CAP_CHOWN)) - RETURN(-EPERM); + if (((uc->uc_fsuid != tmp_la->la_uid) || + (la->la_uid != tmp_la->la_uid)) && + !mdd_capable(uc, CFS_CAP_CHOWN)) + RETURN(-EPERM); /* If the user or group of a non-directory has been * changed by a non-root user, remove the setuid bit. @@ -785,11 +877,11 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, if (la->la_valid & LA_GID) { if (la->la_gid == (gid_t) -1) la->la_gid = tmp_la->la_gid; - if (((uc->mu_fsuid != tmp_la->la_uid) || - ((la->la_gid != tmp_la->la_gid) && - !lustre_in_group_p(uc, la->la_gid))) && - !mdd_capable(uc, CFS_CAP_CHOWN)) - RETURN(-EPERM); + if (((uc->uc_fsuid != tmp_la->la_uid) || + ((la->la_gid != tmp_la->la_gid) && + !lustre_in_group_p(uc, la->la_gid))) && + !mdd_capable(uc, CFS_CAP_CHOWN)) + RETURN(-EPERM); /* Likewise, if the user or group of a non-directory * has been changed by a non-root user, remove the @@ -808,11 +900,11 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, /* For both Size-on-MDS case and truncate case, * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture. - * We distinguish them by "ma->ma_attr_flags & MDS_SOM". + * We distinguish them by "flags & MDS_SOM". * For SOM case, it is true, the MAY_WRITE perm has been checked * when open, no need check again. For truncate case, it is false, * the MAY_WRITE perm should be checked here. */ - if (ma->ma_attr_flags & MDS_SOM) { + if (flags & MDS_SOM) { /* For the "Size-on-MDS" setattr update, merge coming * attributes with the set in the inode. BUG 10641 */ if ((la->la_valid & LA_ATIME) && @@ -826,12 +918,11 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, la->la_valid &= ~(LA_MTIME | LA_CTIME); } else { if (la->la_valid & (LA_SIZE | LA_BLOCKS)) { - if (!((ma->ma_attr_flags & MDS_OPEN_OWNEROVERRIDE) && - (uc->mu_fsuid == tmp_la->la_uid)) && - !(ma->ma_attr_flags & MDS_PERM_BYPASS)) { - rc = mdd_permission_internal_locked(env, obj, - tmp_la, MAY_WRITE, - MOR_TGT_CHILD); + if (!((flags & MDS_OPEN_OWNEROVERRIDE) && + (uc->uc_fsuid == tmp_la->la_uid)) && + !(flags & MDS_PERM_BYPASS)) { + rc = mdd_permission_internal(env, obj, + tmp_la, MAY_WRITE); if (rc) RETURN(rc); } @@ -848,127 +939,266 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, RETURN(0); } +/** Store a data change changelog record + * If this fails, we must fail the whole transaction; we don't + * want the change to commit without the log entry. + * \param mdd_obj - mdd_object of change + * \param handle - transacion handle + */ +static int mdd_changelog_data_store(const struct lu_env *env, + struct mdd_device *mdd, + enum changelog_rec_type type, + int flags, struct mdd_object *mdd_obj, + struct thandle *handle) +{ + const struct lu_fid *tfid = mdo2fid(mdd_obj); + struct llog_changelog_rec *rec; + struct lu_buf *buf; + int reclen; + int rc; + + /* Not recording */ + if (!(mdd->mdd_cl.mc_flags & CLM_ON)) + RETURN(0); + if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) + RETURN(0); + + LASSERT(mdd_obj != NULL); + LASSERT(handle != NULL); + + if ((type >= CL_MTIME) && (type <= CL_ATIME) && + cfs_time_before_64(mdd->mdd_cl.mc_starttime, mdd_obj->mod_cltime)) { + /* Don't need multiple updates in this log */ + /* Don't check under lock - no big deal if we get an extra + entry */ + RETURN(0); + } + + reclen = llog_data_len(sizeof(*rec)); + buf = mdd_buf_alloc(env, reclen); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + rec = buf->lb_buf; + + rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags); + rec->cr.cr_type = (__u32)type; + rec->cr.cr_tfid = *tfid; + rec->cr.cr_namelen = 0; + mdd_obj->mod_cltime = cfs_time_current_64(); + + rc = mdd_changelog_store(env, mdd, rec, handle); + + RETURN(rc); +} + +int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type, + int flags, struct md_object *obj) +{ + struct thandle *handle; + struct mdd_object *mdd_obj = md2mdd_obj(obj); + struct mdd_device *mdd = mdo2mdd(obj); + int rc; + ENTRY; + + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + RETURN(PTR_ERR(handle)); + + rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdd, handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj, + handle); + +stop: + mdd_trans_stop(env, mdd, rc, handle); + + RETURN(rc); +} + +/** + * Save LMA extended attributes with data from \a ma. + * + * HSM and Size-On-MDS data will be extracted from \ma if they are valid, if + * not, LMA EA will be first read from disk, modified and write back. + * + */ +/* Precedence for choosing record type when multiple + * attributes change: setattr > mtime > ctime > atime + * (ctime changes when mtime does, plus chmod/chown. + * atime and ctime are independent.) */ +static int mdd_attr_set_changelog(const struct lu_env *env, + struct md_object *obj, struct thandle *handle, + __u64 valid) +{ + struct mdd_device *mdd = mdo2mdd(obj); + int bits, type = 0; + + bits = (valid & ~(LA_CTIME|LA_MTIME|LA_ATIME)) ? 1 << CL_SETATTR : 0; + bits |= (valid & LA_MTIME) ? 1 << CL_MTIME : 0; + bits |= (valid & LA_CTIME) ? 1 << CL_CTIME : 0; + bits |= (valid & LA_ATIME) ? 1 << CL_ATIME : 0; + bits = bits & mdd->mdd_cl.mc_mask; + if (bits == 0) + return 0; + + /* The record type is the lowest non-masked set bit */ + while (bits && ((bits & 1) == 0)) { + bits = bits >> 1; + type++; + } + + /* FYI we only store the first CLF_FLAGMASK bits of la_valid */ + return mdd_changelog_data_store(env, mdd, type, (int)valid, + md2mdd_obj(obj), handle); +} + +static int mdd_declare_attr_set(const struct lu_env *env, + struct mdd_device *mdd, + struct mdd_object *obj, + const struct lu_attr *attr, + struct thandle *handle) +{ + int rc; + + rc = mdo_declare_attr_set(env, obj, attr, handle); + if (rc) + return rc; + +#ifdef CONFIG_FS_POSIX_ACL + if (attr->la_valid & LA_MODE) { + mdd_read_lock(env, obj, MOR_TGT_CHILD); + rc = mdo_xattr_get(env, obj, &LU_BUF_NULL, + XATTR_NAME_ACL_ACCESS, BYPASS_CAPA); + mdd_read_unlock(env, obj); + if (rc == -EOPNOTSUPP || rc == -ENODATA) + rc = 0; + else if (rc < 0) + return rc; + + if (rc != 0) { + struct lu_buf *buf = mdd_buf_get(env, NULL, rc); + rc = mdo_declare_xattr_set(env, obj, buf, + XATTR_NAME_ACL_ACCESS, 0, + handle); + if (rc) + return rc; + } + } +#endif + + rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + return rc; +} + /* set attr and LOV EA at once, return updated attr */ -static int mdd_attr_set(const struct lu_env *env, struct md_object *obj, - const struct md_attr *ma) +int mdd_attr_set(const struct lu_env *env, struct md_object *obj, + const struct md_attr *ma) { struct mdd_object *mdd_obj = md2mdd_obj(obj); struct mdd_device *mdd = mdo2mdd(obj); struct thandle *handle; - struct lov_mds_md *lmm = NULL; - struct llog_cookie *logcookies = NULL; - int rc, lmm_size = 0, cookie_size = 0; struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix; + const struct lu_attr *la = &ma->ma_attr; + int rc; ENTRY; - mdd_txn_param_build(env, mdd, MDD_TXN_ATTR_SET_OP); - handle = mdd_trans_start(env, mdd); + /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */ + LASSERT((ma->ma_valid & MA_LOV) == 0); + LASSERT((ma->ma_valid & MA_HSM) == 0); + LASSERT((ma->ma_valid & MA_SOM) == 0); + + *la_copy = ma->ma_attr; + rc = mdd_fix_attr(env, mdd_obj, la_copy, ma->ma_attr_flags); + if (rc) + RETURN(rc); + + /* setattr on "close" only change atime, or do nothing */ + if (la->la_valid == LA_ATIME && la_copy->la_valid == 0) + RETURN(0); + + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); - /*TODO: add lock here*/ - /* start a log jounal handle if needed */ - if (S_ISREG(mdd_object_type(mdd_obj)) && - ma->ma_attr.la_valid & (LA_UID | LA_GID)) { - lmm_size = mdd_lov_mdsize(env, mdd); - lmm = mdd_max_lmm_get(env, mdd); - if (lmm == NULL) - GOTO(cleanup, rc = -ENOMEM); - - rc = mdd_get_md_locked(env, mdd_obj, lmm, &lmm_size, - MDS_LOV_MD_NAME); - - if (rc < 0) - GOTO(cleanup, rc); - } - if (ma->ma_attr.la_valid & (ATTR_MTIME | ATTR_CTIME)) - CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n", - ma->ma_attr.la_mtime, ma->ma_attr.la_ctime); + rc = mdd_declare_attr_set(env, mdd, mdd_obj, la, handle); + if (rc) + GOTO(stop, rc); - *la_copy = ma->ma_attr; - rc = mdd_fix_attr(env, mdd_obj, la_copy, ma); + rc = mdd_trans_start(env, mdd, handle); if (rc) - GOTO(cleanup, rc); + GOTO(stop, rc); + + /* permission changes may require sync operation */ + if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID)) + handle->th_sync |= !!mdd->mdd_sync_permission; + + if (la->la_valid & (LA_MTIME | LA_CTIME)) + CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n", + la->la_mtime, la->la_ctime); if (la_copy->la_valid & LA_FLAGS) { - rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy, - handle, 1); + rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1); if (rc == 0) mdd_flags_xlate(mdd_obj, la_copy->la_flags); } else if (la_copy->la_valid) { /* setattr */ - rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy, - handle, 1); - /* journal chown/chgrp in llog, just like unlink */ - if (rc == 0 && lmm_size){ - cookie_size = mdd_lov_cookiesize(env, mdd); - logcookies = mdd_max_cookie_get(env, mdd); - if (logcookies == NULL) - GOTO(cleanup, rc = -ENOMEM); - - if (mdd_setattr_log(env, mdd, ma, lmm, lmm_size, - logcookies, cookie_size) <= 0) - logcookies = NULL; - } + rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1); } - if (rc == 0 && ma->ma_valid & MA_LOV) { - umode_t mode; - - mode = mdd_object_type(mdd_obj); - if (S_ISREG(mode) || S_ISDIR(mode)) { - rc = mdd_lsm_sanity_check(env, mdd_obj); - if (rc) - GOTO(cleanup, rc); - - rc = mdd_lov_set_md(env, NULL, mdd_obj, ma->ma_lmm, - ma->ma_lmm_size, handle, 1); - } - - } -cleanup: + if (rc == 0) + rc = mdd_attr_set_changelog(env, obj, handle, + la->la_valid); +stop: mdd_trans_stop(env, mdd, rc, handle); - if (rc == 0 && (lmm != NULL && lmm_size > 0 )) { - /*set obd attr, if needed*/ - rc = mdd_lov_setattr_async(env, mdd_obj, lmm, lmm_size, - logcookies); - } RETURN(rc); } -int mdd_xattr_set_txn(const struct lu_env *env, struct mdd_object *obj, - const struct lu_buf *buf, const char *name, int fl, - struct thandle *handle) +static int mdd_xattr_sanity_check(const struct lu_env *env, + struct mdd_object *obj) { - int rc; - ENTRY; + struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; + struct lu_ucred *uc = lu_ucred_assert(env); + int rc; + ENTRY; - mdd_write_lock(env, obj, MOR_TGT_CHILD); - rc = __mdd_xattr_set(env, obj, buf, name, fl, handle); - mdd_write_unlock(env, obj); + if (mdd_is_immutable(obj) || mdd_is_append(obj)) + RETURN(-EPERM); - RETURN(rc); + rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA); + if (rc) + RETURN(rc); + + if ((uc->uc_fsuid != tmp_la->la_uid) && + !mdd_capable(uc, CFS_CAP_FOWNER)) + RETURN(-EPERM); + + RETURN(rc); } -static int mdd_xattr_sanity_check(const struct lu_env *env, - struct mdd_object *obj) +static int mdd_declare_xattr_set(const struct lu_env *env, + struct mdd_device *mdd, + struct mdd_object *obj, + const struct lu_buf *buf, + const char *name, + struct thandle *handle) { - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; - struct md_ucred *uc = md_ucred(env); int rc; - ENTRY; - - if (mdd_is_immutable(obj) || mdd_is_append(obj)) - RETURN(-EPERM); - rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA); + rc = mdo_declare_xattr_set(env, obj, buf, name, 0, handle); if (rc) - RETURN(rc); + return rc; - if ((uc->mu_fsuid != tmp_la->la_uid) && - !mdd_capable(uc, CFS_CAP_FOWNER)) - RETURN(-EPERM); + /* Only record user xattr changes */ + if ((strncmp("user.", name, 5) == 0)) + rc = mdd_declare_changelog_store(env, mdd, NULL, handle); - RETURN(rc); + rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + return rc; } /** @@ -985,21 +1215,73 @@ static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj, int rc; ENTRY; + if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) { + rc = mdd_acl_set(env, mdd_obj, buf, fl); + RETURN(rc); + } + rc = mdd_xattr_sanity_check(env, mdd_obj); if (rc) RETURN(rc); - mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP); - handle = mdd_trans_start(env, mdd); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); - rc = mdd_xattr_set_txn(env, mdd_obj, buf, name, fl, handle); + rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdd, handle); + if (rc) + GOTO(stop, rc); + + /* security-replated changes may require sync */ + if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) + handle->th_sync |= !!mdd->mdd_sync_permission; + + mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle, + mdd_object_capa(env, mdd_obj)); + mdd_write_unlock(env, mdd_obj); + if (rc) + GOTO(stop, rc); + + /* Only record system & user xattr changes */ + if (strncmp(XATTR_USER_PREFIX, name, + sizeof(XATTR_USER_PREFIX) - 1) == 0 || + strncmp(POSIX_ACL_XATTR_ACCESS, name, + sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 || + strncmp(POSIX_ACL_XATTR_DEFAULT, name, + sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0) + rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj, + handle); + +stop: mdd_trans_stop(env, mdd, rc, handle); RETURN(rc); } +static int mdd_declare_xattr_del(const struct lu_env *env, + struct mdd_device *mdd, + struct mdd_object *obj, + const char *name, + struct thandle *handle) +{ + int rc; + + rc = mdo_declare_xattr_del(env, obj, name, handle); + if (rc) + return rc; + + /* Only record user xattr changes */ + if ((strncmp("user.", name, 5) == 0)) + rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + + return rc; +} + /** * The caller should guarantee to update the object ctime * after xattr_set if needed. @@ -1017,210 +1299,50 @@ int mdd_xattr_del(const struct lu_env *env, struct md_object *obj, if (rc) RETURN(rc); - mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP); - handle = mdd_trans_start(env, mdd); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = mdo_xattr_del(env, mdd_obj, name, handle, - mdd_object_capa(env, mdd_obj)); - mdd_write_unlock(env, mdd_obj); - mdd_trans_stop(env, mdd, rc, handle); - - RETURN(rc); -} - -/* partial unlink */ -static int mdd_ref_del(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) -{ - struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_obj = md2mdd_obj(obj); - struct mdd_device *mdd = mdo2mdd(obj); - struct thandle *handle; - int rc; - ENTRY; - - /* - * Check -ENOENT early here because we need to get object type - * to calculate credits before transaction start - */ - if (!mdd_object_exists(mdd_obj)) - RETURN(-ENOENT); - - LASSERT(mdd_object_exists(mdd_obj) > 0); - - rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP); + rc = mdd_declare_xattr_del(env, mdd, mdd_obj, name, handle); if (rc) - RETURN(rc); + GOTO(stop, rc); - handle = mdd_trans_start(env, mdd); - if (IS_ERR(handle)) - RETURN(-ENOMEM); - - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - - rc = mdd_unlink_sanity_check(env, NULL, mdd_obj, ma); + rc = mdd_trans_start(env, mdd, handle); if (rc) - GOTO(cleanup, rc); - - __mdd_ref_del(env, mdd_obj, handle, 0); + GOTO(stop, rc); - if (S_ISDIR(lu_object_attr(&obj->mo_lu))) { - /* unlink dot */ - __mdd_ref_del(env, mdd_obj, handle, 1); - } - - LASSERT(ma->ma_attr.la_valid & LA_CTIME); - la_copy->la_ctime = ma->ma_attr.la_ctime; - - la_copy->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal(env, mdd_obj, la_copy, handle, 0); - if (rc) - GOTO(cleanup, rc); - - rc = mdd_finish_unlink(env, mdd_obj, ma, handle); - - EXIT; -cleanup: + mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + rc = mdo_xattr_del(env, mdd_obj, name, handle, + mdd_object_capa(env, mdd_obj)); mdd_write_unlock(env, mdd_obj); + if (rc) + GOTO(stop, rc); + + /* Only record system & user xattr changes */ + if (strncmp(XATTR_USER_PREFIX, name, + sizeof(XATTR_USER_PREFIX) - 1) == 0 || + strncmp(POSIX_ACL_XATTR_ACCESS, name, + sizeof(POSIX_ACL_XATTR_ACCESS) - 1) == 0 || + strncmp(POSIX_ACL_XATTR_DEFAULT, name, + sizeof(POSIX_ACL_XATTR_DEFAULT) - 1) == 0) + rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj, + handle); + +stop: mdd_trans_stop(env, mdd, rc, handle); - return rc; -} - -/* partial operation */ -static int mdd_oc_sanity_check(const struct lu_env *env, - struct mdd_object *obj, - struct md_attr *ma) -{ - int rc; - ENTRY; - switch (ma->ma_attr.la_mode & S_IFMT) { - case S_IFREG: - case S_IFDIR: - case S_IFLNK: - case S_IFCHR: - case S_IFBLK: - case S_IFIFO: - case S_IFSOCK: - rc = 0; - break; - default: - rc = -EINVAL; - break; - } RETURN(rc); } -static int mdd_object_create(const struct lu_env *env, - struct md_object *obj, - const struct md_op_spec *spec, - struct md_attr *ma) -{ - - struct mdd_device *mdd = mdo2mdd(obj); - struct mdd_object *mdd_obj = md2mdd_obj(obj); - const struct lu_fid *pfid = spec->u.sp_pfid; - struct thandle *handle; - int rc; - ENTRY; - - mdd_txn_param_build(env, mdd, MDD_TXN_OBJECT_CREATE_OP); - handle = mdd_trans_start(env, mdd); - if (IS_ERR(handle)) - RETURN(PTR_ERR(handle)); - - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = mdd_oc_sanity_check(env, mdd_obj, ma); - if (rc) - GOTO(unlock, rc); - - rc = mdd_object_create_internal(env, NULL, mdd_obj, ma, handle); - if (rc) - GOTO(unlock, rc); - - if (spec->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) { - /* If creating the slave object, set slave EA here. */ - int lmv_size = spec->u.sp_ea.eadatalen; - struct lmv_stripe_md *lmv; - - lmv = (struct lmv_stripe_md *)spec->u.sp_ea.eadata; - LASSERT(lmv != NULL && lmv_size > 0); - - rc = __mdd_xattr_set(env, mdd_obj, - mdd_buf_get_const(env, lmv, lmv_size), - MDS_LMV_MD_NAME, 0, handle); - if (rc) - GOTO(unlock, rc); - - rc = mdd_attr_set_internal(env, mdd_obj, &ma->ma_attr, - handle, 0); - } else { -#ifdef CONFIG_FS_POSIX_ACL - if (spec->sp_cr_flags & MDS_CREATE_RMT_ACL) { - struct lu_buf *buf = &mdd_env_info(env)->mti_buf; - - buf->lb_buf = (void *)spec->u.sp_ea.eadata; - buf->lb_len = spec->u.sp_ea.eadatalen; - if ((buf->lb_len > 0) && (buf->lb_buf != NULL)) { - rc = __mdd_acl_init(env, mdd_obj, buf, - &ma->ma_attr.la_mode, - handle); - if (rc) - GOTO(unlock, rc); - else - ma->ma_attr.la_valid |= LA_MODE; - } - - pfid = spec->u.sp_ea.fid; - } -#endif - rc = mdd_object_initialize(env, pfid, mdd_obj, ma, handle); - } - EXIT; -unlock: - if (rc == 0) - rc = mdd_attr_get_internal(env, mdd_obj, ma); - mdd_write_unlock(env, mdd_obj); - - mdd_trans_stop(env, mdd, rc, handle); - return rc; -} - -/* partial link */ -static int mdd_ref_add(const struct lu_env *env, struct md_object *obj, - const struct md_attr *ma) +void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent, + struct mdd_object *child, struct lu_attr *attr) { - struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_obj = md2mdd_obj(obj); - struct mdd_device *mdd = mdo2mdd(obj); - struct thandle *handle; - int rc; - ENTRY; - - mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP); - handle = mdd_trans_start(env, mdd); - if (IS_ERR(handle)) - RETURN(-ENOMEM); - - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = mdd_link_sanity_check(env, NULL, NULL, mdd_obj); - if (rc == 0) - __mdd_ref_add(env, mdd_obj, handle); - mdd_write_unlock(env, mdd_obj); - if (rc == 0) { - LASSERT(ma->ma_attr.la_valid & LA_CTIME); - la_copy->la_ctime = ma->ma_attr.la_ctime; - - la_copy->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_obj, la_copy, - handle, 0); - } - mdd_trans_stop(env, mdd, 0, handle); + struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint; + struct dt_object *np = parent ? mdd_object_child(parent) : NULL; + struct dt_object *nc = mdd_object_child(child); - RETURN(rc); + /* @hint will be initialized by underlying device. */ + nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT); } /* @@ -1228,28 +1350,27 @@ static int mdd_ref_add(const struct lu_env *env, struct md_object *obj, */ int accmode(const struct lu_env *env, struct lu_attr *la, int flags) { - int res = 0; - - /* Sadly, NFSD reopens a file repeatedly during operation, so the - * "acc_mode = 0" allowance for newly-created files isn't honoured. - * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file - * owner can write to a file even if it is marked readonly to hide - * its brokenness. (bug 5781) */ - if (flags & MDS_OPEN_OWNEROVERRIDE) { - struct md_ucred *uc = md_ucred(env); - - if ((uc == NULL) || (uc->mu_valid == UCRED_INIT) || - (la->la_uid == uc->mu_fsuid)) - return 0; - } - - if (flags & FMODE_READ) - res |= MAY_READ; - if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND)) - res |= MAY_WRITE; - if (flags & MDS_FMODE_EXEC) - res |= MAY_EXEC; - return res; + int res = 0; + + /* Sadly, NFSD reopens a file repeatedly during operation, so the + * "acc_mode = 0" allowance for newly-created files isn't honoured. + * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file + * owner can write to a file even if it is marked readonly to hide + * its brokenness. (bug 5781) */ + if (flags & MDS_OPEN_OWNEROVERRIDE) { + struct lu_ucred *uc = lu_ucred_check(env); + + if ((uc == NULL) || (la->la_uid == uc->uc_fsuid)) + return 0; + } + + if (flags & FMODE_READ) + res |= MAY_READ; + if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND)) + res |= MAY_WRITE; + if (flags & MDS_FMODE_EXEC) + res = MAY_EXEC; + return res; } static int mdd_open_sanity_check(const struct lu_env *env, @@ -1298,13 +1419,13 @@ static int mdd_open_sanity_check(const struct lu_env *env, * Now, flag -- O_NOATIME does not be packed by client. */ if (flag & O_NOATIME) { - struct md_ucred *uc = md_ucred(env); + struct lu_ucred *uc = lu_ucred(env); - if (uc && ((uc->mu_valid == UCRED_OLD) || - (uc->mu_valid == UCRED_NEW)) && - (uc->mu_fsuid != tmp_la->la_uid) && - !mdd_capable(uc, CFS_CAP_FOWNER)) - RETURN(-EPERM); + if (uc && ((uc->uc_valid == UCRED_OLD) || + (uc->uc_valid == UCRED_NEW)) && + (uc->uc_fsuid != tmp_la->la_uid) && + !mdd_capable(uc, CFS_CAP_FOWNER)) + RETURN(-EPERM); } #endif @@ -1327,56 +1448,175 @@ static int mdd_open(const struct lu_env *env, struct md_object *obj, return rc; } +int mdd_declare_object_kill(const struct lu_env *env, struct mdd_object *obj, + struct md_attr *ma, struct thandle *handle) +{ + return mdo_declare_destroy(env, obj, handle); +} + /* return md_attr back, * if it is last unlink then return lov ea + llog cookie*/ int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj, - struct md_attr *ma) + struct md_attr *ma, struct thandle *handle) { - int rc = 0; + int rc; ENTRY; - if (S_ISREG(mdd_object_type(obj))) { - /* Return LOV & COOKIES unconditionally here. We clean evth up. - * Caller must be ready for that. */ - rc = __mdd_lmm_get(env, obj, ma); - if ((ma->ma_valid & MA_LOV)) - rc = mdd_unlink_log(env, mdo2mdd(&obj->mod_obj), - obj, ma); - } + rc = mdo_destroy(env, obj, handle); + RETURN(rc); } +static int mdd_declare_close(const struct lu_env *env, + struct mdd_object *obj, + struct md_attr *ma, + struct thandle *handle) +{ + int rc; + + rc = orph_declare_index_delete(env, obj, handle); + if (rc) + return rc; + + return mdo_declare_destroy(env, obj, handle); +} + /* * No permission check is needed. */ static int mdd_close(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) + struct md_attr *ma, int mode) { - int rc; struct mdd_object *mdd_obj = md2mdd_obj(obj); - struct thandle *handle; + struct mdd_device *mdd = mdo2mdd(obj); + struct thandle *handle = NULL; + int rc, is_orphan = 0; ENTRY; - rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP); - if (rc) - RETURN(rc); - handle = mdd_trans_start(env, mdo2mdd(obj)); - if (IS_ERR(handle)) - RETURN(PTR_ERR(handle)); + if (ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_KEEP_ORPHAN) { + mdd_obj->mod_count--; + + if (mdd_obj->mod_flags & ORPHAN_OBJ && !mdd_obj->mod_count) + CDEBUG(D_HA, "Object "DFID" is retained in orphan " + "list\n", PFID(mdd_object_fid(mdd_obj))); + RETURN(0); + } + + /* check without any lock */ + if (mdd_obj->mod_count == 1 && + (mdd_obj->mod_flags & (ORPHAN_OBJ | DEAD_OBJ)) != 0) { + again: + handle = mdd_trans_create(env, mdo2mdd(obj)); + if (IS_ERR(handle)) + RETURN(PTR_ERR(handle)); + + rc = mdd_declare_close(env, mdd_obj, ma, handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdo2mdd(obj), handle); + if (rc) + GOTO(stop, rc); + } mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + if (handle == NULL && mdd_obj->mod_count == 1 && + (mdd_obj->mod_flags & ORPHAN_OBJ) != 0) { + mdd_write_unlock(env, mdd_obj); + goto again; + } + /* release open count */ mdd_obj->mod_count --; - rc = mdd_iattr_get(env, mdd_obj, ma); - if (rc == 0 && mdd_obj->mod_count == 0 && ma->ma_attr.la_nlink == 0) - rc = mdd_object_kill(env, mdd_obj, ma); - else - ma->ma_valid &= ~(MA_LOV | MA_COOKIE); - + if (mdd_obj->mod_count == 0 && mdd_obj->mod_flags & ORPHAN_OBJ) { + /* remove link to object from orphan index */ + LASSERT(handle != NULL); + rc = __mdd_orphan_del(env, mdd_obj, handle); + if (rc == 0) { + CDEBUG(D_HA, "Object "DFID" is deleted from orphan " + "list, OSS objects to be destroyed.\n", + PFID(mdd_object_fid(mdd_obj))); + is_orphan = 1; + } else { + CERROR("Object "DFID" can not be deleted from orphan " + "list, maybe cause OST objects can not be " + "destroyed (err: %d).\n", + PFID(mdd_object_fid(mdd_obj)), rc); + /* If object was not deleted from orphan list, do not + * destroy OSS objects, which will be done when next + * recovery. */ + GOTO(out, rc); + } + } + + rc = mdd_la_get(env, mdd_obj, &ma->ma_attr, + mdd_object_capa(env, mdd_obj)); + /* Object maybe not in orphan list originally, it is rare case for + * mdd_finish_unlink() failure. */ + if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_orphan)) { + if (handle == NULL) { + handle = mdd_trans_create(env, mdo2mdd(obj)); + if (IS_ERR(handle)) + GOTO(out, rc = PTR_ERR(handle)); + + rc = mdo_declare_destroy(env, mdd_obj, handle); + if (rc) + GOTO(out, rc); + + rc = mdd_declare_changelog_store(env, mdd, + NULL, handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdo2mdd(obj), handle); + if (rc) + GOTO(out, rc); + } + + rc = mdo_destroy(env, mdd_obj, handle); + + if (rc != 0) + CERROR("Error when prepare to delete Object "DFID" , " + "which will cause OST objects can not be " + "destroyed.\n", PFID(mdd_object_fid(mdd_obj))); + } + EXIT; + +out: + mdd_write_unlock(env, mdd_obj); - mdd_trans_stop(env, mdo2mdd(obj), rc, handle); - RETURN(rc); + + if (rc == 0 && + (mode & (FMODE_WRITE | MDS_OPEN_APPEND | MDS_OPEN_TRUNC)) && + !(ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_RECOV_OPEN)) { + if (handle == NULL) { + handle = mdd_trans_create(env, mdo2mdd(obj)); + if (IS_ERR(handle)) + GOTO(stop, rc = IS_ERR(handle)); + + rc = mdd_declare_changelog_store(env, mdd, NULL, + handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdo2mdd(obj), handle); + if (rc) + GOTO(stop, rc); + } + + mdd_changelog_data_store(env, mdd, CL_CLOSE, mode, + mdd_obj, handle); + } + +stop: + if (handle != NULL) + mdd_trans_stop(env, mdd, rc, handle); + return rc; } /* @@ -1398,175 +1638,93 @@ static int mdd_readpage_sanity_check(const struct lu_env *env, RETURN(rc); } -static int mdd_dir_page_build(const struct lu_env *env, int first, - void *area, int nob, struct dt_it_ops *iops, - struct dt_it *it, __u64 *start, __u64 *end, - struct lu_dirent **last) +static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp, + int nob, const struct dt_it_ops *iops, + struct dt_it *it, __u32 attr, void *arg) { - struct lu_fid *fid = &mdd_env_info(env)->mti_fid2; - struct mdd_thread_info *info = mdd_env_info(env); - struct lu_fid_pack *pack = &info->mti_pack; - int result; - struct lu_dirent *ent; - - if (first) { - memset(area, 0, sizeof (struct lu_dirpage)); - area += sizeof (struct lu_dirpage); - nob -= sizeof (struct lu_dirpage); - } - - LASSERT(nob > sizeof *ent); + struct lu_dirpage *dp = &lp->lp_dir; + void *area = dp; + int result; + __u64 hash = 0; + struct lu_dirent *ent; + struct lu_dirent *last = NULL; + int first = 1; + + memset(area, 0, sizeof (*dp)); + area += sizeof (*dp); + nob -= sizeof (*dp); ent = area; - result = 0; do { - char *name; int len; int recsize; - __u64 hash; - name = (char *)iops->key(env, it); - len = iops->key_size(env, it); + len = iops->key_size(env, it); - pack = (struct lu_fid_pack *)iops->rec(env, it); - result = fid_unpack(pack, fid); - if (result != 0) - break; + /* IAM iterator can return record with zero len. */ + if (len == 0) + goto next; - recsize = (sizeof(*ent) + len + 7) & ~7; hash = iops->store(env, it); - *end = hash; + if (unlikely(first)) { + first = 0; + dp->ldp_hash_start = cpu_to_le64(hash); + } - CDEBUG(D_INFO, "%p %p %d "DFID": "LPU64" (%d) \"%*.*s\"\n", - name, ent, nob, PFID(fid), hash, len, len, len, name); + /* calculate max space required for lu_dirent */ + recsize = lu_dirent_calc_size(len, attr); if (nob >= recsize) { - ent->lde_fid = *fid; - fid_cpu_to_le(&ent->lde_fid, &ent->lde_fid); - ent->lde_hash = hash; - ent->lde_namelen = cpu_to_le16(len); - ent->lde_reclen = cpu_to_le16(recsize); - memcpy(ent->lde_name, name, len); - if (first && ent == area) - *start = hash; - *last = ent; - ent = (void *)ent + recsize; - nob -= recsize; - result = iops->next(env, it); + result = iops->rec(env, it, (struct dt_rec *)ent, attr); + if (result == -ESTALE) + goto next; + if (result != 0) + goto out; + + /* osd might not able to pack all attributes, + * so recheck rec length */ + recsize = le16_to_cpu(ent->lde_reclen); } else { - /* - * record doesn't fit into page, enlarge previous one. - */ - LASSERT(*last != NULL); - (*last)->lde_reclen = - cpu_to_le16(le16_to_cpu((*last)->lde_reclen) + - nob); - break; + result = (last != NULL) ? 0 :-EINVAL; + goto out; } + last = ent; + ent = (void *)ent + recsize; + nob -= recsize; + +next: + result = iops->next(env, it); + if (result == -ESTALE) + goto next; } while (result == 0); - return result; -} - -static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj, - const struct lu_rdpg *rdpg) -{ - struct dt_it *it; - struct dt_object *next = mdd_object_child(obj); - struct dt_it_ops *iops; - struct page *pg; - struct lu_dirent *last = NULL; - int i; - int rc; - int nob; - __u64 hash_start; - __u64 hash_end = 0; - - LASSERT(rdpg->rp_pages != NULL); - LASSERT(next->do_index_ops != NULL); - - if (rdpg->rp_count <= 0) - return -EFAULT; - - /* - * iterate through directory and fill pages from @rdpg - */ - iops = &next->do_index_ops->dio_it; - it = iops->init(env, next, 0, mdd_object_capa(env, obj)); - if (IS_ERR(it)) - return PTR_ERR(it); - - rc = iops->load(env, it, rdpg->rp_hash); - - if (rc == 0) - /* - * Iterator didn't find record with exactly the key requested. - * - * It is currently either - * - * - positioned above record with key less than - * requested---skip it. - * - * - or not positioned at all (is in IAM_IT_SKEWED - * state)---position it on the next item. - */ - rc = iops->next(env, it); - else if (rc > 0) - rc = 0; - - /* - * At this point and across for-loop: - * - * rc == 0 -> ok, proceed. - * rc > 0 -> end of directory. - * rc < 0 -> error. - */ - for (i = 0, nob = rdpg->rp_count; rc == 0 && nob > 0; - i++, nob -= CFS_PAGE_SIZE) { - LASSERT(i < rdpg->rp_npages); - pg = rdpg->rp_pages[i]; - rc = mdd_dir_page_build(env, !i, cfs_kmap(pg), - min_t(int, nob, CFS_PAGE_SIZE), iops, - it, &hash_start, &hash_end, &last); - if (rc != 0 || i == rdpg->rp_npages - 1) - last->lde_reclen = 0; - cfs_kunmap(pg); - } - if (rc > 0) { - /* - * end of directory. - */ - hash_end = DIR_END_OFF; - rc = 0; - } - if (rc == 0) { - struct lu_dirpage *dp; - - dp = cfs_kmap(rdpg->rp_pages[0]); - dp->ldp_hash_start = rdpg->rp_hash; - dp->ldp_hash_end = hash_end; - if (i == 0) - /* - * No pages were processed, mark this. - */ - dp->ldp_flags |= LDF_EMPTY; - dp->ldp_flags = cpu_to_le32(dp->ldp_flags); - cfs_kunmap(rdpg->rp_pages[0]); +out: + dp->ldp_hash_end = cpu_to_le64(hash); + if (last != NULL) { + if (last->lde_hash == dp->ldp_hash_end) + dp->ldp_flags |= cpu_to_le32(LDF_COLLIDE); + last->lde_reclen = 0; /* end mark */ } - iops->put(env, it); - iops->fini(env, it); - - return rc; + if (result > 0) + /* end of directory */ + dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF); + if (result < 0) + CWARN("build page failed: %d!\n", result); + return result; } -static int mdd_readpage(const struct lu_env *env, struct md_object *obj, - const struct lu_rdpg *rdpg) +int mdd_readpage(const struct lu_env *env, struct md_object *obj, + const struct lu_rdpg *rdpg) { struct mdd_object *mdd_obj = md2mdd_obj(obj); int rc; ENTRY; - LASSERT(mdd_object_exists(mdd_obj)); + if (mdd_object_exists(mdd_obj) == 0) { + CERROR("%s: object "DFID" not found: rc = -2\n", + mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj))); + return -ENOENT; + } mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); rc = mdd_readpage_sanity_check(env, mdd_obj); @@ -1577,12 +1735,12 @@ static int mdd_readpage(const struct lu_env *env, struct md_object *obj, struct page *pg; struct lu_dirpage *dp; - /* - * According to POSIX, please do not return any entry to client: - * even dot and dotdot should not be returned. - */ - CWARN("readdir from dead object: "DFID"\n", - PFID(mdd_object_fid(mdd_obj))); + /* + * According to POSIX, please do not return any entry to client: + * even dot and dotdot should not be returned. + */ + CDEBUG(D_INODE, "readdir from dead object: "DFID"\n", + PFID(mdd_object_fid(mdd_obj))); if (rdpg->rp_count <= 0) GOTO(out_unlock, rc = -EFAULT); @@ -1591,17 +1749,32 @@ static int mdd_readpage(const struct lu_env *env, struct md_object *obj, pg = rdpg->rp_pages[0]; dp = (struct lu_dirpage*)cfs_kmap(pg); memset(dp, 0 , sizeof(struct lu_dirpage)); - dp->ldp_hash_start = rdpg->rp_hash; - dp->ldp_hash_end = DIR_END_OFF; - dp->ldp_flags |= LDF_EMPTY; - dp->ldp_flags = cpu_to_le32(dp->ldp_flags); + dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash); + dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF); + dp->ldp_flags = cpu_to_le32(LDF_EMPTY); cfs_kunmap(pg); - GOTO(out_unlock, rc = 0); + GOTO(out_unlock, rc = LU_PAGE_SIZE); } - rc = __mdd_readpage(env, mdd_obj, rdpg); - - EXIT; + rc = dt_index_walk(env, mdd_object_child(mdd_obj), rdpg, + mdd_dir_page_build, NULL); + if (rc >= 0) { + struct lu_dirpage *dp; + + dp = cfs_kmap(rdpg->rp_pages[0]); + dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash); + if (rc == 0) { + /* + * No pages were processed, mark this for first page + * and send back. + */ + dp->ldp_flags = cpu_to_le32(LDF_EMPTY); + rc = min_t(unsigned int, LU_PAGE_SIZE, rdpg->rp_count); + } + cfs_kunmap(rdpg->rp_pages[0]); + } + + GOTO(out_unlock, rc); out_unlock: mdd_read_unlock(env, mdd_obj); return rc; @@ -1610,14 +1783,16 @@ out_unlock: static int mdd_object_sync(const struct lu_env *env, struct md_object *obj) { struct mdd_object *mdd_obj = md2mdd_obj(obj); - struct dt_object *next; - LASSERT(mdd_object_exists(mdd_obj)); - next = mdd_object_child(mdd_obj); - return next->do_ops->do_object_sync(env, next); + if (mdd_object_exists(mdd_obj) == 0) { + CERROR("%s: object "DFID" not found: rc = -2\n", + mdd_obj_dev_name(mdd_obj),PFID(mdd_object_fid(mdd_obj))); + return -ENOENT; + } + return dt_object_sync(env, mdd_object_child(mdd_obj)); } -struct md_object_operations mdd_obj_ops = { +const struct md_object_operations mdd_obj_ops = { .moo_permission = mdd_permission, .moo_attr_get = mdd_attr_get, .moo_attr_set = mdd_attr_set, @@ -1625,13 +1800,12 @@ struct md_object_operations mdd_obj_ops = { .moo_xattr_set = mdd_xattr_set, .moo_xattr_list = mdd_xattr_list, .moo_xattr_del = mdd_xattr_del, - .moo_object_create = mdd_object_create, - .moo_ref_add = mdd_ref_add, - .moo_ref_del = mdd_ref_del, .moo_open = mdd_open, .moo_close = mdd_close, .moo_readpage = mdd_readpage, .moo_readlink = mdd_readlink, + .moo_changelog = mdd_changelog, .moo_capa_get = mdd_capa_get, .moo_object_sync = mdd_object_sync, + .moo_path = mdd_path, };