X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_object.c;h=6cef91896e99ae657f3f94be890772f84aa33ef3;hp=c13cdd60ade589cd383ae0236836e7b963de3d58;hb=624fd0ca0832e92dd2f5486984b6c7f7397619ba;hpb=cefa8cda2ba2d288ccaa4ec077a6c627592503ea diff --git a/lustre/mdd/mdd_object.c b/lustre/mdd/mdd_object.c index c13cdd6..6cef918 100644 --- a/lustre/mdd/mdd_object.c +++ b/lustre/mdd/mdd_object.c @@ -26,7 +26,7 @@ * 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. */ /* @@ -46,7 +46,11 @@ #define DEBUG_SUBSYSTEM S_MDS #include +#ifdef HAVE_EXT4_LDISKFS +#include +#else #include +#endif #include #include #include @@ -54,7 +58,12 @@ /* fid_be_cpu(), fid_cpu_to_be(). */ #include +#include +#ifdef HAVE_EXT4_LDISKFS +#include +#else #include +#endif #include #include @@ -62,6 +71,19 @@ static const struct lu_object_operations mdd_lu_obj_ops; +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) +{ + LASSERTF(mdd_object_exists(obj), "FID is "DFID"\n", + PFID(mdd_object_fid(obj))); + 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) { @@ -81,6 +103,15 @@ 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; @@ -91,6 +122,69 @@ struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len) return buf; } +void mdd_buf_put(struct lu_buf *buf) +{ + if (buf == NULL || buf->lb_buf == NULL) + return; + OBD_FREE_LARGE(buf->lb_buf, buf->lb_len); + buf->lb_buf = NULL; + buf->lb_len = 0; +} + +const struct lu_buf *mdd_buf_get_const(const struct lu_env *env, + const void *area, ssize_t len) +{ + struct lu_buf *buf; + + buf = &mdd_env_info(env)->mti_buf; + buf->lb_buf = (void *)area; + buf->lb_len = len; + return buf; +} + +struct lu_buf *mdd_buf_alloc(const struct lu_env *env, ssize_t len) +{ + 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->lb_buf = NULL; + } + if (buf->lb_buf == NULL) { + buf->lb_len = len; + OBD_ALLOC_LARGE(buf->lb_buf, buf->lb_len); + if (buf->lb_buf == NULL) + buf->lb_len = 0; + } + return buf; +} + +/** 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 llog_cookie *mdd_max_cookie_get(const struct lu_env *env, struct mdd_device *mdd) { @@ -100,12 +194,13 @@ struct llog_cookie *mdd_max_cookie_get(const struct lu_env *env, 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); + OBD_FREE_LARGE(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); + OBD_ALLOC_LARGE(mti->mti_max_cookie, max_cookie_size); if (likely(mti->mti_max_cookie != NULL)) mti->mti_max_cookie_size = max_cookie_size; } @@ -123,38 +218,18 @@ struct lov_mds_md *mdd_max_lmm_get(const struct lu_env *env, 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); + OBD_FREE_LARGE(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)) + OBD_ALLOC_LARGE(mti->mti_max_lmm, max_lmm_size); + if (likely(mti->mti_max_lmm != NULL)) mti->mti_max_lmm_size = max_lmm_size; } return mti->mti_max_lmm; } -const struct lu_buf *mdd_buf_get_const(const struct lu_env *env, - const void *area, ssize_t len) -{ - struct lu_buf *buf; - - buf = &mdd_env_info(env)->mti_buf; - buf->lb_buf = (void *)area; - buf->lb_len = len; - return buf; -} - -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_object *mdd_object_alloc(const struct lu_env *env, const struct lu_object_header *hdr, struct lu_device *d) @@ -178,20 +253,23 @@ 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, - const struct lu_object_conf *_) + const struct lu_object_conf *unused) { 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; + 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(lu2mdd_obj(o)); + mdd_pdlock_init(mdd_obj); if (below == NULL) RETURN(-ENOMEM); lu_object_add(o, below); + RETURN(0); } @@ -211,38 +289,21 @@ static void mdd_object_free(const struct lu_env *env, struct lu_object *o) OBD_FREE_PTR(mdd); } -/* orphan handling is here */ -static void mdd_object_delete(const struct lu_env *env, struct lu_object *o) +static int mdd_object_print(const struct lu_env *env, void *cookie, + lu_printer_t p, const 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); - } - } + 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); } 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_delete = mdd_object_delete + .loo_object_print = mdd_object_print, }; struct mdd_object *mdd_object_find(const struct lu_env *env, @@ -252,6 +313,246 @@ struct mdd_object *mdd_object_find(const struct lu_env *env, return md2mdd_obj(md_object_find_slice(env, &d->mdd_md_dev, f)); } +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; + + /* temp buffer for path element */ + buf = mdd_buf_alloc(env, PATH_MAX); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + lname->ln_name = name = buf->lb_buf; + lname->ln_namelen = 0; + *f = mdd->mdd_root_fid; + + while(1) { + while (*path == '/') + path++; + if (*path == '\0') + break; + while (*path != '/' && *path != '\0') { + *name = *path; + path++; + name++; + lname->ln_namelen++; + } + + *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; + + name = buf->lb_buf; + lname->ln_namelen = 0; + } + + if (!rc) + *fid = *f; +out: + RETURN(rc); +} + +/** 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; + + 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); + } + + /* 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++; + } + + /* 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; + } + + /* 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. */ + cfs_spin_lock(&mdd->mdd_cl.mc_lock); + pli->pli_currec = mdd->mdd_cl.mc_index; + cfs_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); + } + 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); + + 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); + + return rc; +} + +static int mdd_path_historic(const struct lu_env *env, + struct path_lookup_info *pli) +{ + return 0; +} + +/* 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) +{ + struct path_lookup_info *pli; + int tries = 3; + int rc = -EAGAIN; + ENTRY; + + if (pathlen < 3) + RETURN(-EOVERFLOW); + + if (mdd_is_root(mdo2mdd(obj), mdd_object_fid(md2mdd_obj(obj)))) { + path[0] = '\0'; + RETURN(0); + } + + 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; + } + + OBD_FREE_PTR(pli); + + RETURN (rc); +} + int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj) { struct lu_attr *la = &mdd_env_info(env)->mti_la; @@ -284,27 +585,35 @@ int mdd_iattr_get(const struct lu_env *env, struct mdd_object *mdd_obj, RETURN(rc); } -static int mdd_get_default_md(struct mdd_object *mdd_obj, - struct lov_mds_md *lmm, int *size) +int mdd_get_default_md(struct mdd_object *mdd_obj, struct lov_mds_md *lmm) { struct lov_desc *ldesc; struct mdd_device *mdd = mdo2mdd(&mdd_obj->mod_obj); + struct lov_user_md *lum = (struct lov_user_md*)lmm; ENTRY; + if (!lum) + RETURN(0); + ldesc = &mdd->mdd_obd_dev->u.mds.mds_lov_desc; LASSERT(ldesc != NULL); - if (!lmm) - RETURN(0); + lum->lmm_magic = LOV_MAGIC_V1; + lum->lmm_object_seq = FID_SEQ_LOV_DEFAULT; + lum->lmm_pattern = ldesc->ld_pattern; + lum->lmm_stripe_size = ldesc->ld_default_stripe_size; + lum->lmm_stripe_count = ldesc->ld_default_stripe_count; + lum->lmm_stripe_offset = ldesc->ld_default_stripe_offset; - 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); + RETURN(sizeof(*lum)); +} + +static int is_rootdir(struct mdd_object *mdd_obj) +{ + const struct mdd_device *mdd_dev = mdd_obj2mdd_dev(mdd_obj); + const struct lu_fid *fid = mdo2fid(mdd_obj); - RETURN(sizeof(struct lov_mds_md)); + return lu_fid_eq(&mdd_dev->mdd_root_fid, fid); } /* get lov EA only */ @@ -318,20 +627,45 @@ static int __mdd_lmm_get(const struct lu_env *env, RETURN(0); rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &ma->ma_lmm_size, - MDS_LOV_MD_NAME); - - if (rc == 0 && (ma->ma_need & MA_LOV_DEF)) { - rc = mdd_get_default_md(mdd_obj, ma->ma_lmm, - &ma->ma_lmm_size); - } - + XATTR_NAME_LOV); + if (rc == 0 && (ma->ma_need & MA_LOV_DEF) && is_rootdir(mdd_obj)) + rc = mdd_get_default_md(mdd_obj, ma->ma_lmm); if (rc > 0) { + ma->ma_lmm_size = rc; ma->ma_valid |= MA_LOV; rc = 0; } RETURN(rc); } +/* get the first parent fid from link EA */ +static int mdd_pfid_get(const struct lu_env *env, + struct mdd_object *mdd_obj, struct md_attr *ma) +{ + struct lu_buf *buf; + struct link_ea_header *leh; + struct link_ea_entry *lee; + struct lu_fid *pfid = &ma->ma_pfid; + ENTRY; + + if (ma->ma_valid & MA_PFID) + RETURN(0); + + buf = mdd_links_get(env, mdd_obj); + if (IS_ERR(buf)) + RETURN(PTR_ERR(buf)); + + leh = buf->lb_buf; + lee = (struct link_ea_entry *)(leh + 1); + memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid)); + fid_be_to_cpu(pfid, pfid); + ma->ma_valid |= MA_PFID; + if (buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + mdd_buf_put(buf); + RETURN(0); +} + int mdd_lmm_get_locked(const struct lu_env *env, struct mdd_object *mdd_obj, struct md_attr *ma) { @@ -355,7 +689,7 @@ static int __mdd_lmv_get(const struct lu_env *env, RETURN(0); rc = mdd_get_md(env, mdd_obj, ma->ma_lmv, &ma->ma_lmv_size, - MDS_LMV_MD_NAME); + XATTR_NAME_LMV); if (rc > 0) { ma->ma_valid |= MA_LMV; rc = 0; @@ -363,8 +697,58 @@ static int __mdd_lmv_get(const struct lu_env *env, RETURN(rc); } -static int mdd_attr_get_internal(const struct lu_env *env, - struct mdd_object *mdd_obj, +static int __mdd_lma_get(const struct lu_env *env, struct mdd_object *mdd_obj, + struct md_attr *ma) +{ + struct mdd_thread_info *info = mdd_env_info(env); + struct lustre_mdt_attrs *lma = + (struct lustre_mdt_attrs *)info->mti_xattr_buf; + int lma_size; + int rc; + ENTRY; + + /* If all needed data are already valid, nothing to do */ + if ((ma->ma_valid & (MA_HSM | MA_SOM)) == + (ma->ma_need & (MA_HSM | MA_SOM))) + RETURN(0); + + /* Read LMA from disk EA */ + lma_size = sizeof(info->mti_xattr_buf); + rc = mdd_get_md(env, mdd_obj, lma, &lma_size, XATTR_NAME_LMA); + if (rc <= 0) + RETURN(rc); + + /* Useless to check LMA incompatibility because this is already done in + * osd_ea_fid_get(), and this will fail long before this code is + * called. + * So, if we are here, LMA is compatible. + */ + + lustre_lma_swab(lma); + + /* Swab and copy LMA */ + if (ma->ma_need & MA_HSM) { + if (lma->lma_compat & LMAC_HSM) + ma->ma_hsm.mh_flags = lma->lma_flags & HSM_FLAGS_MASK; + else + ma->ma_hsm.mh_flags = 0; + ma->ma_valid |= MA_HSM; + } + + /* Copy SOM */ + if (ma->ma_need & MA_SOM && lma->lma_compat & LMAC_SOM) { + LASSERT(ma->ma_som != NULL); + ma->ma_som->msd_ioepoch = lma->lma_ioepoch; + ma->ma_som->msd_size = lma->lma_som_size; + ma->ma_som->msd_blocks = lma->lma_som_blocks; + ma->ma_som->msd_mountid = lma->lma_som_mountid; + ma->ma_valid |= MA_SOM; + } + + RETURN(0); +} + +int mdd_attr_get_internal(const struct lu_env *env, struct mdd_object *mdd_obj, struct md_attr *ma) { int rc = 0; @@ -378,18 +762,26 @@ static int mdd_attr_get_internal(const struct lu_env *env, S_ISDIR(mdd_object_type(mdd_obj))) rc = __mdd_lmm_get(env, mdd_obj, ma); } + if (rc == 0 && ma->ma_need & MA_PFID && !(ma->ma_valid & MA_LOV)) { + if (S_ISREG(mdd_object_type(mdd_obj))) + rc = mdd_pfid_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 (rc == 0 && ma->ma_need & (MA_HSM | MA_SOM)) { + if (S_ISREG(mdd_object_type(mdd_obj))) + rc = __mdd_lma_get(env, mdd_obj, ma); + } #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); } #endif - CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64"\n", - rc, ma->ma_valid); + CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n", + rc, ma->ma_valid, ma->ma_lmm); RETURN(rc); } @@ -397,7 +789,8 @@ int mdd_attr_get_internal_locked(const struct lu_env *env, struct mdd_object *mdd_obj, struct md_attr *ma) { int rc; - int needlock = ma->ma_need & (MA_LOV | MA_LMV | MA_ACL_DEF); + int needlock = ma->ma_need & + (MA_LOV | MA_LMV | MA_ACL_DEF | MA_HSM | MA_SOM | MA_PFID); if (needlock) mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); @@ -486,10 +879,13 @@ static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj, 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 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; + const struct dt_index_features *feat = spec->sp_feat; int rc; ENTRY; @@ -497,11 +893,19 @@ int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p, struct dt_object *next = mdd_object_child(c); LASSERT(next); + if (feat != &dt_directory_features && feat != NULL) + dof->dof_type = DFT_INDEX; + else + dof->dof_type = dt_mode_to_dft(attr->la_mode); + + dof->u.dof_idx.di_feat = feat; + /* @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); + + rc = mdo_create_obj(env, c, attr, hint, dof, handle); LASSERT(ergo(rc == 0, mdd_object_exists(c))); } else rc = -EEXIST; @@ -605,9 +1009,9 @@ int mdd_attr_check_set_internal_locked(const struct lu_env *env, 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) +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; @@ -632,7 +1036,7 @@ 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 *tmp_la = &mdd_env_info(env)->mti_la; - struct md_ucred *uc = md_ucred(env); + struct md_ucred *uc; int rc; ENTRY; @@ -647,6 +1051,13 @@ 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. */ + if (!env->le_ses) + RETURN(0); + + uc = md_ucred(env); + rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA); if (rc) RETURN(rc); @@ -664,8 +1075,9 @@ 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); } @@ -712,17 +1124,35 @@ static int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj, } } + 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) && + if (!(ma->ma_attr_flags & MDS_PERM_BYPASS) && (uc->mu_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) | @@ -829,6 +1259,193 @@ 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(handle != NULL); + LASSERT(mdd_obj != 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 = (struct llog_changelog_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_llog_write(mdd, rec, handle); + if (rc < 0) { + CERROR("changelog failed: rc=%d op%d t"DFID"\n", + rc, type, PFID(tfid)); + return -EFAULT; + } + + return 0; +} + +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_start(env, mdd); + + if (IS_ERR(handle)) + return(PTR_ERR(handle)); + + rc = mdd_changelog_data_store(env, mdd, type, flags, mdd_obj, + handle); + + mdd_trans_stop(env, mdd, rc, handle); + + RETURN(rc); +} + +/** + * Should be called with write lock held. + * + * \see mdd_lma_set_locked(). + */ +static int __mdd_lma_set(const struct lu_env *env, struct mdd_object *mdd_obj, + const struct md_attr *ma, struct thandle *handle) +{ + struct mdd_thread_info *info = mdd_env_info(env); + struct lu_buf *buf; + struct lustre_mdt_attrs *lma = + (struct lustre_mdt_attrs *) info->mti_xattr_buf; + int lmasize = sizeof(struct lustre_mdt_attrs); + int rc = 0; + + ENTRY; + + /* Either HSM or SOM part is not valid, we need to read it before */ + if ((!ma->ma_valid) & (MA_HSM | MA_SOM)) { + rc = mdd_get_md(env, mdd_obj, lma, &lmasize, XATTR_NAME_LMA); + if (rc <= 0) + RETURN(rc); + + lustre_lma_swab(lma); + } else { + memset(lma, 0, lmasize); + } + + /* Copy HSM data */ + if (ma->ma_valid & MA_HSM) { + lma->lma_flags |= ma->ma_hsm.mh_flags & HSM_FLAGS_MASK; + lma->lma_compat |= LMAC_HSM; + } + + /* Copy SOM data */ + if (ma->ma_valid & MA_SOM) { + LASSERT(ma->ma_som != NULL); + if (ma->ma_som->msd_ioepoch == IOEPOCH_INVAL) { + lma->lma_compat &= ~LMAC_SOM; + } else { + lma->lma_compat |= LMAC_SOM; + lma->lma_ioepoch = ma->ma_som->msd_ioepoch; + lma->lma_som_size = ma->ma_som->msd_size; + lma->lma_som_blocks = ma->ma_som->msd_blocks; + lma->lma_som_mountid = ma->ma_som->msd_mountid; + } + } + + /* Copy FID */ + memcpy(&lma->lma_self_fid, mdo2fid(mdd_obj), sizeof(lma->lma_self_fid)); + + lustre_lma_swab(lma); + buf = mdd_buf_get(env, lma, lmasize); + rc = __mdd_xattr_set(env, mdd_obj, buf, XATTR_NAME_LMA, 0, 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. + * + */ +static int mdd_lma_set_locked(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct md_attr *ma, struct thandle *handle) +{ + int rc; + + mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + rc = __mdd_lma_set(env, mdd_obj, ma, handle); + mdd_write_unlock(env, mdd_obj); + return rc; +} + +/* 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); +} + /* 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) @@ -846,10 +1463,21 @@ static int mdd_attr_set(const struct lu_env *env, struct md_object *obj, unsigned int qnids[MAXQUOTAS] = { 0, 0 }; unsigned int qoids[MAXQUOTAS] = { 0, 0 }; int quota_opc = 0, block_count = 0; - int inode_pending = 0, block_pending = 0; + int inode_pending[MAXQUOTAS] = { 0, 0 }; + int block_pending[MAXQUOTAS] = { 0, 0 }; #endif ENTRY; + *la_copy = ma->ma_attr; + rc = mdd_fix_attr(env, mdd_obj, la_copy, ma); + if (rc != 0) + RETURN(rc); + + /* setattr on "close" only change atime, or do nothing */ + if (ma->ma_valid == MA_INODE && + ma->ma_attr.la_valid == LA_ATIME && la_copy->la_valid == 0) + RETURN(0); + mdd_setattr_txn_param_build(env, obj, (struct md_attr *)ma, MDD_TXN_ATTR_SET_OP); handle = mdd_trans_start(env, mdd); @@ -865,23 +1493,19 @@ static int mdd_attr_set(const struct lu_env *env, struct md_object *obj, GOTO(cleanup, rc = -ENOMEM); rc = mdd_get_md_locked(env, mdd_obj, lmm, &lmm_size, - MDS_LOV_MD_NAME); + XATTR_NAME_LOV); if (rc < 0) GOTO(cleanup, rc); } - if (ma->ma_attr.la_valid & (ATTR_MTIME | ATTR_CTIME)) + if (ma->ma_attr.la_valid & (LA_MTIME | LA_CTIME)) CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n", ma->ma_attr.la_mtime, ma->ma_attr.la_ctime); - *la_copy = ma->ma_attr; - rc = mdd_fix_attr(env, mdd_obj, la_copy, ma); - if (rc) - GOTO(cleanup, rc); - #ifdef HAVE_QUOTA_SUPPORT if (mds->mds_quota && la_copy->la_valid & (LA_UID | LA_GID)) { + struct obd_export *exp = md_quota(env)->mq_exp; struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA); @@ -890,17 +1514,19 @@ static int mdd_attr_set(const struct lu_env *env, struct md_object *obj, mdd_quota_wrapper(la_copy, qnids); mdd_quota_wrapper(la_tmp, qoids); /* get file quota for new owner */ - lquota_chkquota(mds_quota_interface_ref, obd, - qnids[USRQUOTA], qnids[GRPQUOTA], 1, - &inode_pending, NULL, 0); + lquota_chkquota(mds_quota_interface_ref, obd, exp, + qnids, inode_pending, 1, NULL, 0, + NULL, 0); block_count = (la_tmp->la_blocks + 7) >> 3; - if (block_count) + if (block_count) { + void *data = NULL; + mdd_data_get(env, mdd_obj, &data); /* get block quota for new owner */ lquota_chkquota(mds_quota_interface_ref, obd, - qnids[USRQUOTA], - qnids[GRPQUOTA], - block_count, &block_pending, - NULL, LQUOTA_FLAGS_BLK); + exp, qnids, block_pending, + block_count, NULL, + LQUOTA_FLAGS_BLK, data, 1); + } } } #endif @@ -927,7 +1553,7 @@ static int mdd_attr_set(const struct lu_env *env, struct md_object *obj, } if (rc == 0 && ma->ma_valid & MA_LOV) { - umode_t mode; + cfs_umode_t mode; mode = mdd_object_type(mdd_obj); if (S_ISREG(mode) || S_ISDIR(mode)) { @@ -940,7 +1566,18 @@ static int mdd_attr_set(const struct lu_env *env, struct md_object *obj, } } + if (rc == 0 && ma->ma_valid & (MA_HSM | MA_SOM)) { + cfs_umode_t mode; + + mode = mdd_object_type(mdd_obj); + if (S_ISREG(mode)) + rc = mdd_lma_set_locked(env, mdd_obj, ma, handle); + + } cleanup: + if (rc == 0) + rc = mdd_attr_set_changelog(env, obj, handle, + ma->ma_attr.la_valid); mdd_trans_stop(env, mdd, rc, handle); if (rc == 0 && (lmm != NULL && lmm_size > 0 )) { /*set obd attr, if needed*/ @@ -949,14 +1586,10 @@ cleanup: } #ifdef HAVE_QUOTA_SUPPORT if (quota_opc) { - if (inode_pending) - lquota_pending_commit(mds_quota_interface_ref, obd, - qnids[USRQUOTA], qnids[GRPQUOTA], - 1, 0); - if (block_pending) - lquota_pending_commit(mds_quota_interface_ref, obd, - qnids[USRQUOTA], qnids[GRPQUOTA], - block_count, 1); + lquota_pending_commit(mds_quota_interface_ref, obd, qnids, + inode_pending, 0); + lquota_pending_commit(mds_quota_interface_ref, obd, qnids, + block_pending, 1); /* Trigger dqrel/dqacq for original owner and new owner. * If failed, the next call for lquota_chkquota will * process it. */ @@ -1022,11 +1655,21 @@ static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj, RETURN(rc); mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP); + /* security-replated changes may require sync */ + if (!strcmp(name, XATTR_NAME_ACL_ACCESS) && + mdd->mdd_sync_permission == 1) + txn_param_sync(&mdd_env_info(env)->mti_param); + handle = mdd_trans_start(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); rc = mdd_xattr_set_txn(env, mdd_obj, buf, name, fl, handle); + + /* Only record user xattr changes */ + if ((rc == 0) && (strncmp("user.", name, 5) == 0)) + rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj, + handle); mdd_trans_stop(env, mdd, rc, handle); RETURN(rc); @@ -1058,6 +1701,12 @@ int mdd_xattr_del(const struct lu_env *env, struct md_object *obj, rc = mdo_xattr_del(env, mdd_obj, name, handle, mdd_object_capa(env, mdd_obj)); mdd_write_unlock(env, mdd_obj); + + /* Only record user xattr changes */ + if ((rc == 0) && (strncmp("user.", name, 5) != 0)) + rc = mdd_changelog_data_store(env, mdd, CL_XATTR, 0, mdd_obj, + handle); + mdd_trans_stop(env, mdd, rc, handle); RETURN(rc); @@ -1179,10 +1828,12 @@ static int mdd_object_create(const struct lu_env *env, struct thandle *handle; #ifdef HAVE_QUOTA_SUPPORT struct obd_device *obd = mdd->mdd_obd_dev; + struct obd_export *exp = md_quota(env)->mq_exp; struct mds_obd *mds = &obd->u.mds; unsigned int qids[MAXQUOTAS] = { 0, 0 }; int quota_opc = 0, block_count = 0; - int inode_pending = 0, block_pending = 0; + int inode_pending[MAXQUOTAS] = { 0, 0 }; + int block_pending[MAXQUOTAS] = { 0, 0 }; #endif int rc = 0; ENTRY; @@ -1192,8 +1843,9 @@ static int mdd_object_create(const struct lu_env *env, quota_opc = FSFILT_OP_CREATE_PARTIAL_CHILD; mdd_quota_wrapper(&ma->ma_attr, qids); /* get file quota for child */ - lquota_chkquota(mds_quota_interface_ref, obd, qids[USRQUOTA], - qids[GRPQUOTA], 1, &inode_pending, NULL, 0); + lquota_chkquota(mds_quota_interface_ref, obd, exp, + qids, inode_pending, 1, NULL, 0, + NULL, 0); switch (ma->ma_attr.la_mode & S_IFMT) { case S_IFLNK: case S_IFDIR: @@ -1205,10 +1857,9 @@ static int mdd_object_create(const struct lu_env *env, } /* get block quota for child */ if (block_count) - lquota_chkquota(mds_quota_interface_ref, obd, - qids[USRQUOTA], qids[GRPQUOTA], - block_count, &block_pending, NULL, - LQUOTA_FLAGS_BLK); + lquota_chkquota(mds_quota_interface_ref, obd, exp, + qids, block_pending, block_count, + NULL, LQUOTA_FLAGS_BLK, NULL, 0); } #endif @@ -1222,7 +1873,7 @@ static int mdd_object_create(const struct lu_env *env, if (rc) GOTO(unlock, rc); - rc = mdd_object_create_internal(env, NULL, mdd_obj, ma, handle); + rc = mdd_object_create_internal(env, NULL, mdd_obj, ma, handle, spec); if (rc) GOTO(unlock, rc); @@ -1236,7 +1887,7 @@ static int mdd_object_create(const struct lu_env *env, rc = __mdd_xattr_set(env, mdd_obj, mdd_buf_get_const(env, lmv, lmv_size), - MDS_LMV_MD_NAME, 0, handle); + XATTR_NAME_LMV, 0, handle); if (rc) GOTO(unlock, rc); @@ -1262,7 +1913,8 @@ static int mdd_object_create(const struct lu_env *env, pfid = spec->u.sp_ea.fid; } #endif - rc = mdd_object_initialize(env, pfid, mdd_obj, ma, handle); + rc = mdd_object_initialize(env, pfid, NULL, mdd_obj, ma, handle, + spec); } EXIT; unlock: @@ -1274,18 +1926,14 @@ unlock: out_pending: #ifdef HAVE_QUOTA_SUPPORT if (quota_opc) { - if (inode_pending) - lquota_pending_commit(mds_quota_interface_ref, obd, - qids[USRQUOTA], qids[GRPQUOTA], - 1, 0); - if (block_pending) - lquota_pending_commit(mds_quota_interface_ref, obd, - qids[USRQUOTA], qids[GRPQUOTA], - block_count, 1); + lquota_pending_commit(mds_quota_interface_ref, obd, qids, + inode_pending, 0); + lquota_pending_commit(mds_quota_interface_ref, obd, qids, + block_pending, 1); /* Trigger dqacq on the owner of child. If failed, * the next call for lquota_chkquota will process it. */ lquota_adjust(mds_quota_interface_ref, obd, qids, 0, rc, - FSFILT_OP_CREATE_PARTIAL_CHILD); + quota_opc); } #endif return rc; @@ -1350,7 +1998,7 @@ int accmode(const struct lu_env *env, struct lu_attr *la, int flags) if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND)) res |= MAY_WRITE; if (flags & MDS_FMODE_EXEC) - res |= MAY_EXEC; + res = MAY_EXEC; return res; } @@ -1440,6 +2088,7 @@ int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj, 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), @@ -1454,9 +2103,12 @@ int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj, static int mdd_close(const struct lu_env *env, struct md_object *obj, struct md_attr *ma) { - 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; + int reset = 1; + #ifdef HAVE_QUOTA_SUPPORT struct obd_device *obd = mdo2mdd(obj)->mdd_obd_dev; struct mds_obd *mds = &obd->u.mds; @@ -1465,32 +2117,90 @@ static int mdd_close(const struct lu_env *env, struct md_object *obj, #endif 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: + 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)); + } 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 --; + if (mdd_obj->mod_count == 0 && mdd_obj->mod_flags & ORPHAN_OBJ) { + /* remove link to object from orphan index */ + 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))); + } 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_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); + /* 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) { #ifdef HAVE_QUOTA_SUPPORT if (mds->mds_quota) { quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD; mdd_quota_wrapper(&ma->ma_attr, qids); } #endif - } else { - ma->ma_valid &= ~(MA_LOV | MA_COOKIE); + /* MDS_CLOSE_CLEANUP means destroy OSS objects by MDS. */ + if (ma->ma_valid & MA_FLAGS && + ma->ma_attr_flags & MDS_CLOSE_CLEANUP) { + rc = mdd_lov_destroy(env, mdd, mdd_obj, &ma->ma_attr); + } else { + rc = mdd_object_kill(env, mdd_obj, ma); + if (rc == 0) + reset = 0; + } + + 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: + if (reset) + ma->ma_valid &= ~(MA_LOV | MA_COOKIE); mdd_write_unlock(env, mdd_obj); - mdd_trans_stop(env, mdo2mdd(obj), rc, handle); + if (handle != NULL) + mdd_trans_stop(env, mdo2mdd(obj), rc, handle); #ifdef HAVE_QUOTA_SUPPORT if (quota_opc) /* Trigger dqrel on the owner of child. If failed, @@ -1498,7 +2208,7 @@ static int mdd_close(const struct lu_env *env, struct md_object *obj, lquota_adjust(mds_quota_interface_ref, obd, qids, 0, rc, quota_opc); #endif - RETURN(rc); + return rc; } /* @@ -1520,15 +2230,14 @@ 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, const 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, struct mdd_device *mdd, + int first, void *area, int nob, + const struct dt_it_ops *iops, struct dt_it *it, + __u64 *start, __u64 *end, + struct lu_dirent **last, __u32 attr) { - 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; + __u64 hash = 0; struct lu_dirent *ent; if (first) { @@ -1537,56 +2246,62 @@ static int mdd_dir_page_build(const struct lu_env *env, int first, nob -= sizeof (struct lu_dirpage); } - LASSERT(nob > sizeof *ent); - 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); - 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; + *start = 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, 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; + if (*last) { + (*last)->lde_reclen = + cpu_to_le16(le16_to_cpu((*last)->lde_reclen) + + nob); + result = 0; + } else + result = -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); +out: + *end = hash; return result; } @@ -1598,6 +2313,7 @@ static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj, const struct dt_it_ops *iops; struct page *pg; struct lu_dirent *last = NULL; + struct mdd_device *mdd = mdo2mdd(&obj->mod_obj); int i; int rc; int nob; @@ -1614,13 +2330,13 @@ static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj, * 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)); + it = iops->init(env, next, rdpg->rp_attrs, mdd_object_capa(env, obj)); if (IS_ERR(it)) return PTR_ERR(it); rc = iops->load(env, it, rdpg->rp_hash); - if (rc == 0) + if (rc == 0){ /* * Iterator didn't find record with exactly the key requested. * @@ -1633,7 +2349,7 @@ static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj, * state)---position it on the next item. */ rc = iops->next(env, it); - else if (rc > 0) + } else if (rc > 0) rc = 0; /* @@ -1647,31 +2363,35 @@ static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj, 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), + rc = mdd_dir_page_build(env, mdd, !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; + it, &hash_start, &hash_end, &last, + rdpg->rp_attrs); + if (rc != 0 || i == rdpg->rp_npages - 1) { + if (last) + last->lde_reclen = 0; + } cfs_kunmap(pg); } if (rc > 0) { /* * end of directory. */ - hash_end = DIR_END_OFF; + hash_end = MDS_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; + dp->ldp_hash_start = cpu_to_le64(rdpg->rp_hash); + dp->ldp_hash_end = cpu_to_le64(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]); } @@ -1681,8 +2401,8 @@ static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj, return rc; } -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; @@ -1713,8 +2433,8 @@ 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_hash_start = cpu_to_le64(rdpg->rp_hash); + dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF); dp->ldp_flags |= LDF_EMPTY; dp->ldp_flags = cpu_to_le32(dp->ldp_flags); cfs_kunmap(pg); @@ -1739,6 +2459,24 @@ static int mdd_object_sync(const struct lu_env *env, struct md_object *obj) return next->do_ops->do_object_sync(env, next); } +static dt_obj_version_t mdd_version_get(const struct lu_env *env, + struct md_object *obj) +{ + struct mdd_object *mdd_obj = md2mdd_obj(obj); + + LASSERT(mdd_object_exists(mdd_obj)); + return do_version_get(env, mdd_object_child(mdd_obj)); +} + +static void mdd_version_set(const struct lu_env *env, struct md_object *obj, + dt_obj_version_t version) +{ + struct mdd_object *mdd_obj = md2mdd_obj(obj); + + LASSERT(mdd_object_exists(mdd_obj)); + do_version_set(env, mdd_object_child(mdd_obj), version); +} + const struct md_object_operations mdd_obj_ops = { .moo_permission = mdd_permission, .moo_attr_get = mdd_attr_get, @@ -1754,6 +2492,12 @@ const struct md_object_operations mdd_obj_ops = { .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_version_get = mdd_version_get, + .moo_version_set = mdd_version_set, + .moo_path = mdd_path, + .moo_file_lock = mdd_file_lock, + .moo_file_unlock = mdd_file_unlock, };