X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_object.c;h=b6c8126440868ef90d58c7a01fea8e2435cd8f59;hp=c3f695847c08ea06644fba22b68da44ef26deaf3;hb=473bc6473eb8d6d78069a561dbe035e795d8b7e2;hpb=8f249ade28ac5cd3a9d51e1a641e2ea9e8d98084 diff --git a/lustre/mdd/mdd_object.c b/lustre/mdd/mdd_object.c index c3f6958..b6c8126 100644 --- a/lustre/mdd/mdd_object.c +++ b/lustre/mdd/mdd_object.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel Corporation. + * Copyright (c) 2011, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -49,8 +49,7 @@ #include /* fid_be_cpu(), fid_cpu_to_be(). */ #include -#include - +#include #include #include #include @@ -58,50 +57,28 @@ #include "mdd_internal.h" 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) + struct lu_attr *la) { 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); -} -static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags) -{ - obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ); - - if (flags & LUSTRE_APPEND_FL) - obj->mod_flags |= APPEND_OBJ; - - if (flags & LUSTRE_IMMUTABLE_FL) - obj->mod_flags |= IMMUTE_OBJ; + return mdo_attr_get(env, obj, la); } struct mdd_thread_info *mdd_env_info(const struct lu_env *env) { struct mdd_thread_info *info; + lu_env_refill((struct lu_env *)env); info = lu_context_key_get(&env->le_ctx, &mdd_thread_key); LASSERT(info != NULL); return info; @@ -109,96 +86,45 @@ struct mdd_thread_info *mdd_env_info(const struct lu_env *env) struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len) { - struct lu_buf *buf; - - buf = &mdd_env_info(env)->mti_buf; - buf->lb_buf = area; - buf->lb_len = len; - return buf; -} + struct lu_buf *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 = LU_BUF_NULL; + buf = &mdd_env_info(env)->mti_buf[0]; + buf->lb_buf = area; + buf->lb_len = len; + return buf; } 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; + struct lu_buf *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; - } + buf = &mdd_env_info(env)->mti_buf[0]; + buf->lb_buf = (void *)area; + buf->lb_len = len; 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 lu_object *mdd_object_alloc(const struct lu_env *env, - const struct lu_object_header *hdr, - struct lu_device *d) + const struct lu_object_header *hdr, + struct lu_device *d) { - struct mdd_object *mdd_obj; - - OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, CFS_ALLOC_IO); - if (mdd_obj != NULL) { - struct lu_object *o; - - o = mdd2lu_obj(mdd_obj); - lu_object_init(o, NULL, d); - mdd_obj->mod_obj.mo_ops = &mdd_obj_ops; - mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops; - mdd_obj->mod_count = 0; - o->lo_ops = &mdd_lu_obj_ops; - return o; - } else { - return NULL; - } + struct mdd_object *mdd_obj; + + OBD_SLAB_ALLOC_PTR_GFP(mdd_obj, mdd_object_kmem, GFP_NOFS); + if (mdd_obj != NULL) { + struct lu_object *o; + + o = mdd2lu_obj(mdd_obj); + lu_object_init(o, NULL, d); + mdd_obj->mod_obj.mo_ops = &mdd_obj_ops; + mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops; + mdd_obj->mod_count = 0; + o->lo_ops = &mdd_lu_obj_ops; + return o; + } else { + return NULL; + } } static int mdd_object_init(const struct lu_env *env, struct lu_object *o, @@ -213,9 +139,8 @@ static int mdd_object_init(const struct lu_env *env, struct lu_object *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); + if (IS_ERR(below)) + RETURN(PTR_ERR(below)); lu_object_add(o, below); @@ -224,10 +149,16 @@ static int mdd_object_init(const struct lu_env *env, struct lu_object *o, static int mdd_object_start(const struct lu_env *env, struct lu_object *o) { - if (lu_object_exists(o)) - return mdd_get_flags(env, lu2mdd_obj(o)); - else - return 0; + int rc = 0; + + if (lu_object_exists(o)) { + struct mdd_object *mdd_obj = lu2mdd_obj(o); + struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start); + + rc = mdd_la_get(env, mdd_obj, attr); + } + + return rc; } static void mdd_object_free(const struct lu_env *env, struct lu_object *o) @@ -262,271 +193,22 @@ 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. */ - 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); - } - 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; - int 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. */ int mdd_attr_get(const struct lu_env *env, struct md_object *obj, struct md_attr *ma) { - int rc; - ENTRY; + struct mdd_object *mdd_obj = md2mdd_obj(obj); + int rc; - return mdd_la_get(env, md2mdd_obj(obj), &ma->ma_attr, - mdd_object_capa(env, md2mdd_obj(obj))); - RETURN(rc); + ENTRY; + + rc = mdd_la_get(env, mdd_obj, &ma->ma_attr); + if ((ma->ma_need & MA_INODE) != 0 && mdd_is_dead_obj(mdd_obj)) + ma->ma_attr.la_nlink = 0; + + RETURN(rc); } /* @@ -547,9 +229,14 @@ static int mdd_xattr_get(const struct lu_env *env, return -ENOENT; } + /* If the object has been delete from the namespace, then + * get linkEA should return -ENOENT as well */ + if (unlikely((mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ)) && + strcmp(name, XATTR_NAME_LINK) == 0)) + RETURN(-ENOENT); + mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = mdo_xattr_get(env, mdd_obj, buf, name, - mdd_object_capa(env, mdd_obj)); + rc = mdo_xattr_get(env, mdd_obj, buf, name); mdd_read_unlock(env, mdd_obj); RETURN(rc); @@ -559,8 +246,8 @@ static int mdd_xattr_get(const struct lu_env *env, * Permission check is done when open, * no need check again. */ -static int mdd_readlink(const struct lu_env *env, struct md_object *obj, - struct lu_buf *buf) +int mdd_readlink(const struct lu_env *env, struct md_object *obj, + struct lu_buf *buf) { struct mdd_object *mdd_obj = md2mdd_obj(obj); struct dt_object *next; @@ -575,9 +262,11 @@ static int mdd_readlink(const struct lu_env *env, struct md_object *obj, } next = mdd_object_child(mdd_obj); - mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = next->do_body_ops->dbo_read(env, next, buf, &pos, - mdd_object_capa(env, mdd_obj)); + LASSERT(next != NULL); + LASSERT(next->do_body_ops != NULL); + LASSERT(next->do_body_ops->dbo_read != NULL); + mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); + rc = dt_read(env, next, buf, &pos); mdd_read_unlock(env, mdd_obj); RETURN(rc); } @@ -594,10 +283,38 @@ static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj, ENTRY; mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = mdo_xattr_list(env, mdd_obj, buf, mdd_object_capa(env, mdd_obj)); + rc = mdo_xattr_list(env, mdd_obj, buf); mdd_read_unlock(env, mdd_obj); - RETURN(rc); + if (rc < 0) + RETURN(rc); + + /* + * Filter out XATTR_NAME_LINK if this is an orphan object. See + * mdd_xattr_get(). + */ + if (unlikely(mdd_obj->mod_flags & (DEAD_OBJ | ORPHAN_OBJ))) { + char *end = (char *)buf->lb_buf + rc; + char *p = buf->lb_buf; + + while (p < end) { + char *next = p + strlen(p) + 1; + + if (strcmp(p, XATTR_NAME_LINK) == 0) { + if (end - next > 0) + memmove(p, next, end - next); + rc -= next - p; + CDEBUG(D_INFO, "Filtered out "XATTR_NAME_LINK + " of orphan "DFID"\n", + PFID(mdd_object_fid(mdd_obj))); + break; + } + + p = next; + } + } + + RETURN(rc); } int mdd_declare_object_create_internal(const struct lu_env *env, @@ -605,7 +322,8 @@ int mdd_declare_object_create_internal(const struct lu_env *env, struct mdd_object *c, struct lu_attr *attr, struct thandle *handle, - const struct md_op_spec *spec) + const struct md_op_spec *spec, + struct dt_allocation_hint *hint) { struct dt_object_format *dof = &mdd_env_info(env)->mti_dof; const struct dt_index_features *feat = spec->sp_feat; @@ -629,17 +347,17 @@ int mdd_declare_object_create_internal(const struct lu_env *env, } } - rc = mdo_declare_create_obj(env, c, attr, NULL, dof, handle); + rc = mdo_declare_create_obj(env, c, attr, hint, dof, handle); RETURN(rc); } int mdd_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 thandle *handle, + const struct md_op_spec *spec, + struct dt_allocation_hint *hint) { - 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; @@ -648,44 +366,17 @@ int mdd_object_create_internal(const struct lu_env *env, struct mdd_object *p, rc = mdo_create_obj(env, c, attr, hint, dof, handle); - LASSERT(ergo(rc == 0, mdd_object_exists(c))); - RETURN(rc); } -/** - * Make sure the ctime is increased only. - */ -static inline int mdd_attr_check(const struct lu_env *env, - struct mdd_object *obj, - struct lu_attr *attr) -{ - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; - int rc; - ENTRY; - - if (attr->la_valid & LA_CTIME) { - rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA); - if (rc) - RETURN(rc); - - if (attr->la_ctime < tmp_la->la_ctime) - attr->la_valid &= ~(LA_MTIME | LA_CTIME); - else if (attr->la_valid == LA_CTIME && - attr->la_ctime == tmp_la->la_ctime) - attr->la_valid &= ~LA_CTIME; - } - RETURN(0); -} - int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj, - struct lu_attr *attr, struct thandle *handle, + const struct lu_attr *attr, struct thandle *handle, int needacl) { int rc; ENTRY; - rc = mdo_attr_set(env, obj, attr, handle, mdd_object_capa(env, obj)); + rc = mdo_attr_set(env, obj, attr, handle); #ifdef CONFIG_FS_POSIX_ACL if (!rc && (attr->la_valid & LA_MODE) && needacl) rc = mdd_acl_chmod(env, obj, attr->la_mode, handle); @@ -693,20 +384,29 @@ int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *obj, RETURN(rc); } -int mdd_attr_check_set_internal(const struct lu_env *env, - struct mdd_object *obj, struct lu_attr *attr, - struct thandle *handle, int needacl) +int mdd_update_time(const struct lu_env *env, struct mdd_object *obj, + const struct lu_attr *oattr, struct lu_attr *attr, + struct thandle *handle) { - int rc; - ENTRY; - - rc = mdd_attr_check(env, obj, attr); - if (rc) - RETURN(rc); + int rc = 0; + ENTRY; - if (attr->la_valid) - rc = mdd_attr_set_internal(env, obj, attr, handle, needacl); - RETURN(rc); + LASSERT(attr->la_valid & LA_CTIME); + LASSERT(oattr != NULL); + + /* Make sure the ctime is increased only, however, it's not strictly + * reliable at here because there is not guarantee to hold lock on + * object, so we just bypass some unnecessary cmtime setting first + * and OSD has to check it again. */ + if (attr->la_ctime < oattr->la_ctime) + attr->la_valid &= ~(LA_MTIME | LA_CTIME); + else if (attr->la_valid == LA_CTIME && + attr->la_ctime == oattr->la_ctime) + attr->la_valid &= ~LA_CTIME; + + if (attr->la_valid != 0) + rc = mdd_attr_set_internal(env, obj, attr, handle, 0); + RETURN(rc); } /* @@ -717,241 +417,214 @@ int mdd_attr_check_set_internal(const struct lu_env *env, * 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 unsigned long flags) + const struct lu_attr *oattr, struct lu_attr *la, + const unsigned long flags) { - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; struct lu_ucred *uc; - int rc; - ENTRY; + int rc = 0; + ENTRY; - if (!la->la_valid) - RETURN(0); + if (!la->la_valid) + RETURN(0); + + /* Do not permit change file type */ + if (la->la_valid & LA_TYPE) + RETURN(-EPERM); - /* Do not permit change file type */ - if (la->la_valid & LA_TYPE) - RETURN(-EPERM); + /* They should not be processed by setattr */ + if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE)) + RETURN(-EPERM); - /* They should not be processed by setattr */ - if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE)) - RETURN(-EPERM); + LASSERT(oattr != NULL); - /* 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 (la->la_valid == LA_CTIME) { 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, tmp_la, NULL, 1, 0); - if (rc == 0 && la->la_ctime <= tmp_la->la_ctime) - la->la_valid &= ~LA_CTIME; - RETURN(rc); - } + /* This is only for set ctime when rename's source is + * on remote MDS. */ + rc = mdd_may_delete(env, NULL, NULL, obj, oattr, NULL, + 1, 0); + if (rc == 0 && la->la_ctime <= oattr->la_ctime) + la->la_valid &= ~LA_CTIME; + RETURN(rc); + } - 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 && - la->la_atime < (tmp_la->la_atime + - mdd_obj2mdd_dev(obj)->mdd_atime_diff)) - la->la_valid &= ~LA_ATIME; - RETURN(0); - } + if (la->la_valid == LA_ATIME) { + /* This is atime only set for read atime update on close. */ + if (la->la_atime >= oattr->la_atime && + la->la_atime < (oattr->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); + /* Check if flags change. */ + if (la->la_valid & LA_FLAGS) { + unsigned int oldflags = oattr->la_flags & + (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL); + unsigned int newflags = la->la_flags & + (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL); - if ((uc->uc_fsuid != tmp_la->la_uid) && - !mdd_capable(uc, CFS_CAP_FOWNER)) + if ((uc->uc_fsuid != oattr->la_uid) && + !md_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; - if ((oldflags ^ newflags) && - !mdd_capable(uc, CFS_CAP_LINUX_IMMUTABLE)) - RETURN(-EPERM); - - if (!S_ISDIR(tmp_la->la_mode)) - la->la_flags &= ~LUSTRE_DIRSYNC_FL; - } + /* The IMMUTABLE and APPEND_ONLY flags can + * only be changed by the relevant capability. */ + if ((oldflags ^ newflags) && + !md_capable(uc, CFS_CAP_LINUX_IMMUTABLE)) + RETURN(-EPERM); + + if (!S_ISDIR(oattr->la_mode)) + la->la_flags &= ~LUSTRE_DIRSYNC_FL; + } - if ((mdd_is_immutable(obj) || mdd_is_append(obj)) && - (la->la_valid & ~LA_FLAGS) && + if (oattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL) && + (la->la_valid & ~LA_FLAGS) && !(flags & MDS_PERM_BYPASS)) - RETURN(-EPERM); + 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->uc_fsuid != tmp_la->la_uid) && - !mdd_capable(uc, CFS_CAP_FOWNER)) { - rc = mdd_permission_internal(env, obj, tmp_la, + if ((uc->uc_fsuid != oattr->la_uid) && + !md_capable(uc, CFS_CAP_FOWNER)) { + rc = mdd_permission_internal(env, obj, oattr, 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_SUID) { + la->la_valid &= ~LA_KILL_SUID; + if ((oattr->la_mode & S_ISUID) && + !(la->la_valid & LA_MODE)) { + la->la_mode = oattr->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; - } + if (la->la_valid & LA_KILL_SGID) { + la->la_valid &= ~LA_KILL_SGID; + if (((oattr->la_mode & (S_ISGID | S_IXGRP)) == + (S_ISGID | S_IXGRP)) && + !(la->la_valid & LA_MODE)) { + la->la_mode = oattr->la_mode; + la->la_valid |= LA_MODE; + } + la->la_mode &= ~S_ISGID; + } - /* Make sure a caller can chmod. */ - if (la->la_valid & LA_MODE) { + /* Make sure a caller can chmod. */ + if (la->la_valid & LA_MODE) { if (!(flags & MDS_PERM_BYPASS) && - (uc->uc_fsuid != tmp_la->la_uid) && - !mdd_capable(uc, CFS_CAP_FOWNER)) + (uc->uc_fsuid != oattr->la_uid) && + !md_capable(uc, CFS_CAP_FOWNER)) RETURN(-EPERM); - if (la->la_mode == (cfs_umode_t) -1) - la->la_mode = tmp_la->la_mode; - else - la->la_mode = (la->la_mode & S_IALLUGO) | - (tmp_la->la_mode & ~S_IALLUGO); - - /* Also check the setgid bit! */ - if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ? - la->la_gid : tmp_la->la_gid) && - !mdd_capable(uc, CFS_CAP_FSETID)) - la->la_mode &= ~S_ISGID; - } else { - la->la_mode = tmp_la->la_mode; - } + if (la->la_mode == (umode_t) -1) + la->la_mode = oattr->la_mode; + else + la->la_mode = (la->la_mode & S_IALLUGO) | + (oattr->la_mode & ~S_IALLUGO); + + /* Also check the setgid bit! */ + if (!lustre_in_group_p(uc, (la->la_valid & LA_GID) ? + la->la_gid : oattr->la_gid) && + !md_capable(uc, CFS_CAP_FSETID)) + la->la_mode &= ~S_ISGID; + } else { + la->la_mode = oattr->la_mode; + } - /* Make sure a caller can chown. */ - if (la->la_valid & LA_UID) { - if (la->la_uid == (uid_t) -1) - la->la_uid = tmp_la->la_uid; - if (((uc->uc_fsuid != tmp_la->la_uid) || - (la->la_uid != tmp_la->la_uid)) && - !mdd_capable(uc, CFS_CAP_CHOWN)) + /* Make sure a caller can chown. */ + if (la->la_valid & LA_UID) { + if (la->la_uid == (uid_t) -1) + la->la_uid = oattr->la_uid; + if (((uc->uc_fsuid != oattr->la_uid) || + (la->la_uid != oattr->la_uid)) && + !md_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. - * 19981026 David C Niemi - * - * Changed this to apply to all users, including root, - * to avoid some races. This is the behavior we had in - * 2.0. The check for non-root was definitely wrong - * for 2.2 anyway, as it should have been using - * CAP_FSETID rather than fsuid -- 19990830 SD. */ - if (((tmp_la->la_mode & S_ISUID) == S_ISUID) && - !S_ISDIR(tmp_la->la_mode)) { - la->la_mode &= ~S_ISUID; - la->la_valid |= LA_MODE; - } - } + /* If the user or group of a non-directory has been + * changed by a non-root user, remove the setuid bit. + * 19981026 David C Niemi + * + * Changed this to apply to all users, including root, + * to avoid some races. This is the behavior we had in + * 2.0. The check for non-root was definitely wrong + * for 2.2 anyway, as it should have been using + * CAP_FSETID rather than fsuid -- 19990830 SD. */ + if (((oattr->la_mode & S_ISUID) == S_ISUID) && + !S_ISDIR(oattr->la_mode)) { + la->la_mode &= ~S_ISUID; + la->la_valid |= LA_MODE; + } + } - /* Make sure caller can chgrp. */ - if (la->la_valid & LA_GID) { - if (la->la_gid == (gid_t) -1) - la->la_gid = tmp_la->la_gid; - if (((uc->uc_fsuid != tmp_la->la_uid) || - ((la->la_gid != tmp_la->la_gid) && + /* Make sure caller can chgrp. */ + if (la->la_valid & LA_GID) { + if (la->la_gid == (gid_t) -1) + la->la_gid = oattr->la_gid; + if (((uc->uc_fsuid != oattr->la_uid) || + ((la->la_gid != oattr->la_gid) && !lustre_in_group_p(uc, la->la_gid))) && - !mdd_capable(uc, CFS_CAP_CHOWN)) + !md_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 - * setgid bit UNLESS there is no group execute bit - * (this would be a file marked for mandatory - * locking). 19981026 David C Niemi - * - * Removed the fsuid check (see the comment above) -- - * 19990830 SD. */ - if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) == - (S_ISGID | S_IXGRP)) && !S_ISDIR(tmp_la->la_mode)) { - la->la_mode &= ~S_ISGID; - la->la_valid |= LA_MODE; - } - } + /* Likewise, if the user or group of a non-directory + * has been changed by a non-root user, remove the + * setgid bit UNLESS there is no group execute bit + * (this would be a file marked for mandatory + * locking). 19981026 David C Niemi + * + * Removed the fsuid check (see the comment above) -- + * 19990830 SD. */ + if (((oattr->la_mode & (S_ISGID | S_IXGRP)) == + (S_ISGID | S_IXGRP)) && !S_ISDIR(oattr->la_mode)) { + la->la_mode &= ~S_ISGID; + la->la_valid |= LA_MODE; + } + } - /* For both Size-on-MDS case and truncate case, - * "la->la_valid & (LA_SIZE | LA_BLOCKS)" are ture. - * 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 (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) && - (la->la_atime <= tmp_la->la_atime)) - la->la_valid &= ~LA_ATIME; - - /* OST attributes do not have a priority over MDS attributes, - * so drop times if ctime is equal. */ - if ((la->la_valid & LA_CTIME) && - (la->la_ctime <= tmp_la->la_ctime)) - la->la_valid &= ~(LA_MTIME | LA_CTIME); - } else { - if (la->la_valid & (LA_SIZE | LA_BLOCKS)) { - 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); - } - } - if (la->la_valid & LA_CTIME) { - /* The pure setattr, it has the priority over what is - * already set, do not drop it if ctime is equal. */ - if (la->la_ctime < tmp_la->la_ctime) - la->la_valid &= ~(LA_ATIME | LA_MTIME | - LA_CTIME); - } - } + if (la->la_valid & (LA_SIZE | LA_BLOCKS)) { + if (!((flags & MDS_OWNEROVERRIDE) && + (uc->uc_fsuid == oattr->la_uid)) && + !(flags & MDS_PERM_BYPASS)) { + rc = mdd_permission_internal(env, obj, oattr, + MAY_WRITE); + if (rc != 0) + RETURN(rc); + } + } - RETURN(0); + if (la->la_valid & LA_CTIME) { + /* The pure setattr, it has the priority over what is + * already set, do not drop it if ctime is equal. */ + if (la->la_ctime < oattr->la_ctime) + la->la_valid &= ~(LA_ATIME | LA_MTIME | LA_CTIME); + } + + 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 + * \param handle - transaction 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) +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); + const struct lu_ucred *uc = lu_ucred(env); + const struct lu_fid *tfid; struct llog_changelog_rec *rec; struct lu_buf *buf; int reclen; @@ -966,6 +639,8 @@ static int mdd_changelog_data_store(const struct lu_env *env, LASSERT(mdd_obj != NULL); LASSERT(handle != NULL); + tfid = mdo2fid(mdd_obj); + 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 */ @@ -974,25 +649,32 @@ static int mdd_changelog_data_store(const struct lu_env *env, RETURN(0); } - reclen = llog_data_len(sizeof(*rec)); - buf = mdd_buf_alloc(env, reclen); - if (buf->lb_buf == NULL) - RETURN(-ENOMEM); + flags = (flags & CLF_FLAGMASK) | CLF_VERSION; + if (uc != NULL && uc->uc_jobid[0] != '\0') + flags |= CLF_JOBID; + + reclen = llog_data_len(changelog_rec_offset(flags & CLF_SUPPORTED)); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, 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(); + rec->cr.cr_flags = 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(); + + if (flags & CLF_JOBID) + mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid); 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) +static 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); @@ -1004,9 +686,9 @@ int mdd_changelog(const struct lu_env *env, enum changelog_rec_type type, if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); - rc = mdd_declare_changelog_store(env, mdd, NULL, handle); - if (rc) - GOTO(stop, rc); + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle); + if (rc) + GOTO(stop, rc); rc = mdd_trans_start(env, mdd, handle); if (rc) @@ -1039,19 +721,19 @@ static int mdd_attr_set_changelog(const struct lu_env *env, 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_SIZE) ? 1 << CL_TRUNC : 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; + /* This is an implementation limit rather than a protocol limit */ + CLASSERT(CL_LAST <= sizeof(int) * 8); if (bits == 0) return 0; /* The record type is the lowest non-masked set bit */ - while (bits && ((bits & 1) == 0)) { - bits = bits >> 1; - type++; - } + type = __ffs(bits); /* FYI we only store the first CLF_FLAGMASK bits of la_valid */ return mdd_changelog_data_store(env, mdd, type, (int)valid, @@ -1074,7 +756,7 @@ static int mdd_declare_attr_set(const struct lu_env *env, 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); + XATTR_NAME_ACL_ACCESS); mdd_read_unlock(env, obj); if (rc == -EOPNOTSUPP || rc == -ENODATA) rc = 0; @@ -1092,31 +774,66 @@ static int mdd_declare_attr_set(const struct lu_env *env, } #endif - rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle); return rc; } +/* + * LU-3671 + * + * permission changes may require sync operation, to mitigate performance + * impact, only do this for dir and when permission is reduced. + * + * For regular files, version is updated with permission change (see VBR), async + * permission won't cause any issue, while missing permission change on + * directory may affect accessibility of other objects after recovery. + */ +static inline bool permission_needs_sync(const struct lu_attr *old, + const struct lu_attr *new) +{ + if (!S_ISDIR(old->la_mode)) + return false; + + if (new->la_valid & (LA_UID | LA_GID)) + return true; + + if (new->la_valid & LA_MODE && + new->la_mode & (S_ISUID | S_ISGID | S_ISVTX)) + return true; + + if ((new->la_valid & LA_MODE) && + ((new->la_mode & old->la_mode) & S_IRWXUGO) != + (old->la_mode & S_IRWXUGO)) + return true; + + return false; +} + /* set attr and LOV EA at once, return updated attr */ 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 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; + struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix; + struct lu_attr *attr = MDD_ENV_VAR(env, cattr); const struct lu_attr *la = &ma->ma_attr; int rc; - ENTRY; + ENTRY; - /* we do not use ->attr_set() for LOV/SOM/HSM EA any more */ + /* we do not use ->attr_set() for LOV/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); + rc = mdd_la_get(env, mdd_obj, attr); if (rc) - RETURN(rc); + RETURN(rc); + + *la_copy = ma->ma_attr; + rc = mdd_fix_attr(env, mdd_obj, attr, 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) @@ -1134,50 +851,44 @@ int mdd_attr_set(const struct lu_env *env, struct md_object *obj, if (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 (mdd->mdd_sync_permission && permission_needs_sync(attr, la)) + handle->th_sync = 1; 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) { + mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); + if (la_copy->la_valid & LA_FLAGS) 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 */ + else if (la_copy->la_valid) /* setattr */ rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 1); - } + mdd_write_unlock(env, mdd_obj); + + if (rc == 0) + rc = mdd_attr_set_changelog(env, obj, handle, la->la_valid); + + GOTO(stop, rc); - if (rc == 0) - rc = mdd_attr_set_changelog(env, obj, handle, - la->la_valid); stop: - mdd_trans_stop(env, mdd, rc, handle); - RETURN(rc); + mdd_trans_stop(env, mdd, rc, handle); + return rc; } static int mdd_xattr_sanity_check(const struct lu_env *env, - struct mdd_object *obj) + struct mdd_object *obj, + const struct lu_attr *attr) { - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; struct lu_ucred *uc = lu_ucred_assert(env); - int rc; ENTRY; - if (mdd_is_immutable(obj) || mdd_is_append(obj)) + if (attr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL)) RETURN(-EPERM); - 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)) + if ((uc->uc_fsuid != attr->la_uid) && !md_capable(uc, CFS_CAP_FOWNER)) RETURN(-EPERM); - RETURN(rc); + RETURN(0); } static int mdd_declare_xattr_set(const struct lu_env *env, @@ -1193,14 +904,91 @@ static int mdd_declare_xattr_set(const struct lu_env *env, if (rc) return rc; - /* Only record user xattr changes */ - if ((strncmp("user.", name, 5) == 0)) - rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + /* Only record user and layout xattr changes */ + if (strncmp(XATTR_USER_PREFIX, name, + sizeof(XATTR_USER_PREFIX) - 1) == 0 || + strcmp(XATTR_NAME_LOV, name) == 0) { + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle); + if (rc) + return rc; + } + + /* If HSM data is modified, this could add a changelog */ + if (strcmp(XATTR_NAME_HSM, name) == 0) { + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle); + if (rc) + return rc; + } - rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle); return rc; } +/* + * Compare current and future data of HSM EA and add a changelog if needed. + * + * Caller should have write-locked \param obj. + * + * \param buf - Future HSM EA content. + * \retval 0 if no changelog is needed or changelog was added properly. + * \retval -ve errno if there was a problem + */ +static int mdd_hsm_update_locked(const struct lu_env *env, + struct md_object *obj, + const struct lu_buf *buf, + struct thandle *handle) +{ + struct mdd_thread_info *info = mdd_env_info(env); + struct mdd_device *mdd = mdo2mdd(obj); + struct mdd_object *mdd_obj = md2mdd_obj(obj); + struct lu_buf *current_buf; + struct md_hsm *current_mh; + struct md_hsm *new_mh; + int rc; + ENTRY; + + OBD_ALLOC_PTR(current_mh); + if (current_mh == NULL) + RETURN(-ENOMEM); + + /* Read HSM attrs from disk */ + CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf)); + current_buf = mdd_buf_get(env, info->mti_xattr_buf, + sizeof(info->mti_xattr_buf)); + rc = mdo_xattr_get(env, mdd_obj, current_buf, XATTR_NAME_HSM); + rc = lustre_buf2hsm(current_buf->lb_buf, rc, current_mh); + if (rc < 0 && rc != -ENODATA) + GOTO(free, rc); + else if (rc == -ENODATA) + current_mh->mh_flags = 0; + + /* Map future HSM xattr */ + OBD_ALLOC_PTR(new_mh); + if (new_mh == NULL) + GOTO(free, rc = -ENOMEM); + lustre_buf2hsm(buf->lb_buf, buf->lb_len, new_mh); + + /* If HSM flags are different, add a changelog */ + rc = 0; + if (current_mh->mh_flags != new_mh->mh_flags) { + int flags = 0; + hsm_set_cl_event(&flags, HE_STATE); + if (new_mh->mh_flags & HS_DIRTY) + hsm_set_cl_flags(&flags, CLF_HSM_DIRTY); + + rc = mdd_changelog_data_store(env, mdd, CL_HSM, flags, mdd_obj, + handle); + } + + OBD_FREE_PTR(new_mh); +free: + OBD_FREE_PTR(current_mh); + return(rc); +} + +static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj, + const char *name); + /** * The caller should guarantee to update the object ctime * after xattr_set if needed. @@ -1210,25 +998,45 @@ static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj, int fl) { struct mdd_object *mdd_obj = md2mdd_obj(obj); + struct lu_attr *attr = MDD_ENV_VAR(env, cattr); struct mdd_device *mdd = mdo2mdd(obj); struct thandle *handle; int rc; - ENTRY; + ENTRY; - if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) { - rc = mdd_acl_set(env, mdd_obj, buf, fl); + rc = mdd_la_get(env, mdd_obj, attr); + if (rc) RETURN(rc); - } - rc = mdd_xattr_sanity_check(env, mdd_obj); + rc = mdd_xattr_sanity_check(env, mdd_obj, attr); if (rc) RETURN(rc); + if (strcmp(name, XATTR_NAME_ACL_ACCESS) == 0 || + strcmp(name, XATTR_NAME_ACL_DEFAULT) == 0) { + struct posix_acl *acl; + + /* user may set empty ACL, which should be treated as removing + * ACL. */ + acl = posix_acl_from_xattr(&init_user_ns, buf->lb_buf, + buf->lb_len); + if (acl == NULL) { + rc = mdd_xattr_del(env, obj, name); + RETURN(rc); + } + posix_acl_release(acl); + } + + if (!strcmp(name, XATTR_NAME_ACL_ACCESS)) { + rc = mdd_acl_set(env, mdd_obj, attr, buf, fl); + RETURN(rc); + } + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); - rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, 0, handle); + rc = mdd_declare_xattr_set(env, mdd, mdd_obj, buf, name, fl, handle); if (rc) GOTO(stop, rc); @@ -1236,19 +1044,25 @@ static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj, 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)); + + if (strcmp(XATTR_NAME_HSM, name) == 0) { + rc = mdd_hsm_update_locked(env, obj, buf, handle); + if (rc) { + mdd_write_unlock(env, mdd_obj); + GOTO(stop, rc); + } + } + + rc = mdo_xattr_set(env, mdd_obj, buf, name, fl, handle); mdd_write_unlock(env, mdd_obj); if (rc) GOTO(stop, rc); - /* Only record system & user xattr changes */ - if (strncmp(XATTR_USER_PREFIX, name, + if (strcmp(XATTR_NAME_LOV, name) == 0) + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, mdd_obj, + handle); + else 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 || @@ -1269,35 +1083,41 @@ static int mdd_declare_xattr_del(const struct lu_env *env, const char *name, struct thandle *handle) { - int rc; + int rc; - rc = mdo_declare_xattr_del(env, obj, name, handle); - if (rc) - return 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); + /* Only record user xattr changes */ + if ((strncmp(XATTR_USER_PREFIX, name, + sizeof(XATTR_USER_PREFIX) - 1) == 0)) + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle); - return rc; + return rc; } /** * The caller should guarantee to update the object ctime * after xattr_set if needed. */ -int mdd_xattr_del(const struct lu_env *env, struct md_object *obj, - const char *name) +static int mdd_xattr_del(const struct lu_env *env, struct md_object *obj, + const char *name) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); - struct mdd_device *mdd = mdo2mdd(obj); - struct thandle *handle; - int rc; - ENTRY; + struct mdd_object *mdd_obj = md2mdd_obj(obj); + struct lu_attr *attr = MDD_ENV_VAR(env, cattr); + struct mdd_device *mdd = mdo2mdd(obj); + struct thandle *handle; + int rc; + ENTRY; - rc = mdd_xattr_sanity_check(env, mdd_obj); - if (rc) - RETURN(rc); + rc = mdd_la_get(env, mdd_obj, attr); + if (rc) + RETURN(rc); + + rc = mdd_xattr_sanity_check(env, mdd_obj, attr); + if (rc) + RETURN(rc); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) @@ -1312,8 +1132,7 @@ int mdd_xattr_del(const struct lu_env *env, struct md_object *obj, GOTO(stop, rc); mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = mdo_xattr_del(env, mdd_obj, name, handle, - mdd_object_capa(env, mdd_obj)); + rc = mdo_xattr_del(env, mdd_obj, name, handle); mdd_write_unlock(env, mdd_obj); if (rc) GOTO(stop, rc); @@ -1338,92 +1157,121 @@ stop: * read lov EA of an object * return the lov EA in an allocated lu_buf */ -static struct lu_buf *mdd_get_lov_ea(const struct lu_env *env, - struct mdd_object *obj) +int mdd_get_lov_ea(const struct lu_env *env, struct mdd_object *obj, + struct lu_buf *lmm_buf) { struct lu_buf *buf = &mdd_env_info(env)->mti_big_buf; - struct lu_buf *lmm_buf = NULL; - int rc, sz; + int rc, bufsize; ENTRY; repeat: - rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV, - mdd_object_capa(env, obj)); + rc = mdo_xattr_get(env, obj, buf, XATTR_NAME_LOV); + + if (rc == -ERANGE) { + /* mti_big_buf is allocated but is too small + * we need to increase it */ + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, + buf->lb_len * 2); + if (buf->lb_buf == NULL) + GOTO(out, rc = -ENOMEM); + goto repeat; + } + if (rc < 0) - GOTO(out, rc); + RETURN(rc); if (rc == 0) - GOTO(out, rc = -ENODATA); + RETURN(-ENODATA); - sz = rc; + bufsize = rc; if (memcmp(buf, &LU_BUF_NULL, sizeof(*buf)) == 0) { /* mti_big_buf was not allocated, so we have to * allocate it based on the ea size */ - buf = mdd_buf_alloc(env, sz); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, + bufsize); if (buf->lb_buf == NULL) GOTO(out, rc = -ENOMEM); goto repeat; } - OBD_ALLOC_PTR(lmm_buf); - if (!lmm_buf) + lu_buf_alloc(lmm_buf, bufsize); + if (lmm_buf->lb_buf == NULL) GOTO(out, rc = -ENOMEM); - OBD_ALLOC(lmm_buf->lb_buf, sz); - if (!lmm_buf->lb_buf) - GOTO(free, rc = -ENOMEM); - - memcpy(lmm_buf->lb_buf, buf->lb_buf, sz); - lmm_buf->lb_len = sz; - - GOTO(out, rc = 0); + memcpy(lmm_buf->lb_buf, buf->lb_buf, bufsize); + rc = 0; + EXIT; -free: - if (lmm_buf) - OBD_FREE_PTR(lmm_buf); out: - if (rc) - return ERR_PTR(rc); - return lmm_buf; + if (rc < 0) + lu_buf_free(lmm_buf); + return rc; } +static int mdd_xattr_hsm_replace(const struct lu_env *env, + struct mdd_object *o, struct lu_buf *buf, + struct thandle *handle) +{ + struct hsm_attrs *attrs; + __u32 hsm_flags; + int flags = 0; + int rc; + ENTRY; + + rc = mdo_xattr_set(env, o, buf, XATTR_NAME_HSM, LU_XATTR_REPLACE, + handle); + if (rc != 0) + RETURN(rc); + + attrs = buf->lb_buf; + hsm_flags = le32_to_cpu(attrs->hsm_flags); + if (!(hsm_flags & HS_RELEASED) || mdd_is_dead_obj(o)) + RETURN(0); + + /* Add a changelog record for release. */ + hsm_set_cl_event(&flags, HE_RELEASE); + rc = mdd_changelog_data_store(env, mdo2mdd(&o->mod_obj), CL_HSM, + flags, o, handle); + RETURN(rc); +} /* * check if layout swapping between 2 objects is allowed * the rules are: + * - only normal FIDs or non-system IGIFs * - same type of objects * - same owner/group (so quotas are still valid) */ static int mdd_layout_swap_allowed(const struct lu_env *env, struct mdd_object *o1, - struct mdd_object *o2) + const struct lu_attr *attr1, + struct mdd_object *o2, + const struct lu_attr *attr2) { const struct lu_fid *fid1, *fid2; - __u32 uid, gid; - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; - int rc; ENTRY; fid1 = mdo2fid(o1); fid2 = mdo2fid(o2); - if (!fid_is_norm(fid1) || !fid_is_norm(fid2) || - (mdd_object_type(o1) != mdd_object_type(o2))) - RETURN(-EPERM); + if (!fid_is_norm(fid1) && + (!fid_is_igif(fid1) || IS_ERR(mdd_links_get(env, o1)))) + RETURN(-EBADF); - tmp_la->la_valid = 0; - rc = mdd_la_get(env, o1, tmp_la, BYPASS_CAPA); - if (rc) - RETURN(rc); - uid = tmp_la->la_uid; - gid = tmp_la->la_gid; + if (!fid_is_norm(fid2) && + (!fid_is_igif(fid2) || IS_ERR(mdd_links_get(env, o2)))) + RETURN(-EBADF); - tmp_la->la_valid = 0; - rc = mdd_la_get(env, o2, tmp_la, BYPASS_CAPA); - if (rc) - RETURN(rc); + if (mdd_object_type(o1) != mdd_object_type(o2)) { + if (S_ISDIR(mdd_object_type(o1))) + RETURN(-ENOTDIR); + if (S_ISREG(mdd_object_type(o1))) + RETURN(-EISDIR); + RETURN(-EBADF); + } - if ((uid != tmp_la->la_uid) || (gid != tmp_la->la_gid)) + if ((attr1->la_uid != attr2->la_uid) || + (attr1->la_gid != attr2->la_gid)) RETURN(-EPERM); RETURN(0); @@ -1435,35 +1283,48 @@ static int mdd_layout_swap_allowed(const struct lu_env *env, static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, struct md_object *obj2, __u64 flags) { - struct mdd_object *o1, *o2, *fst_o, *snd_o; - struct lu_buf *lmm1_buf = NULL, *lmm2_buf = NULL; - struct lu_buf *fst_buf, *snd_buf; - struct lov_mds_md *fst_lmm, *snd_lmm, *old_fst_lmm = NULL; - struct thandle *handle; + struct mdd_thread_info *info = mdd_env_info(env); + struct mdd_object *fst_o = md2mdd_obj(obj1); + struct mdd_object *snd_o = md2mdd_obj(obj2); + struct lu_attr *fst_la = MDD_ENV_VAR(env, cattr); + struct lu_attr *snd_la = MDD_ENV_VAR(env, tattr); struct mdd_device *mdd = mdo2mdd(obj1); - int rc; + struct lov_mds_md *fst_lmm, *snd_lmm; + struct lu_buf *fst_buf = &info->mti_buf[0]; + struct lu_buf *snd_buf = &info->mti_buf[1]; + struct lu_buf *fst_hsm_buf = &info->mti_buf[2]; + struct lu_buf *snd_hsm_buf = &info->mti_buf[3]; + struct ost_id *saved_oi = NULL; + struct thandle *handle; __u16 fst_gen, snd_gen; + int fst_fl; + int rc; + int rc2; ENTRY; + CLASSERT(ARRAY_SIZE(info->mti_buf) >= 4); + memset(info->mti_buf, 0, sizeof(info->mti_buf)); + /* we have to sort the 2 obj, so locking will always * be in the same order, even in case of 2 concurrent swaps */ - rc = lu_fid_cmp(mdo2fid(md2mdd_obj(obj1)), - mdo2fid(md2mdd_obj(obj2))); - /* same fid ? */ - if (rc == 0) + rc = lu_fid_cmp(mdo2fid(fst_o), mdo2fid(snd_o)); + if (rc == 0) /* same fid ? */ RETURN(-EPERM); - if (rc > 0) { - o1 = md2mdd_obj(obj1); - o2 = md2mdd_obj(obj2); - } else { - o1 = md2mdd_obj(obj2); - o2 = md2mdd_obj(obj1); - } + if (rc < 0) + swap(fst_o, snd_o); + + rc = mdd_la_get(env, fst_o, fst_la); + if (rc != 0) + RETURN(rc); + + rc = mdd_la_get(env, snd_o, snd_la); + if (rc != 0) + RETURN(rc); /* check if layout swapping is allowed */ - rc = mdd_layout_swap_allowed(env, o1, o2); - if (rc) + rc = mdd_layout_swap_allowed(env, fst_o, fst_la, snd_o, snd_la); + if (rc != 0) RETURN(rc); handle = mdd_trans_create(env, mdd); @@ -1471,133 +1332,178 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, RETURN(PTR_ERR(handle)); /* objects are already sorted */ - mdd_write_lock(env, o1, MOR_TGT_CHILD); - mdd_write_lock(env, o2, MOR_TGT_CHILD); - - lmm1_buf = mdd_get_lov_ea(env, o1); - if (IS_ERR(lmm1_buf)) { - rc = PTR_ERR(lmm1_buf); - lmm1_buf = NULL; - if (rc != -ENODATA) - GOTO(unlock, rc); - } + mdd_write_lock(env, fst_o, MOR_TGT_CHILD); + mdd_write_lock(env, snd_o, MOR_TGT_CHILD); - lmm2_buf = mdd_get_lov_ea(env, o2); - if (IS_ERR(lmm2_buf)) { - rc = PTR_ERR(lmm2_buf); - lmm2_buf = NULL; - if (rc != -ENODATA) - GOTO(unlock, rc); - } + rc = mdd_get_lov_ea(env, fst_o, fst_buf); + if (rc < 0 && rc != -ENODATA) + GOTO(stop, rc); + + rc = mdd_get_lov_ea(env, snd_o, snd_buf); + if (rc < 0 && rc != -ENODATA) + GOTO(stop, rc); /* swapping 2 non existant layouts is a success */ - if ((lmm1_buf == NULL) && (lmm2_buf == NULL)) - GOTO(unlock, rc = 0); + if (fst_buf->lb_buf == NULL && snd_buf->lb_buf == NULL) + GOTO(stop, rc = 0); /* to help inode migration between MDT, it is better to * start by the no layout file (if one), so we order the swap */ - if (lmm1_buf == NULL) { - fst_o = o1; - fst_buf = lmm1_buf; - snd_o = o2; - snd_buf = lmm2_buf; - } else { - fst_o = o2; - fst_buf = lmm2_buf; - snd_o = o1; - snd_buf = lmm1_buf; + if (snd_buf->lb_buf == NULL) { + swap(fst_o, snd_o); + swap(fst_buf, snd_buf); } /* lmm and generation layout initialization */ - if (fst_buf) { + if (fst_buf->lb_buf != NULL) { fst_lmm = fst_buf->lb_buf; fst_gen = le16_to_cpu(fst_lmm->lmm_layout_gen); + fst_fl = LU_XATTR_REPLACE; } else { fst_lmm = NULL; fst_gen = 0; + fst_fl = LU_XATTR_CREATE; } - if (snd_buf) { - snd_lmm = snd_buf->lb_buf; - snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen); - } else { - snd_lmm = NULL; - snd_gen = 0; - } - - /* save the orignal lmm common header of first file - * to be able to roll back */ - OBD_ALLOC_PTR(old_fst_lmm); - if (old_fst_lmm == NULL) - GOTO(unlock, rc = -ENOMEM); - - memcpy(old_fst_lmm, fst_lmm, sizeof(*old_fst_lmm)); + snd_lmm = snd_buf->lb_buf; + snd_gen = le16_to_cpu(snd_lmm->lmm_layout_gen); /* increase the generation layout numbers */ snd_gen++; fst_gen++; /* set the file specific informations in lmm */ - if (fst_lmm) { + if (fst_lmm != NULL) { + saved_oi = &info->mti_oa.o_oi; + + *saved_oi = fst_lmm->lmm_oi; fst_lmm->lmm_layout_gen = cpu_to_le16(snd_gen); - fst_lmm->lmm_object_seq = snd_lmm->lmm_object_seq; - fst_lmm->lmm_object_id = snd_lmm->lmm_object_id; + fst_lmm->lmm_oi = snd_lmm->lmm_oi; + snd_lmm->lmm_oi = *saved_oi; + } else { + if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) + snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEF); + else if (snd_lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) + snd_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V3_DEF); + else + GOTO(stop, rc = -EPROTO); } - - if (snd_lmm) { - snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen); - snd_lmm->lmm_object_seq = old_fst_lmm->lmm_object_seq; - snd_lmm->lmm_object_id = old_fst_lmm->lmm_object_id; + snd_lmm->lmm_layout_gen = cpu_to_le16(fst_gen); + + /* Prepare HSM attribute if it's required */ + if (flags & SWAP_LAYOUTS_MDS_HSM) { + const int buflen = sizeof(struct hsm_attrs); + + lu_buf_alloc(fst_hsm_buf, buflen); + lu_buf_alloc(snd_hsm_buf, buflen); + if (fst_hsm_buf->lb_buf == NULL || snd_hsm_buf->lb_buf == NULL) + GOTO(stop, rc = -ENOMEM); + + /* Read HSM attribute */ + rc = mdo_xattr_get(env, fst_o, fst_hsm_buf, XATTR_NAME_HSM); + if (rc < 0) + GOTO(stop, rc); + + rc = mdo_xattr_get(env, snd_o, snd_hsm_buf, XATTR_NAME_HSM); + if (rc < 0) + GOTO(stop, rc); + + rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_hsm_buf, + XATTR_NAME_HSM, LU_XATTR_REPLACE, + handle); + if (rc < 0) + GOTO(stop, rc); + + rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_hsm_buf, + XATTR_NAME_HSM, LU_XATTR_REPLACE, + handle); + if (rc < 0) + GOTO(stop, rc); } /* prepare transaction */ rc = mdd_declare_xattr_set(env, mdd, fst_o, snd_buf, XATTR_NAME_LOV, - LU_XATTR_REPLACE, handle); - if (rc) + fst_fl, handle); + if (rc != 0) GOTO(stop, rc); - rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf, XATTR_NAME_LOV, - LU_XATTR_REPLACE, handle); - if (rc) + if (fst_buf->lb_buf != NULL) + rc = mdd_declare_xattr_set(env, mdd, snd_o, fst_buf, + XATTR_NAME_LOV, LU_XATTR_REPLACE, + handle); + else + rc = mdd_declare_xattr_del(env, mdd, snd_o, XATTR_NAME_LOV, + handle); + if (rc != 0) GOTO(stop, rc); rc = mdd_trans_start(env, mdd, handle); - if (rc) + if (rc != 0) GOTO(stop, rc); - rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, - LU_XATTR_REPLACE, handle, - mdd_object_capa(env, fst_o)); - if (rc) + if (flags & SWAP_LAYOUTS_MDS_HSM) { + rc = mdd_xattr_hsm_replace(env, fst_o, snd_hsm_buf, handle); + if (rc < 0) + GOTO(stop, rc); + + rc = mdd_xattr_hsm_replace(env, snd_o, fst_hsm_buf, handle); + if (rc < 0) { + rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, + handle); + if (rc2 < 0) + CERROR("%s: restore "DFID" HSM error: %d/%d\n", + mdd_obj_dev_name(fst_o), + PFID(mdo2fid(fst_o)), rc, rc2); + GOTO(stop, rc); + } + } + + rc = mdo_xattr_set(env, fst_o, snd_buf, XATTR_NAME_LOV, fst_fl, handle); + if (rc != 0) GOTO(stop, rc); - rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV, - LU_XATTR_REPLACE, handle, - mdd_object_capa(env, snd_o)); - if (rc) { - int rc2; + if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_MDS_HSM_SWAP_LAYOUTS))) { + rc = -EOPNOTSUPP; + } else { + if (fst_buf->lb_buf != NULL) + rc = mdo_xattr_set(env, snd_o, fst_buf, XATTR_NAME_LOV, + LU_XATTR_REPLACE, handle); + else + rc = mdo_xattr_del(env, snd_o, XATTR_NAME_LOV, handle); + } + + if (rc != 0) { + int steps = 0; /* failure on second file, but first was done, so we have - * to roll back first */ - /* restore object_id, object_seq and generation number - * on first file */ - if (fst_lmm) { - fst_lmm->lmm_object_id = old_fst_lmm->lmm_object_id; - fst_lmm->lmm_object_seq = old_fst_lmm->lmm_object_seq; - fst_lmm->lmm_layout_gen = old_fst_lmm->lmm_layout_gen; + * to roll back first. */ + if (fst_buf->lb_buf != NULL) { + fst_lmm->lmm_oi = *saved_oi; + fst_lmm->lmm_layout_gen = cpu_to_le16(fst_gen - 1); + rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV, + LU_XATTR_REPLACE, handle); + } else { + rc2 = mdo_xattr_del(env, fst_o, XATTR_NAME_LOV, handle); } + if (rc2 < 0) + goto do_lbug; - rc2 = mdo_xattr_set(env, fst_o, fst_buf, XATTR_NAME_LOV, - LU_XATTR_REPLACE, handle, - mdd_object_capa(env, fst_o)); - if (rc2) { + ++steps; + rc2 = mdd_xattr_hsm_replace(env, fst_o, fst_hsm_buf, handle); + if (rc2 < 0) + goto do_lbug; + + ++steps; + rc2 = mdd_xattr_hsm_replace(env, snd_o, snd_hsm_buf, handle); + + do_lbug: + if (rc2 < 0) { /* very bad day */ - CERROR("%s: unable to roll back after swap layouts" - " failure between "DFID" and "DFID - " rc2 = %d rc = %d)\n", - mdd2obd_dev(mdd)->obd_name, + CERROR("%s: unable to roll back layout swap. FIDs: " + DFID" and "DFID "error: %d/%d, steps: %d\n", + mdd_obj_dev_name(fst_o), PFID(mdo2fid(snd_o)), PFID(mdo2fid(fst_o)), - rc2, rc); + rc, rc2, steps); /* a solution to avoid journal commit is to panic, * but it has strong consequences so we use LBUG to * allow sysdamin to choose to panic or not @@ -1606,37 +1512,52 @@ static int mdd_swap_layouts(const struct lu_env *env, struct md_object *obj1, } GOTO(stop, rc); } + + /* Issue one changelog record per file */ + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, fst_o, handle); + if (rc) + GOTO(stop, rc); + + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, snd_o, handle); + if (rc) + GOTO(stop, rc); EXIT; stop: mdd_trans_stop(env, mdd, rc, handle); -unlock: - mdd_write_unlock(env, o2); - mdd_write_unlock(env, o1); - - if (lmm1_buf && lmm1_buf->lb_buf) - OBD_FREE(lmm1_buf->lb_buf, lmm1_buf->lb_len); - if (lmm1_buf) - OBD_FREE_PTR(lmm1_buf); - - if (lmm2_buf && lmm2_buf->lb_buf) - OBD_FREE(lmm2_buf->lb_buf, lmm2_buf->lb_len); - if (lmm2_buf) - OBD_FREE_PTR(lmm2_buf); - - if (old_fst_lmm) - OBD_FREE_PTR(old_fst_lmm); + mdd_write_unlock(env, snd_o); + mdd_write_unlock(env, fst_o); + lu_buf_free(fst_buf); + lu_buf_free(snd_buf); + lu_buf_free(fst_hsm_buf); + lu_buf_free(snd_hsm_buf); return rc; } void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent, - struct mdd_object *child, struct lu_attr *attr) + struct mdd_object *child, const struct lu_attr *attr, + const struct md_op_spec *spec, + struct dt_allocation_hint *hint) { - struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint; - struct dt_object *np = parent ? mdd_object_child(parent) : NULL; + struct dt_object *np = parent ? mdd_object_child(parent) : NULL; struct dt_object *nc = mdd_object_child(child); + memset(hint, 0, sizeof(*hint)); + + /* For striped directory, give striping EA to lod_ah_init, which will + * decide the stripe_offset and stripe count by it. */ + if (S_ISDIR(attr->la_mode) && + unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA)) { + hint->dah_eadata = spec->u.sp_ea.eadata; + hint->dah_eadata_len = spec->u.sp_ea.eadatalen; + } else { + hint->dah_eadata = NULL; + hint->dah_eadata_len = 0; + } + + CDEBUG(D_INFO, DFID" eadata %p len %d\n", PFID(mdd_object_fid(child)), + hint->dah_eadata, hint->dah_eadata_len); /* @hint will be initialized by underlying device. */ nc->do_ops->do_ah_init(env, hint, np, nc, attr->la_mode & S_IFMT); } @@ -1644,7 +1565,7 @@ void mdd_object_make_hint(const struct lu_env *env, struct mdd_object *parent, /* * do NOT or the MAY_*'s, you'll get the weakest */ -int accmode(const struct lu_env *env, struct lu_attr *la, int flags) +int accmode(const struct lu_env *env, const struct lu_attr *la, int flags) { int res = 0; @@ -1670,45 +1591,41 @@ int accmode(const struct lu_env *env, struct lu_attr *la, int flags) } static int mdd_open_sanity_check(const struct lu_env *env, - struct mdd_object *obj, int flag) + struct mdd_object *obj, + const struct lu_attr *attr, int flag) { - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; - int mode, rc; - ENTRY; - - /* EEXIST check */ - if (mdd_is_dead_obj(obj)) - RETURN(-ENOENT); + int mode, rc; + ENTRY; - rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA); - if (rc) - RETURN(rc); + /* EEXIST check */ + if (mdd_is_dead_obj(obj)) + RETURN(-ENOENT); - if (S_ISLNK(tmp_la->la_mode)) - RETURN(-ELOOP); + if (S_ISLNK(attr->la_mode)) + RETURN(-ELOOP); - mode = accmode(env, tmp_la, flag); + mode = accmode(env, attr, flag); - if (S_ISDIR(tmp_la->la_mode) && (mode & MAY_WRITE)) - RETURN(-EISDIR); + if (S_ISDIR(attr->la_mode) && (mode & MAY_WRITE)) + RETURN(-EISDIR); - if (!(flag & MDS_OPEN_CREATED)) { - rc = mdd_permission_internal(env, obj, tmp_la, mode); - if (rc) - RETURN(rc); - } + if (!(flag & MDS_OPEN_CREATED)) { + rc = mdd_permission_internal(env, obj, attr, mode); + if (rc) + RETURN(rc); + } - if (S_ISFIFO(tmp_la->la_mode) || S_ISSOCK(tmp_la->la_mode) || - S_ISBLK(tmp_la->la_mode) || S_ISCHR(tmp_la->la_mode)) - flag &= ~MDS_OPEN_TRUNC; + if (S_ISFIFO(attr->la_mode) || S_ISSOCK(attr->la_mode) || + S_ISBLK(attr->la_mode) || S_ISCHR(attr->la_mode)) + flag &= ~MDS_OPEN_TRUNC; - /* For writing append-only file must open it with append mode. */ - if (mdd_is_append(obj)) { - if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND)) - RETURN(-EPERM); - if (flag & MDS_OPEN_TRUNC) - RETURN(-EPERM); - } + /* For writing append-only file must open it with append mode. */ + if (attr->la_flags & LUSTRE_APPEND_FL) { + if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND)) + RETURN(-EPERM); + if (flag & MDS_OPEN_TRUNC) + RETURN(-EPERM); + } #if 0 /* @@ -1719,48 +1636,37 @@ static int mdd_open_sanity_check(const struct lu_env *env, 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)) + (uc->uc_fsuid != attr->la_uid) && + !md_capable(uc, CFS_CAP_FOWNER)) RETURN(-EPERM); } #endif - RETURN(0); + RETURN(0); } static int mdd_open(const struct lu_env *env, struct md_object *obj, - int flags) + int flags) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); - int rc = 0; - - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - - rc = mdd_open_sanity_check(env, mdd_obj, flags); - if (rc == 0) - mdd_obj->mod_count++; - - mdd_write_unlock(env, mdd_obj); - return rc; -} + struct mdd_object *mdd_obj = md2mdd_obj(obj); + struct lu_attr *attr = MDD_ENV_VAR(env, cattr); + int rc = 0; -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); -} + mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); -/* 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 thandle *handle) -{ - int rc; - ENTRY; + rc = mdd_la_get(env, mdd_obj, attr); + if (rc != 0) + GOTO(out, rc); - rc = mdo_destroy(env, obj, handle); + rc = mdd_open_sanity_check(env, mdd_obj, attr, flags); + if (rc != 0) + GOTO(out, rc); - RETURN(rc); + mdd_obj->mod_count++; + EXIT; +out: + mdd_write_unlock(env, mdd_obj); + return rc; } static int mdd_declare_close(const struct lu_env *env, @@ -1816,9 +1722,9 @@ again: if (rc) GOTO(stop, rc); - rc = mdd_declare_changelog_store(env, mdd, NULL, handle); - if (rc) - GOTO(stop, rc); + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle); + if (rc) + GOTO(stop, rc); rc = mdd_trans_start(env, mdo2mdd(obj), handle); if (rc) @@ -1826,8 +1732,7 @@ again: } mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - rc = mdd_la_get(env, mdd_obj, &ma->ma_attr, - mdd_object_capa(env, mdd_obj)); + rc = mdd_la_get(env, mdd_obj, &ma->ma_attr); if (rc != 0) { CERROR("Failed to get lu_attr of "DFID": %d\n", PFID(mdd_object_fid(mdd_obj)), rc); @@ -1890,12 +1795,12 @@ out: if (handle == NULL) { handle = mdd_trans_create(env, mdo2mdd(obj)); if (IS_ERR(handle)) - GOTO(stop, rc = IS_ERR(handle)); + GOTO(stop, rc = PTR_ERR(handle)); - rc = mdd_declare_changelog_store(env, mdd, NULL, - handle); - if (rc) - GOTO(stop, rc); + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, + handle); + if (rc) + GOTO(stop, rc); rc = mdd_trans_start(env, mdo2mdd(obj), handle); if (rc) @@ -1907,9 +1812,10 @@ out: } stop: - if (handle != NULL) - mdd_trans_stop(env, mdd, rc, handle); - return rc; + if (handle != NULL && !IS_ERR(handle)) + mdd_trans_stop(env, mdd, rc, handle); + + return rc; } /* @@ -1932,7 +1838,7 @@ static int mdd_readpage_sanity_check(const struct lu_env *env, } static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp, - int nob, const struct dt_it_ops *iops, + size_t nob, const struct dt_it_ops *iops, struct dt_it *it, __u32 attr, void *arg) { struct lu_dirpage *dp = &lp->lp_dir; @@ -1941,8 +1847,12 @@ static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp, __u64 hash = 0; struct lu_dirent *ent; struct lu_dirent *last = NULL; + struct lu_fid fid; int first = 1; + if (nob < sizeof(*dp)) + return -EINVAL; + memset(area, 0, sizeof (*dp)); area += sizeof (*dp); nob -= sizeof (*dp); @@ -1950,7 +1860,7 @@ static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp, ent = area; do { int len; - int recsize; + size_t recsize; len = iops->key_size(env, it); @@ -1977,6 +1887,12 @@ static int mdd_dir_page_build(const struct lu_env *env, union lu_page *lp, /* osd might not able to pack all attributes, * so recheck rec length */ recsize = le16_to_cpu(ent->lde_reclen); + + if (le32_to_cpu(ent->lde_attrs) & LUDA_FID) { + fid_le_to_cpu(&fid, &ent->lde_fid); + if (fid_is_dot_lustre(&fid)) + goto next; + } } else { result = (last != NULL) ? 0 :-EINVAL; goto out; @@ -2001,7 +1917,7 @@ out: if (result > 0) /* end of directory */ dp->ldp_hash_end = cpu_to_le64(MDS_DIR_END_OFF); - if (result < 0) + else if (result < 0) CWARN("build page failed: %d!\n", result); return result; } @@ -2040,12 +1956,12 @@ int mdd_readpage(const struct lu_env *env, struct md_object *obj, LASSERT(rdpg->rp_pages != NULL); pg = rdpg->rp_pages[0]; - dp = (struct lu_dirpage*)cfs_kmap(pg); + dp = (struct lu_dirpage *)kmap(pg); memset(dp, 0 , sizeof(struct lu_dirpage)); 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); + kunmap(pg); GOTO(out_unlock, rc = LU_PAGE_SIZE); } @@ -2054,17 +1970,18 @@ int mdd_readpage(const struct lu_env *env, struct md_object *obj, if (rc >= 0) { struct lu_dirpage *dp; - dp = cfs_kmap(rdpg->rp_pages[0]); + dp = 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_hash_end = cpu_to_le64(MDS_DIR_END_OFF); 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]); + kunmap(rdpg->rp_pages[0]); } GOTO(out_unlock, rc); @@ -2075,14 +1992,38 @@ 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 mdd_object *mdd_obj = md2mdd_obj(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; - } - return dt_object_sync(env, mdd_object_child(mdd_obj)); + if (mdd_object_exists(mdd_obj) == 0) { + int rc = -ENOENT; + + CERROR("%s: object "DFID" not found: rc = %d\n", + mdd_obj_dev_name(mdd_obj), + PFID(mdd_object_fid(mdd_obj)), rc); + return rc; + } + return dt_object_sync(env, mdd_object_child(mdd_obj), + 0, OBD_OBJECT_EOF); +} + +static int mdd_object_lock(const struct lu_env *env, + struct md_object *obj, + struct lustre_handle *lh, + struct ldlm_enqueue_info *einfo, + ldlm_policy_data_t *policy) +{ + struct mdd_object *mdd_obj = md2mdd_obj(obj); + return dt_object_lock(env, mdd_object_child(mdd_obj), lh, + einfo, policy); +} + +static int mdd_object_unlock(const struct lu_env *env, + struct md_object *obj, + struct ldlm_enqueue_info *einfo, + ldlm_policy_data_t *policy) +{ + struct mdd_object *mdd_obj = md2mdd_obj(obj); + return dt_object_unlock(env, mdd_object_child(mdd_obj), einfo, policy); } const struct md_object_operations mdd_obj_ops = { @@ -2099,7 +2040,7 @@ const struct md_object_operations mdd_obj_ops = { .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, + .moo_object_lock = mdd_object_lock, + .moo_object_unlock = mdd_object_unlock, };