* 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.
*/
/*
{
if (buf == NULL || buf->lb_buf == NULL)
return;
- if (buf->lb_vmalloc)
- OBD_VFREE(buf->lb_buf, buf->lb_len);
- else
- OBD_FREE(buf->lb_buf, buf->lb_len);
+ OBD_FREE_LARGE(buf->lb_buf, buf->lb_len);
buf->lb_buf = NULL;
buf->lb_len = 0;
}
return buf;
}
-#define BUF_VMALLOC_SIZE (CFS_PAGE_SIZE<<2) /* 16k */
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)) {
- if (buf->lb_vmalloc)
- OBD_VFREE(buf->lb_buf, buf->lb_len);
- else
- OBD_FREE(buf->lb_buf, buf->lb_len);
+ OBD_FREE_LARGE(buf->lb_buf, buf->lb_len);
buf->lb_buf = NULL;
}
if (buf->lb_buf == NULL) {
buf->lb_len = len;
- if (buf->lb_len <= BUF_VMALLOC_SIZE) {
- OBD_ALLOC(buf->lb_buf, buf->lb_len);
- buf->lb_vmalloc = 0;
- }
- if (buf->lb_buf == NULL) {
- OBD_VMALLOC(buf->lb_buf, buf->lb_len);
- buf->lb_vmalloc = 1;
- }
+ OBD_ALLOC_LARGE(buf->lb_buf, buf->lb_len);
if (buf->lb_buf == NULL)
buf->lb_len = 0;
}
struct lu_buf buf;
LASSERT(len >= oldbuf->lb_len);
- if (len > BUF_VMALLOC_SIZE) {
- OBD_VMALLOC(buf.lb_buf, len);
- buf.lb_vmalloc = 1;
- } else {
- OBD_ALLOC(buf.lb_buf, len);
- buf.lb_vmalloc = 0;
- }
+ 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);
- if (oldbuf->lb_vmalloc)
- OBD_VFREE(oldbuf->lb_buf, oldbuf->lb_len);
- else
- OBD_FREE(oldbuf->lb_buf, oldbuf->lb_len);
+ OBD_FREE_LARGE(oldbuf->lb_buf, oldbuf->lb_len);
memcpy(oldbuf, &buf, sizeof(buf));
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;
}
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;
if (obj == NULL)
GOTO(out, rc = -EREMOTE);
if (IS_ERR(obj))
- GOTO(out, rc = -PTR_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 (mdd_obj == NULL)
GOTO(out, rc = -EREMOTE);
if (IS_ERR(mdd_obj))
- GOTO(out, rc = -PTR_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);
EXIT;
out:
- if (buf && !IS_ERR(buf) && buf->lb_vmalloc)
+ 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);
}
-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);
- LASSERT(size != NULL);
- if (!lmm) {
- *size = 0;
- 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 */
rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &ma->ma_lmm_size,
XATTR_NAME_LOV);
- if (rc == 0 && ma->ma_need & MA_LOV_DEF)
- rc = mdd_get_default_md(mdd_obj, ma->ma_lmm,
- &ma->ma_lmm_size);
+ 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)
{
/* Swab and copy LMA */
if (ma->ma_need & MA_HSM) {
if (lma->lma_compat & LMAC_HSM)
- ma->ma_hsm_flags = lma->lma_flags & HSM_FLAGS_MASK;
+ ma->ma_hsm.mh_flags = lma->lma_flags & HSM_FLAGS_MASK;
else
- ma->ma_hsm_flags = 0;
+ ma->ma_hsm.mh_flags = 0;
ma->ma_valid |= MA_HSM;
}
RETURN(0);
}
-static int mdd_attr_get_internal(const struct lu_env *env,
- struct mdd_object *mdd_obj,
+int mdd_attr_get_internal(const struct lu_env *env, struct mdd_object *mdd_obj,
struct md_attr *ma)
{
int rc = 0;
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);
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);
}
{
int rc;
int needlock = ma->ma_need &
- (MA_LOV | MA_LMV | MA_ACL_DEF | MA_HSM | MA_SOM);
+ (MA_LOV | MA_LMV | MA_ACL_DEF | MA_HSM | MA_SOM | MA_PFID);
if (needlock)
mdd_read_lock(env, mdd_obj, MOR_TGT_CHILD);
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);
}
}
}
+ 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);
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 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_TIME) &&
+ 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
RETURN(-ENOMEM);
rec = (struct llog_changelog_rec *)buf->lb_buf;
- rec->cr.cr_flags = CLF_VERSION;
+ 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;
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.
*
/* Copy HSM data */
if (ma->ma_valid & MA_HSM) {
- lma->lma_flags |= ma->ma_hsm_flags & HSM_FLAGS_MASK;
+ lma->lma_flags |= ma->ma_hsm.mh_flags & HSM_FLAGS_MASK;
lma->lma_compat |= LMAC_HSM;
}
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)
#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);
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;
}
cleanup:
if (rc == 0)
- rc = mdd_changelog_data_store(env, mdd,
- (ma->ma_attr.la_valid &
- ~(LA_MTIME|LA_CTIME|LA_ATIME)) ?
- CL_SETATTR : CL_TIME,
- mdd_obj, handle);
+ 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*/
rc = mdd_xattr_set_txn(env, mdd_obj, buf, name, fl, handle);
/* Only record user xattr changes */
- if ((rc == 0) && (mdd->mdd_cl.mc_flags & CLM_ON) &&
- (strncmp("user.", name, 5) == 0))
- rc = mdd_changelog_data_store(env, mdd, CL_XATTR, mdd_obj,
+ 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);
mdd_write_unlock(env, mdd_obj);
/* Only record user xattr changes */
- if ((rc == 0) && (mdd->mdd_cl.mc_flags & CLM_ON) &&
- (strncmp("user.", name, 5) != 0))
- rc = mdd_changelog_data_store(env, mdd, CL_XATTR, mdd_obj,
+ 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);
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;
}
{
struct mdd_object *mdd_obj = md2mdd_obj(obj);
struct mdd_device *mdd = mdo2mdd(obj);
- struct thandle *handle;
+ struct thandle *handle = NULL;
int rc;
int reset = 1;
#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 --;
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)
+ reset = 0;
}
if (rc != 0)
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,
* iterate through directory and fill pages from @rdpg
*/
iops = &next->do_index_ops->dio_it;
- it = iops->init(env, next, 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);
/*
* end of directory.
*/
- hash_end = DIR_END_OFF;
+ hash_end = MDS_DIR_END_OFF;
rc = 0;
}
if (rc == 0) {
dp = (struct lu_dirpage*)cfs_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(DIR_END_OFF);
+ 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);
.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,
};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff