static void vvp_object_size_unlock(struct cl_object *obj)
{
struct inode *inode = vvp_object_inode(obj);
cl_object_attr_unlock(obj);
ll_inode_size_unlock(inode);
}
/**
* Implementation of struct cl_req_operations::cro_attr_set() for VVP
* layer. VVP is responsible for
*
* - o_[mac]time
*
* - o_mode
*
* - o_parent_seq
*
* - o_[ug]id
*
* - o_parent_oid
*
* - o_parent_ver
*
* - o_ioepoch,
*
*/
static void vvp_req_attr_set(const struct lu_env *env,
const struct cl_req_slice *slice,
const struct cl_object *obj,
struct cl_req_attr *attr, u64 flags)
{
struct inode *inode;
struct obdo *oa;
u32 valid_flags;
oa = attr->cra_oa;
inode = vvp_object_inode(obj);
valid_flags = OBD_MD_FLTYPE;
if (slice->crs_req->crq_type == CRT_WRITE) {
if (flags & OBD_MD_FLEPOCH) {
oa->o_valid |= OBD_MD_FLEPOCH;
oa->o_ioepoch = ll_i2info(inode)->lli_ioepoch;
valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
OBD_MD_FLUID | OBD_MD_FLGID;
}
}
obdo_from_inode(oa, inode, valid_flags & flags);
obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
oa->o_parent_oid++;
memcpy(attr->cra_jobid, ll_i2info(inode)->lli_jobid,
LUSTRE_JOBID_SIZE);
}
int vvp_object_invariant(const struct cl_object *obj)
{
struct inode *inode = vvp_object_inode(obj);
struct ll_inode_info *lli = ll_i2info(inode);
return (S_ISREG(inode->i_mode) || inode->i_mode == 0) &&
lli->lli_clob == obj;
}
/**
* Helper function that if necessary adjusts file size (inode->i_size), when
* position at the offset \a pos is accessed. File size can be arbitrary stale
* on a Lustre client, but client at least knows KMS. If accessed area is
* inside [0, KMS], set file size to KMS, otherwise glimpse file size.
*
* Locking: cl_isize_lock is used to serialize changes to inode size and to
* protect consistency between inode size and cl_object
* attributes. cl_object_size_lock() protects consistency between cl_attr's of
* top-object and sub-objects.
*/
static int vvp_prep_size(const struct lu_env *env, struct cl_object *obj,
struct cl_io *io, loff_t start, size_t count,
int *exceed)
{
struct cl_attr *attr = vvp_env_thread_attr(env);
struct inode *inode = vvp_object_inode(obj);
loff_t pos = start + count - 1;
loff_t kms;
int result;
/*
* Consistency guarantees: following possibilities exist for the
* relation between region being accessed and real file size at this
* moment:
*
* (A): the region is completely inside of the file;
*
* (B-x): x bytes of region are inside of the file, the rest is
* outside;
*
* (C): the region is completely outside of the file.
*
* This classification is stable under DLM lock already acquired by
* the caller, because to change the class, other client has to take
* DLM lock conflicting with our lock. Also, any updates to ->i_size
* by other threads on this client are serialized by
* ll_inode_size_lock(). This guarantees that short reads are handled
* correctly in the face of concurrent writes and truncates.
*/
vvp_object_size_lock(obj);
result = cl_object_attr_get(env, obj, attr);
if (result == 0) {
kms = attr->cat_kms;
if (pos > kms) {
/*
* A glimpse is necessary to determine whether we
* return a short read (B) or some zeroes at the end
* of the buffer (C)
*/
vvp_object_size_unlock(obj);
result = cl_glimpse_lock(env, io, inode, obj, 0);
if (result == 0 && exceed) {
/* If objective page index exceed end-of-file
* page index, return directly. Do not expect
* kernel will check such case correctly.
* linux-2.6.18-128.1.1 miss to do that.
* --bug 17336
*/
loff_t size = i_size_read(inode);
loff_t cur_index = start >> PAGE_SHIFT;
loff_t size_index = (size - 1) >> PAGE_SHIFT;
if ((size == 0 && cur_index != 0) ||
size_index < cur_index)
*exceed = 1;
}
static int vvp_prune(const struct lu_env *env, struct cl_object *obj)
{
struct inode *inode = vvp_object_inode(obj);
int rc;
rc = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
if (rc < 0) {
CDEBUG(D_VFSTRACE, DFID ": writeback failed: %d\n",
PFID(lu_object_fid(&obj->co_lu)), rc);
return rc;
}
truncate_inode_pages(inode->i_mapping, 0);
return 0;
}
static int vvp_object_glimpse(const struct lu_env *env,
const struct cl_object *obj, struct ost_lvb *lvb)
{
struct inode *inode = vvp_object_inode(obj);
lvb->lvb_mtime = LTIME_S(inode->i_mtime);
lvb->lvb_atime = LTIME_S(inode->i_atime);
lvb->lvb_ctime = LTIME_S(inode->i_ctime);
/*
* LU-417: Add dirty pages block count lest i_blocks reports 0, some
* "cp" or "tar" on remote node may think it's a completely sparse file
* and skip it.
*/
if (lvb->lvb_size > 0 && lvb->lvb_blocks == 0)
lvb->lvb_blocks = dirty_cnt(inode);
return 0;
}
static void vvp_req_attr_set(const struct lu_env *env, struct cl_object *obj,
struct cl_req_attr *attr)
{
struct inode *inode;
struct obdo *oa;
u64 valid_flags = OBD_MD_FLTYPE;
oa = attr->cra_oa;
inode = vvp_object_inode(obj);
if (attr->cra_type == CRT_WRITE)
valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
OBD_MD_FLUID | OBD_MD_FLGID;
obdo_from_inode(oa, inode, valid_flags & attr->cra_flags);
obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
oa->o_parent_oid++;
memcpy(attr->cra_jobid, ll_i2info(inode)->lli_jobid, LUSTRE_JOBID_SIZE);
}
static int vvp_attr_get(const struct lu_env *env, struct cl_object *obj,
struct cl_attr *attr)
{
struct inode *inode = vvp_object_inode(obj);
/*
* lov overwrites most of these fields in
* lov_attr_get()->...lov_merge_lvb_kms(), except when inode
* attributes are newer.
*/
attr->cat_size = i_size_read(inode);
attr->cat_mtime = inode->i_mtime.tv_sec;
attr->cat_atime = inode->i_atime.tv_sec;
attr->cat_ctime = inode->i_ctime.tv_sec;
attr->cat_blocks = inode->i_blocks;
attr->cat_uid = from_kuid(&init_user_ns, inode->i_uid);
attr->cat_gid = from_kgid(&init_user_ns, inode->i_gid);
/* KMS is not known by this layer */
return 0; /* layers below have to fill in the rest */
}
static int vvp_attr_update(const struct lu_env *env, struct cl_object *obj,
const struct cl_attr *attr, unsigned valid)
{
struct inode *inode = vvp_object_inode(obj);
if (valid & CAT_UID)
inode->i_uid = make_kuid(&init_user_ns, attr->cat_uid);
if (valid & CAT_GID)
inode->i_gid = make_kgid(&init_user_ns, attr->cat_gid);
if (valid & CAT_ATIME)
inode->i_atime.tv_sec = attr->cat_atime;
if (valid & CAT_MTIME)
inode->i_mtime.tv_sec = attr->cat_mtime;
if (valid & CAT_CTIME)
inode->i_ctime.tv_sec = attr->cat_ctime;
if (0 && valid & CAT_SIZE)
i_size_write(inode, attr->cat_size);
/* not currently necessary */
if (0 && valid & (CAT_UID|CAT_GID|CAT_SIZE))
mark_inode_dirty(inode);
return 0;
}
static int vvp_prune(const struct lu_env *env, struct cl_object *obj)
{
struct inode *inode = vvp_object_inode(obj);
int rc;
ENTRY;
rc = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
if (rc < 0) {
CDEBUG(D_VFSTRACE, DFID ": writeback failed: %d\n",
PFID(lu_object_fid(&obj->co_lu)), rc);
RETURN(rc);
}
truncate_inode_pages(inode->i_mapping, 0);
if (inode->i_mapping->nrpages) {
CDEBUG(D_VFSTRACE, DFID ": still has %lu pages remaining\n",
PFID(lu_object_fid(&obj->co_lu)),
inode->i_mapping->nrpages);
RETURN(-EIO);
}
RETURN(0);
}
/* Sharing code of page_mkwrite method for rhel5 and rhel6 */
static int ll_page_mkwrite0(struct vm_area_struct *vma, struct page *vmpage,
bool *retry)
{
struct lu_env *env;
struct cl_io *io;
struct vvp_io *vio;
int result;
u16 refcheck;
sigset_t set;
struct inode *inode;
struct ll_inode_info *lli;
env = cl_env_get(&refcheck);
if (IS_ERR(env))
return PTR_ERR(env);
io = ll_fault_io_init(env, vma, vmpage->index, NULL);
if (IS_ERR(io)) {
result = PTR_ERR(io);
goto out;
}
result = io->ci_result;
if (result < 0)
goto out_io;
io->u.ci_fault.ft_mkwrite = 1;
io->u.ci_fault.ft_writable = 1;
vio = vvp_env_io(env);
vio->u.fault.ft_vma = vma;
vio->u.fault.ft_vmpage = vmpage;
set = cfs_block_sigsinv(sigmask(SIGKILL) | sigmask(SIGTERM));
inode = vvp_object_inode(io->ci_obj);
lli = ll_i2info(inode);
result = cl_io_loop(env, io);
cfs_restore_sigs(set);
if (result == 0) {
struct inode *inode = file_inode(vma->vm_file);
struct ll_inode_info *lli = ll_i2info(inode);
lock_page(vmpage);
if (!vmpage->mapping) {
unlock_page(vmpage);
/* page was truncated and lock was cancelled, return
* ENODATA so that VM_FAULT_NOPAGE will be returned
* to handle_mm_fault().
*/
if (result == 0)
result = -ENODATA;
} else if (!PageDirty(vmpage)) {
/* race, the page has been cleaned by ptlrpcd after
* it was unlocked, it has to be added into dirty
* cache again otherwise this soon-to-dirty page won't
* consume any grants, even worse if this page is being
* transferred because it will break RPC checksum.
*/
unlock_page(vmpage);
CDEBUG(D_MMAP,
"Race on page_mkwrite %p/%lu, page has been written out, retry.\n",
vmpage, vmpage->index);
*retry = true;
result = -EAGAIN;
}
if (!result)
set_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
}
out_io:
cl_io_fini(env, io);
out:
cl_env_put(env, &refcheck);
CDEBUG(D_MMAP, "%s mkwrite with %d\n", current->comm, result);
LASSERT(ergo(result == 0, PageLocked(vmpage)));
return result;
}