static struct slp_io *cl2slp_io(const struct lu_env *env,
const struct cl_io_slice *slice)
{
/* We call it just for assertion here */
cl2ccc_io(env, slice);
return slp_env_io(env);
}
static void vvp_io_read_fini(const struct lu_env *env, const struct cl_io_slice *ios)
{
struct vvp_io *vio = cl2vvp_io(env, ios);
struct ccc_io *cio = cl2ccc_io(env, ios);
if (vio->cui_ra_window_set)
ll_ra_read_ex(cio->cui_fd->fd_file, &vio->cui_bead);
vvp_io_fini(env, ios);
}
static int vvp_io_fault_iter_init(const struct lu_env *env,
const struct cl_io_slice *ios)
{
struct vvp_io *vio = cl2vvp_io(env, ios);
struct inode *inode = ccc_object_inode(ios->cis_obj);
LASSERT(inode ==
cl2ccc_io(env, ios)->cui_fd->fd_file->f_dentry->d_inode);
vio->u.fault.ft_mtime = LTIME_S(inode->i_mtime);
return 0;
}
static int vvp_io_read_page(const struct lu_env *env,
const struct cl_io_slice *ios,
const struct cl_page_slice *slice)
{
struct cl_io *io = ios->cis_io;
struct cl_object *obj = slice->cpl_obj;
struct ccc_page *cp = cl2ccc_page(slice);
struct cl_page *page = slice->cpl_page;
struct inode *inode = ccc_object_inode(obj);
struct ll_sb_info *sbi = ll_i2sbi(inode);
struct ll_file_data *fd = cl2ccc_io(env, ios)->cui_fd;
struct ll_readahead_state *ras = &fd->fd_ras;
struct page *vmpage = cp->cpg_page;
struct cl_2queue *queue = &io->ci_queue;
int rc;
CLOBINVRNT(env, obj, ccc_object_invariant(obj));
LASSERT(slice->cpl_obj == obj);
ENTRY;
if (sbi->ll_ra_info.ra_max_pages_per_file &&
sbi->ll_ra_info.ra_max_pages)
ras_update(sbi, inode, ras, page->cp_index,
cp->cpg_defer_uptodate);
/* Sanity check whether the page is protected by a lock. */
rc = cl_page_is_under_lock(env, io, page);
if (rc != -EBUSY) {
CL_PAGE_HEADER(D_WARNING, env, page, "%s: %d\n",
rc == -ENODATA ? "without a lock" :
"match failed", rc);
if (rc != -ENODATA)
RETURN(rc);
}
if (cp->cpg_defer_uptodate) {
cp->cpg_ra_used = 1;
cl_page_export(env, page, 1);
}
/*
* Add page into the queue even when it is marked uptodate above.
* this will unlock it automatically as part of cl_page_list_disown().
*/
cl_2queue_add(queue, page);
if (sbi->ll_ra_info.ra_max_pages_per_file &&
sbi->ll_ra_info.ra_max_pages)
ll_readahead(env, io, ras,
vmpage->mapping, &queue->c2_qin, fd->fd_flags);
RETURN(0);
}
static int vvp_io_write_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
struct ccc_io *cio = cl2ccc_io(env, ios);
struct cl_io *io = ios->cis_io;
struct cl_object *obj = io->ci_obj;
struct inode *inode = ccc_object_inode(obj);
struct file *file = cio->cui_fd->fd_file;
ssize_t result = 0;
loff_t pos = io->u.ci_wr.wr.crw_pos;
size_t cnt = io->u.ci_wr.wr.crw_count;
ENTRY;
if (!can_populate_pages(env, io, inode))
return 0;
if (cl_io_is_append(io)) {
/*
* PARALLEL IO This has to be changed for parallel IO doing
* out-of-order writes.
*/
pos = io->u.ci_wr.wr.crw_pos = i_size_read(inode);
cio->cui_iocb->ki_pos = pos;
}
CDEBUG(D_VFSTRACE, "write: [%lli, %lli)\n", pos, pos + (long long)cnt);
if (cio->cui_iov == NULL) /* from a temp io in ll_cl_init(). */
result = 0;
else
result = lustre_generic_file_write(file, cio, &pos);
if (result > 0) {
if (result < cnt)
io->ci_continue = 0;
io->ci_nob += result;
ll_rw_stats_tally(ll_i2sbi(inode), current->pid,
cio->cui_fd, pos, result, WRITE);
result = 0;
}
RETURN(result);
}
static int slp_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
{
struct ccc_io *cio = cl2ccc_io(env, ios);
struct cl_io *io = ios->cis_io;
struct cl_object *obj = io->ci_obj;
struct inode *inode = ccc_object_inode(obj);
int err, ret;
loff_t pos;
long cnt;
struct llu_io_group *iogroup;
struct lustre_rw_params p = {0};
int iovidx;
struct intnl_stat *st = llu_i2stat(inode);
struct llu_inode_info *lli = llu_i2info(inode);
struct llu_io_session *session = cl2slp_io(env, ios)->sio_session;
int write = io->ci_type == CIT_WRITE;
int exceed = 0;
CLOBINVRNT(env, obj, ccc_object_invariant(obj));
if (write) {
pos = io->u.ci_wr.wr.crw_pos;
cnt = io->u.ci_wr.wr.crw_count;
} else {
pos = io->u.ci_rd.rd.crw_pos;
cnt = io->u.ci_rd.rd.crw_count;
}
if (io->u.ci_wr.wr_append) {
p.lrp_lock_mode = LCK_PW;
} else {
p.lrp_brw_flags = OBD_BRW_SRVLOCK;
p.lrp_lock_mode = LCK_NL;
}
iogroup = get_io_group(inode, max_io_pages(cnt, cio->cui_nrsegs), &p);
if (IS_ERR(iogroup))
RETURN(PTR_ERR(iogroup));
err = ccc_prep_size(env, obj, io, pos, cnt, &exceed);
if (err != 0 || (write == 0 && exceed != 0))
GOTO(out, err);
CDEBUG(D_INODE,
"%s ino %lu, %lu bytes, offset "LPU64", i_size "LPU64"\n",
write ? "Write" : "Read", (unsigned long)st->st_ino,
cnt, (__u64)pos, (__u64)st->st_size);
if (write && io->u.ci_wr.wr_append)
pos = io->u.ci_wr.wr.crw_pos = st->st_size; /* XXX? Do we need to change io content too here? */
/* XXX What about if one write syscall writes at 2 different offsets? */
for (iovidx = 0; iovidx < cio->cui_nrsegs; iovidx++) {
char *buf = (char *) cio->cui_iov[iovidx].iov_base;
long count = cio->cui_iov[iovidx].iov_len;
if (!count)
continue;
if (cnt < count)
count = cnt;
if (IS_BAD_PTR(buf) || IS_BAD_PTR(buf + count)) {
GOTO(out, err = -EFAULT);
}
if (io->ci_type == CIT_READ) {
if (/* local_lock && */ pos >= st->st_size)
break;
} else if (io->ci_type == CIT_WRITE) {
if (pos >= lli->lli_maxbytes) {
GOTO(out, err = -EFBIG);
}
if (pos + count >= lli->lli_maxbytes)
count = lli->lli_maxbytes - pos;
} else {
LBUG();
}
ret = llu_queue_pio(env, io, iogroup, buf, count, pos);
if (ret < 0) {
GOTO(out, err = ret);
} else {
io->ci_nob += ret;
pos += ret;
cnt -= ret;
if (io->ci_type == CIT_WRITE) {
// obd_adjust_kms(exp, lsm, pos, 0); // XXX
if (pos > st->st_size)
st->st_size = pos;
}
if (!cnt)
break;
}
}
LASSERT(cnt == 0 || io->ci_type == CIT_READ); /* libsysio should guarantee this */
if (!iogroup->lig_rc)
session->lis_rwcount += iogroup->lig_rwcount;
else if (!session->lis_rc)
session->lis_rc = iogroup->lig_rc;
err = 0;
out:
put_io_group(iogroup);
return err;
}
static int vvp_io_read_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
struct vvp_io *vio = cl2vvp_io(env, ios);
struct ccc_io *cio = cl2ccc_io(env, ios);
struct cl_io *io = ios->cis_io;
struct cl_object *obj = io->ci_obj;
struct inode *inode = ccc_object_inode(obj);
struct ll_ra_read *bead = &vio->cui_bead;
struct file *file = cio->cui_fd->fd_file;
int result;
loff_t pos = io->u.ci_rd.rd.crw_pos;
long cnt = io->u.ci_rd.rd.crw_count;
long tot = cio->cui_tot_count;
int exceed = 0;
CLOBINVRNT(env, obj, ccc_object_invariant(obj));
CDEBUG(D_VFSTRACE, "read: -> [%lli, %lli)\n", pos, pos + cnt);
if (!can_populate_pages(env, io, inode))
return 0;
result = ccc_prep_size(env, obj, io, pos, tot, &exceed);
if (result != 0)
return result;
else if (exceed != 0)
goto out;
LU_OBJECT_HEADER(D_INODE, env, &obj->co_lu,
"Read ino %lu, %lu bytes, offset %lld, size %llu\n",
inode->i_ino, cnt, pos, i_size_read(inode));
/* turn off the kernel's read-ahead */
cio->cui_fd->fd_file->f_ra.ra_pages = 0;
/* initialize read-ahead window once per syscall */
if (!vio->cui_ra_window_set) {
vio->cui_ra_window_set = 1;
bead->lrr_start = cl_index(obj, pos);
bead->lrr_count = cl_index(obj, tot + PAGE_CACHE_SIZE - 1);
ll_ra_read_in(file, bead);
}
/* BUG: 5972 */
file_accessed(file);
switch (vio->cui_io_subtype) {
case IO_NORMAL:
result = lustre_generic_file_read(file, cio, &pos);
break;
case IO_SPLICE:
result = generic_file_splice_read(file, &pos,
vio->u.splice.cui_pipe, cnt,
vio->u.splice.cui_flags);
/* LU-1109: do splice read stripe by stripe otherwise if it
* may make nfsd stuck if this read occupied all internal pipe
* buffers. */
io->ci_continue = 0;
break;
default:
CERROR("Wrong IO type %u\n", vio->cui_io_subtype);
LBUG();
}
out:
if (result >= 0) {
if (result < cnt)
io->ci_continue = 0;
io->ci_nob += result;
ll_rw_stats_tally(ll_i2sbi(inode), current->pid, cio->cui_fd,
pos, result, READ);
result = 0;
}
return result;
}
static void vvp_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
{
struct cl_io *io = ios->cis_io;
struct cl_object *obj = io->ci_obj;
struct ccc_io *cio = cl2ccc_io(env, ios);
struct inode *inode = ccc_object_inode(obj);
CLOBINVRNT(env, obj, ccc_object_invariant(obj));
CDEBUG(D_VFSTRACE, DFID" ignore/verify layout %d/%d, layout version %d "
"restore needed %d\n",
PFID(lu_object_fid(&obj->co_lu)),
io->ci_ignore_layout, io->ci_verify_layout,
cio->cui_layout_gen, io->ci_restore_needed);
if (io->ci_restore_needed == 1) {
int rc;
/* file was detected release, we need to restore it
* before finishing the io
*/
rc = ll_layout_restore(inode, 0, OBD_OBJECT_EOF);
/* if restore registration failed, no restart,
* we will return -ENODATA */
/* The layout will change after restore, so we need to
* block on layout lock hold by the MDT
* as MDT will not send new layout in lvb (see LU-3124)
* we have to explicitly fetch it, all this will be done
* by ll_layout_refresh()
*/
if (rc == 0) {
io->ci_restore_needed = 0;
io->ci_need_restart = 1;
io->ci_verify_layout = 1;
} else {
io->ci_restore_needed = 1;
io->ci_need_restart = 0;
io->ci_verify_layout = 0;
io->ci_result = rc;
}
}
if (!io->ci_ignore_layout && io->ci_verify_layout) {
__u32 gen = 0;
/* check layout version */
ll_layout_refresh(inode, &gen);
io->ci_need_restart = cio->cui_layout_gen != gen;
if (io->ci_need_restart) {
CDEBUG(D_VFSTRACE,
DFID" layout changed from %d to %d.\n",
PFID(lu_object_fid(&obj->co_lu)),
cio->cui_layout_gen, gen);
/* today successful restore is the only possible
* case */
/* restore was done, clear restoring state */
ll_i2info(ccc_object_inode(obj))->lli_flags &=
~LLIF_FILE_RESTORING;
}
}
}