static int vvp_io_fault_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
struct vvp_io *vio = cl2vvp_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 cl_fault_io *fio = &io->u.ci_fault;
struct vvp_fault_io *cfio = &vio->u.fault;
loff_t offset;
int result = 0;
struct page *vmpage = NULL;
struct cl_page *page;
loff_t size;
pgoff_t last; /* last page in a file data region */
if (fio->ft_executable &&
LTIME_S(inode->i_mtime) != vio->u.fault.ft_mtime)
CWARN("binary "DFID
" changed while waiting for the page fault lock\n",
PFID(lu_object_fid(&obj->co_lu)));
/* offset of the last byte on the page */
offset = cl_offset(obj, fio->ft_index + 1) - 1;
LASSERT(cl_index(obj, offset) == fio->ft_index);
result = ccc_prep_size(env, obj, io, 0, offset + 1, NULL);
if (result != 0)
return result;
/* must return locked page */
if (fio->ft_mkwrite) {
LASSERT(cfio->ft_vmpage != NULL);
lock_page(cfio->ft_vmpage);
} else {
result = vvp_io_kernel_fault(cfio);
if (result != 0)
return result;
}
vmpage = cfio->ft_vmpage;
LASSERT(PageLocked(vmpage));
if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_FAULT_TRUNC_RACE))
ll_invalidate_page(vmpage);
size = i_size_read(inode);
/* Though we have already held a cl_lock upon this page, but
* it still can be truncated locally. */
if (unlikely((vmpage->mapping != inode->i_mapping) ||
(page_offset(vmpage) > size))) {
CDEBUG(D_PAGE, "llite: fault and truncate race happened!\n");
/* return +1 to stop cl_io_loop() and ll_fault() will catch
* and retry. */
GOTO(out, result = +1);
}
if (fio->ft_mkwrite ) {
pgoff_t last_index;
/*
* Capture the size while holding the lli_trunc_sem from above
* we want to make sure that we complete the mkwrite action
* while holding this lock. We need to make sure that we are
* not past the end of the file.
*/
last_index = cl_index(obj, size - 1);
if (last_index < fio->ft_index) {
CDEBUG(D_PAGE,
"llite: mkwrite and truncate race happened: "
"%p: 0x%lx 0x%lx\n",
vmpage->mapping,fio->ft_index,last_index);
/*
* We need to return if we are
* passed the end of the file. This will propagate
* up the call stack to ll_page_mkwrite where
* we will return VM_FAULT_NOPAGE. Any non-negative
* value returned here will be silently
* converted to 0. If the vmpage->mapping is null
* the error code would be converted back to ENODATA
* in ll_page_mkwrite0. Thus we return -ENODATA
* to handle both cases
*/
GOTO(out, result = -ENODATA);
}
}
page = cl_page_find(env, obj, fio->ft_index, vmpage, CPT_CACHEABLE);
if (IS_ERR(page))
GOTO(out, result = PTR_ERR(page));
/* if page is going to be written, we should add this page into cache
* earlier. */
if (fio->ft_mkwrite) {
wait_on_page_writeback(vmpage);
if (set_page_dirty(vmpage)) {
struct ccc_page *cp;
/* vvp_page_assume() calls wait_on_page_writeback(). */
cl_page_assume(env, io, page);
cp = cl2ccc_page(cl_page_at(page, &vvp_device_type));
vvp_write_pending(cl2ccc(obj), cp);
/* Do not set Dirty bit here so that in case IO is
* started before the page is really made dirty, we
* still have chance to detect it. */
result = cl_page_cache_add(env, io, page, CRT_WRITE);
LASSERT(cl_page_is_owned(page, io));
vmpage = NULL;
if (result < 0) {
cl_page_unmap(env, io, page);
cl_page_discard(env, io, page);
cl_page_disown(env, io, page);
cl_page_put(env, page);
/* we're in big trouble, what can we do now? */
if (result == -EDQUOT)
result = -ENOSPC;
GOTO(out, result);
} else
cl_page_disown(env, io, page);
}
}
last = cl_index(obj, size - 1);
/*
* The ft_index is only used in the case of
* a mkwrite action. We need to check
* our assertions are correct, since
* we should have caught this above
*/
LASSERT(!fio->ft_mkwrite || fio->ft_index <= last);
if (fio->ft_index == last)
/*
* Last page is mapped partially.
*/
fio->ft_nob = size - cl_offset(obj, fio->ft_index);
else
fio->ft_nob = cl_page_size(obj);
lu_ref_add(&page->cp_reference, "fault", io);
fio->ft_page = page;
EXIT;
out:
/* return unlocked vmpage to avoid deadlocking */
if (vmpage != NULL)
unlock_page(vmpage);
cfio->fault.ft_flags &= ~VM_FAULT_LOCKED;
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;
}
