static int vvp_io_prepare_write(const struct lu_env *env,
const struct cl_io_slice *ios,
const struct cl_page_slice *slice,
unsigned from, unsigned to)
{
struct cl_object *obj = slice->cpl_obj;
struct ccc_page *cp = cl2ccc_page(slice);
struct cl_page *pg = slice->cpl_page;
struct page *vmpage = cp->cpg_page;
int result;
ENTRY;
LINVRNT(cl_page_is_vmlocked(env, pg));
LASSERT(vmpage->mapping->host == ccc_object_inode(obj));
result = 0;
CL_PAGE_HEADER(D_PAGE, env, pg, "preparing: [%d, %d]\n", from, to);
if (!PageUptodate(vmpage)) {
/*
* We're completely overwriting an existing page, so _don't_
* set it up to date until commit_write
*/
if (from == 0 && to == PAGE_CACHE_SIZE) {
CL_PAGE_HEADER(D_PAGE, env, pg, "full page write\n");
POISON_PAGE(page, 0x11);
} else
result = vvp_io_prepare_partial(env, ios->cis_io, obj,
pg, cp, from, to);
} else
CL_PAGE_HEADER(D_PAGE, env, pg, "uptodate\n");
RETURN(result);
}
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_prune(const struct lu_env *env, struct cl_object *obj)
{
struct inode *inode = ccc_object_inode(obj);
int rc;
ENTRY;
rc = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_ALL, 1);
if (rc == 0)
truncate_inode_pages(inode->i_mapping, 0);
RETURN(rc);
}
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 void vvp_io_setattr_end(const struct lu_env *env,
const struct cl_io_slice *ios)
{
struct cl_io *io = ios->cis_io;
struct inode *inode = ccc_object_inode(io->ci_obj);
if (cl_io_is_trunc(io)) {
/* Truncate in memory pages - they must be clean pages
* because osc has already notified to destroy osc_extents. */
vvp_do_vmtruncate(inode, io->u.ci_setattr.sa_attr.lvb_size);
inode_dio_write_done(inode);
}
mutex_unlock(&inode->i_mutex);
}
static int slp_attr_get(const struct lu_env *env, struct cl_object *obj,
struct cl_attr *attr)
{
struct inode *inode = ccc_object_inode(obj);
struct intnl_stat *st = llu_i2stat(inode);
attr->cat_size = st->st_size;
attr->cat_blocks = st->st_blocks;
attr->cat_mtime = st->st_mtime;
attr->cat_atime = st->st_atime;
attr->cat_ctime = st->st_ctime;
/* KMS is not known by this layer */
return 0; /* layers below have to fill in the rest */
}
static int vvp_io_setattr_start(const struct lu_env *env,
const struct cl_io_slice *ios)
{
struct cl_io *io = ios->cis_io;
struct inode *inode = ccc_object_inode(io->ci_obj);
int result = 0;
mutex_lock(&inode->i_mutex);
if (cl_io_is_trunc(io))
result = vvp_io_setattr_trunc(env, ios, inode,
io->u.ci_setattr.sa_attr.lvb_size);
if (result == 0)
result = vvp_io_setattr_time(env, ios);
return result;
}
static int vvp_io_read_lock(const struct lu_env *env,
const struct cl_io_slice *ios)
{
struct cl_io *io = ios->cis_io;
struct ll_inode_info *lli = ll_i2info(ccc_object_inode(io->ci_obj));
int result;
ENTRY;
/* XXX: Layer violation, we shouldn't see lsm at llite level. */
if (lli->lli_has_smd) /* lsm-less file doesn't need to lock */
result = vvp_io_rw_lock(env, io, CLM_READ,
io->u.ci_rd.rd.crw_pos,
io->u.ci_rd.rd.crw_pos +
io->u.ci_rd.rd.crw_count - 1);
else
result = 0;
RETURN(result);
}
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 vvp_attr_get(const struct lu_env *env, struct cl_object *obj,
struct cl_attr *attr)
{
struct inode *inode = ccc_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 = LTIME_S(inode->i_mtime);
attr->cat_atime = LTIME_S(inode->i_atime);
attr->cat_ctime = LTIME_S(inode->i_ctime);
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_set(const struct lu_env *env, struct cl_object *obj,
const struct cl_attr *attr, unsigned valid)
{
struct inode *inode = ccc_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)
LTIME_S(inode->i_atime) = attr->cat_atime;
if (valid & CAT_MTIME)
LTIME_S(inode->i_mtime) = attr->cat_mtime;
if (valid & CAT_CTIME)
LTIME_S(inode->i_ctime) = attr->cat_ctime;
if (0 && valid & CAT_SIZE)
cl_isize_write_nolock(inode, attr->cat_size);
/* not currently necessary */
if (0 && valid & (CAT_UID|CAT_GID|CAT_SIZE))
mark_inode_dirty(inode);
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;
struct cl_env_nest nest;
int result;
sigset_t set;
struct inode *inode;
struct ll_inode_info *lli;
io = ll_fault_io_init(vma, &env, &nest, 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));
/* we grab lli_trunc_sem to exclude truncate case.
* Otherwise, we could add dirty pages into osc cache
* while truncate is on-going.
*/
inode = ccc_object_inode(io->ci_obj);
lli = ll_i2info(inode);
down_read(&lli->lli_trunc_sem);
result = cl_io_loop(env, io);
up_read(&lli->lli_trunc_sem);
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 == 0) {
spin_lock(&lli->lli_lock);
lli->lli_flags |= LLIF_DATA_MODIFIED;
spin_unlock(&lli->lli_lock);
}
}
out_io:
cl_io_fini(env, io);
cl_env_nested_put(&nest, env);
out:
CDEBUG(D_MMAP, "%s mkwrite with %d\n", current->comm, result);
LASSERT(ergo(result == 0, PageLocked(vmpage)));
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 llu_queue_pio(const struct lu_env *env, struct cl_io *io,
struct llu_io_group *group,
char *buf, size_t count, loff_t pos)
{
struct cl_object *obj = io->ci_obj;
struct inode *inode = ccc_object_inode(obj);
struct intnl_stat *st = llu_i2stat(inode);
struct obd_export *exp = llu_i2obdexp(inode);
struct page *page;
int rc = 0, ret_bytes = 0;
struct cl_page *clp;
struct cl_2queue *queue;
ENTRY;
if (!exp)
RETURN(-EINVAL);
queue = &io->ci_queue;
cl_2queue_init(queue);
/* prepare the pages array */
do {
unsigned long index, offset, bytes;
offset = (pos & ~CFS_PAGE_MASK);
index = pos >> PAGE_CACHE_SHIFT;
bytes = PAGE_CACHE_SIZE - offset;
if (bytes > count)
bytes = count;
/* prevent read beyond file range */
if (/* local_lock && */
io->ci_type == CIT_READ && pos + bytes >= st->st_size) {
if (pos >= st->st_size)
break;
bytes = st->st_size - pos;
}
/* prepare page for this index */
page = llu_get_user_page(index, buf - offset, offset, bytes);
if (!page) {
rc = -ENOMEM;
break;
}
clp = cl_page_find(env, obj,
cl_index(obj, pos),
page, CPT_TRANSIENT);
if (IS_ERR(clp)) {
rc = PTR_ERR(clp);
break;
}
rc = cl_page_own(env, io, clp);
if (rc) {
LASSERT(clp->cp_state == CPS_FREEING);
cl_page_put(env, clp);
break;
}
cl_2queue_add(queue, clp);
/* drop the reference count for cl_page_find, so that the page
* will be freed in cl_2queue_fini. */
cl_page_put(env, clp);
cl_page_clip(env, clp, offset, offset+bytes);
count -= bytes;
pos += bytes;
buf += bytes;
group->lig_rwcount += bytes;
ret_bytes += bytes;
page++;
} while (count);
if (rc == 0) {
enum cl_req_type iot;
iot = io->ci_type == CIT_READ ? CRT_READ : CRT_WRITE;
rc = cl_io_submit_sync(env, io, iot, queue, 0);
}
group->lig_rc = rc;
cl_2queue_discard(env, io, queue);
cl_2queue_disown(env, io, queue);
cl_2queue_fini(env, queue);
RETURN(ret_bytes);
}
static int vvp_io_commit_write(const struct lu_env *env,
const struct cl_io_slice *ios,
const struct cl_page_slice *slice,
unsigned from, unsigned to)
{
struct cl_object *obj = slice->cpl_obj;
struct cl_io *io = ios->cis_io;
struct ccc_page *cp = cl2ccc_page(slice);
struct cl_page *pg = slice->cpl_page;
struct inode *inode = ccc_object_inode(obj);
struct ll_sb_info *sbi = ll_i2sbi(inode);
struct ll_inode_info *lli = ll_i2info(inode);
struct page *vmpage = cp->cpg_page;
int result;
int tallyop;
loff_t size;
ENTRY;
LINVRNT(cl_page_is_vmlocked(env, pg));
LASSERT(vmpage->mapping->host == inode);
LU_OBJECT_HEADER(D_INODE, env, &obj->co_lu, "commiting page write\n");
CL_PAGE_HEADER(D_PAGE, env, pg, "committing: [%d, %d]\n", from, to);
/*
* queue a write for some time in the future the first time we
* dirty the page.
*
* This is different from what other file systems do: they usually
* just mark page (and some of its buffers) dirty and rely on
* balance_dirty_pages() to start a write-back. Lustre wants write-back
* to be started earlier for the following reasons:
*
* (1) with a large number of clients we need to limit the amount
* of cached data on the clients a lot;
*
* (2) large compute jobs generally want compute-only then io-only
* and the IO should complete as quickly as possible;
*
* (3) IO is batched up to the RPC size and is async until the
* client max cache is hit
* (/proc/fs/lustre/osc/OSC.../max_dirty_mb)
*
*/
if (!PageDirty(vmpage)) {
tallyop = LPROC_LL_DIRTY_MISSES;
result = cl_page_cache_add(env, io, pg, CRT_WRITE);
if (result == 0) {
/* page was added into cache successfully. */
set_page_dirty(vmpage);
vvp_write_pending(cl2ccc(obj), cp);
} else if (result == -EDQUOT) {
pgoff_t last_index = i_size_read(inode) >> PAGE_CACHE_SHIFT;
bool need_clip = true;
/*
* Client ran out of disk space grant. Possible
* strategies are:
*
* (a) do a sync write, renewing grant;
*
* (b) stop writing on this stripe, switch to the
* next one.
*
* (b) is a part of "parallel io" design that is the
* ultimate goal. (a) is what "old" client did, and
* what the new code continues to do for the time
* being.
*/
if (last_index > pg->cp_index) {
to = PAGE_CACHE_SIZE;
need_clip = false;
} else if (last_index == pg->cp_index) {
int size_to = i_size_read(inode) & ~CFS_PAGE_MASK;
if (to < size_to)
to = size_to;
}
if (need_clip)
cl_page_clip(env, pg, 0, to);
result = vvp_page_sync_io(env, io, pg, cp, CRT_WRITE);
if (result)
CERROR("Write page %lu of inode %p failed %d\n",
pg->cp_index, inode, 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;
}
}
}
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 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;
}
