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 int lov_init_sub(const struct lu_env *env, struct lov_object *lov,
struct cl_object *stripe, struct lov_layout_raid0 *r0,
int idx)
{
struct cl_object_header *hdr;
struct cl_object_header *subhdr;
struct cl_object_header *parent;
struct lov_oinfo *oinfo;
int result;
if (OBD_FAIL_CHECK(OBD_FAIL_LOV_INIT)) {
/* For sanity:test_206.
* Do not leave the object in cache to avoid accessing
* freed memory. This is because osc_object is referring to
* lov_oinfo of lsm_stripe_data which will be freed due to
* this failure. */
cl_object_kill(env, stripe);
cl_object_put(env, stripe);
return -EIO;
}
hdr = cl_object_header(lov2cl(lov));
subhdr = cl_object_header(stripe);
oinfo = lov->lo_lsm->lsm_oinfo[idx];
CDEBUG(D_INODE, DFID"@%p[%d] -> "DFID"@%p: ostid: "DOSTID
" idx: %d gen: %d\n",
PFID(&subhdr->coh_lu.loh_fid), subhdr, idx,
PFID(&hdr->coh_lu.loh_fid), hdr, POSTID(&oinfo->loi_oi),
oinfo->loi_ost_idx, oinfo->loi_ost_gen);
/* reuse ->coh_attr_guard to protect coh_parent change */
spin_lock(&subhdr->coh_attr_guard);
parent = subhdr->coh_parent;
if (parent == NULL) {
subhdr->coh_parent = hdr;
spin_unlock(&subhdr->coh_attr_guard);
subhdr->coh_nesting = hdr->coh_nesting + 1;
lu_object_ref_add(&stripe->co_lu, "lov-parent", lov);
r0->lo_sub[idx] = cl2lovsub(stripe);
r0->lo_sub[idx]->lso_super = lov;
r0->lo_sub[idx]->lso_index = idx;
result = 0;
} else {
struct lu_object *old_obj;
struct lov_object *old_lov;
unsigned int mask = D_INODE;
spin_unlock(&subhdr->coh_attr_guard);
old_obj = lu_object_locate(&parent->coh_lu, &lov_device_type);
LASSERT(old_obj != NULL);
old_lov = cl2lov(lu2cl(old_obj));
if (old_lov->lo_layout_invalid) {
/* the object's layout has already changed but isn't
* refreshed */
lu_object_unhash(env, &stripe->co_lu);
result = -EAGAIN;
} else {
mask = D_ERROR;
result = -EIO;
}
LU_OBJECT_DEBUG(mask, env, &stripe->co_lu,
"stripe %d is already owned.\n", idx);
LU_OBJECT_DEBUG(mask, env, old_obj, "owned.\n");
LU_OBJECT_HEADER(mask, env, lov2lu(lov), "try to own.\n");
cl_object_put(env, stripe);
}
return result;
}
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;
}