static int osc_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 osc_io *oio = cl2osc_io(env, ios);
struct osc_page *opg = cl2osc_page(slice);
struct osc_object *obj = cl2osc(opg->ops_cl.cpl_obj);
struct osc_async_page *oap = &opg->ops_oap;
LASSERT(to > 0);
/*
* XXX instead of calling osc_page_touch() here and in
* osc_io_fault_start() it might be more logical to introduce
* cl_page_touch() method, that generic cl_io_commit_write() and page
* fault code calls.
*/
osc_page_touch(env, cl2osc_page(slice), to);
if (!client_is_remote(osc_export(obj)) &&
capable(CFS_CAP_SYS_RESOURCE))
oap->oap_brw_flags |= OBD_BRW_NOQUOTA;
if (oio->oi_lockless)
/* see osc_io_prepare_write() for lockless io handling. */
cl_page_clip(env, slice->cpl_page, from, to);
return 0;
}
ssize_t ll_direct_rw_pages(const struct lu_env *env, struct cl_io *io,
int rw, struct inode *inode,
struct ll_dio_pages *pv)
{
struct cl_page *clp;
struct cl_2queue *queue;
struct cl_object *obj = io->ci_obj;
int i;
ssize_t rc = 0;
loff_t file_offset = pv->ldp_start_offset;
size_t size = pv->ldp_size;
int page_count = pv->ldp_nr;
struct page **pages = pv->ldp_pages;
size_t page_size = cl_page_size(obj);
bool do_io;
int io_pages = 0;
ENTRY;
queue = &io->ci_queue;
cl_2queue_init(queue);
for (i = 0; i < page_count; i++) {
if (pv->ldp_offsets)
file_offset = pv->ldp_offsets[i];
LASSERT(!(file_offset & (page_size - 1)));
clp = cl_page_find(env, obj, cl_index(obj, file_offset),
pv->ldp_pages[i], 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;
}
do_io = true;
/* check the page type: if the page is a host page, then do
* write directly */
if (clp->cp_type == CPT_CACHEABLE) {
struct page *vmpage = cl_page_vmpage(clp);
struct page *src_page;
struct page *dst_page;
void *src;
void *dst;
src_page = (rw == WRITE) ? pages[i] : vmpage;
dst_page = (rw == WRITE) ? vmpage : pages[i];
src = ll_kmap_atomic(src_page, KM_USER0);
dst = ll_kmap_atomic(dst_page, KM_USER1);
memcpy(dst, src, min(page_size, size));
ll_kunmap_atomic(dst, KM_USER1);
ll_kunmap_atomic(src, KM_USER0);
/* make sure page will be added to the transfer by
* cl_io_submit()->...->vvp_page_prep_write(). */
if (rw == WRITE)
set_page_dirty(vmpage);
if (rw == READ) {
/* do not issue the page for read, since it
* may reread a ra page which has NOT uptodate
* bit set. */
cl_page_disown(env, io, clp);
do_io = false;
}
}
if (likely(do_io)) {
cl_2queue_add(queue, clp);
/*
* Set page clip to tell transfer formation engine
* that page has to be sent even if it is beyond KMS.
*/
cl_page_clip(env, clp, 0, min(size, page_size));
++io_pages;
}
/* drop the reference count for cl_page_find */
cl_page_put(env, clp);
size -= page_size;
file_offset += page_size;
}
if (rc == 0 && io_pages) {
rc = cl_io_submit_sync(env, io,
rw == READ ? CRT_READ : CRT_WRITE,
queue, 0);
}
if (rc == 0)
rc = pv->ldp_size;
cl_2queue_discard(env, io, queue);
cl_2queue_disown(env, io, queue);
cl_2queue_fini(env, queue);
RETURN(rc);
}
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 int osc_io_commit_async(const struct lu_env *env,
const struct cl_io_slice *ios,
struct cl_page_list *qin, int from, int to,
cl_commit_cbt cb)
{
struct cl_io *io = ios->cis_io;
struct osc_io *oio = cl2osc_io(env, ios);
struct osc_object *osc = cl2osc(ios->cis_obj);
struct cl_page *page;
struct cl_page *last_page;
struct osc_page *opg;
int result = 0;
ENTRY;
LASSERT(qin->pl_nr > 0);
/* Handle partial page cases */
last_page = cl_page_list_last(qin);
if (oio->oi_lockless) {
page = cl_page_list_first(qin);
if (page == last_page) {
cl_page_clip(env, page, from, to);
} else {
if (from != 0)
cl_page_clip(env, page, from, PAGE_SIZE);
if (to != PAGE_SIZE)
cl_page_clip(env, last_page, 0, to);
}
}
while (qin->pl_nr > 0) {
struct osc_async_page *oap;
page = cl_page_list_first(qin);
opg = osc_cl_page_osc(page, osc);
oap = &opg->ops_oap;
if (!list_empty(&oap->oap_rpc_item)) {
CDEBUG(D_CACHE, "Busy oap %p page %p for submit.\n",
oap, opg);
result = -EBUSY;
break;
}
/* The page may be already in dirty cache. */
if (list_empty(&oap->oap_pending_item)) {
result = osc_page_cache_add(env, &opg->ops_cl, io);
if (result != 0)
break;
}
osc_page_touch_at(env, osc2cl(osc), osc_index(opg),
page == last_page ? to : PAGE_SIZE);
cl_page_list_del(env, qin, page);
(*cb)(env, io, page);
/* Can't access page any more. Page can be in transfer and
* complete at any time. */
}
/* for sync write, kernel will wait for this page to be flushed before
* osc_io_end() is called, so release it earlier.
* for mkwrite(), it's known there is no further pages. */
if (cl_io_is_sync_write(io) && oio->oi_active != NULL) {
osc_extent_release(env, oio->oi_active);
oio->oi_active = NULL;
}
CDEBUG(D_INFO, "%d %d\n", qin->pl_nr, result);
RETURN(result);
}
