Whamcloud - gitweb
LU-10810 clio: SEEK_HOLE/SEEK_DATA on client side
[fs/lustre-release.git] / lustre / obdclass / cl_io.c
index c4a0fd7..21fa939 100644 (file)
@@ -45,6 +45,7 @@
 #include <lustre_fid.h>
 #include <cl_object.h>
 #include "cl_internal.h"
+#include <libcfs/crypto/llcrypt.h>
 
 /*****************************************************************************
  *
@@ -125,6 +126,7 @@ void cl_io_fini(const struct lu_env *env, struct cl_io *io)
        case CIT_GLIMPSE:
                break;
        case CIT_LADVISE:
+       case CIT_LSEEK:
                break;
        default:
                LBUG();
@@ -562,7 +564,9 @@ int cl_io_read_ahead(const struct lu_env *env, struct cl_io *io,
        const struct cl_io_slice *scan;
        int                       result = 0;
 
-       LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_FAULT);
+       LINVRNT(io->ci_type == CIT_READ ||
+               io->ci_type == CIT_FAULT ||
+               io->ci_type == CIT_WRITE);
        LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED);
        LINVRNT(cl_io_invariant(io));
        ENTRY;
@@ -648,6 +652,7 @@ int cl_io_submit_sync(const struct lu_env *env, struct cl_io *io,
        struct cl_sync_io *anchor = &cl_env_info(env)->clt_anchor;
        struct cl_page *pg;
        int rc;
+       ENTRY;
 
        cl_page_list_for_each(pg, &queue->c2_qin) {
                LASSERT(pg->cp_sync_io == NULL);
@@ -676,30 +681,11 @@ int cl_io_submit_sync(const struct lu_env *env, struct cl_io *io,
                cl_page_list_for_each(pg, &queue->c2_qin)
                        pg->cp_sync_io = NULL;
        }
-       return rc;
+       RETURN(rc);
 }
 EXPORT_SYMBOL(cl_io_submit_sync);
 
 /**
- * Cancel an IO which has been submitted by cl_io_submit_rw.
- */
-int cl_io_cancel(const struct lu_env *env, struct cl_io *io,
-                 struct cl_page_list *queue)
-{
-        struct cl_page *page;
-        int result = 0;
-
-        CERROR("Canceling ongoing page trasmission\n");
-        cl_page_list_for_each(page, queue) {
-                int rc;
-
-                rc = cl_page_cancel(env, page);
-                result = result ?: rc;
-        }
-        return result;
-}
-
-/**
  * Main io loop.
  *
  * Pumps io through iterations calling
@@ -720,7 +706,8 @@ int cl_io_cancel(const struct lu_env *env, struct cl_io *io,
  */
 int cl_io_loop(const struct lu_env *env, struct cl_io *io)
 {
-       int result   = 0;
+       int result = 0;
+       int rc = 0;
 
        LINVRNT(cl_io_is_loopable(io));
        ENTRY;
@@ -754,7 +741,13 @@ int cl_io_loop(const struct lu_env *env, struct cl_io *io)
                        }
                }
                cl_io_iter_fini(env, io);
-       } while (result == 0 && io->ci_continue);
+               if (result)
+                       rc = result;
+       } while ((result == 0 || result == -EIOCBQUEUED) &&
+                io->ci_continue);
+
+       if (rc && !result)
+               result = rc;
 
        if (result == -EWOULDBLOCK && io->ci_ndelay) {
                io->ci_need_restart = 1;
@@ -803,7 +796,6 @@ void cl_page_list_init(struct cl_page_list *plist)
        ENTRY;
        plist->pl_nr = 0;
        INIT_LIST_HEAD(&plist->pl_pages);
-       plist->pl_owner = current;
        EXIT;
 }
 EXPORT_SYMBOL(cl_page_list_init);
@@ -817,7 +809,6 @@ void cl_page_list_add(struct cl_page_list *plist, struct cl_page *page)
        /* it would be better to check that page is owned by "current" io, but
         * it is not passed here. */
        LASSERT(page->cp_owner != NULL);
-       LINVRNT(plist->pl_owner == current);
 
        LASSERT(list_empty(&page->cp_batch));
        list_add_tail(&page->cp_batch, &plist->pl_pages);
@@ -836,7 +827,6 @@ void cl_page_list_del(const struct lu_env *env,
 {
        LASSERT(plist->pl_nr > 0);
        LASSERT(cl_page_is_vmlocked(env, page));
-       LINVRNT(plist->pl_owner == current);
 
        ENTRY;
        list_del_init(&page->cp_batch);
@@ -854,8 +844,6 @@ void cl_page_list_move(struct cl_page_list *dst, struct cl_page_list *src,
                       struct cl_page *page)
 {
        LASSERT(src->pl_nr > 0);
-       LINVRNT(dst->pl_owner == current);
-       LINVRNT(src->pl_owner == current);
 
        ENTRY;
        list_move_tail(&page->cp_batch, &dst->pl_pages);
@@ -874,8 +862,6 @@ void cl_page_list_move_head(struct cl_page_list *dst, struct cl_page_list *src,
                            struct cl_page *page)
 {
        LASSERT(src->pl_nr > 0);
-       LINVRNT(dst->pl_owner == current);
-       LINVRNT(src->pl_owner == current);
 
        ENTRY;
        list_move(&page->cp_batch, &dst->pl_pages);
@@ -895,8 +881,6 @@ void cl_page_list_splice(struct cl_page_list *list, struct cl_page_list *head)
        struct cl_page *page;
        struct cl_page *tmp;
 
-       LINVRNT(list->pl_owner == current);
-       LINVRNT(head->pl_owner == current);
 
        ENTRY;
        cl_page_list_for_each_safe(page, tmp, list)
@@ -914,7 +898,6 @@ void cl_page_list_disown(const struct lu_env *env,
        struct cl_page *page;
        struct cl_page *temp;
 
-       LINVRNT(plist->pl_owner == current);
 
        ENTRY;
        cl_page_list_for_each_safe(page, temp, plist) {
@@ -947,7 +930,6 @@ void cl_page_list_fini(const struct lu_env *env, struct cl_page_list *plist)
        struct cl_page *page;
        struct cl_page *temp;
 
-       LINVRNT(plist->pl_owner == current);
 
        ENTRY;
        cl_page_list_for_each_safe(page, temp, plist)
@@ -965,7 +947,6 @@ void cl_page_list_assume(const struct lu_env *env,
 {
        struct cl_page *page;
 
-       LINVRNT(plist->pl_owner == current);
 
        cl_page_list_for_each(page, plist)
                cl_page_assume(env, io, page);
@@ -979,7 +960,6 @@ void cl_page_list_discard(const struct lu_env *env, struct cl_io *io,
 {
        struct cl_page *page;
 
-       LINVRNT(plist->pl_owner == current);
        ENTRY;
        cl_page_list_for_each(page, plist)
                cl_page_discard(env, io, page);
@@ -1122,7 +1102,7 @@ EXPORT_SYMBOL(cl_req_attr_set);
  */
 
 void cl_sync_io_init_notify(struct cl_sync_io *anchor, int nr,
-                           cl_sync_io_end_t *end)
+                           struct cl_dio_aio *aio, cl_sync_io_end_t *end)
 {
        ENTRY;
        memset(anchor, 0, sizeof(*anchor));
@@ -1130,6 +1110,7 @@ void cl_sync_io_init_notify(struct cl_sync_io *anchor, int nr,
        atomic_set(&anchor->csi_sync_nr, nr);
        anchor->csi_sync_rc = 0;
        anchor->csi_end_io = end;
+       anchor->csi_aio = aio;
        EXIT;
 }
 EXPORT_SYMBOL(cl_sync_io_init_notify);
@@ -1141,25 +1122,25 @@ EXPORT_SYMBOL(cl_sync_io_init_notify);
 int cl_sync_io_wait(const struct lu_env *env, struct cl_sync_io *anchor,
                    long timeout)
 {
-       struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(timeout),
-                                                 NULL, NULL, NULL);
-       int rc;
+       int rc = 0;
        ENTRY;
 
        LASSERT(timeout >= 0);
 
-       rc = l_wait_event(anchor->csi_waitq,
-                         atomic_read(&anchor->csi_sync_nr) == 0,
-                         &lwi);
-       if (rc < 0) {
+       if (timeout > 0 &&
+           wait_event_idle_timeout(anchor->csi_waitq,
+                                   atomic_read(&anchor->csi_sync_nr) == 0,
+                                   cfs_time_seconds(timeout)) == 0) {
+               rc = -ETIMEDOUT;
                CERROR("IO failed: %d, still wait for %d remaining entries\n",
                       rc, atomic_read(&anchor->csi_sync_nr));
+       }
 
-               wait_event_idle(anchor->csi_waitq,
-                               atomic_read(&anchor->csi_sync_nr) == 0);
-       } else {
+       wait_event_idle(anchor->csi_waitq,
+                       atomic_read(&anchor->csi_sync_nr) == 0);
+       if (!rc)
                rc = anchor->csi_sync_rc;
-       }
+
        /* We take the lock to ensure that cl_sync_io_note() has finished */
        spin_lock(&anchor->csi_waitq.lock);
        LASSERT(atomic_read(&anchor->csi_sync_nr) == 0);
@@ -1169,6 +1150,76 @@ int cl_sync_io_wait(const struct lu_env *env, struct cl_sync_io *anchor,
 }
 EXPORT_SYMBOL(cl_sync_io_wait);
 
+#ifndef HAVE_AIO_COMPLETE
+static inline void aio_complete(struct kiocb *iocb, ssize_t res, ssize_t res2)
+{
+       if (iocb->ki_complete)
+               iocb->ki_complete(iocb, res, res2);
+}
+#endif
+
+static void cl_aio_end(const struct lu_env *env, struct cl_sync_io *anchor)
+{
+       struct cl_dio_aio *aio = container_of(anchor, typeof(*aio), cda_sync);
+       ssize_t ret = anchor->csi_sync_rc;
+
+       ENTRY;
+
+       /* release pages */
+       while (aio->cda_pages.pl_nr > 0) {
+               struct cl_page *page = cl_page_list_first(&aio->cda_pages);
+               struct page *vmpage = cl_page_vmpage(page);
+               struct inode *inode = vmpage ? page2inode(vmpage) : NULL;
+
+               cl_page_get(page);
+               /* We end up here in case of Direct IO only. For encrypted file,
+                * mapping was set on pages in ll_direct_rw_pages(), so it has
+                * to be cleared now before page cleanup.
+                * PageChecked flag was also set there, so we clean up here.
+                */
+               if (inode && IS_ENCRYPTED(inode)) {
+                       vmpage->mapping = NULL;
+                       ClearPageChecked(vmpage);
+               }
+               cl_page_list_del(env, &aio->cda_pages, page);
+               cl_page_delete(env, page);
+               cl_page_put(env, page);
+       }
+
+       if (!is_sync_kiocb(aio->cda_iocb) && !aio->cda_no_aio_complete)
+               aio_complete(aio->cda_iocb, ret ?: aio->cda_bytes, 0);
+
+       EXIT;
+}
+
+struct cl_dio_aio *cl_aio_alloc(struct kiocb *iocb)
+{
+       struct cl_dio_aio *aio;
+
+       OBD_SLAB_ALLOC_PTR_GFP(aio, cl_dio_aio_kmem, GFP_NOFS);
+       if (aio != NULL) {
+               /*
+                * Hold one ref so that it won't be released until
+                * every pages is added.
+                */
+               cl_sync_io_init_notify(&aio->cda_sync, 1, is_sync_kiocb(iocb) ?
+                                      NULL : aio, cl_aio_end);
+               cl_page_list_init(&aio->cda_pages);
+               aio->cda_iocb = iocb;
+               aio->cda_no_aio_complete = 0;
+       }
+       return aio;
+}
+EXPORT_SYMBOL(cl_aio_alloc);
+
+void cl_aio_free(struct cl_dio_aio *aio)
+{
+       if (aio)
+               OBD_SLAB_FREE_PTR(aio, cl_dio_aio_kmem);
+}
+EXPORT_SYMBOL(cl_aio_free);
+
+
 /**
  * Indicate that transfer of a single page completed.
  */
@@ -1186,6 +1237,8 @@ void cl_sync_io_note(const struct lu_env *env, struct cl_sync_io *anchor,
        LASSERT(atomic_read(&anchor->csi_sync_nr) > 0);
        if (atomic_dec_and_lock(&anchor->csi_sync_nr,
                                &anchor->csi_waitq.lock)) {
+               struct cl_dio_aio *aio = NULL;
+
                cl_sync_io_end_t *end_io = anchor->csi_end_io;
 
                /*
@@ -1199,9 +1252,16 @@ void cl_sync_io_note(const struct lu_env *env, struct cl_sync_io *anchor,
                wake_up_all_locked(&anchor->csi_waitq);
                if (end_io)
                        end_io(env, anchor);
+               if (anchor->csi_aio)
+                       aio = anchor->csi_aio;
+
                spin_unlock(&anchor->csi_waitq.lock);
 
-               /* Can't access anchor any more */
+               /**
+                * If anchor->csi_aio is set, we are responsible for freeing
+                * memory here rather than when cl_sync_io_wait() completes.
+                */
+               cl_aio_free(aio);
        }
        EXIT;
 }