Whamcloud - gitweb
minor fix for buffer offsets.
[fs/lustre-release.git] / lustre / obdfilter / filter_io.c
index 8cf3f59..10cc687 100644 (file)
@@ -33,6 +33,8 @@
 
 #include <linux/obd_class.h>
 #include <linux/lustre_fsfilt.h>
+#include <linux/lustre_smfs.h>
+#include <linux/lustre_snap.h>
 #include "filter_internal.h"
 
 static int filter_start_page_read(struct obd_device *obd, struct inode *inode,
@@ -52,11 +54,6 @@ static int filter_start_page_read(struct obd_device *obd, struct inode *inode,
 
         lnb->page = page;
 
-        if (inode->i_size < lnb->offset + lnb->len - 1)
-                lnb->rc = inode->i_size - lnb->offset;
-        else
-                lnb->rc = lnb->len;
-
         return 0;
 }
 
@@ -362,6 +359,11 @@ static int filter_preprw_read(int cmd, struct obd_export *exp, struct obdo *oa,
                                 GOTO(cleanup, rc);
                         }
 
+                        if (inode->i_size < lnb->offset + lnb->len - 1)
+                                lnb->rc = inode->i_size - lnb->offset;
+                        else
+                                lnb->rc = lnb->len;
+
                         tot_bytes += lnb->rc;
                         if (lnb->rc < lnb->len) {
                                 /* short read, be sure to wait on it */
@@ -523,25 +525,55 @@ static int filter_grant_check(struct obd_export *exp, int objcount,
         return rc;
 }
 
-static int filter_start_page_write(struct inode *inode,
+static int filter_start_page_write(struct obd_device *obd, struct inode *inode,
                                    struct niobuf_local *lnb)
 {
-        struct page *page = alloc_pages(GFP_HIGHUSER, 0);
+        struct page *page;
+
+        if (lnb->len != PAGE_SIZE)
+                return filter_start_page_read(obd, inode, lnb);
+
+        page = alloc_pages(GFP_HIGHUSER, 0);
         if (page == NULL) {
                 CERROR("no memory for a temp page\n");
                 RETURN(lnb->rc = -ENOMEM);
         }
+#if 0
         POISON_PAGE(page, 0xf1);
         if (lnb->len != PAGE_SIZE) {
                 memset(kmap(page) + lnb->len, 0, PAGE_SIZE - lnb->len);
                 kunmap(page);
         }
+#endif
         page->index = lnb->offset >> PAGE_SHIFT;
         lnb->page = page;
 
         return 0;
 }
 
+static void filter_abort_page_write(struct niobuf_local *lnb)
+{
+        LASSERT(lnb->page != NULL);
+
+        if (lnb->len != PAGE_SIZE)
+                page_cache_release(lnb->page);
+        else
+                __free_pages(lnb->page, 0);
+}
+
+/* a helper for both the 2.4 and 2.6 commitrw paths which are both built
+ * up by our shared filter_preprw_write() */
+void filter_release_write_page(struct filter_obd *filter, struct inode *inode,
+                               struct niobuf_local *lnb, int rc)
+{
+        if (lnb->len != PAGE_SIZE)
+                return filter_release_read_page(filter, inode, lnb->page);
+
+        if (rc == 0)
+                flip_into_page_cache(inode, lnb->page);
+        __free_page(lnb->page);
+}
+
 /* If we ever start to support multi-object BRW RPCs, we will need to get locks
  * on mulitple inodes.  That isn't all, because there still exists the
  * possibility of a truncate starting a new transaction while holding the ext3
@@ -623,13 +655,13 @@ static int filter_preprw_write(int cmd, struct obd_export *exp, struct obdo *oa,
                 lnb->len    = rnb->len;
                 lnb->flags  = rnb->flags;
 
-                rc = filter_start_page_write(dentry->d_inode, lnb);
+                rc = filter_start_page_write(exp->exp_obd, dentry->d_inode,lnb);
                 if (rc) {
                         CERROR("page err %u@"LPU64" %u/%u %p: rc %d\n",
                                lnb->len, lnb->offset,
                                i, obj->ioo_bufcnt, dentry, rc);
                         while (lnb-- > res)
-                                __free_pages(lnb->page, 0);
+                                filter_abort_page_write(lnb);
                         f_dput(dentry);
                         GOTO(cleanup, rc);
                 }
@@ -637,6 +669,17 @@ static int filter_preprw_write(int cmd, struct obd_export *exp, struct obdo *oa,
                         tot_bytes += lnb->len;
         }
 
+        while (lnb-- > res) {
+                if (lnb->len == PAGE_SIZE)
+                        continue;
+                rc = filter_finish_page_read(lnb);
+                if (rc) {
+                        CERROR("error page %u@"LPU64" %u %p: rc %d\n", lnb->len,
+                               lnb->offset, (int)(lnb - res), lnb->dentry, rc);
+                        GOTO(cleanup, rc);
+                }
+        }
+
         if (time_after(jiffies, now + 15 * HZ))
                 CERROR("slow start_page_write %lus\n", (jiffies - now) / HZ);
         else
@@ -676,6 +719,24 @@ int filter_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
         return -EPROTO;
 }
 
+void filter_release_read_page(struct filter_obd *filter, struct inode *inode,
+                              struct page *page)
+{
+        int drop = 0;
+
+        if (inode != NULL &&
+            (inode->i_size > filter->fo_readcache_max_filesize))
+                drop = 1;
+
+        /* drop from cache like truncate_list_pages() */
+        if (drop && !TryLockPage(page)) {
+                if (page->mapping)
+                        ll_truncate_complete_page(page);
+                unlock_page(page);
+        }
+        page_cache_release(page);
+}
+
 static int filter_commitrw_read(struct obd_export *exp, struct obdo *oa,
                                 int objcount, struct obd_ioobj *obj,
                                 int niocount, struct niobuf_local *res,
@@ -683,24 +744,19 @@ static int filter_commitrw_read(struct obd_export *exp, struct obdo *oa,
 {
         struct obd_ioobj *o;
         struct niobuf_local *lnb;
-        int i, j, drop = 0;
+        int i, j;
+        struct inode *inode = NULL;
         ENTRY;
 
         if (res->dentry != NULL)
-                drop = (res->dentry->d_inode->i_size >
-                        exp->exp_obd->u.filter.fo_readcache_max_filesize);
+                inode = res->dentry->d_inode;
 
         for (i = 0, o = obj, lnb = res; i < objcount; i++, o++) {
                 for (j = 0 ; j < o->ioo_bufcnt ; j++, lnb++) {
                         if (lnb->page == NULL)
                                 continue;
-                        /* drop from cache like truncate_list_pages() */
-                        if (drop && !TryLockPage(lnb->page)) {
-                                if (lnb->page->mapping)
-                                        ll_truncate_complete_page(lnb->page);
-                                unlock_page(lnb->page);
-                        }
-                        page_cache_release(lnb->page);
+                        filter_release_read_page(&exp->exp_obd->u.filter,
+                                                 inode, lnb->page);
                 }
         }
 
@@ -778,6 +834,131 @@ void filter_grant_commit(struct obd_export *exp, int niocount,
 
         spin_unlock(&exp->exp_obd->obd_osfs_lock);
 }
+int filter_do_cow(struct obd_export *exp, struct obd_ioobj *obj,
+                  int nioo, struct niobuf_remote *rnb)
+{
+        struct dentry *dentry;
+        struct lvfs_run_ctxt saved;
+        struct write_extents *extents = NULL;
+        int j, rc = 0, numexts = 0, flags = 0;
+
+        ENTRY;
+
+        LASSERT(nioo == 1);
+
+        push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
+        
+        dentry = filter_fid2dentry(exp->exp_obd, NULL, obj->ioo_gr,
+                                   obj->ioo_id);
+        if (IS_ERR(dentry)) {
+                pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
+                RETURN (PTR_ERR(dentry));
+        }
+
+        if (dentry->d_inode == NULL) {
+                CERROR("trying to write extents to non-existent file "LPU64"\n",
+                       obj->ioo_id);
+                GOTO(cleanup, rc = -ENOENT);
+        }
+        
+        flags = fsfilt_get_fs_flags(exp->exp_obd, dentry);
+        if (!(flags & SM_DO_COW)) {
+                GOTO(cleanup, rc);
+        }
+        OBD_ALLOC(extents, obj->ioo_bufcnt * sizeof(struct write_extents)); 
+        if (!extents) {
+                CERROR("No Memory\n");
+                GOTO(cleanup, rc = -ENOMEM);
+        }
+        for (j = 0; j < obj->ioo_bufcnt; j++) {
+                if (rnb[j].len != 0) {
+                        extents[numexts].w_count = rnb[j].len;
+                        extents[numexts].w_pos = rnb[j].offset;
+                        numexts++;
+                } 
+        } 
+        rc = fsfilt_do_write_cow(exp->exp_obd, dentry, extents, numexts);
+        if (rc) {
+                CERROR("Do cow error id "LPU64" rc:%d \n",
+                        obj->ioo_id, rc);
+                GOTO(cleanup, rc); 
+        }
+        
+cleanup:
+        if (extents) {
+                OBD_FREE(extents, obj->ioo_bufcnt * sizeof(struct write_extents));
+        }
+        f_dput(dentry);
+        pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
+        RETURN(rc);
+
+}
+int filter_write_extents(struct obd_export *exp, struct obd_ioobj *obj, int nobj,
+                         int niocount, struct niobuf_local *local, int rc)
+{
+        struct lvfs_run_ctxt saved;
+        struct dentry *dentry;
+        struct niobuf_local *lnb;
+        __u64  offset = 0;
+        __u32  len = 0;
+        int    i, flags; 
+        ENTRY;
+
+        LASSERT(nobj == 1);
+
+        push_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
+
+        dentry = filter_fid2dentry(exp->exp_obd, NULL, obj->ioo_gr,
+                                   obj->ioo_id);
+        if (IS_ERR(dentry)) {
+                pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
+                RETURN (PTR_ERR(dentry));
+        }
+
+        if (dentry->d_inode == NULL) {
+                CERROR("trying to write extents to non-existent file "LPU64"\n",
+                       obj->ioo_id);
+                GOTO(cleanup, rc = -ENOENT);
+        }
+        
+        flags = fsfilt_get_fs_flags(exp->exp_obd, dentry);
+        if (!(flags & SM_DO_REC)) {
+                GOTO(cleanup, rc);
+        }
+
+        for (i = 0, lnb = local; i < obj->ioo_bufcnt; i++, lnb++) {
+                if (len == 0) {
+                        offset = lnb->offset;
+                        len = lnb->len;
+                } else if (lnb->offset == (offset + len)) {
+                        len += lnb->len;
+                } else {
+                        rc = fsfilt_write_extents(exp->exp_obd, dentry, 
+                                                  offset, len);
+                        if (rc) {
+                                CERROR("write exts off "LPU64" num %u rc:%d\n",
+                                        offset, len, rc);
+                                GOTO(cleanup, rc);
+                        }
+                        offset = lnb->offset;
+                        len = lnb->len; 
+                } 
+        }
+        if (len > 0) {
+                rc = fsfilt_write_extents(exp->exp_obd, dentry, 
+                                          offset, len);
+                if (rc) {
+                        CERROR("write exts off "LPU64" num %u rc:%d\n",
+                                offset, len, rc);
+                        GOTO(cleanup, rc);
+                }
+        }
+cleanup:
+        f_dput(dentry);
+        pop_ctxt(&saved, &exp->exp_obd->obd_lvfs_ctxt, NULL);
+        RETURN(rc);
+}
 
 int filter_commitrw(int cmd, struct obd_export *exp, struct obdo *oa,
                     int objcount, struct obd_ioobj *obj, int niocount,