/*
* OBDFS Super operations
*
+ * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
* Copryright (C) 1999 Stelias Computing Inc,
* (author Peter J. Braam <braam@stelias.com>)
* Copryright (C) 1999 Seagate Technology Inc.
- */
+*/
#include <linux/config.h>
#include <linux/obd_ext2.h>
#include <linux/obdfs.h>
-int console_loglevel;
-/* VFS super_block ops */
-
-#if 0
-int obdfs_brw(struct inode *dir, int rw, struct page *page, int create)
+/* SYNCHRONOUS I/O for an inode */
+static int obdfs_brw(int rw, struct inode *inode, struct page *page, int create)
{
- return iops(dir)->o_brw(rw, iid(dir), dir, page, create);
-}
-#endif
+ obd_count num_obdo = 1;
+ obd_count bufs_per_obdo = 1;
+ struct obdo *oa;
+ char *buf = (char *)page_address(page);
+ obd_size count = PAGE_SIZE;
+ obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
+ obd_flag flags = create ? OBD_BRW_CREATE : 0;
+ int err;
+
+ ENTRY;
+ if (IOPS(inode, brw) == NULL) {
+ printk(KERN_ERR __FUNCTION__ ": no brw method!\n");
+ EXIT;
+ return -EIO;
+ }
+
+ oa = obdo_fromid(IID(inode), inode->i_ino, OBD_MD_FLNOTOBD);
+ if ( IS_ERR(oa) ) {
+ EXIT;
+ return PTR_ERR(oa);
+ }
+ obdfs_from_inode(oa, inode);
+
+ err = IOPS(inode, brw)(rw, IID(inode), num_obdo, &oa, &bufs_per_obdo,
+ &buf, &count, &offset, &flags);
+
+ if ( !err )
+ obdfs_to_inode(inode, oa); /* copy o_blocks to i_blocks */
+
+ obdo_free(oa);
+
+ EXIT;
+ return err;
+} /* obdfs_brw */
/* returns the page unlocked, but with a reference */
-int obdfs_readpage(struct dentry *dentry, struct page *page)
+int obdfs_readpage(struct file *file, struct page *page)
{
- struct inode *inode = dentry->d_inode;
- struct obdfs_wreq *wreq;
- int rc = 0;
+ struct dentry *dentry = file->f_dentry;
+ struct inode *inode = dentry->d_inode;
+ int rc;
ENTRY;
- PDEBUG(page, "READ");
- rc = iops(inode)->o_brw(READ, iid(inode),inode, page, 0);
- if (rc == PAGE_SIZE ) {
- SetPageUptodate(page);
- UnlockPage(page);
- }
- PDEBUG(page, "READ");
- if ( rc == PAGE_SIZE )
- rc = 0;
- }
- EXIT;
- return rc;
+ /* PDEBUG(page, "READ"); */
+ rc = obdfs_brw(READ, inode, page, 0);
+ if ( !rc ) {
+ SetPageUptodate(page);
+ obd_unlock_page(page);
+ }
+ /* PDEBUG(page, "READ"); */
+ EXIT;
+ return rc;
+} /* obdfs_readpage */
+
+int obdfs_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
+{
+ struct inode *inode = page->mapping->host;
+ obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
+ int rc;
+ ENTRY;
+
+ /* PDEBUG(page, "READ"); */
+ if (Page_Uptodate(page)) {
+ EXIT;
+ return 0;
+ }
+
+ if ( (from <= offset) && (to >= offset + PAGE_SIZE) ) {
+ EXIT;
+ return 0;
+ }
+
+ rc = obdfs_brw(READ, inode, page, 0);
+ if ( !rc ) {
+ SetPageUptodate(page);
+ /* obd_unlock_page(page); */
+ }
+ /* PDEBUG(page, "READ"); */
+ EXIT;
+ return rc;
}
-static kmem_cache_t *obdfs_wreq_cachep;
-int obdfs_init_wreqcache(void)
+
+
+static kmem_cache_t *obdfs_pgrq_cachep = NULL;
+
+int obdfs_init_pgrqcache(void)
+{
+ ENTRY;
+ if (obdfs_pgrq_cachep == NULL) {
+ CDEBUG(D_CACHE, "allocating obdfs_pgrq_cache\n");
+ obdfs_pgrq_cachep = kmem_cache_create("obdfs_pgrq",
+ sizeof(struct obdfs_pgrq),
+ 0, SLAB_HWCACHE_ALIGN,
+ NULL, NULL);
+ if (obdfs_pgrq_cachep == NULL) {
+ EXIT;
+ return -ENOMEM;
+ } else {
+ CDEBUG(D_CACHE, "allocated cache at %p\n",
+ obdfs_pgrq_cachep);
+ }
+ } else {
+ CDEBUG(D_CACHE, "using existing cache at %p\n",
+ obdfs_pgrq_cachep);
+ }
+ EXIT;
+ return 0;
+} /* obdfs_init_wreqcache */
+
+inline void obdfs_pgrq_del(struct obdfs_pgrq *pgrq)
{
- /* XXX need to free this somewhere? */
- ENTRY;
- obdfs_wreq_cachep = kmem_cache_create("obdfs_wreq",
- sizeof(struct obdfs_wreq),
- 0, SLAB_HWCACHE_ALIGN,
- NULL, NULL);
- if (obdfs_wreq_cachep == NULL) {
- EXIT;
- return -ENOMEM;
- }
- EXIT;
- return 0;
+ --obdfs_cache_count;
+ CDEBUG(D_INFO, "deleting page %p from list [count %ld]\n",
+ pgrq->rq_page, obdfs_cache_count);
+ list_del(&pgrq->rq_plist);
+ OBDClearCachePage(pgrq->rq_page);
+ kmem_cache_free(obdfs_pgrq_cachep, pgrq);
}
-/*
- * Find a specific page in the page cache. If it is found, we return
- * the write request struct associated with it, if not found return NULL.
- */
-static struct obdfs_wreq *
-obdfs_find_in_page_cache(struct inode *inode, struct page *page)
+void obdfs_cleanup_pgrqcache(void)
{
- struct list_head *list_head = &OBD_LIST(inode);
- struct obdfs_wreq *head, *wreq;
-
- ENTRY;
- CDEBUG(D_INODE, "looking for inode %ld page %p\n", inode->i_ino, page);
- if (list_empty(list_head)) {
- CDEBUG(D_INODE, "empty list\n");
- EXIT;
- return NULL;
- }
- wreq = head = WREQ(list_head->next);
- do {
- CDEBUG(D_INODE, "checking page %p\n", wreq->wb_page);
- if (wreq->wb_page == page) {
- CDEBUG(D_INODE, "found page %p in list\n", page);
- EXIT;
- return wreq;
- }
- } while ((wreq = WB_NEXT(wreq)) != head);
-
- EXIT;
- return NULL;
-}
+ ENTRY;
+ if (obdfs_pgrq_cachep != NULL) {
+ CDEBUG(D_CACHE, "destroying obdfs_pgrqcache at %p, count %ld\n",
+ obdfs_pgrq_cachep, obdfs_cache_count);
+ if (kmem_cache_destroy(obdfs_pgrq_cachep))
+ printk(KERN_INFO __FUNCTION__
+ ": unable to free all of cache\n");
+ obdfs_pgrq_cachep = NULL;
+ } else
+ printk(KERN_INFO __FUNCTION__ ": called with NULL pointer\n");
+
+ EXIT;
+} /* obdfs_cleanup_wreqcache */
+
+
+/* called with the list lock held */
+static struct page *obdfs_find_page_index(struct inode *inode,
+ unsigned long index)
+{
+ struct list_head *page_list = obdfs_iplist(inode);
+ struct list_head *tmp;
+ struct page *page;
+ ENTRY;
-/*
- * Remove a writeback request from a list
- */
-static inline int
-obdfs_remove_from_page_cache(struct obdfs_wreq *wreq)
+ CDEBUG(D_INFO, "looking for inode %ld pageindex %ld\n",
+ inode->i_ino, index);
+ OIDEBUG(inode);
+
+ if (list_empty(page_list)) {
+ EXIT;
+ return NULL;
+ }
+ tmp = page_list;
+ while ( (tmp = tmp->next) != page_list ) {
+ struct obdfs_pgrq *pgrq;
+
+ pgrq = list_entry(tmp, struct obdfs_pgrq, rq_plist);
+ page = pgrq->rq_page;
+ if (index == page->index) {
+ CDEBUG(D_INFO,
+ "INDEX SEARCH found page %p, index %ld\n",
+ page, index);
+ EXIT;
+ return page;
+ }
+ }
+
+ EXIT;
+ return NULL;
+} /* obdfs_find_page_index */
+
+
+/* call and free pages from Linux page cache: called with io lock on inodes */
+int obdfs_do_vec_wr(struct inode **inodes, obd_count num_io,
+ obd_count num_obdos, struct obdo **obdos,
+ obd_count *oa_bufs, struct page **pages, char **bufs,
+ obd_size *counts, obd_off *offsets, obd_flag *flags)
{
- struct inode *inode = wreq->wb_inode;
- struct page *page = wreq->wb_page;
- int rc;
-
- ENTRY;
- CDEBUG(D_INODE, "removing inode %ld page %p, wreq: %p\n",
- inode->i_ino, page, wreq);
- rc = iops(inode)->o_brw(WRITE, iid(inode), inode, page, 1);
- /* XXX probably should handle error here somehow. I think that
- * ext2 also does the same thing - discard write even if error?
- */
- put_page(page);
- list_del(&wreq->wb_list);
- kmem_cache_free(obdfs_wreq_cachep, wreq);
-
- EXIT;
- return rc;
+ int err;
+
+ ENTRY;
+ if (IOPS(inodes[0], brw) == NULL) {
+ printk(KERN_ERR __FUNCTION__ ": no brw method!\n");
+ EXIT;
+ return -EIO;
+ }
+
+ CDEBUG(D_INFO, "writing %d page(s), %d obdo(s) in vector\n",
+ num_io, num_obdos);
+ if (obd_debug_level & D_INFO) { /* DEBUGGING */
+ int i;
+ printk("OBDOS: ");
+ for (i = 0; i < num_obdos; i++)
+ printk("%ld:0x%p ", (long)obdos[i]->o_id, obdos[i]);
+
+ printk("\nPAGES: ");
+ for (i = 0; i < num_io; i++)
+ printk("0x%p ", pages[i]);
+ printk("\n");
+ }
+
+ err = IOPS(inodes[0], brw)(WRITE, IID(inodes[0]), num_obdos, obdos,
+ oa_bufs, bufs, counts, offsets, flags);
+
+ CDEBUG(D_INFO, "BRW done\n");
+ /* release the pages from the page cache */
+ while ( num_io > 0 ) {
+ --num_io;
+ CDEBUG(D_INFO, "calling put_page for %p, index %ld\n",
+ pages[num_io], pages[num_io]->index);
+ /* PDEBUG(pages[num_io], "do_vec_wr"); */
+ put_page(pages[num_io]);
+ /* PDEBUG(pages[num_io], "do_vec_wr"); */
+ }
+ CDEBUG(D_INFO, "put_page done\n");
+
+ while ( num_obdos > 0) {
+ --num_obdos;
+ CDEBUG(D_INFO, "free obdo %ld\n",(long)obdos[num_obdos]->o_id);
+ /* copy o_blocks to i_blocks */
+ obdfs_to_inode(inodes[num_obdos], obdos[num_obdos]);
+ obdo_free(obdos[num_obdos]);
+ }
+ CDEBUG(D_INFO, "obdo_free done\n");
+ EXIT;
+ return err;
}
+
/*
- * Add a page to the write request cache list for later writing
+ * Add a page to the write request cache list for later writing.
+ * ASYNCHRONOUS write method.
*/
-static int
-obdfs_add_to_page_cache(struct inode *inode, struct page *page)
+static int obdfs_add_page_to_cache(struct inode *inode, struct page *page)
{
- struct obdfs_wreq *wreq;
-
- ENTRY;
- wreq = kmem_cache_alloc(obdfs_wreq_cachep, SLAB_KERNEL);
- CDEBUG(D_INODE, "adding inode %ld page %p, wreq: %p\n",
- inode->i_ino, page, wreq);
- if (!wreq) {
- EXIT;
- return -ENOMEM;
- }
- memset(wreq, 0, sizeof(*wreq));
-
- wreq->wb_page = page;
- wreq->wb_inode = inode;
-
- get_page(wreq->wb_page);
- list_add(&wreq->wb_list, &OBD_LIST(inode));
-
- /* For testing purposes, we write out the page here.
- * In the future, a flush daemon will write out the page.
- */
- wreq = obdfs_find_in_page_cache(inode, page);
- if (!wreq) {
- CDEBUG(D_INODE, "XXXX Can't find page after adding it!!!\n");
- return -EINVAL;
- } else
- return obdfs_remove_from_page_cache(wreq);
-
- return 0;
-}
+ int err = 0;
+ ENTRY;
+
+ /* The PG_obdcache bit is cleared by obdfs_pgrq_del() BEFORE the page
+ * is written, so at worst we will write the page out twice.
+ *
+ * If the page has the PG_obdcache bit set, then the inode MUST be
+ * on the superblock dirty list so we don't need to check this.
+ * Dirty inodes are removed from the superblock list ONLY when they
+ * don't have any more cached pages. It is possible to have an inode
+ * with no dirty pages on the superblock list, but not possible to
+ * have an inode with dirty pages NOT on the superblock dirty list.
+ */
+ if (!OBDAddCachePage(page)) {
+ struct obdfs_pgrq *pgrq;
+ pgrq = kmem_cache_alloc(obdfs_pgrq_cachep, SLAB_KERNEL);
+ if (!pgrq) {
+ OBDClearCachePage(page);
+ EXIT;
+ return -ENOMEM;
+ }
+ /* not really necessary since we set all pgrq fields here
+ memset(pgrq, 0, sizeof(*pgrq));
+ */
+
+ pgrq->rq_page = page;
+ pgrq->rq_jiffies = jiffies;
+ get_page(pgrq->rq_page);
+
+ obd_down(&obdfs_i2sbi(inode)->osi_list_mutex);
+ list_add(&pgrq->rq_plist, obdfs_iplist(inode));
+ obdfs_cache_count++;
+
+ /* If inode isn't already on superblock inodes list, add it.
+ *
+ * We increment the reference count on the inode to keep it
+ * from being freed from memory. This _should_ be an iget()
+ * with an iput() in both flush_reqs() and put_inode(), but
+ * since put_inode() is called from iput() we can't call iput()
+ * again there. Instead we just increment/decrement i_count,
+ * which is mostly what iget/iput do for an inode in memory.
+ */
+ if ( list_empty(obdfs_islist(inode)) ) {
+ atomic_inc(&inode->i_count);
+ CDEBUG(D_INFO,
+ "adding inode %ld to superblock list %p\n",
+ inode->i_ino, obdfs_slist(inode));
+ list_add(obdfs_islist(inode), obdfs_slist(inode));
+ }
+ obd_up(&obdfs_i2sbi(inode)->osi_list_mutex);
+ }
+
+ /* XXX For testing purposes, we can write out the page here.
+ err = obdfs_flush_reqs(obdfs_slist(inode), ~0UL);
+ */
+
+ EXIT;
+ return err;
+} /* obdfs_add_page_to_cache */
+
+
+/* select between SYNC and ASYNC I/O methods */
+int obdfs_do_writepage(struct page *page, int sync)
+{
+ struct inode *inode = page->mapping->host;
+ int err;
+
+ ENTRY;
+ /* PDEBUG(page, "WRITEPAGE"); */
+ if ( sync )
+ err = obdfs_brw(WRITE, inode, page, 1);
+ else {
+ err = obdfs_add_page_to_cache(inode, page);
+ CDEBUG(D_INFO, "DO_WR ino: %ld, page %p, err %d, uptodate %d\n",
+ inode->i_ino, page, err, Page_Uptodate(page));
+ }
+
+ if ( !err )
+ SetPageUptodate(page);
+ /* PDEBUG(page,"WRITEPAGE"); */
+ EXIT;
+ return err;
+} /* obdfs_do_writepage */
+
/* returns the page unlocked, but with a reference */
-int obdfs_writepage(struct dentry *dentry, struct page *page)
+int obdfs_writepage(struct page *page)
{
- struct inode *inode = dentry->d_inode;
- int rc;
+ return obdfs_do_writepage(page, 0);
+}
- ENTRY;
- PDEBUG(page, "WRITEPAGE");
- /* XXX flush stuff */
- rc = obdfs_add_to_page_cache(inode, page);
-
- if (!rc)
- SetPageUptodate(page);
- PDEBUG(page,"WRITEPAGE");
- return rc;
+int obdfs_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
+{
+ int rc;
+ struct inode *inode = page->mapping->host;
+ loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
+ rc = obdfs_writepage(page);
+ kunmap(page);
+ if (pos > inode->i_size) {
+ inode->i_size = pos;
+ mark_inode_dirty(inode);
+ }
+ return 0;
}
+
/*
* This does the "real" work of the write. The generic routine has
* allocated the page, locked it, done all the page alignment stuff
*
* If the writer ends up delaying the write, the writer needs to
* increment the page use counts until he is done with the page.
+ *
+ * Return value is the number of bytes written.
*/
-int obdfs_write_one_page(struct file *file, struct page *page, unsigned long offset, unsigned long bytes, const char * buf)
+int obdfs_write_one_page(struct file *file, struct page *page,
+ unsigned long offset, unsigned long bytes,
+ const char * buf)
{
- long status;
struct inode *inode = file->f_dentry->d_inode;
+ int err;
- ENTRY;
- if ( !Page_Uptodate(page) ) {
- status = iops(inode)->o_brw(READ, iid(inode), inode, page, 1);
- if (status == PAGE_SIZE ) {
- SetPageUptodate(page);
- } else {
- return status;
- }
- }
- bytes -= copy_from_user((u8*)page_address(page) + offset, buf, bytes);
- status = -EFAULT;
-
- if (bytes) {
- lock_kernel();
- status = obdfs_writepage(file->f_dentry, page);
- unlock_kernel();
- }
- EXIT;
- if ( status != PAGE_SIZE )
- return status;
- else
- return bytes;
-}
+ ENTRY;
+ /* We check for complete page writes here, as we then don't have to
+ * get the page before writing over everything anyways.
+ */
+ if ( !Page_Uptodate(page) && (offset != 0 || bytes != PAGE_SIZE) ) {
+ err = obdfs_brw(READ, inode, page, 0);
+ if ( err )
+ return err;
+ SetPageUptodate(page);
+ }
+
+ if (copy_from_user((u8*)page_address(page) + offset, buf, bytes))
+ return -EFAULT;
+
+ lock_kernel();
+ err = obdfs_writepage(page);
+ unlock_kernel();
+
+ return (err < 0 ? err : bytes);
+} /* obdfs_write_one_page */
+/*
+ * return an up to date page:
+ * - if locked is true then is returned locked
+ * - if create is true the corresponding disk blocks are created
+ * - page is held, i.e. caller must release the page
+ *
+ * modeled on NFS code.
+ */
+struct page *obdfs_getpage(struct inode *inode, unsigned long offset,
+ int create, int locked)
+{
+ struct page * page;
+ int index;
+ int err;
+ ENTRY;
+ offset = offset & PAGE_CACHE_MASK;
+ CDEBUG(D_INFO, "ino: %ld, offset %ld, create %d, locked %d\n",
+ inode->i_ino, offset, create, locked);
+ index = offset >> PAGE_CACHE_SHIFT;
+
+ page = grab_cache_page(&inode->i_data, index);
+
+ /* Yuck, no page */
+ if (! page) {
+ printk(KERN_WARNING " grab_cache_page says no dice ...\n");
+ EXIT;
+ return NULL;
+ }
+
+ /* PDEBUG(page, "GETPAGE: got page - before reading\n"); */
+ /* now check if the data in the page is up to date */
+ if ( Page_Uptodate(page)) {
+ if (!locked) {
+ if (PageLocked(page))
+ obd_unlock_page(page);
+ } else {
+ printk("file %s, line %d: expecting locked page\n",
+ __FILE__, __LINE__);
+ }
+ EXIT;
+ return page;
+ }
+
+
+#ifdef EXT2_OBD_DEBUG
+ if ((obd_debug_level & D_INFO) && obdfs_find_page_index(inode, index)) {
+ CDEBUG(D_INFO, "OVERWRITE: found dirty page %p, index %ld\n",
+ page, page->index);
+ }
+#endif
+ err = obdfs_brw(READ, inode, page, create);
-void report_inode(struct page * page) {
- struct inode *inode = (struct inode *)0;
- int offset = (int)&inode->i_data;
- inode = (struct inode *)( (char *)page->mapping - offset);
- if ( inode->i_sb->s_magic == 0x4711 )
- printk("----> ino %ld , dev %d\n", inode->i_ino, inode->i_dev);
-}
+ if ( err ) {
+ SetPageError(page);
+ obd_unlock_page(page);
+ EXIT;
+ return page;
+ }
-/*
- return an up to date page:
- - if locked is true then is returned locked
- - if create is true the corresponding disk blocks are created
- - page is held, i.e. caller must release the page
+ if ( !locked )
+ obd_unlock_page(page);
+ SetPageUptodate(page);
+ /* PDEBUG(page,"GETPAGE - after reading"); */
+ EXIT;
+ return page;
+} /* obdfs_getpage */
- modeled on NFS code.
-*/
-struct page *obdfs_getpage(struct inode *inode, unsigned long offset, int create, int locked)
-{
- struct page *page_cache;
- struct page ** hash;
- struct page * page;
- int rc;
+void obdfs_truncate(struct inode *inode)
+{
+ struct obdo *oa;
+ int err;
ENTRY;
- offset = offset & PAGE_CACHE_MASK;
- CDEBUG(D_INODE, "\n");
-
- page = NULL;
- page_cache = page_cache_alloc();
- if ( ! page_cache )
- return NULL;
- CDEBUG(D_INODE, "page_cache %p\n", page_cache);
-
- hash = page_hash(&inode->i_data, offset);
- page = grab_cache_page(&inode->i_data, offset);
-
- PDEBUG(page, "GETPAGE: got page - before reading\n");
- /* now check if the data in the page is up to date */
- if ( Page_Uptodate(page)) {
- if (!locked)
- UnlockPage(page);
- EXIT;
- return page;
- }
-
- /* it's not: read it */
- if (! page) {
- printk("get_page_map says no dice ...\n");
- return 0;
- }
-
- rc = iops(inode)->o_brw(READ, iid(inode), inode, page, create);
- if ( rc != PAGE_SIZE ) {
- SetPageError(page);
- UnlockPage(page);
- return page;
- }
-
- if ( !locked )
- UnlockPage(page);
- SetPageUptodate(page);
- PDEBUG(page,"GETPAGE - after reading");
- EXIT;
- return page;
-}
-
-
+ obdfs_dequeue_pages(inode);
+
+ if (IOPS(inode, punch) == NULL) {
+ printk(KERN_ERR __FUNCTION__ ": no punch method!\n");
+ EXIT;
+ return;
+ }
+ oa = obdo_alloc();
+ if ( !oa ) {
+ /* XXX This would give an inconsistent FS, so deal with it as
+ * best we can for now - an obdo on the stack is not pretty.
+ */
+ struct obdo obdo;
+
+ printk(__FUNCTION__ ": obdo_alloc failed - using stack!\n");
+
+ obdo.o_valid = OBD_MD_FLNOTOBD;
+ obdfs_from_inode(&obdo, inode);
+
+ err = IOPS(inode, punch)(IID(inode), &obdo, obdo.o_size, 0);
+ } else {
+ oa->o_valid = OBD_MD_FLNOTOBD;
+ obdfs_from_inode(oa, inode);
+
+ CDEBUG(D_INFO, "calling punch for %ld (%Lu bytes at 0)\n",
+ (long)oa->o_id, oa->o_size);
+ err = IOPS(inode, punch)(IID(inode), oa, oa->o_size, 0);
+
+ obdo_free(oa);
+ }
+
+ if (err) {
+ printk(__FUNCTION__ ": obd_truncate fails (%d)\n", err);
+ EXIT;
+ return;
+ }
+ EXIT;
+} /* obdfs_truncate */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff