/*
* 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/pagemap.h>
#include <linux/smp_lock.h>
-#include <../obd/linux/obd_support.h>
-#include <../obd/linux/obd_sim.h>
-#include <obdfs.h>
+#include <linux/obd_support.h>
+#include <linux/obd_ext2.h>
+#include <linux/obdfs.h>
-/* VFS super_block ops */
+int console_loglevel;
+
+/* SYNCHRONOUS I/O for an inode */
+static int obdfs_brw(int rw, struct inode *inode, struct page *page, int create)
+{
+ struct obdo *obdo;
+ obd_size count = PAGE_SIZE;
+ int err;
+
+ ENTRY;
+ obdo = obdo_alloc();
+ if ( ! obdo ) {
+ EXIT;
+ return -ENOMEM;
+ }
+
+ obdo->o_id = inode->i_ino;
+
+ err = IOPS(inode, brw)(rw, IID(inode), obdo, (char *)page_address(page),
+ &count, (page->index) >> PAGE_SHIFT, create);
+
+ obdo_to_inode(inode, obdo); /* copy o_blocks to i_blocks */
+ obdo_free(obdo);
+
+ EXIT;
+ return err;
+} /* obdfs_brw */
/* returns the page unlocked, but with a reference */
-int obdfs_readpage(struct file *file, struct page *page)
+int obdfs_readpage(struct dentry *dentry, struct page *page)
{
- struct obdfs_sb_info *sbi;
- struct super_block *sb = file->f_dentry->d_inode->i_sb;
+ struct inode *inode = dentry->d_inode;
int rc;
- ENTRY;
-
- /* XXX flush stuff */
- sbi = sb->u.generic_sbp;
- PDEBUG(page, READ);
- rc = sbi->osi_ops->o_brw(READ, sbi->osi_conn_info.conn_id,
- file->f_dentry->d_inode->i_ino, page);
- if (rc == PAGE_SIZE ) {
+ ENTRY;
+ PDEBUG(page, "READ");
+ rc = obdfs_brw(READ, inode, page, 0);
+ if ( !rc ) {
SetPageUptodate(page);
UnlockPage(page);
}
- PDEBUG(page, READ);
- if ( rc == PAGE_SIZE )
- rc = 0;
+ PDEBUG(page, "READ");
+ EXIT;
+ return rc;
+} /* obdfs_readpage */
+
+static kmem_cache_t *obdfs_pgrq_cachep;
+
+int obdfs_init_pgrqcache(void)
+{
+ ENTRY;
+ 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;
+ }
+
+ EXIT;
+ return 0;
+} /* obdfs_init_wreqcache */
+
+void obdfs_cleanup_pgrqcache(void)
+{
+ if (obdfs_pgrq_cachep != NULL)
+ kmem_cache_destroy(obdfs_pgrq_cachep);
+ obdfs_pgrq_cachep = NULL;
+ EXIT;
+} /* obdfs_cleanup_wreqcache */
+
+
+/*
+ * 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_pgrq *
+obdfs_find_in_page_cache(struct inode *inode, struct page *page)
+{
+ struct list_head *page_list = &OBD_LIST(inode);
+ struct list_head *tmp;
+ struct obdfs_pgrq *pgrq;
+
+ ENTRY;
+ CDEBUG(D_INODE, "looking for inode %ld page %p\n", inode->i_ino, page);
+ if (list_empty(page_list)) {
+ CDEBUG(D_INODE, "empty list\n");
+ EXIT;
+ return NULL;
+ }
+ tmp = page_list;
+ while ( (tmp = tmp->next) != page_list ) {
+ pgrq = list_entry(tmp, struct obdfs_pgrq, rq_list);
+ CDEBUG(D_INODE, "checking page %p\n", pgrq->rq_page);
+ if (pgrq->rq_page == page) {
+ CDEBUG(D_INODE, "found page %p in list\n", page);
+ EXIT;
+ return pgrq;
+ }
+ }
+
+ EXIT;
+ return NULL;
+} /* obdfs_find_in_page_cache */
+
+
+/*
+ * Remove a writeback request from a list
+ */
+static inline int
+obdfs_remove_from_page_cache(struct obdfs_pgrq *pgrq)
+{
+ struct inode *inode = pgrq->rq_inode;
+ struct page *page = pgrq->rq_page;
+ int rc;
+
+ ENTRY;
+ CDEBUG(D_INODE, "removing inode %ld page %p, pgrq: %p\n",
+ inode->i_ino, page, pgrq);
+ rc = obdfs_brw(WRITE, 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(&pgrq->rq_list);
+ kmem_cache_free(obdfs_pgrq_cachep, pgrq);
+
+ EXIT;
return rc;
+} /* obdfs_remove_from_page_cache */
+
+/*
+ * Add a page to the write request cache list for later writing
+ */
+static int obdfs_add_to_page_cache(struct inode *inode, struct page *page)
+{
+ struct obdfs_pgrq *pgrq;
+
+ ENTRY;
+ pgrq = kmem_cache_alloc(obdfs_pgrq_cachep, SLAB_KERNEL);
+ CDEBUG(D_INODE, "adding inode %ld page %p, pgrq: %p\n",
+ inode->i_ino, page, pgrq);
+ if (!pgrq) {
+ EXIT;
+ return -ENOMEM;
+ }
+ memset(pgrq, 0, sizeof(*pgrq));
+
+ pgrq->rq_page = page;
+ pgrq->rq_inode = inode;
+
+ get_page(pgrq->rq_page);
+ list_add(&pgrq->rq_list, &OBD_LIST(inode));
+
+ /* For testing purposes, we write out the page here.
+ * In the future, a flush daemon will write out the page.
+ return 0;
+ */
+ pgrq = obdfs_find_in_page_cache(inode, page);
+ if (!pgrq) {
+ CDEBUG(D_INODE, "XXXX Can't find page after adding it!!!\n");
+ EXIT;
+ return -EINVAL;
+ }
+
+ return obdfs_remove_from_page_cache(pgrq);
+} /* obdfs_add_to_page_cache */
+
+
+int obdfs_do_writepage(struct inode *inode, struct page *page, int sync)
+{
+ int err;
+
+ ENTRY;
+ PDEBUG(page, "WRITEPAGE");
+ if ( sync )
+ err = obdfs_brw(WRITE, inode, page, 1);
+ else
+ err = obdfs_add_to_page_cache(inode, page);
+
+ if ( !err )
+ SetPageUptodate(page);
+ PDEBUG(page,"WRITEPAGE");
+ return err;
+} /* obdfs_do_writepage */
+/* returns the page unlocked, but with a reference */
+int obdfs_writepage(struct dentry *dentry, struct page *page)
+{
+ return obdfs_do_writepage(dentry->d_inode, page, 0);
}
/*
* If the writer ends up delaying the write, the writer needs to
* increment the page use counts until he is done with the page.
*/
-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) ) {
+ err = obdfs_brw(READ, inode, page, 1);
+ if ( !err )
+ SetPageUptodate(page);
+ else
+ return err;
+ }
bytes -= copy_from_user((u8*)page_address(page) + offset, buf, bytes);
- status = -EFAULT;
- CDEBUG(D_INODE, "page offset %ld, bytes %ld, offset %ld, page addr %lx, writing: %s, beg of page %s\n", page->offset, bytes, offset, page_address(page), ((char *) page_address(page)) + offset, (char *)page_address(page));
+ err = -EFAULT;
+
if (bytes) {
lock_kernel();
- status = obdfs_writepage(file, page);
+ err = obdfs_writepage(file->f_dentry, page);
unlock_kernel();
}
- if ( status != PAGE_SIZE )
- return status;
- else
- return bytes;
-}
-
-
-
-
-/* returns the page unlocked, but with a reference */
-int obdfs_writepage(struct file *file, struct page *page)
-{
- struct obdfs_sb_info *sbi;
- struct super_block *sb = file->f_dentry->d_inode->i_sb;
- int rc;
-
- ENTRY;
- PDEBUG(page,WRITE);
- /* XXX flush stuff */
- sbi = sb->u.generic_sbp;
-
- rc = sbi->osi_ops->o_brw(WRITE, sbi->osi_conn_info.conn_id,
- file->f_dentry->d_inode->i_ino, page);
- SetPageUptodate(page);
- PDEBUG(page,WRITE);
- return rc;
-}
+ return err;
+} /* obdfs_write_one_page */
/*
- page is returned unlocked, with the up to date flag set,
- and held, i.e. caller must do a page_put
+ 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)
+struct page *obdfs_getpage(struct inode *inode, unsigned long offset, int create, int locked)
{
- unsigned long new_page;
+ struct page *page_cache;
struct page ** hash;
- struct page * page;
- struct obdfs_sb_info *sbi;
- struct super_block *sb = inode->i_sb;
+ struct page * page;
+ int err;
- ENTRY;
+ ENTRY;
- sbi = sb->u.generic_sbp;
+ offset = offset & PAGE_CACHE_MASK;
+ CDEBUG(D_INODE, "\n");
- page = find_get_page(inode, offset);
- if (page && Page_Uptodate(page)) {
- PDEBUG(page,READ);
+ page = NULL;
+ page_cache = page_cache_alloc();
+ if ( ! page_cache ) {
+ EXIT;
+ 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);
+
+ /* Yuck, no page */
+ if (! page) {
+ printk("grab_cache_page says no dice ...\n");
+ EXIT;
+ return 0;
+ }
+
+ 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;
}
-
- if (page && !Page_Uptodate(page) ) {
- CDEBUG(D_INODE, "Page found but not up to date\n");
- }
- /* page_cache_alloc returns address of page */
- new_page = page_cache_alloc();
- if (!new_page)
- return NULL;
-
- /* corresponding struct page in the mmap */
- hash = page_hash(inode, offset);
- page = page_cache_entry(new_page);
- if (!add_to_page_cache_unique(page, inode, offset, hash)) {
- CDEBUG(D_INODE, "Page not found. Reading it.\n");
- PDEBUG(page,READ);
- sbi->osi_ops->o_brw(READ, sbi->osi_conn_info.conn_id,
- inode->i_ino, page);
+ err = obdfs_brw(READ, inode, page, create);
+
+ if ( err ) {
+ SetPageError(page);
UnlockPage(page);
- SetPageUptodate(page);
+ EXIT;
return page;
}
- /*
- * We arrive here in the unlikely event that someone
- * raced with us and added our page to the cache first.
- */
- CDEBUG(D_INODE, "Page not found. Someone raced us.\n");
- PDEBUG(page,READ);
- return page;
-}
+ if ( !locked )
+ UnlockPage(page);
+ SetPageUptodate(page);
+ PDEBUG(page,"GETPAGE - after reading");
+ EXIT;
+ return page;
+} /* obdfs_getpage */
-struct file_operations obdfs_file_operations = {
- NULL, /* lseek - default */
- generic_file_read, /* read */
- obdfs_file_write, /* write - bad */
- obdfs_readdir, /* readdir */
- NULL, /* poll - default */
- NULL, /* ioctl */
- NULL, /* mmap */
- NULL, /* no special open code */
- NULL, /* flush */
- NULL, /* no special release code */
- NULL, /* fsync */
- NULL, /* fasync */
- NULL, /* check_media_change */
- NULL /* revalidate */
-};
-
-struct inode_operations obdfs_inode_ops = {
- &obdfs_file_operations, /* default directory file-ops */
- NULL, /* create */
- obdfs_lookup, /* lookup */
- NULL, /* link */
- NULL, /* unlink */
- NULL, /* symlink */
- NULL, /* mkdir */
- NULL, /* rmdir */
- NULL, /* mknod */
- NULL, /* rename */
- NULL, /* readlink */
- NULL, /* follow_link */
- NULL, /* get_block */
- obdfs_readpage, /* readpage */
- obdfs_writepage, /* writepage */
- NULL, /* flushpage */
- NULL, /* truncate */
- NULL, /* permission */
- NULL, /* smap */
- NULL /* revalidate */
-};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff