2 * OBDFS Super operations
4 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
5 * Copryright (C) 1999 Stelias Computing Inc,
6 * (author Peter J. Braam <braam@stelias.com>)
7 * Copryright (C) 1999 Seagate Technology Inc.
11 #include <linux/config.h>
12 #include <linux/kernel.h>
14 #include <linux/string.h>
15 #include <linux/stat.h>
16 #include <linux/errno.h>
17 #include <linux/locks.h>
18 #include <linux/unistd.h>
20 #include <asm/system.h>
21 #include <asm/uaccess.h>
24 #include <linux/stat.h>
25 #include <asm/uaccess.h>
26 #include <linux/vmalloc.h>
27 #include <asm/segment.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp_lock.h>
32 #include <linux/obd_support.h>
33 #include <linux/obd_ext2.h>
34 #include <linux/obdfs.h>
37 /* SYNCHRONOUS I/O for an inode */
38 static int obdfs_brw(int rw, struct inode *inode, struct page *page, int create)
40 obd_count num_obdo = 1;
41 obd_count bufs_per_obdo = 1;
43 char *buf = (char *)page_address(page);
44 obd_size count = PAGE_SIZE;
45 obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
46 obd_flag flags = create ? OBD_BRW_CREATE : 0;
50 oa = obdo_fromid(IID(inode), inode->i_ino, OBD_MD_FLNOTOBD);
55 obdfs_from_inode(oa, inode);
57 err = IOPS(inode, brw)(rw, IID(inode), num_obdo, &oa, &bufs_per_obdo,
58 &buf, &count, &offset, &flags);
61 obdfs_to_inode(inode, oa); /* copy o_blocks to i_blocks */
69 /* returns the page unlocked, but with a reference */
70 int obdfs_readpage(struct dentry *dentry, struct page *page)
72 struct inode *inode = dentry->d_inode;
77 rc = obdfs_brw(READ, inode, page, 0);
79 SetPageUptodate(page);
85 } /* obdfs_readpage */
87 static kmem_cache_t *obdfs_pgrq_cachep = NULL;
89 /* XXX should probably have one of these per superblock */
90 static int obdfs_cache_count = 0;
92 int obdfs_init_pgrqcache(void)
95 if (obdfs_pgrq_cachep == NULL) {
96 CDEBUG(D_INODE, "allocating obdfs_pgrq_cache\n");
97 obdfs_pgrq_cachep = kmem_cache_create("obdfs_pgrq",
98 sizeof(struct obdfs_pgrq),
99 0, SLAB_HWCACHE_ALIGN,
101 if (obdfs_pgrq_cachep == NULL) {
105 CDEBUG(D_INODE, "allocated cache at %p\n",
109 CDEBUG(D_INODE, "using existing cache at %p\n",
114 } /* obdfs_init_wreqcache */
116 inline void obdfs_pgrq_del(struct obdfs_pgrq *pgrq)
119 CDEBUG(D_INODE, "deleting page %p from list [count %d]\n",
120 pgrq->rq_page, obdfs_cache_count);
121 list_del(&pgrq->rq_plist);
122 kmem_cache_free(obdfs_pgrq_cachep, pgrq);
125 void obdfs_cleanup_pgrqcache(void)
128 if (obdfs_pgrq_cachep != NULL) {
129 CDEBUG(D_INODE, "destroying obdfs_pgrqcache at %p, count %d\n",
130 obdfs_pgrq_cachep, obdfs_cache_count);
131 if (kmem_cache_destroy(obdfs_pgrq_cachep))
132 printk(KERN_INFO "obd_cleanup_pgrqcache: unable to free all of cache\n");
134 printk(KERN_INFO "obd_cleanup_pgrqcache: called with NULL cache pointer\n");
137 } /* obdfs_cleanup_wreqcache */
141 * Find a specific page in the page cache. If it is found, we return
142 * the write request struct associated with it, if not found return NULL.
143 * Called with the list lock held.
145 static struct obdfs_pgrq *
146 obdfs_find_in_page_list(struct inode *inode, struct page *page)
148 struct list_head *page_list = obdfs_iplist(inode);
149 struct list_head *tmp;
153 CDEBUG(D_INODE, "looking for inode %ld page %p\n", inode->i_ino, page);
156 if (list_empty(page_list)) {
157 CDEBUG(D_INODE, "empty list\n");
162 while ( (tmp = tmp->next) != page_list ) {
163 struct obdfs_pgrq *pgrq;
165 pgrq = list_entry(tmp, struct obdfs_pgrq, rq_plist);
166 if (pgrq->rq_page == page) {
167 CDEBUG(D_INODE, "found page %p in list\n", page);
175 } /* obdfs_find_in_page_list */
178 /* called with the list lock held */
179 static struct page* obdfs_find_page_index(struct inode *inode,
182 struct list_head *page_list = obdfs_iplist(inode);
183 struct list_head *tmp;
188 CDEBUG(D_INODE, "looking for inode %ld pageindex %ld\n",
189 inode->i_ino, index);
192 if (list_empty(page_list)) {
197 while ( (tmp = tmp->next) != page_list ) {
198 struct obdfs_pgrq *pgrq;
200 pgrq = list_entry(tmp, struct obdfs_pgrq, rq_plist);
201 page = pgrq->rq_page;
202 if (index == page->index) {
204 "INDEX SEARCH found page %p, index %ld\n",
213 } /* obdfs_find_page_index */
216 /* call and free pages from Linux page cache: called with io lock on inodes */
217 int obdfs_do_vec_wr(struct inode **inodes, obd_count num_io,
218 obd_count num_obdos, struct obdo **obdos,
219 obd_count *oa_bufs, struct page **pages, char **bufs,
220 obd_size *counts, obd_off *offsets, obd_flag *flags)
222 struct super_block *sb = inodes[0]->i_sb;
223 struct obdfs_sb_info *sbi = (struct obdfs_sb_info *)&sb->u.generic_sbp;
227 CDEBUG(D_INODE, "writing %d page(s), %d obdo(s) in vector\n",
229 err = OPS(sb, brw)(WRITE, &sbi->osi_conn, num_obdos, obdos, oa_bufs,
230 bufs, counts, offsets, flags);
232 /* release the pages from the page cache */
233 while ( num_io > 0 ) {
235 CDEBUG(D_INODE, "calling put_page for %p, index %ld\n",
236 pages[num_io], pages[num_io]->index);
237 put_page(pages[num_io]);
240 while ( num_obdos > 0) {
242 CDEBUG(D_INODE, "copy/free obdo %ld\n",
243 (long)obdos[num_obdos]->o_id);
244 obdfs_to_inode(inodes[num_obdos], obdos[num_obdos]);
245 obdo_free(obdos[num_obdos]);
253 * Add a page to the write request cache list for later writing
254 * ASYNCHRONOUS write method.
256 static int obdfs_add_page_to_cache(struct inode *inode, struct page *page)
262 /* If this page isn't already in the inode page list, add it */
263 obd_down(&obdfs_i2sbi(inode)->osi_list_mutex);
264 if ( !obdfs_find_in_page_list(inode, page) ) {
265 struct obdfs_pgrq *pgrq;
266 pgrq = kmem_cache_alloc(obdfs_pgrq_cachep, SLAB_KERNEL);
267 CDEBUG(D_INODE, "adding inode %ld page %p, pgrq: %p, cache count [%d]\n",
268 inode->i_ino, page, pgrq, obdfs_cache_count);
271 obd_up(&obdfs_i2sbi(inode)->osi_list_mutex);
274 memset(pgrq, 0, sizeof(*pgrq));
276 pgrq->rq_page = page;
277 get_page(pgrq->rq_page);
278 list_add(&pgrq->rq_plist, obdfs_iplist(inode));
282 /* If inode isn't already on the superblock inodes list, add it,
283 * and increase ref count on inode so it doesn't disappear on us.
285 if ( list_empty(obdfs_islist(inode)) ) {
286 iget(inode->i_sb, inode->i_ino);
287 CDEBUG(D_INODE, "adding inode %ld to superblock list %p\n",
288 inode->i_ino, obdfs_slist(inode));
289 list_add(obdfs_islist(inode), obdfs_slist(inode));
292 /* XXX For testing purposes, we write out the page here.
293 * In the future, a flush daemon will write out the page.
294 res = obdfs_flush_reqs(obdfs_slist(inode), 0);
295 obdfs_flush_dirty_pages(1);
297 obd_up(&obdfs_i2sbi(inode)->osi_list_mutex);
301 } /* obdfs_add_page_to_cache */
304 /* select between SYNC and ASYNC I/O methods */
305 int obdfs_do_writepage(struct inode *inode, struct page *page, int sync)
310 /* PDEBUG(page, "WRITEPAGE"); */
312 err = obdfs_brw(WRITE, inode, page, 1);
314 err = obdfs_add_page_to_cache(inode, page);
315 CDEBUG(D_IOCTL, "DO_WR ino: %ld, page %p, err %d, uptodata %d\n", inode->i_ino, page, err, Page_Uptodate(page));
319 SetPageUptodate(page);
320 /* PDEBUG(page,"WRITEPAGE"); */
323 } /* obdfs_do_writepage */
325 /* returns the page unlocked, but with a reference */
326 int obdfs_writepage(struct dentry *dentry, struct page *page)
328 return obdfs_do_writepage(dentry->d_inode, page, 0);
332 * This does the "real" work of the write. The generic routine has
333 * allocated the page, locked it, done all the page alignment stuff
334 * calculations etc. Now we should just copy the data from user
335 * space and write it back to the real medium..
337 * If the writer ends up delaying the write, the writer needs to
338 * increment the page use counts until he is done with the page.
340 * Return value is the number of bytes written.
342 int obdfs_write_one_page(struct file *file, struct page *page,
343 unsigned long offset, unsigned long bytes,
346 struct inode *inode = file->f_dentry->d_inode;
350 if ( !Page_Uptodate(page) ) {
351 err = obdfs_brw(READ, inode, page, 1);
353 SetPageUptodate(page);
358 if (copy_from_user((u8*)page_address(page) + offset, buf, bytes))
362 err = obdfs_writepage(file->f_dentry, page);
365 return (err < 0 ? err : bytes);
366 } /* obdfs_write_one_page */
369 * return an up to date page:
370 * - if locked is true then is returned locked
371 * - if create is true the corresponding disk blocks are created
372 * - page is held, i.e. caller must release the page
374 * modeled on NFS code.
376 struct page *obdfs_getpage(struct inode *inode, unsigned long offset,
377 int create, int locked)
379 struct page *page_cache;
387 offset = offset & PAGE_CACHE_MASK;
388 CDEBUG(D_INODE, "ino: %ld, offset %ld, create %d, locked %d\n",
389 inode->i_ino, offset, create, locked);
390 index = offset >> PAGE_CACHE_SHIFT;
394 page_cache = page_cache_alloc();
395 if ( ! page_cache ) {
399 CDEBUG(D_INODE, "page_cache %p\n", page_cache);
401 hash = page_hash(&inode->i_data, index);
402 page = grab_cache_page(&inode->i_data, index);
406 printk("grab_cache_page says no dice ...\n");
411 PDEBUG(page, "GETPAGE: got page - before reading\n");
412 /* now check if the data in the page is up to date */
413 if ( Page_Uptodate(page)) {
421 if ( obdfs_find_page_index(inode, index) ) {
422 CDEBUG(D_INODE, "OVERWRITE: found dirty page %p, index %ld\n",
426 err = obdfs_brw(READ, inode, page, create);
437 SetPageUptodate(page);
438 PDEBUG(page,"GETPAGE - after reading");
441 } /* obdfs_getpage */