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 if (IOPS(inode, brw) == NULL) {
51 printk(KERN_ERR __FUNCTION__ ": no brw method!\n");
56 oa = obdo_fromid(IID(inode), inode->i_ino, OBD_MD_FLNOTOBD);
61 obdfs_from_inode(oa, inode);
63 err = IOPS(inode, brw)(rw, IID(inode), num_obdo, &oa, &bufs_per_obdo,
64 &buf, &count, &offset, &flags);
67 obdfs_to_inode(inode, oa); /* copy o_blocks to i_blocks */
75 /* returns the page unlocked, but with a reference */
76 int obdfs_readpage(struct dentry *dentry, struct page *page)
78 struct inode *inode = dentry->d_inode;
82 /* PDEBUG(page, "READ"); */
83 rc = obdfs_brw(READ, inode, page, 0);
85 SetPageUptodate(page);
88 /* PDEBUG(page, "READ"); */
91 } /* obdfs_readpage */
93 static kmem_cache_t *obdfs_pgrq_cachep = NULL;
95 int obdfs_init_pgrqcache(void)
98 if (obdfs_pgrq_cachep == NULL) {
99 CDEBUG(D_CACHE, "allocating obdfs_pgrq_cache\n");
100 obdfs_pgrq_cachep = kmem_cache_create("obdfs_pgrq",
101 sizeof(struct obdfs_pgrq),
102 0, SLAB_HWCACHE_ALIGN,
104 if (obdfs_pgrq_cachep == NULL) {
108 CDEBUG(D_CACHE, "allocated cache at %p\n",
112 CDEBUG(D_CACHE, "using existing cache at %p\n",
117 } /* obdfs_init_wreqcache */
119 inline void obdfs_pgrq_del(struct obdfs_pgrq *pgrq)
122 CDEBUG(D_INFO, "deleting page %p from list [count %ld]\n",
123 pgrq->rq_page, obdfs_cache_count);
124 list_del(&pgrq->rq_plist);
125 OBDClearCachePage(pgrq->rq_page);
126 kmem_cache_free(obdfs_pgrq_cachep, pgrq);
129 void obdfs_cleanup_pgrqcache(void)
132 if (obdfs_pgrq_cachep != NULL) {
133 CDEBUG(D_CACHE, "destroying obdfs_pgrqcache at %p, count %ld\n",
134 obdfs_pgrq_cachep, obdfs_cache_count);
135 if (kmem_cache_destroy(obdfs_pgrq_cachep))
136 printk(KERN_INFO __FUNCTION__
137 ": unable to free all of cache\n");
138 obdfs_pgrq_cachep = NULL;
140 printk(KERN_INFO __FUNCTION__ ": called with NULL pointer\n");
143 } /* obdfs_cleanup_wreqcache */
146 /* called with the list lock held */
147 static struct page *obdfs_find_page_index(struct inode *inode,
150 struct list_head *page_list = obdfs_iplist(inode);
151 struct list_head *tmp;
156 CDEBUG(D_INFO, "looking for inode %ld pageindex %ld\n",
157 inode->i_ino, index);
160 if (list_empty(page_list)) {
165 while ( (tmp = tmp->next) != page_list ) {
166 struct obdfs_pgrq *pgrq;
168 pgrq = list_entry(tmp, struct obdfs_pgrq, rq_plist);
169 page = pgrq->rq_page;
170 if (index == page->index) {
172 "INDEX SEARCH found page %p, index %ld\n",
181 } /* obdfs_find_page_index */
184 /* call and free pages from Linux page cache: called with io lock on inodes */
185 int obdfs_do_vec_wr(struct inode **inodes, obd_count num_io,
186 obd_count num_obdos, struct obdo **obdos,
187 obd_count *oa_bufs, struct page **pages, char **bufs,
188 obd_size *counts, obd_off *offsets, obd_flag *flags)
193 if (IOPS(inodes[0], brw) == NULL) {
194 printk(KERN_ERR __FUNCTION__ ": no brw method!\n");
199 CDEBUG(D_INFO, "writing %d page(s), %d obdo(s) in vector\n",
201 if (obd_debug_level & D_INFO) { /* DEBUGGING */
204 for (i = 0; i < num_obdos; i++)
205 printk("%ld:0x%p ", (long)obdos[i]->o_id, obdos[i]);
208 for (i = 0; i < num_io; i++)
209 printk("0x%p ", pages[i]);
213 err = IOPS(inodes[0], brw)(WRITE, IID(inodes[0]), num_obdos, obdos,
214 oa_bufs, bufs, counts, offsets, flags);
216 CDEBUG(D_INFO, "BRW done\n");
217 /* release the pages from the page cache */
218 while ( num_io > 0 ) {
220 CDEBUG(D_INFO, "calling put_page for %p, index %ld\n",
221 pages[num_io], pages[num_io]->index);
222 /* PDEBUG(pages[num_io], "do_vec_wr"); */
223 put_page(pages[num_io]);
224 /* PDEBUG(pages[num_io], "do_vec_wr"); */
226 CDEBUG(D_INFO, "put_page done\n");
228 while ( num_obdos > 0) {
230 CDEBUG(D_INFO, "free obdo %ld\n",(long)obdos[num_obdos]->o_id);
231 /* copy o_blocks to i_blocks */
232 obdfs_to_inode(inodes[num_obdos], obdos[num_obdos]);
233 obdo_free(obdos[num_obdos]);
235 CDEBUG(D_INFO, "obdo_free done\n");
242 * Add a page to the write request cache list for later writing.
243 * ASYNCHRONOUS write method.
245 static int obdfs_add_page_to_cache(struct inode *inode, struct page *page)
250 /* The PG_obdcache bit is cleared by obdfs_pgrq_del() BEFORE the page
251 * is written, so at worst we will write the page out twice.
253 * If the page has the PG_obdcache bit set, then the inode MUST be
254 * on the superblock dirty list so we don't need to check this.
255 * Dirty inodes are removed from the superblock list ONLY when they
256 * don't have any more cached pages. It is possible to have an inode
257 * with no dirty pages on the superblock list, but not possible to
258 * have an inode with dirty pages NOT on the superblock dirty list.
260 if (!OBDAddCachePage(page)) {
261 struct obdfs_pgrq *pgrq;
262 pgrq = kmem_cache_alloc(obdfs_pgrq_cachep, SLAB_KERNEL);
264 OBDClearCachePage(page);
268 /* not really necessary since we set all pgrq fields here
269 memset(pgrq, 0, sizeof(*pgrq));
272 pgrq->rq_page = page;
273 pgrq->rq_jiffies = jiffies;
274 get_page(pgrq->rq_page);
276 obd_down(&obdfs_i2sbi(inode)->osi_list_mutex);
277 list_add(&pgrq->rq_plist, obdfs_iplist(inode));
280 /* If inode isn't already on superblock inodes list, add it.
282 * We increment the reference count on the inode to keep it
283 * from being freed from memory. This _should_ be an iget()
284 * with an iput() in both flush_reqs() and put_inode(), but
285 * since put_inode() is called from iput() we can't call iput()
286 * again there. Instead we just increment/decrement i_count,
287 * which is mostly what iget/iput do for an inode in memory.
289 if ( list_empty(obdfs_islist(inode)) ) {
292 "adding inode %ld to superblock list %p\n",
293 inode->i_ino, obdfs_slist(inode));
294 list_add(obdfs_islist(inode), obdfs_slist(inode));
296 obd_up(&obdfs_i2sbi(inode)->osi_list_mutex);
299 /* XXX For testing purposes, we can write out the page here.
300 err = obdfs_flush_reqs(obdfs_slist(inode), ~0UL);
305 } /* obdfs_add_page_to_cache */
308 /* select between SYNC and ASYNC I/O methods */
309 int obdfs_do_writepage(struct inode *inode, struct page *page, int sync)
314 /* PDEBUG(page, "WRITEPAGE"); */
316 err = obdfs_brw(WRITE, inode, page, 1);
318 err = obdfs_add_page_to_cache(inode, page);
319 CDEBUG(D_INFO, "DO_WR ino: %ld, page %p, err %d, uptodate %d\n",
320 inode->i_ino, page, err, Page_Uptodate(page));
324 SetPageUptodate(page);
325 /* PDEBUG(page,"WRITEPAGE"); */
328 } /* obdfs_do_writepage */
330 /* returns the page unlocked, but with a reference */
331 int obdfs_writepage(struct dentry *dentry, struct page *page)
333 return obdfs_do_writepage(dentry->d_inode, page, 0);
337 * This does the "real" work of the write. The generic routine has
338 * allocated the page, locked it, done all the page alignment stuff
339 * calculations etc. Now we should just copy the data from user
340 * space and write it back to the real medium..
342 * If the writer ends up delaying the write, the writer needs to
343 * increment the page use counts until he is done with the page.
345 * Return value is the number of bytes written.
347 int obdfs_write_one_page(struct file *file, struct page *page,
348 unsigned long offset, unsigned long bytes,
351 struct inode *inode = file->f_dentry->d_inode;
355 /* We check for complete page writes here, as we then don't have to
356 * get the page before writing over everything anyways.
358 if ( !Page_Uptodate(page) && (offset != 0 || bytes != PAGE_SIZE) ) {
359 err = obdfs_brw(READ, inode, page, 0);
362 SetPageUptodate(page);
365 if (copy_from_user((u8*)page_address(page) + offset, buf, bytes))
369 err = obdfs_writepage(file->f_dentry, page);
372 return (err < 0 ? err : bytes);
373 } /* obdfs_write_one_page */
376 * return an up to date page:
377 * - if locked is true then is returned locked
378 * - if create is true the corresponding disk blocks are created
379 * - page is held, i.e. caller must release the page
381 * modeled on NFS code.
383 struct page *obdfs_getpage(struct inode *inode, unsigned long offset,
384 int create, int locked)
392 offset = offset & PAGE_CACHE_MASK;
393 CDEBUG(D_INFO, "ino: %ld, offset %ld, create %d, locked %d\n",
394 inode->i_ino, offset, create, locked);
395 index = offset >> PAGE_CACHE_SHIFT;
397 page = grab_cache_page(&inode->i_data, index);
401 printk(KERN_WARNING " grab_cache_page says no dice ...\n");
406 /* PDEBUG(page, "GETPAGE: got page - before reading\n"); */
407 /* now check if the data in the page is up to date */
408 if ( Page_Uptodate(page)) {
416 #ifdef EXT2_OBD_DEBUG
417 if ((obd_debug_level & D_INFO) && obdfs_find_page_index(inode, index)) {
418 CDEBUG(D_INFO, "OVERWRITE: found dirty page %p, index %ld\n",
423 err = obdfs_brw(READ, inode, page, create);
434 SetPageUptodate(page);
435 /* PDEBUG(page,"GETPAGE - after reading"); */
438 } /* obdfs_getpage */
441 void obdfs_truncate(struct inode *inode)
447 obdfs_dequeue_pages(inode);
449 if (IOPS(inode, punch) == NULL) {
450 printk(KERN_ERR __FUNCTION__ ": no punch method!\n");
456 /* XXX This would give an inconsistent FS, so deal with it as
457 * best we can for now - an obdo on the stack is not pretty.
461 printk(__FUNCTION__ ": obdo_alloc failed - using stack!\n");
463 obdo.o_valid = OBD_MD_FLNOTOBD;
464 obdfs_from_inode(&obdo, inode);
466 err = IOPS(inode, punch)(IID(inode), &obdo, obdo.o_size, 0);
468 oa->o_valid = OBD_MD_FLNOTOBD;
469 obdfs_from_inode(oa, inode);
471 CDEBUG(D_INFO, "calling punch for %ld (%Lu bytes at 0)\n",
472 (long)oa->o_id, oa->o_size);
473 err = IOPS(inode, punch)(IID(inode), oa, oa->o_size, 0);
479 printk(__FUNCTION__ ": obd_truncate fails (%d)\n", err);
484 } /* obdfs_truncate */