1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Lite I/O Page Cache
6 * Copyright (c) 2001, 2002 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #include <linux/config.h>
25 #include <linux/kernel.h>
27 #include <linux/string.h>
28 #include <linux/stat.h>
29 #include <linux/iobuf.h>
30 #include <linux/errno.h>
31 #include <linux/smp_lock.h>
32 #include <linux/unistd.h>
33 #include <linux/version.h>
34 #include <asm/system.h>
35 #include <asm/uaccess.h>
38 #include <linux/stat.h>
39 #include <asm/uaccess.h>
40 #include <asm/segment.h>
42 #include <linux/pagemap.h>
43 #include <linux/smp_lock.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <linux/lustre_mds.h>
48 #include <linux/lustre_lite.h>
49 #include <linux/lustre_lib.h>
52 * Remove page from dirty list
54 static void __set_page_clean(struct page *page)
56 struct address_space *mapping = page->mapping;
62 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,5,0))
63 spin_lock(&pagecache_lock);
66 list_del(&page->list);
67 list_add(&page->list, &mapping->clean_pages);
69 inode = mapping->host;
70 if (list_empty(&mapping->dirty_pages)) {
71 CDEBUG(D_INODE, "inode clean\n");
72 inode->i_state &= ~I_DIRTY_PAGES;
74 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,5,0))
75 spin_unlock(&pagecache_lock);
80 inline void set_page_clean(struct page *page)
82 if (PageDirty(page)) {
84 __set_page_clean(page);
88 /* SYNCHRONOUS I/O to object storage for an inode */
89 static int ll_brw(int cmd, struct inode *inode, struct page *page, int create)
91 struct ll_inode_info *lli = ll_i2info(inode);
92 struct lov_stripe_md *lsm = lli->lli_smd;
93 struct obd_brw_set *set;
98 set = obd_brw_set_new();
103 pg.off = ((obd_off)page->index) << PAGE_SHIFT;
105 if (cmd == OBD_BRW_WRITE && (pg.off + PAGE_SIZE > inode->i_size))
106 pg.count = inode->i_size % PAGE_SIZE;
108 pg.count = PAGE_SIZE;
110 pg.flag = create ? OBD_BRW_CREATE : 0;
112 set->brw_callback = ll_brw_sync_wait;
113 rc = obd_brw(cmd, ll_i2obdconn(inode), lsm, 1, &pg, set);
116 CERROR("error from obd_brw: rc = %d\n", rc);
118 rc = ll_brw_sync_wait(set, CB_PHASE_START);
120 CERROR("error from callback: rc = %d\n", rc);
122 obd_brw_set_free(set);
127 /* returns the page unlocked, but with a reference */
128 static int ll_readpage(struct file *file, struct page *page)
130 struct inode *inode = page->mapping->host;
131 obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
135 if (!PageLocked(page))
138 if (inode->i_size <= offset) {
139 memset(kmap(page), 0, PAGE_SIZE);
141 GOTO(readpage_out, rc);
144 if (PageUptodate(page)) {
145 CERROR("Explain this please?\n");
146 GOTO(readpage_out, rc);
149 rc = ll_brw(OBD_BRW_READ, inode, page, 0);
154 SetPageUptodate(page);
159 void ll_truncate(struct inode *inode)
161 struct obdo oa = {0};
162 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
163 struct lustre_handle *lockhs = NULL;
168 /* object not yet allocated */
169 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
173 oa.o_id = lsm->lsm_object_id;
174 oa.o_mode = inode->i_mode;
175 oa.o_valid = OBD_MD_FLID | OBD_MD_FLMODE | OBD_MD_FLTYPE;
177 CDEBUG(D_INFO, "calling punch for "LPX64" (all bytes after "LPD64")\n",
178 oa.o_id, inode->i_size);
180 err = ll_size_lock(inode, lsm, inode->i_size, LCK_PW, &lockhs);
182 CERROR("ll_size_lock failed: %d\n", err);
183 /* FIXME: What to do here? It's too late to back out... */
187 /* truncate == punch from new size to absolute end of file */
188 err = obd_punch(ll_i2obdconn(inode), &oa, lsm, inode->i_size,
192 CERROR("obd_truncate fails (%d) ino %lu\n", err,
195 obdo_to_inode(inode, &oa, oa.o_valid);
197 err = ll_size_unlock(inode, lsm, LCK_PW, lockhs);
199 CERROR("ll_size_unlock failed: %d\n", err);
205 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
207 static int ll_prepare_write(struct file *file, struct page *page, unsigned from,
210 struct inode *inode = page->mapping->host;
211 obd_off offset = ((obd_off)page->index) << PAGE_SHIFT;
217 if (!PageLocked(page))
220 if (Page_Uptodate(page))
221 GOTO(prepare_done, rc);
223 /* We're completely overwriting an existing page, so _don't_ set it up
224 * to date until commit_write */
225 if (from == 0 && to == PAGE_SIZE)
228 /* We are writing to a new page, no need to read old data */
229 if (inode->i_size <= offset) {
230 memset(addr, 0, PAGE_SIZE);
231 GOTO(prepare_done, rc=0);
234 rc = ll_brw(OBD_BRW_READ, inode, page, 0);
239 SetPageUptodate(page);
244 /* Write a page from kupdated or kswapd.
246 * We unlock the page even in the face of an error, otherwise dirty
247 * pages could OOM the system if they cannot be written. Also, there
248 * is nobody to return an error code to from here - the application
249 * may not even be running anymore.
251 * Returns the page unlocked, but with a reference.
253 static int ll_writepage(struct page *page) {
254 struct inode *inode = page->mapping->host; int err; ENTRY;
256 LASSERT(PageLocked(page));
258 /* XXX need to make sure we have LDLM lock on this page */
259 err = ll_brw(OBD_BRW_WRITE, inode, page, 1);
261 CERROR("ll_brw failure %d\n", err);
263 set_page_clean(page);
270 /* SYNCHRONOUS I/O to object storage for an inode -- object attr will be updated
272 static int ll_commit_write(struct file *file, struct page *page,
273 unsigned from, unsigned to)
275 struct inode *inode = page->mapping->host;
276 struct ll_inode_info *lli = ll_i2info(inode);
277 struct lov_stripe_md *md = lli->lli_smd;
279 struct obd_brw_set *set;
286 pg.off = (((obd_off)page->index) << PAGE_SHIFT);
287 pg.flag = create ? OBD_BRW_CREATE : 0;
289 set = obd_brw_set_new();
293 SetPageUptodate(page);
295 if (!PageLocked(page))
298 CDEBUG(D_INODE, "commit_page writing (off "LPD64"), count "LPD64"\n",
301 set->brw_callback = ll_brw_sync_wait;
302 rc = obd_brw(OBD_BRW_WRITE, ll_i2obdconn(inode), md, 1, &pg, set);
304 CERROR("error from obd_brw: rc = %d\n", rc);
306 rc = ll_brw_sync_wait(set, CB_PHASE_START);
308 CERROR("error from callback: rc = %d\n", rc);
310 obd_brw_set_free(set);
313 size = pg.off + pg.count;
314 /* do NOT truncate when writing in the middle of a file */
315 if (size > inode->i_size)
316 inode->i_size = size;
319 } /* ll_commit_write */
322 static int ll_direct_IO(int rw, struct inode *inode, struct kiobuf *iobuf,
323 unsigned long blocknr, int blocksize)
325 obd_count bufs_per_obdo = iobuf->nr_pages;
326 struct ll_inode_info *lli = ll_i2info(inode);
327 struct lov_stripe_md *lsm = lli->lli_smd;
328 struct brw_page *pga;
329 struct obd_brw_set *set;
333 if (!lsm || !lsm->lsm_object_id)
336 if (blocksize != PAGE_SIZE) {
337 CERROR("direct_IO blocksize != PAGE_SIZE\n");
341 set = obd_brw_set_new();
345 OBD_ALLOC(pga, sizeof(*pga) * bufs_per_obdo);
347 obd_brw_set_free(set);
351 /* NB: we can't use iobuf->maplist[i]->index for the offset
352 * instead of "blocknr" because ->index contains garbage.
354 for (i = 0; i < bufs_per_obdo; i++, blocknr++) {
355 pga[i].pg = iobuf->maplist[i];
356 pga[i].count = PAGE_SIZE;
357 pga[i].off = (obd_off)blocknr << PAGE_SHIFT;
358 pga[i].flag = OBD_BRW_CREATE;
361 set->brw_callback = ll_brw_sync_wait;
362 rc = obd_brw(rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
363 ll_i2obdconn(inode), lsm, bufs_per_obdo, pga, set);
365 CERROR("error from obd_brw: rc = %d\n", rc);
367 rc = ll_brw_sync_wait(set, CB_PHASE_START);
369 CERROR("error from callback: rc = %d\n", rc);
371 obd_brw_set_free(set);
373 rc = bufs_per_obdo * PAGE_SIZE;
375 OBD_FREE(pga, sizeof(*pga) * bufs_per_obdo);
379 int ll_flush_inode_pages(struct inode * inode)
381 obd_count bufs_per_obdo = 0;
382 obd_size *count = NULL;
383 obd_off *offset = NULL;
384 obd_flag *flags = NULL;
389 spin_lock(&pagecache_lock);
391 spin_unlock(&pagecache_lock);
394 OBD_ALLOC(count, sizeof(*count) * bufs_per_obdo);
395 OBD_ALLOC(offset, sizeof(*offset) * bufs_per_obdo);
396 OBD_ALLOC(flags, sizeof(*flags) * bufs_per_obdo);
397 if (!count || !offset || !flags)
398 GOTO(out, err=-ENOMEM);
401 for (i = 0 ; i < bufs_per_obdo ; i++) {
402 count[i] = PAGE_SIZE;
403 offset[i] = ((obd_off)(iobuf->maplist[i])->index) << PAGE_SHIFT;
404 flags[i] = OBD_BRW_CREATE;
407 err = obd_brw(OBD_BRW_WRITE, ll_i2obdconn(inode),
408 ll_i2info(inode)->lli_smd, bufs_per_obdo,
409 iobuf->maplist, count, offset, flags, NULL, NULL);
411 err = bufs_per_obdo * 4096;
414 OBD_FREE(flags, sizeof(*flags) * bufs_per_obdo);
415 OBD_FREE(count, sizeof(*count) * bufs_per_obdo);
416 OBD_FREE(offset, sizeof(*offset) * bufs_per_obdo);
423 struct address_space_operations ll_aops = {
424 readpage: ll_readpage,
425 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,5,0))
426 direct_IO: ll_direct_IO,
427 writepage: ll_writepage,
428 sync_page: block_sync_page,
429 prepare_write: ll_prepare_write,
430 commit_write: ll_commit_write,