1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Lustre Lite I/O page cache routines for the 2.5/2.6 kernel version
41 #ifndef AUTOCONF_INCLUDED
42 #include <linux/config.h>
44 #include <linux/kernel.h>
46 #include <linux/string.h>
47 #include <linux/stat.h>
48 #include <linux/errno.h>
49 #include <linux/smp_lock.h>
50 #include <linux/unistd.h>
51 #include <linux/version.h>
52 #include <asm/system.h>
53 #include <asm/uaccess.h>
56 #include <linux/buffer_head.h>
57 #include <linux/writeback.h>
58 #include <linux/stat.h>
59 #include <asm/uaccess.h>
61 # include <asm/segment.h>
64 #include <linux/pagemap.h>
65 #include <linux/smp_lock.h>
67 #define DEBUG_SUBSYSTEM S_LLITE
69 #include <lustre_lite.h>
70 #include "llite_internal.h"
71 #include <linux/lustre_compat25.h>
73 static int ll_writepage_26(struct page *page, struct writeback_control *wbc)
75 return ll_writepage(page);
78 /* It is safe to not check anything in invalidatepage/releasepage below
79 because they are run with page locked and all our io is happening with
81 #ifdef HAVE_INVALIDATEPAGE_RETURN_INT
82 static int ll_invalidatepage(struct page *page, unsigned long offset)
86 if (PagePrivate(page))
91 static void ll_invalidatepage(struct page *page, unsigned long offset)
95 if (PagePrivate(page))
100 #ifdef HAVE_RELEASEPAGE_WITH_GFP
101 #define RELEASEPAGE_ARG_TYPE gfp_t
103 #define RELEASEPAGE_ARG_TYPE int
105 static int ll_releasepage(struct page *page, RELEASEPAGE_ARG_TYPE gfp_mask)
107 if (PagePrivate(page))
112 #define MAX_DIRECTIO_SIZE 2*1024*1024*1024UL
114 static inline int ll_get_user_pages(int rw, unsigned long user_addr,
115 size_t size, struct page ***pages)
117 int result = -ENOMEM;
120 /* set an arbitrary limit to prevent arithmetic overflow */
121 if (size > MAX_DIRECTIO_SIZE) {
126 page_count = ((user_addr + size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT)-
127 (user_addr >> CFS_PAGE_SHIFT);
129 OBD_ALLOC_WAIT(*pages, page_count * sizeof(**pages));
131 down_read(¤t->mm->mmap_sem);
132 result = get_user_pages(current, current->mm, user_addr,
133 page_count, (rw == READ), 0, *pages,
135 up_read(¤t->mm->mmap_sem);
137 OBD_FREE(*pages, page_count * sizeof(**pages));
143 /* ll_free_user_pages - tear down page struct array
144 * @pages: array of page struct pointers underlying target buffer */
145 static void ll_free_user_pages(struct page **pages, int npages, int do_dirty)
149 for (i = 0; i < npages; i++) {
151 set_page_dirty_lock(pages[i]);
152 page_cache_release(pages[i]);
155 OBD_FREE(pages, npages * sizeof(*pages));
158 static ssize_t ll_direct_IO_26_seg(int rw, struct inode *inode,
159 struct address_space *mapping,
160 struct obd_info *oinfo,
161 struct ptlrpc_request_set *set,
162 size_t size, loff_t file_offset,
163 struct page **pages, int page_count)
165 struct brw_page *pga;
170 OBD_ALLOC(pga, sizeof(*pga) * page_count);
172 CDEBUG(D_VFSTRACE, "sizeof(*pga) = %u page_count = %u\n",
173 (int)sizeof(*pga), page_count);
177 for (i = 0, length = size; length > 0;
178 length -=pga[i].count, file_offset +=pga[i].count,i++) {/*i last!*/
179 pga[i].pg = pages[i];
180 pga[i].off = file_offset;
181 /* To the end of the page, or the length, whatever is less */
182 pga[i].count = min_t(int, CFS_PAGE_SIZE -(file_offset & ~CFS_PAGE_MASK),
184 pga[i].flag = OBD_BRW_SYNC;
186 POISON_PAGE(pages[i], 0x0d);
189 rc = obd_brw_async(rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
190 ll_i2obdexp(inode), oinfo, page_count,
195 OBD_FREE(pga, sizeof(*pga) * page_count);
199 /* This is the maximum size of a single O_DIRECT request, based on a 128kB
200 * kmalloc limit. We need to fit all of the brw_page structs, each one
201 * representing PAGE_SIZE worth of user data, into a single buffer, and
202 * then truncate this to be a full-sized RPC. This is 22MB for 4kB pages. */
203 #define MAX_DIO_SIZE ((128 * 1024 / sizeof(struct brw_page) * CFS_PAGE_SIZE) & \
204 ~(PTLRPC_MAX_BRW_SIZE - 1))
205 static ssize_t ll_direct_IO_26(int rw, struct kiocb *iocb,
206 const struct iovec *iov, loff_t file_offset,
207 unsigned long nr_segs)
209 struct file *file = iocb->ki_filp;
210 struct inode *inode = file->f_mapping->host;
211 ssize_t count = iov_length(iov, nr_segs), tot_bytes = 0;
212 struct ll_inode_info *lli = ll_i2info(inode);
213 struct lov_stripe_md *lsm = lli->lli_smd;
214 struct ptlrpc_request_set *set;
215 struct obd_info oinfo;
218 size_t size = MAX_DIO_SIZE;
221 if (!lli->lli_smd || !lli->lli_smd->lsm_object_id)
224 /* FIXME: io smaller than CFS_PAGE_SIZE is broken on ia64 ??? */
225 if ((file_offset & (~CFS_PAGE_MASK)) || (count & ~CFS_PAGE_MASK))
228 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), size="LPSZ" (max %lu), "
229 "offset=%lld=%llx, pages "LPSZ" (max %lu)\n",
230 inode->i_ino, inode->i_generation, inode, count, MAX_DIO_SIZE,
231 file_offset, file_offset, count >> CFS_PAGE_SHIFT,
232 MAX_DIO_SIZE >> CFS_PAGE_SHIFT);
235 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_DIRECT_WRITE, count);
237 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_DIRECT_READ, count);
239 /* Check that all user buffers are aligned as well */
240 for (seg = 0; seg < nr_segs; seg++) {
241 if (((unsigned long)iov[seg].iov_base & ~CFS_PAGE_MASK) ||
242 (iov[seg].iov_len & ~CFS_PAGE_MASK))
246 set = ptlrpc_prep_set();
250 ll_inode_fill_obdo(inode, rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ, &oa);
254 /* need locking between buffered and direct access. and race with
255 *size changing by concurrent truncates and writes. */
257 LOCK_INODE_MUTEX(inode);
259 for (seg = 0; seg < nr_segs; seg++) {
260 size_t iov_left = iov[seg].iov_len;
261 unsigned long user_addr = (unsigned long)iov[seg].iov_base;
264 if (file_offset >= inode->i_size)
266 if (file_offset + iov_left > inode->i_size)
267 iov_left = inode->i_size - file_offset;
270 while (iov_left > 0) {
275 page_count = ll_get_user_pages(rw, user_addr,
278 LASSERT(page_count != 0);
279 if (page_count > 0) {
280 result = ll_direct_IO_26_seg(rw, inode,
286 ll_free_user_pages(pages, page_count, rw==READ);
290 if (page_count < 0 || result <= 0) {
291 /* If we can't allocate a large enough buffer
292 * for the request, shrink it to a smaller
293 * PAGE_SIZE multiple and try again.
294 * We should always be able to kmalloc for a
295 * page worth of page pointers = 4MB on i386. */
296 if ((page_count == -ENOMEM||result == -ENOMEM)&&
297 size > (CFS_PAGE_SIZE / sizeof(*pages)) *
299 size = ((((size / 2) - 1) |
300 ~CFS_PAGE_MASK) + 1) &
302 CDEBUG(D_VFSTRACE, "DIO size now %u\n",
307 GOTO(wait_io, tot_bytes);
308 GOTO(out, tot_bytes = page_count < 0 ? page_count : result);
312 file_offset += result;
321 rc = ptlrpc_set_wait(set);
323 GOTO(out, tot_bytes = rc);
325 lov_stripe_lock(lsm);
326 obd_adjust_kms(ll_i2obdexp(inode), lsm, file_offset, 0);
327 lov_stripe_unlock(lsm);
332 UNLOCK_INODE_MUTEX(inode);
334 ptlrpc_set_destroy(set);
338 struct address_space_operations ll_aops = {
339 .readpage = ll_readpage,
340 // .readpages = ll_readpages,
341 .direct_IO = ll_direct_IO_26,
342 .writepage = ll_writepage_26,
343 .writepages = generic_writepages,
344 .set_page_dirty = __set_page_dirty_nobuffers,
346 .prepare_write = ll_prepare_write,
347 .commit_write = ll_commit_write,
348 .invalidatepage = ll_invalidatepage,
349 .releasepage = ll_releasepage,