Whamcloud - gitweb
b=2103
[fs/lustre-release.git] / lustre / llite / rw24.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * Lustre Lite I/O page cache for the 2.4 kernel version
5  *
6  *  Copyright (c) 2001-2003 Cluster File Systems, Inc.
7  *
8  *   This file is part of Lustre, http://www.lustre.org.
9  *
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.
13  *
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.
18  *
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.
22  */
23
24 #include <linux/config.h>
25 #include <linux/kernel.h>
26 #include <linux/mm.h>
27 #include <linux/string.h>
28 #include <linux/stat.h>
29 #include <linux/errno.h>
30 #include <linux/smp_lock.h>
31 #include <linux/unistd.h>
32 #include <linux/version.h>
33 #include <asm/system.h>
34 #include <asm/uaccess.h>
35
36 #include <linux/fs.h>
37 #include <linux/iobuf.h>
38 #include <linux/stat.h>
39 #include <asm/uaccess.h>
40 #include <asm/segment.h>
41 #include <linux/mm.h>
42 #include <linux/pagemap.h>
43 #include <linux/smp_lock.h>
44
45 #define DEBUG_SUBSYSTEM S_LLITE
46
47 #include <linux/lustre_mds.h>
48 #include <linux/lustre_lite.h>
49 #include "llite_internal.h"
50 #include <linux/lustre_compat25.h>
51
52 /* called for each page in a completed rpc.*/
53 void ll_ap_completion_24(void *data, int cmd, int rc)
54 {
55         struct ll_async_page *llap;
56         struct page *page;
57
58         llap = llap_from_cookie(data);
59         if (IS_ERR(llap)) {
60                 EXIT;
61                 return;
62         }
63
64         page = llap->llap_page;
65         LASSERT(PageLocked(page));
66
67         if (rc == 0)  {
68                 if (cmd == OBD_BRW_READ) {
69                         if (!llap->llap_defer_uptodate)
70                                 SetPageUptodate(page);
71                 } else {
72                         llap->llap_write_queued = 0;
73                 }
74         } else { 
75                 SetPageError(page);
76         }
77
78         LL_CDEBUG_PAGE(page, "io complete, unlocking\n");
79
80         unlock_page(page);
81
82         if (0 && cmd == OBD_BRW_WRITE) {
83                 llap_write_complete(page->mapping->host, llap);
84                 ll_try_done_writing(page->mapping->host);
85         }
86
87         page_cache_release(page);
88 }
89
90 static int ll_writepage_24(struct page *page)
91 {
92         struct inode *inode = page->mapping->host;
93         struct obd_export *exp;
94         struct ll_async_page *llap;
95         int rc = 0;
96         ENTRY;
97
98         LASSERT(!PageDirty(page));
99         LASSERT(PageLocked(page));
100
101         exp = ll_i2obdexp(inode);
102         if (exp == NULL)
103                 GOTO(out, rc = -EINVAL);
104
105         llap = llap_from_page(page);
106         if (IS_ERR(llap))
107                 GOTO(out, rc = PTR_ERR(llap));
108
109         page_cache_get(page);
110         if (llap->llap_write_queued) {
111                 LL_CDEBUG_PAGE(page, "marking urgent\n");
112                 rc = obd_set_async_flags(exp, ll_i2info(inode)->lli_smd, NULL, 
113                                          llap->llap_cookie, ASYNC_READY | 
114                                          ASYNC_URGENT);
115         } else {
116                 llap->llap_write_queued = 1;
117                 rc = obd_queue_async_io(exp, ll_i2info(inode)->lli_smd, NULL, 
118                                         llap->llap_cookie, OBD_BRW_WRITE, 0, 0, 
119                                         OBD_BRW_CREATE, ASYNC_READY | 
120                                         ASYNC_URGENT);
121                 if (rc == 0)
122                         LL_CDEBUG_PAGE(page, "mmap write queued\n");
123                 else 
124                         llap->llap_write_queued = 0;
125         }
126         if (rc)
127                 page_cache_release(page);
128 out:
129         if (rc)
130                 unlock_page(page);
131         RETURN(rc);
132 }
133
134 static int ll_direct_IO_24(int rw, struct inode *inode, struct kiobuf *iobuf,
135                            unsigned long blocknr, int blocksize)
136 {
137         struct ll_inode_info *lli = ll_i2info(inode);
138         struct lov_stripe_md *lsm = lli->lli_smd;
139         struct brw_page *pga;
140         struct ptlrpc_request_set *set;
141         struct obdo oa;
142         int length, i, flags, rc = 0;
143         loff_t offset;
144         ENTRY;
145
146         if (!lsm || !lsm->lsm_object_id)
147                 RETURN(-EBADF);
148
149         /* FIXME: io smaller than PAGE_SIZE is broken on ia64 */
150         if ((iobuf->offset & (PAGE_SIZE - 1)) ||
151             (iobuf->length & (PAGE_SIZE - 1)))
152                 RETURN(-EINVAL);
153
154         set = ptlrpc_prep_set();
155         if (set == NULL)
156                 RETURN(-ENOMEM);
157
158         OBD_ALLOC(pga, sizeof(*pga) * iobuf->nr_pages);
159         if (!pga) {
160                 ptlrpc_set_destroy(set);
161                 RETURN(-ENOMEM);
162         }
163
164         flags = (rw == WRITE ? OBD_BRW_CREATE : 0) /* | OBD_BRW_DIRECTIO */;
165         offset = ((obd_off)blocknr << inode->i_blkbits);
166         length = iobuf->length;
167
168         for (i = 0, length = iobuf->length; length > 0;
169              length -= pga[i].count, offset += pga[i].count, i++) { /*i last!*/
170                 pga[i].pg = iobuf->maplist[i];
171                 pga[i].off = offset;
172                 /* To the end of the page, or the length, whatever is less */
173                 pga[i].count = min_t(int, PAGE_SIZE - (offset & ~PAGE_MASK),
174                                      length);
175                 pga[i].flag = flags;
176                 if (rw == READ)
177                         POISON_PAGE(iobuf->maplist[i], 0x0d);
178         }
179
180         ll_inode_fill_obdo(inode, rw, &oa);
181
182         if (rw == WRITE)
183                 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats,
184                                     LPROC_LL_DIRECT_WRITE, iobuf->length);
185         else
186                 lprocfs_counter_add(ll_i2sbi(inode)->ll_stats,
187                                     LPROC_LL_DIRECT_READ, iobuf->length);
188         rc = obd_brw_async(rw == WRITE ? OBD_BRW_WRITE : OBD_BRW_READ,
189                            ll_i2obdexp(inode), &oa, lsm, iobuf->nr_pages, pga,
190                            set, NULL);
191         if (rc) {
192                 CDEBUG(rc == -ENOSPC ? D_INODE : D_ERROR,
193                        "error from obd_brw_async: rc = %d\n", rc);
194         } else {
195                 rc = ptlrpc_set_wait(set);
196                 if (rc)
197                         CERROR("error from callback: rc = %d\n", rc);
198         }
199         ptlrpc_set_destroy(set);
200         if (rc == 0) {
201                 rc = iobuf->length;
202                 obdo_to_inode(inode, &oa, OBD_MD_FLBLOCKS);
203         }
204
205         OBD_FREE(pga, sizeof(*pga) * iobuf->nr_pages);
206         RETURN(rc);
207 }
208
209 struct address_space_operations ll_aops = {
210         readpage: ll_readpage,
211         direct_IO: ll_direct_IO_24,
212         writepage: ll_writepage_24,
213         prepare_write: ll_prepare_write,
214         commit_write: ll_commit_write,
215         removepage: ll_removepage,
216         sync_page: ll_sync_page,
217         bmap: NULL
218 };