Whamcloud - gitweb
b=4834
[fs/lustre-release.git] / lustre / llite / llite_close.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * Lustre Lite routines to issue a secondary close after writeback
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/module.h>
25
26 #define DEBUG_SUBSYSTEM S_LLITE
27
28 #include <linux/lustre_mds.h>
29 #include <linux/lustre_lite.h>
30 #include "llite_internal.h"
31
32 /* record that a write is in flight */
33 void llap_write_pending(struct inode *inode, struct ll_async_page *llap)
34 {
35         struct ll_inode_info *lli = ll_i2info(inode);
36         spin_lock(&lli->lli_lock);
37         list_add(&llap->llap_pending_write, &lli->lli_pending_write_llaps);
38         spin_unlock(&lli->lli_lock);
39 }
40
41 /* record that a write has completed */
42 void llap_write_complete(struct inode *inode, struct ll_async_page *llap)
43 {
44         struct ll_inode_info *lli = ll_i2info(inode);
45         spin_lock(&lli->lli_lock);
46         list_del_init(&llap->llap_pending_write);
47         spin_unlock(&lli->lli_lock);
48 }
49
50 void ll_open_complete(struct inode *inode)
51 {
52         struct ll_inode_info *lli = ll_i2info(inode);
53         spin_lock(&lli->lli_lock);
54         lli->lli_send_done_writing = 0;
55         spin_unlock(&lli->lli_lock);
56 }
57
58 /* if we close with writes in flight then we want the completion or cancelation
59  * of those writes to send a DONE_WRITING rpc to the MDS */
60 int ll_is_inode_dirty(struct inode *inode)
61 {
62         struct ll_inode_info *lli = ll_i2info(inode);
63         int rc = 0;
64         ENTRY;
65
66         spin_lock(&lli->lli_lock);
67         if (!list_empty(&lli->lli_pending_write_llaps))
68                 rc = 1;
69         spin_unlock(&lli->lli_lock);
70         RETURN(rc);
71 }
72
73 void ll_try_done_writing(struct inode *inode)
74 {
75         struct ll_inode_info *lli = ll_i2info(inode);
76         struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;
77
78         spin_lock(&lli->lli_lock);
79
80         if (lli->lli_send_done_writing &&
81             list_empty(&lli->lli_pending_write_llaps)) {
82
83                 spin_lock(&lcq->lcq_lock);
84                 if (list_empty(&lli->lli_close_item)) {
85                         CDEBUG(D_INODE, "adding inode %lu/%u to close list\n",
86                                inode->i_ino, inode->i_generation);
87                         igrab(inode);
88                         list_add_tail(&lli->lli_close_item, &lcq->lcq_list);
89                         wake_up(&lcq->lcq_waitq);
90                 }
91                 spin_unlock(&lcq->lcq_lock);
92         }
93
94         spin_unlock(&lli->lli_lock);
95 }
96
97 /* The MDS needs us to get the real file attributes, then send a DONE_WRITING */
98 void ll_queue_done_writing(struct inode *inode)
99 {
100         struct ll_inode_info *lli = ll_i2info(inode);
101         ENTRY;
102
103         spin_lock(&lli->lli_lock);
104         lli->lli_send_done_writing = 1;
105         spin_unlock(&lli->lli_lock);
106
107         ll_try_done_writing(inode);
108         EXIT;
109 }
110
111 #if 0
112 /* If we know the file size and have the cookies:
113  *  - send a DONE_WRITING rpc
114  *
115  * Otherwise:
116  *  - get a whole-file lock
117  *  - get the authoritative size and all cookies with GETATTRs
118  *  - send a DONE_WRITING rpc
119  */
120 static void ll_close_done_writing(struct inode *inode)
121 {
122         struct ll_inode_info *lli = ll_i2info(inode);
123         ldlm_policy_data_t policy = { .l_extent = {0, OBD_OBJECT_EOF } };
124         struct lustre_handle lockh = { 0 };
125         struct obdo obdo;
126         obd_flag valid;
127         int rc, ast_flags = 0;
128         ENTRY;
129
130         memset(&obdo, 0, sizeof(obdo));
131         if (test_bit(LLI_F_HAVE_OST_SIZE_LOCK, &lli->lli_flags))
132                 goto rpc;
133
134         rc = ll_extent_lock(NULL, inode, lli->lli_smd, LCK_PW, &policy, &lockh,
135                             ast_flags);
136         if (rc != 0) {
137                 CERROR("lock acquisition failed (%d): unable to send "
138                        "DONE_WRITING for inode %lu/%u\n", rc, inode->i_ino,
139                        inode->i_generation);
140                 GOTO(out, rc);
141         }
142
143         rc = ll_lsm_getattr(ll_i2obdexp(inode), lli->lli_smd, &obdo);
144         if (rc) {
145                 CERROR("inode_getattr failed (%d): unable to send DONE_WRITING "
146                        "for inode %lu/%u\n", rc, inode->i_ino,
147                        inode->i_generation);
148                 ll_extent_unlock(NULL, inode, lli->lli_smd, LCK_PW, &lockh);
149                 GOTO(out, rc);
150         }
151
152         obdo_refresh_inode(inode, &obdo, valid);
153
154         CDEBUG(D_INODE, "objid "LPX64" size %Lu, blocks %lu, blksize %lu\n",
155                lli->lli_smd->lsm_object_id, inode->i_size, inode->i_blocks,
156                inode->i_blksize);
157
158         set_bit(LLI_F_HAVE_OST_SIZE_LOCK, &lli->lli_flags);
159
160         rc = ll_extent_unlock(NULL, inode, lli->lli_smd, LCK_PW, &lockh);
161         if (rc != ELDLM_OK)
162                 CERROR("unlock failed (%d)?  proceeding anyways...\n", rc);
163
164  rpc:
165         obdo.o_id = inode->i_ino;
166         obdo.o_size = inode->i_size;
167         obdo.o_blocks = inode->i_blocks;
168         obdo.o_valid = OBD_MD_FLID | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
169
170         rc = mdc_done_writing(ll_i2sbi(inode)->ll_mdc_exp, &obdo);
171  out:
172 }
173 #endif
174
175 static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
176 {
177         struct ll_inode_info *lli = NULL;
178
179         spin_lock(&lcq->lcq_lock);
180
181         if (lcq->lcq_list.next == NULL)
182                 lli = ERR_PTR(-1);
183         else if (!list_empty(&lcq->lcq_list)) {
184                 lli = list_entry(lcq->lcq_list.next, struct ll_inode_info,
185                                  lli_close_item);
186                 list_del(&lli->lli_close_item);
187         }
188
189         spin_unlock(&lcq->lcq_lock);
190         return lli;
191 }
192
193 static int ll_close_thread(void *arg)
194 {
195         struct ll_close_queue *lcq = arg;
196         ENTRY;
197
198         /* XXX boiler-plate */
199         {
200                 char name[sizeof(current->comm)];
201                 unsigned long flags;
202                 snprintf(name, sizeof(name) - 1, "ll_close");
203                 kportal_daemonize(name);
204                 SIGNAL_MASK_LOCK(current, flags);
205                 sigfillset(&current->blocked);
206                 RECALC_SIGPENDING;
207                 SIGNAL_MASK_UNLOCK(current, flags);
208         }
209
210         complete(&lcq->lcq_comp);
211
212         while (1) {
213                 struct l_wait_info lwi = { 0 };
214                 struct ll_inode_info *lli;
215                 //struct inode *inode;
216
217                 l_wait_event_exclusive(lcq->lcq_waitq,
218                                        (lli = ll_close_next_lli(lcq)) != NULL,
219                                        &lwi);
220                 if (IS_ERR(lli))
221                         break;
222
223                 //inode = ll_info2i(lli);
224                 //ll_close_done_writing(inode);
225                 //iput(inode);
226         }
227
228         complete(&lcq->lcq_comp);
229         RETURN(0);
230 }
231
232 int ll_close_thread_start(struct ll_close_queue **lcq_ret)
233 {
234         struct ll_close_queue *lcq;
235         pid_t pid;
236
237         OBD_ALLOC(lcq, sizeof(*lcq));
238         if (lcq == NULL)
239                 return -ENOMEM;
240
241         spin_lock_init(&lcq->lcq_lock);
242         INIT_LIST_HEAD(&lcq->lcq_list);
243         init_waitqueue_head(&lcq->lcq_waitq);
244         init_completion(&lcq->lcq_comp);
245
246         pid = kernel_thread(ll_close_thread, lcq, 0);
247         if (pid < 0) {
248                 OBD_FREE(lcq, sizeof(*lcq));
249                 return pid;
250         }
251
252         wait_for_completion(&lcq->lcq_comp);
253         *lcq_ret = lcq;
254         return 0;
255 }
256
257 void ll_close_thread_shutdown(struct ll_close_queue *lcq)
258 {
259         init_completion(&lcq->lcq_comp);
260         lcq->lcq_list.next = NULL;
261         wake_up(&lcq->lcq_waitq);
262         wait_for_completion(&lcq->lcq_comp);
263         OBD_FREE(lcq, sizeof(*lcq));
264 }