1 |
1275 |
phoenix |
/*
|
2 |
|
|
* linux/fs/nfs/write.c
|
3 |
|
|
*
|
4 |
|
|
* Writing file data over NFS.
|
5 |
|
|
*
|
6 |
|
|
* We do it like this: When a (user) process wishes to write data to an
|
7 |
|
|
* NFS file, a write request is allocated that contains the RPC task data
|
8 |
|
|
* plus some info on the page to be written, and added to the inode's
|
9 |
|
|
* write chain. If the process writes past the end of the page, an async
|
10 |
|
|
* RPC call to write the page is scheduled immediately; otherwise, the call
|
11 |
|
|
* is delayed for a few seconds.
|
12 |
|
|
*
|
13 |
|
|
* Just like readahead, no async I/O is performed if wsize < PAGE_SIZE.
|
14 |
|
|
*
|
15 |
|
|
* Write requests are kept on the inode's writeback list. Each entry in
|
16 |
|
|
* that list references the page (portion) to be written. When the
|
17 |
|
|
* cache timeout has expired, the RPC task is woken up, and tries to
|
18 |
|
|
* lock the page. As soon as it manages to do so, the request is moved
|
19 |
|
|
* from the writeback list to the writelock list.
|
20 |
|
|
*
|
21 |
|
|
* Note: we must make sure never to confuse the inode passed in the
|
22 |
|
|
* write_page request with the one in page->inode. As far as I understand
|
23 |
|
|
* it, these are different when doing a swap-out.
|
24 |
|
|
*
|
25 |
|
|
* To understand everything that goes on here and in the NFS read code,
|
26 |
|
|
* one should be aware that a page is locked in exactly one of the following
|
27 |
|
|
* cases:
|
28 |
|
|
*
|
29 |
|
|
* - A write request is in progress.
|
30 |
|
|
* - A user process is in generic_file_write/nfs_update_page
|
31 |
|
|
* - A user process is in generic_file_read
|
32 |
|
|
*
|
33 |
|
|
* Also note that because of the way pages are invalidated in
|
34 |
|
|
* nfs_revalidate_inode, the following assertions hold:
|
35 |
|
|
*
|
36 |
|
|
* - If a page is dirty, there will be no read requests (a page will
|
37 |
|
|
* not be re-read unless invalidated by nfs_revalidate_inode).
|
38 |
|
|
* - If the page is not uptodate, there will be no pending write
|
39 |
|
|
* requests, and no process will be in nfs_update_page.
|
40 |
|
|
*
|
41 |
|
|
* FIXME: Interaction with the vmscan routines is not optimal yet.
|
42 |
|
|
* Either vmscan must be made nfs-savvy, or we need a different page
|
43 |
|
|
* reclaim concept that supports something like FS-independent
|
44 |
|
|
* buffer_heads with a b_ops-> field.
|
45 |
|
|
*
|
46 |
|
|
* Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
|
47 |
|
|
*/
|
48 |
|
|
|
49 |
|
|
#include <linux/config.h>
|
50 |
|
|
#include <linux/types.h>
|
51 |
|
|
#include <linux/slab.h>
|
52 |
|
|
#include <linux/swap.h>
|
53 |
|
|
#include <linux/pagemap.h>
|
54 |
|
|
#include <linux/file.h>
|
55 |
|
|
|
56 |
|
|
#include <linux/sunrpc/clnt.h>
|
57 |
|
|
#include <linux/nfs_fs.h>
|
58 |
|
|
#include <linux/nfs_mount.h>
|
59 |
|
|
#include <linux/nfs_flushd.h>
|
60 |
|
|
#include <linux/nfs_page.h>
|
61 |
|
|
#include <asm/uaccess.h>
|
62 |
|
|
#include <linux/smp_lock.h>
|
63 |
|
|
|
64 |
|
|
#define NFSDBG_FACILITY NFSDBG_PAGECACHE
|
65 |
|
|
|
66 |
|
|
/*
|
67 |
|
|
* Local structures
|
68 |
|
|
*
|
69 |
|
|
* This is the struct where the WRITE/COMMIT arguments go.
|
70 |
|
|
*/
|
71 |
|
|
struct nfs_write_data {
|
72 |
|
|
struct rpc_task task;
|
73 |
|
|
struct inode *inode;
|
74 |
|
|
struct rpc_cred *cred;
|
75 |
|
|
struct nfs_writeargs args; /* argument struct */
|
76 |
|
|
struct nfs_writeres res; /* result struct */
|
77 |
|
|
struct nfs_fattr fattr;
|
78 |
|
|
struct nfs_writeverf verf;
|
79 |
|
|
struct list_head pages; /* Coalesced requests we wish to flush */
|
80 |
|
|
struct page *pagevec[NFS_WRITE_MAXIOV];
|
81 |
|
|
};
|
82 |
|
|
|
83 |
|
|
/*
|
84 |
|
|
* Local function declarations
|
85 |
|
|
*/
|
86 |
|
|
static struct nfs_page * nfs_update_request(struct file*, struct inode *,
|
87 |
|
|
struct page *,
|
88 |
|
|
unsigned int, unsigned int);
|
89 |
|
|
static void nfs_strategy(struct inode *inode);
|
90 |
|
|
static void nfs_writeback_done(struct rpc_task *);
|
91 |
|
|
#ifdef CONFIG_NFS_V3
|
92 |
|
|
static void nfs_commit_done(struct rpc_task *);
|
93 |
|
|
#endif
|
94 |
|
|
|
95 |
|
|
/* Hack for future NFS swap support */
|
96 |
|
|
#ifndef IS_SWAPFILE
|
97 |
|
|
# define IS_SWAPFILE(inode) (0)
|
98 |
|
|
#endif
|
99 |
|
|
|
100 |
|
|
static kmem_cache_t *nfs_wdata_cachep;
|
101 |
|
|
|
102 |
|
|
static __inline__ struct nfs_write_data *nfs_writedata_alloc(void)
|
103 |
|
|
{
|
104 |
|
|
struct nfs_write_data *p;
|
105 |
|
|
p = kmem_cache_alloc(nfs_wdata_cachep, SLAB_NOFS);
|
106 |
|
|
if (p) {
|
107 |
|
|
memset(p, 0, sizeof(*p));
|
108 |
|
|
INIT_LIST_HEAD(&p->pages);
|
109 |
|
|
p->args.pages = p->pagevec;
|
110 |
|
|
}
|
111 |
|
|
return p;
|
112 |
|
|
}
|
113 |
|
|
|
114 |
|
|
static __inline__ void nfs_writedata_free(struct nfs_write_data *p)
|
115 |
|
|
{
|
116 |
|
|
kmem_cache_free(nfs_wdata_cachep, p);
|
117 |
|
|
}
|
118 |
|
|
|
119 |
|
|
static void nfs_writedata_release(struct rpc_task *task)
|
120 |
|
|
{
|
121 |
|
|
struct nfs_write_data *wdata = (struct nfs_write_data *)task->tk_calldata;
|
122 |
|
|
nfs_writedata_free(wdata);
|
123 |
|
|
}
|
124 |
|
|
|
125 |
|
|
/*
|
126 |
|
|
* Write a page synchronously.
|
127 |
|
|
* Offset is the data offset within the page.
|
128 |
|
|
*/
|
129 |
|
|
static int
|
130 |
|
|
nfs_writepage_sync(struct file *file, struct inode *inode, struct page *page,
|
131 |
|
|
unsigned int offset, unsigned int count)
|
132 |
|
|
{
|
133 |
|
|
struct rpc_cred *cred = NULL;
|
134 |
|
|
loff_t base;
|
135 |
|
|
unsigned int wsize = NFS_SERVER(inode)->wsize;
|
136 |
|
|
int result, refresh = 0, written = 0, flags;
|
137 |
|
|
u8 *buffer;
|
138 |
|
|
struct nfs_fattr fattr;
|
139 |
|
|
struct nfs_writeverf verf;
|
140 |
|
|
|
141 |
|
|
|
142 |
|
|
if (file)
|
143 |
|
|
cred = get_rpccred(nfs_file_cred(file));
|
144 |
|
|
if (!cred)
|
145 |
|
|
cred = get_rpccred(NFS_I(inode)->mm_cred);
|
146 |
|
|
|
147 |
|
|
dprintk("NFS: nfs_writepage_sync(%x/%Ld %d@%Ld)\n",
|
148 |
|
|
inode->i_dev, (long long)NFS_FILEID(inode),
|
149 |
|
|
count, (long long)(page_offset(page) + offset));
|
150 |
|
|
|
151 |
|
|
base = page_offset(page) + offset;
|
152 |
|
|
|
153 |
|
|
flags = ((IS_SWAPFILE(inode)) ? NFS_RW_SWAP : 0) | NFS_RW_SYNC;
|
154 |
|
|
|
155 |
|
|
do {
|
156 |
|
|
if (count < wsize && !IS_SWAPFILE(inode))
|
157 |
|
|
wsize = count;
|
158 |
|
|
|
159 |
|
|
result = NFS_PROTO(inode)->write(inode, cred, &fattr, flags,
|
160 |
|
|
offset, wsize, page, &verf);
|
161 |
|
|
nfs_write_attributes(inode, &fattr);
|
162 |
|
|
|
163 |
|
|
if (result < 0) {
|
164 |
|
|
/* Must mark the page invalid after I/O error */
|
165 |
|
|
ClearPageUptodate(page);
|
166 |
|
|
goto io_error;
|
167 |
|
|
}
|
168 |
|
|
if (result != wsize)
|
169 |
|
|
printk("NFS: short write, wsize=%u, result=%d\n",
|
170 |
|
|
wsize, result);
|
171 |
|
|
refresh = 1;
|
172 |
|
|
buffer += wsize;
|
173 |
|
|
base += wsize;
|
174 |
|
|
offset += wsize;
|
175 |
|
|
written += wsize;
|
176 |
|
|
count -= wsize;
|
177 |
|
|
/*
|
178 |
|
|
* If we've extended the file, update the inode
|
179 |
|
|
* now so we don't invalidate the cache.
|
180 |
|
|
*/
|
181 |
|
|
if (base > inode->i_size)
|
182 |
|
|
inode->i_size = base;
|
183 |
|
|
} while (count);
|
184 |
|
|
|
185 |
|
|
if (PageError(page))
|
186 |
|
|
ClearPageError(page);
|
187 |
|
|
|
188 |
|
|
io_error:
|
189 |
|
|
if (cred)
|
190 |
|
|
put_rpccred(cred);
|
191 |
|
|
|
192 |
|
|
return written? written : result;
|
193 |
|
|
}
|
194 |
|
|
|
195 |
|
|
static int
|
196 |
|
|
nfs_writepage_async(struct file *file, struct inode *inode, struct page *page,
|
197 |
|
|
unsigned int offset, unsigned int count)
|
198 |
|
|
{
|
199 |
|
|
struct nfs_page *req;
|
200 |
|
|
loff_t end;
|
201 |
|
|
int status;
|
202 |
|
|
|
203 |
|
|
req = nfs_update_request(file, inode, page, offset, count);
|
204 |
|
|
status = (IS_ERR(req)) ? PTR_ERR(req) : 0;
|
205 |
|
|
if (status < 0)
|
206 |
|
|
goto out;
|
207 |
|
|
if (!req->wb_cred)
|
208 |
|
|
req->wb_cred = get_rpccred(NFS_I(inode)->mm_cred);
|
209 |
|
|
nfs_unlock_request(req);
|
210 |
|
|
nfs_strategy(inode);
|
211 |
|
|
end = ((loff_t)page->index<<PAGE_CACHE_SHIFT) + (loff_t)(offset + count);
|
212 |
|
|
if (inode->i_size < end)
|
213 |
|
|
inode->i_size = end;
|
214 |
|
|
|
215 |
|
|
out:
|
216 |
|
|
return status;
|
217 |
|
|
}
|
218 |
|
|
|
219 |
|
|
/*
|
220 |
|
|
* Write an mmapped page to the server.
|
221 |
|
|
*/
|
222 |
|
|
int
|
223 |
|
|
nfs_writepage(struct page *page)
|
224 |
|
|
{
|
225 |
|
|
struct inode *inode = page->mapping->host;
|
226 |
|
|
unsigned long end_index;
|
227 |
|
|
unsigned offset = PAGE_CACHE_SIZE;
|
228 |
|
|
int inode_referenced = 0;
|
229 |
|
|
int err;
|
230 |
|
|
|
231 |
|
|
/*
|
232 |
|
|
* Note: We need to ensure that we have a reference to the inode
|
233 |
|
|
* if we are to do asynchronous writes. If not, waiting
|
234 |
|
|
* in nfs_wait_on_request() may deadlock with clear_inode().
|
235 |
|
|
*
|
236 |
|
|
* If igrab() fails here, then it is in any case safe to
|
237 |
|
|
* call nfs_wb_page(), since there will be no pending writes.
|
238 |
|
|
*/
|
239 |
|
|
if (igrab(inode) != 0)
|
240 |
|
|
inode_referenced = 1;
|
241 |
|
|
end_index = inode->i_size >> PAGE_CACHE_SHIFT;
|
242 |
|
|
|
243 |
|
|
/* Ensure we've flushed out any previous writes */
|
244 |
|
|
nfs_wb_page(inode,page);
|
245 |
|
|
|
246 |
|
|
/* easy case */
|
247 |
|
|
if (page->index < end_index)
|
248 |
|
|
goto do_it;
|
249 |
|
|
/* things got complicated... */
|
250 |
|
|
offset = inode->i_size & (PAGE_CACHE_SIZE-1);
|
251 |
|
|
|
252 |
|
|
/* OK, are we completely out? */
|
253 |
|
|
err = -EIO;
|
254 |
|
|
if (page->index >= end_index+1 || !offset)
|
255 |
|
|
goto out;
|
256 |
|
|
do_it:
|
257 |
|
|
lock_kernel();
|
258 |
|
|
if (NFS_SERVER(inode)->wsize >= PAGE_CACHE_SIZE && !IS_SYNC(inode) &&
|
259 |
|
|
inode_referenced) {
|
260 |
|
|
err = nfs_writepage_async(NULL, inode, page, 0, offset);
|
261 |
|
|
if (err >= 0)
|
262 |
|
|
err = 0;
|
263 |
|
|
} else {
|
264 |
|
|
err = nfs_writepage_sync(NULL, inode, page, 0, offset);
|
265 |
|
|
if (err == offset)
|
266 |
|
|
err = 0;
|
267 |
|
|
}
|
268 |
|
|
unlock_kernel();
|
269 |
|
|
out:
|
270 |
|
|
UnlockPage(page);
|
271 |
|
|
if (inode_referenced)
|
272 |
|
|
iput(inode);
|
273 |
|
|
return err;
|
274 |
|
|
}
|
275 |
|
|
|
276 |
|
|
/*
|
277 |
|
|
* Check whether the file range we want to write to is locked by
|
278 |
|
|
* us.
|
279 |
|
|
*/
|
280 |
|
|
static int
|
281 |
|
|
region_locked(struct inode *inode, struct nfs_page *req)
|
282 |
|
|
{
|
283 |
|
|
struct file_lock *fl;
|
284 |
|
|
loff_t rqstart, rqend;
|
285 |
|
|
|
286 |
|
|
/* Don't optimize writes if we don't use NLM */
|
287 |
|
|
if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
|
288 |
|
|
return 0;
|
289 |
|
|
|
290 |
|
|
rqstart = page_offset(req->wb_page) + req->wb_offset;
|
291 |
|
|
rqend = rqstart + req->wb_bytes;
|
292 |
|
|
for (fl = inode->i_flock; fl; fl = fl->fl_next) {
|
293 |
|
|
if (fl->fl_owner == current->files && (fl->fl_flags & FL_POSIX)
|
294 |
|
|
&& fl->fl_type == F_WRLCK
|
295 |
|
|
&& fl->fl_start <= rqstart && rqend <= fl->fl_end) {
|
296 |
|
|
return 1;
|
297 |
|
|
}
|
298 |
|
|
}
|
299 |
|
|
|
300 |
|
|
return 0;
|
301 |
|
|
}
|
302 |
|
|
|
303 |
|
|
/*
|
304 |
|
|
* Insert a write request into an inode
|
305 |
|
|
* Note: we sort the list in order to be able to optimize nfs_find_request()
|
306 |
|
|
* & co. for the 'write append' case. For 2.5 we may want to consider
|
307 |
|
|
* some form of hashing so as to perform well on random writes.
|
308 |
|
|
*/
|
309 |
|
|
static inline void
|
310 |
|
|
nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
|
311 |
|
|
{
|
312 |
|
|
struct list_head *pos, *head;
|
313 |
|
|
unsigned long pg_idx = page_index(req->wb_page);
|
314 |
|
|
|
315 |
|
|
if (!list_empty(&req->wb_hash))
|
316 |
|
|
return;
|
317 |
|
|
if (!NFS_WBACK_BUSY(req))
|
318 |
|
|
printk(KERN_ERR "NFS: unlocked request attempted hashed!\n");
|
319 |
|
|
head = &inode->u.nfs_i.writeback;
|
320 |
|
|
if (list_empty(head))
|
321 |
|
|
igrab(inode);
|
322 |
|
|
list_for_each_prev(pos, head) {
|
323 |
|
|
struct nfs_page *entry = nfs_inode_wb_entry(pos);
|
324 |
|
|
if (page_index(entry->wb_page) < pg_idx)
|
325 |
|
|
break;
|
326 |
|
|
}
|
327 |
|
|
inode->u.nfs_i.npages++;
|
328 |
|
|
list_add(&req->wb_hash, pos);
|
329 |
|
|
req->wb_count++;
|
330 |
|
|
}
|
331 |
|
|
|
332 |
|
|
/*
|
333 |
|
|
* Insert a write request into an inode
|
334 |
|
|
*/
|
335 |
|
|
static inline void
|
336 |
|
|
nfs_inode_remove_request(struct nfs_page *req)
|
337 |
|
|
{
|
338 |
|
|
struct inode *inode;
|
339 |
|
|
spin_lock(&nfs_wreq_lock);
|
340 |
|
|
if (list_empty(&req->wb_hash)) {
|
341 |
|
|
spin_unlock(&nfs_wreq_lock);
|
342 |
|
|
return;
|
343 |
|
|
}
|
344 |
|
|
if (!NFS_WBACK_BUSY(req))
|
345 |
|
|
printk(KERN_ERR "NFS: unlocked request attempted unhashed!\n");
|
346 |
|
|
inode = req->wb_inode;
|
347 |
|
|
list_del(&req->wb_hash);
|
348 |
|
|
INIT_LIST_HEAD(&req->wb_hash);
|
349 |
|
|
inode->u.nfs_i.npages--;
|
350 |
|
|
if ((inode->u.nfs_i.npages == 0) != list_empty(&inode->u.nfs_i.writeback))
|
351 |
|
|
printk(KERN_ERR "NFS: desynchronized value of nfs_i.npages.\n");
|
352 |
|
|
if (list_empty(&inode->u.nfs_i.writeback)) {
|
353 |
|
|
spin_unlock(&nfs_wreq_lock);
|
354 |
|
|
iput(inode);
|
355 |
|
|
} else
|
356 |
|
|
spin_unlock(&nfs_wreq_lock);
|
357 |
|
|
nfs_clear_request(req);
|
358 |
|
|
nfs_release_request(req);
|
359 |
|
|
}
|
360 |
|
|
|
361 |
|
|
/*
|
362 |
|
|
* Find a request
|
363 |
|
|
*/
|
364 |
|
|
static inline struct nfs_page *
|
365 |
|
|
_nfs_find_request(struct inode *inode, struct page *page)
|
366 |
|
|
{
|
367 |
|
|
struct list_head *head, *pos;
|
368 |
|
|
unsigned long pg_idx = page_index(page);
|
369 |
|
|
|
370 |
|
|
head = &inode->u.nfs_i.writeback;
|
371 |
|
|
list_for_each_prev(pos, head) {
|
372 |
|
|
struct nfs_page *req = nfs_inode_wb_entry(pos);
|
373 |
|
|
unsigned long found_idx = page_index(req->wb_page);
|
374 |
|
|
|
375 |
|
|
if (pg_idx < found_idx)
|
376 |
|
|
continue;
|
377 |
|
|
if (pg_idx != found_idx)
|
378 |
|
|
break;
|
379 |
|
|
req->wb_count++;
|
380 |
|
|
return req;
|
381 |
|
|
}
|
382 |
|
|
return NULL;
|
383 |
|
|
}
|
384 |
|
|
|
385 |
|
|
static struct nfs_page *
|
386 |
|
|
nfs_find_request(struct inode *inode, struct page *page)
|
387 |
|
|
{
|
388 |
|
|
struct nfs_page *req;
|
389 |
|
|
|
390 |
|
|
spin_lock(&nfs_wreq_lock);
|
391 |
|
|
req = _nfs_find_request(inode, page);
|
392 |
|
|
spin_unlock(&nfs_wreq_lock);
|
393 |
|
|
return req;
|
394 |
|
|
}
|
395 |
|
|
|
396 |
|
|
/*
|
397 |
|
|
* Add a request to the inode's dirty list.
|
398 |
|
|
*/
|
399 |
|
|
static inline void
|
400 |
|
|
nfs_mark_request_dirty(struct nfs_page *req)
|
401 |
|
|
{
|
402 |
|
|
struct inode *inode = req->wb_inode;
|
403 |
|
|
|
404 |
|
|
spin_lock(&nfs_wreq_lock);
|
405 |
|
|
nfs_list_add_request(req, &inode->u.nfs_i.dirty);
|
406 |
|
|
inode->u.nfs_i.ndirty++;
|
407 |
|
|
__nfs_del_lru(req);
|
408 |
|
|
__nfs_add_lru(&NFS_SERVER(inode)->lru_dirty, req);
|
409 |
|
|
spin_unlock(&nfs_wreq_lock);
|
410 |
|
|
mark_inode_dirty(inode);
|
411 |
|
|
}
|
412 |
|
|
|
413 |
|
|
/*
|
414 |
|
|
* Check if a request is dirty
|
415 |
|
|
*/
|
416 |
|
|
static inline int
|
417 |
|
|
nfs_dirty_request(struct nfs_page *req)
|
418 |
|
|
{
|
419 |
|
|
struct inode *inode = req->wb_inode;
|
420 |
|
|
return !list_empty(&req->wb_list) && req->wb_list_head == &inode->u.nfs_i.dirty;
|
421 |
|
|
}
|
422 |
|
|
|
423 |
|
|
#ifdef CONFIG_NFS_V3
|
424 |
|
|
/*
|
425 |
|
|
* Add a request to the inode's commit list.
|
426 |
|
|
*/
|
427 |
|
|
static inline void
|
428 |
|
|
nfs_mark_request_commit(struct nfs_page *req)
|
429 |
|
|
{
|
430 |
|
|
struct inode *inode = req->wb_inode;
|
431 |
|
|
|
432 |
|
|
spin_lock(&nfs_wreq_lock);
|
433 |
|
|
nfs_list_add_request(req, &inode->u.nfs_i.commit);
|
434 |
|
|
inode->u.nfs_i.ncommit++;
|
435 |
|
|
__nfs_del_lru(req);
|
436 |
|
|
__nfs_add_lru(&NFS_SERVER(inode)->lru_commit, req);
|
437 |
|
|
spin_unlock(&nfs_wreq_lock);
|
438 |
|
|
mark_inode_dirty(inode);
|
439 |
|
|
}
|
440 |
|
|
#endif
|
441 |
|
|
|
442 |
|
|
/*
|
443 |
|
|
* Wait for a request to complete.
|
444 |
|
|
*
|
445 |
|
|
* Interruptible by signals only if mounted with intr flag.
|
446 |
|
|
*/
|
447 |
|
|
static int
|
448 |
|
|
nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, unsigned int npages)
|
449 |
|
|
{
|
450 |
|
|
struct list_head *p, *head;
|
451 |
|
|
unsigned long idx_end;
|
452 |
|
|
unsigned int res = 0;
|
453 |
|
|
int error;
|
454 |
|
|
|
455 |
|
|
if (npages == 0)
|
456 |
|
|
idx_end = ~0;
|
457 |
|
|
else
|
458 |
|
|
idx_end = idx_start + npages - 1;
|
459 |
|
|
|
460 |
|
|
head = &inode->u.nfs_i.writeback;
|
461 |
|
|
restart:
|
462 |
|
|
spin_lock(&nfs_wreq_lock);
|
463 |
|
|
list_for_each_prev(p, head) {
|
464 |
|
|
unsigned long pg_idx;
|
465 |
|
|
struct nfs_page *req = nfs_inode_wb_entry(p);
|
466 |
|
|
|
467 |
|
|
pg_idx = page_index(req->wb_page);
|
468 |
|
|
if (pg_idx < idx_start)
|
469 |
|
|
break;
|
470 |
|
|
if (pg_idx > idx_end)
|
471 |
|
|
continue;
|
472 |
|
|
|
473 |
|
|
if (!NFS_WBACK_BUSY(req))
|
474 |
|
|
continue;
|
475 |
|
|
req->wb_count++;
|
476 |
|
|
spin_unlock(&nfs_wreq_lock);
|
477 |
|
|
error = nfs_wait_on_request(req);
|
478 |
|
|
nfs_release_request(req);
|
479 |
|
|
if (error < 0)
|
480 |
|
|
return error;
|
481 |
|
|
res++;
|
482 |
|
|
goto restart;
|
483 |
|
|
}
|
484 |
|
|
spin_unlock(&nfs_wreq_lock);
|
485 |
|
|
return res;
|
486 |
|
|
}
|
487 |
|
|
|
488 |
|
|
/**
|
489 |
|
|
* nfs_scan_lru_dirty_timeout - Scan LRU list for timed out dirty requests
|
490 |
|
|
* @server: NFS superblock data
|
491 |
|
|
* @dst: destination list
|
492 |
|
|
*
|
493 |
|
|
* Moves a maximum of 'wpages' requests from the NFS dirty page LRU list.
|
494 |
|
|
* The elements are checked to ensure that they form a contiguous set
|
495 |
|
|
* of pages, and that they originated from the same file.
|
496 |
|
|
*/
|
497 |
|
|
int
|
498 |
|
|
nfs_scan_lru_dirty_timeout(struct nfs_server *server, struct list_head *dst)
|
499 |
|
|
{
|
500 |
|
|
struct inode *inode;
|
501 |
|
|
int npages;
|
502 |
|
|
|
503 |
|
|
npages = nfs_scan_lru_timeout(&server->lru_dirty, dst, server->wpages);
|
504 |
|
|
if (npages) {
|
505 |
|
|
inode = nfs_list_entry(dst->next)->wb_inode;
|
506 |
|
|
inode->u.nfs_i.ndirty -= npages;
|
507 |
|
|
}
|
508 |
|
|
return npages;
|
509 |
|
|
}
|
510 |
|
|
|
511 |
|
|
/**
|
512 |
|
|
* nfs_scan_lru_dirty - Scan LRU list for dirty requests
|
513 |
|
|
* @server: NFS superblock data
|
514 |
|
|
* @dst: destination list
|
515 |
|
|
*
|
516 |
|
|
* Moves a maximum of 'wpages' requests from the NFS dirty page LRU list.
|
517 |
|
|
* The elements are checked to ensure that they form a contiguous set
|
518 |
|
|
* of pages, and that they originated from the same file.
|
519 |
|
|
*/
|
520 |
|
|
int
|
521 |
|
|
nfs_scan_lru_dirty(struct nfs_server *server, struct list_head *dst)
|
522 |
|
|
{
|
523 |
|
|
struct inode *inode;
|
524 |
|
|
int npages;
|
525 |
|
|
|
526 |
|
|
npages = nfs_scan_lru(&server->lru_dirty, dst, server->wpages);
|
527 |
|
|
if (npages) {
|
528 |
|
|
inode = nfs_list_entry(dst->next)->wb_inode;
|
529 |
|
|
inode->u.nfs_i.ndirty -= npages;
|
530 |
|
|
}
|
531 |
|
|
return npages;
|
532 |
|
|
}
|
533 |
|
|
|
534 |
|
|
/*
|
535 |
|
|
* nfs_scan_dirty - Scan an inode for dirty requests
|
536 |
|
|
* @inode: NFS inode to scan
|
537 |
|
|
* @dst: destination list
|
538 |
|
|
* @idx_start: lower bound of page->index to scan.
|
539 |
|
|
* @npages: idx_start + npages sets the upper bound to scan.
|
540 |
|
|
*
|
541 |
|
|
* Moves requests from the inode's dirty page list.
|
542 |
|
|
* The requests are *not* checked to ensure that they form a contiguous set.
|
543 |
|
|
*/
|
544 |
|
|
static int
|
545 |
|
|
nfs_scan_dirty(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages)
|
546 |
|
|
{
|
547 |
|
|
int res;
|
548 |
|
|
res = nfs_scan_list(&inode->u.nfs_i.dirty, dst, idx_start, npages);
|
549 |
|
|
inode->u.nfs_i.ndirty -= res;
|
550 |
|
|
if ((inode->u.nfs_i.ndirty == 0) != list_empty(&inode->u.nfs_i.dirty))
|
551 |
|
|
printk(KERN_ERR "NFS: desynchronized value of nfs_i.ndirty.\n");
|
552 |
|
|
return res;
|
553 |
|
|
}
|
554 |
|
|
|
555 |
|
|
#ifdef CONFIG_NFS_V3
|
556 |
|
|
/**
|
557 |
|
|
* nfs_scan_lru_commit_timeout - Scan LRU list for timed out commit requests
|
558 |
|
|
* @server: NFS superblock data
|
559 |
|
|
* @dst: destination list
|
560 |
|
|
*
|
561 |
|
|
* Finds the first a timed out request in the NFS commit LRU list and moves it
|
562 |
|
|
* to the list dst. If such an element is found, we move all other commit
|
563 |
|
|
* requests that apply to the same inode.
|
564 |
|
|
* The assumption is that doing everything in a single commit-to-disk is
|
565 |
|
|
* the cheaper alternative.
|
566 |
|
|
*/
|
567 |
|
|
int
|
568 |
|
|
nfs_scan_lru_commit_timeout(struct nfs_server *server, struct list_head *dst)
|
569 |
|
|
{
|
570 |
|
|
struct inode *inode;
|
571 |
|
|
int npages;
|
572 |
|
|
|
573 |
|
|
npages = nfs_scan_lru_timeout(&server->lru_commit, dst, 1);
|
574 |
|
|
if (npages) {
|
575 |
|
|
inode = nfs_list_entry(dst->next)->wb_inode;
|
576 |
|
|
npages += nfs_scan_list(&inode->u.nfs_i.commit, dst, 0, 0);
|
577 |
|
|
inode->u.nfs_i.ncommit -= npages;
|
578 |
|
|
}
|
579 |
|
|
return npages;
|
580 |
|
|
}
|
581 |
|
|
|
582 |
|
|
|
583 |
|
|
/**
|
584 |
|
|
* nfs_scan_lru_commit_timeout - Scan LRU list for timed out commit requests
|
585 |
|
|
* @server: NFS superblock data
|
586 |
|
|
* @dst: destination list
|
587 |
|
|
*
|
588 |
|
|
* Finds the first request in the NFS commit LRU list and moves it
|
589 |
|
|
* to the list dst. If such an element is found, we move all other commit
|
590 |
|
|
* requests that apply to the same inode.
|
591 |
|
|
* The assumption is that doing everything in a single commit-to-disk is
|
592 |
|
|
* the cheaper alternative.
|
593 |
|
|
*/
|
594 |
|
|
int
|
595 |
|
|
nfs_scan_lru_commit(struct nfs_server *server, struct list_head *dst)
|
596 |
|
|
{
|
597 |
|
|
struct inode *inode;
|
598 |
|
|
int npages;
|
599 |
|
|
|
600 |
|
|
npages = nfs_scan_lru(&server->lru_commit, dst, 1);
|
601 |
|
|
if (npages) {
|
602 |
|
|
inode = nfs_list_entry(dst->next)->wb_inode;
|
603 |
|
|
npages += nfs_scan_list(&inode->u.nfs_i.commit, dst, 0, 0);
|
604 |
|
|
inode->u.nfs_i.ncommit -= npages;
|
605 |
|
|
}
|
606 |
|
|
return npages;
|
607 |
|
|
}
|
608 |
|
|
|
609 |
|
|
/*
|
610 |
|
|
* nfs_scan_commit - Scan an inode for commit requests
|
611 |
|
|
* @inode: NFS inode to scan
|
612 |
|
|
* @dst: destination list
|
613 |
|
|
* @idx_start: lower bound of page->index to scan.
|
614 |
|
|
* @npages: idx_start + npages sets the upper bound to scan.
|
615 |
|
|
*
|
616 |
|
|
* Moves requests from the inode's 'commit' request list.
|
617 |
|
|
* The requests are *not* checked to ensure that they form a contiguous set.
|
618 |
|
|
*/
|
619 |
|
|
static int
|
620 |
|
|
nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages)
|
621 |
|
|
{
|
622 |
|
|
int res;
|
623 |
|
|
res = nfs_scan_list(&inode->u.nfs_i.commit, dst, idx_start, npages);
|
624 |
|
|
inode->u.nfs_i.ncommit -= res;
|
625 |
|
|
if ((inode->u.nfs_i.ncommit == 0) != list_empty(&inode->u.nfs_i.commit))
|
626 |
|
|
printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n");
|
627 |
|
|
return res;
|
628 |
|
|
}
|
629 |
|
|
#endif
|
630 |
|
|
|
631 |
|
|
|
632 |
|
|
/*
|
633 |
|
|
* Try to update any existing write request, or create one if there is none.
|
634 |
|
|
* In order to match, the request's credentials must match those of
|
635 |
|
|
* the calling process.
|
636 |
|
|
*
|
637 |
|
|
* Note: Should always be called with the Page Lock held!
|
638 |
|
|
*/
|
639 |
|
|
static struct nfs_page *
|
640 |
|
|
nfs_update_request(struct file* file, struct inode *inode, struct page *page,
|
641 |
|
|
unsigned int offset, unsigned int bytes)
|
642 |
|
|
{
|
643 |
|
|
struct nfs_page *req, *new = NULL;
|
644 |
|
|
unsigned long rqend, end;
|
645 |
|
|
|
646 |
|
|
end = offset + bytes;
|
647 |
|
|
|
648 |
|
|
for (;;) {
|
649 |
|
|
/* Loop over all inode entries and see if we find
|
650 |
|
|
* A request for the page we wish to update
|
651 |
|
|
*/
|
652 |
|
|
spin_lock(&nfs_wreq_lock);
|
653 |
|
|
req = _nfs_find_request(inode, page);
|
654 |
|
|
if (req) {
|
655 |
|
|
if (!nfs_lock_request_dontget(req)) {
|
656 |
|
|
int error;
|
657 |
|
|
spin_unlock(&nfs_wreq_lock);
|
658 |
|
|
error = nfs_wait_on_request(req);
|
659 |
|
|
nfs_release_request(req);
|
660 |
|
|
if (error < 0)
|
661 |
|
|
return ERR_PTR(error);
|
662 |
|
|
continue;
|
663 |
|
|
}
|
664 |
|
|
spin_unlock(&nfs_wreq_lock);
|
665 |
|
|
if (new)
|
666 |
|
|
nfs_release_request(new);
|
667 |
|
|
break;
|
668 |
|
|
}
|
669 |
|
|
|
670 |
|
|
if (new) {
|
671 |
|
|
nfs_lock_request_dontget(new);
|
672 |
|
|
nfs_inode_add_request(inode, new);
|
673 |
|
|
spin_unlock(&nfs_wreq_lock);
|
674 |
|
|
nfs_mark_request_dirty(new);
|
675 |
|
|
return new;
|
676 |
|
|
}
|
677 |
|
|
spin_unlock(&nfs_wreq_lock);
|
678 |
|
|
|
679 |
|
|
new = nfs_create_request(nfs_file_cred(file), inode, page, offset, bytes);
|
680 |
|
|
if (IS_ERR(new))
|
681 |
|
|
return new;
|
682 |
|
|
if (file) {
|
683 |
|
|
new->wb_file = file;
|
684 |
|
|
get_file(file);
|
685 |
|
|
}
|
686 |
|
|
/* If the region is locked, adjust the timeout */
|
687 |
|
|
if (region_locked(inode, new))
|
688 |
|
|
new->wb_timeout = jiffies + NFS_WRITEBACK_LOCKDELAY;
|
689 |
|
|
else
|
690 |
|
|
new->wb_timeout = jiffies + NFS_WRITEBACK_DELAY;
|
691 |
|
|
}
|
692 |
|
|
|
693 |
|
|
/* We have a request for our page.
|
694 |
|
|
* If the creds don't match, or the
|
695 |
|
|
* page addresses don't match,
|
696 |
|
|
* tell the caller to wait on the conflicting
|
697 |
|
|
* request.
|
698 |
|
|
*/
|
699 |
|
|
rqend = req->wb_offset + req->wb_bytes;
|
700 |
|
|
if (req->wb_file != file
|
701 |
|
|
|| req->wb_page != page
|
702 |
|
|
|| !nfs_dirty_request(req)
|
703 |
|
|
|| offset > rqend || end < req->wb_offset) {
|
704 |
|
|
nfs_unlock_request(req);
|
705 |
|
|
return ERR_PTR(-EBUSY);
|
706 |
|
|
}
|
707 |
|
|
|
708 |
|
|
/* Okay, the request matches. Update the region */
|
709 |
|
|
if (offset < req->wb_offset) {
|
710 |
|
|
req->wb_offset = offset;
|
711 |
|
|
req->wb_bytes = rqend - req->wb_offset;
|
712 |
|
|
}
|
713 |
|
|
|
714 |
|
|
if (end > rqend)
|
715 |
|
|
req->wb_bytes = end - req->wb_offset;
|
716 |
|
|
|
717 |
|
|
return req;
|
718 |
|
|
}
|
719 |
|
|
|
720 |
|
|
/*
|
721 |
|
|
* This is the strategy routine for NFS.
|
722 |
|
|
* It is called by nfs_updatepage whenever the user wrote up to the end
|
723 |
|
|
* of a page.
|
724 |
|
|
*
|
725 |
|
|
* We always try to submit a set of requests in parallel so that the
|
726 |
|
|
* server's write code can gather writes. This is mainly for the benefit
|
727 |
|
|
* of NFSv2.
|
728 |
|
|
*
|
729 |
|
|
* We never submit more requests than we think the remote can handle.
|
730 |
|
|
* For UDP sockets, we make sure we don't exceed the congestion window;
|
731 |
|
|
* for TCP, we limit the number of requests to 8.
|
732 |
|
|
*
|
733 |
|
|
* NFS_STRATEGY_PAGES gives the minimum number of requests for NFSv2 that
|
734 |
|
|
* should be sent out in one go. This is for the benefit of NFSv2 servers
|
735 |
|
|
* that perform write gathering.
|
736 |
|
|
*
|
737 |
|
|
* FIXME: Different servers may have different sweet spots.
|
738 |
|
|
* Record the average congestion window in server struct?
|
739 |
|
|
*/
|
740 |
|
|
#define NFS_STRATEGY_PAGES 8
|
741 |
|
|
static void
|
742 |
|
|
nfs_strategy(struct inode *inode)
|
743 |
|
|
{
|
744 |
|
|
unsigned int dirty, wpages;
|
745 |
|
|
|
746 |
|
|
dirty = inode->u.nfs_i.ndirty;
|
747 |
|
|
wpages = NFS_SERVER(inode)->wpages;
|
748 |
|
|
#ifdef CONFIG_NFS_V3
|
749 |
|
|
if (NFS_PROTO(inode)->version == 2) {
|
750 |
|
|
if (dirty >= NFS_STRATEGY_PAGES * wpages)
|
751 |
|
|
nfs_flush_file(inode, 0, 0, 0);
|
752 |
|
|
} else if (dirty >= wpages)
|
753 |
|
|
nfs_flush_file(inode, 0, 0, 0);
|
754 |
|
|
#else
|
755 |
|
|
if (dirty >= NFS_STRATEGY_PAGES * wpages)
|
756 |
|
|
nfs_flush_file(inode, 0, 0, 0);
|
757 |
|
|
#endif
|
758 |
|
|
}
|
759 |
|
|
|
760 |
|
|
int
|
761 |
|
|
nfs_flush_incompatible(struct file *file, struct page *page)
|
762 |
|
|
{
|
763 |
|
|
struct rpc_cred *cred = nfs_file_cred(file);
|
764 |
|
|
struct inode *inode = page->mapping->host;
|
765 |
|
|
struct nfs_page *req;
|
766 |
|
|
int status = 0;
|
767 |
|
|
/*
|
768 |
|
|
* Look for a request corresponding to this page. If there
|
769 |
|
|
* is one, and it belongs to another file, we flush it out
|
770 |
|
|
* before we try to copy anything into the page. Do this
|
771 |
|
|
* due to the lack of an ACCESS-type call in NFSv2.
|
772 |
|
|
* Also do the same if we find a request from an existing
|
773 |
|
|
* dropped page.
|
774 |
|
|
*/
|
775 |
|
|
req = nfs_find_request(inode,page);
|
776 |
|
|
if (req) {
|
777 |
|
|
if (req->wb_file != file || req->wb_cred != cred || req->wb_page != page)
|
778 |
|
|
status = nfs_wb_page(inode, page);
|
779 |
|
|
nfs_release_request(req);
|
780 |
|
|
}
|
781 |
|
|
return (status < 0) ? status : 0;
|
782 |
|
|
}
|
783 |
|
|
|
784 |
|
|
/*
|
785 |
|
|
* Update and possibly write a cached page of an NFS file.
|
786 |
|
|
*
|
787 |
|
|
* XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
|
788 |
|
|
* things with a page scheduled for an RPC call (e.g. invalidate it).
|
789 |
|
|
*/
|
790 |
|
|
int
|
791 |
|
|
nfs_updatepage(struct file *file, struct page *page, unsigned int offset, unsigned int count)
|
792 |
|
|
{
|
793 |
|
|
struct dentry *dentry = file->f_dentry;
|
794 |
|
|
struct inode *inode = page->mapping->host;
|
795 |
|
|
struct nfs_page *req;
|
796 |
|
|
loff_t end;
|
797 |
|
|
int status = 0;
|
798 |
|
|
|
799 |
|
|
dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n",
|
800 |
|
|
dentry->d_parent->d_name.name, dentry->d_name.name,
|
801 |
|
|
count, (long long)(page_offset(page) +offset));
|
802 |
|
|
|
803 |
|
|
/*
|
804 |
|
|
* If wsize is smaller than page size, update and write
|
805 |
|
|
* page synchronously.
|
806 |
|
|
*/
|
807 |
|
|
if (NFS_SERVER(inode)->wsize < PAGE_CACHE_SIZE || IS_SYNC(inode)) {
|
808 |
|
|
status = nfs_writepage_sync(file, inode, page, offset, count);
|
809 |
|
|
if (status > 0) {
|
810 |
|
|
if (offset == 0 && status == PAGE_CACHE_SIZE)
|
811 |
|
|
SetPageUptodate(page);
|
812 |
|
|
return 0;
|
813 |
|
|
}
|
814 |
|
|
return status;
|
815 |
|
|
}
|
816 |
|
|
|
817 |
|
|
/*
|
818 |
|
|
* Try to find an NFS request corresponding to this page
|
819 |
|
|
* and update it.
|
820 |
|
|
* If the existing request cannot be updated, we must flush
|
821 |
|
|
* it out now.
|
822 |
|
|
*/
|
823 |
|
|
do {
|
824 |
|
|
req = nfs_update_request(file, inode, page, offset, count);
|
825 |
|
|
status = (IS_ERR(req)) ? PTR_ERR(req) : 0;
|
826 |
|
|
if (status != -EBUSY)
|
827 |
|
|
break;
|
828 |
|
|
/* Request could not be updated. Flush it out and try again */
|
829 |
|
|
status = nfs_wb_page(inode, page);
|
830 |
|
|
} while (status >= 0);
|
831 |
|
|
if (status < 0)
|
832 |
|
|
goto done;
|
833 |
|
|
|
834 |
|
|
status = 0;
|
835 |
|
|
end = ((loff_t)page->index<<PAGE_CACHE_SHIFT) + (loff_t)(offset + count);
|
836 |
|
|
if (inode->i_size < end)
|
837 |
|
|
inode->i_size = end;
|
838 |
|
|
|
839 |
|
|
/* If we wrote past the end of the page.
|
840 |
|
|
* Call the strategy routine so it can send out a bunch
|
841 |
|
|
* of requests.
|
842 |
|
|
*/
|
843 |
|
|
if (req->wb_offset == 0 && req->wb_bytes == PAGE_CACHE_SIZE) {
|
844 |
|
|
SetPageUptodate(page);
|
845 |
|
|
nfs_unlock_request(req);
|
846 |
|
|
nfs_strategy(inode);
|
847 |
|
|
} else
|
848 |
|
|
nfs_unlock_request(req);
|
849 |
|
|
done:
|
850 |
|
|
dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n",
|
851 |
|
|
status, (long long)inode->i_size);
|
852 |
|
|
if (status < 0)
|
853 |
|
|
ClearPageUptodate(page);
|
854 |
|
|
return status;
|
855 |
|
|
}
|
856 |
|
|
|
857 |
|
|
/*
|
858 |
|
|
* Set up the argument/result storage required for the RPC call.
|
859 |
|
|
*/
|
860 |
|
|
static void
|
861 |
|
|
nfs_write_rpcsetup(struct list_head *head, struct nfs_write_data *data)
|
862 |
|
|
{
|
863 |
|
|
struct nfs_page *req;
|
864 |
|
|
struct page **pages;
|
865 |
|
|
unsigned int count;
|
866 |
|
|
|
867 |
|
|
/* Set up the RPC argument and reply structs
|
868 |
|
|
* NB: take care not to mess about with data->commit et al. */
|
869 |
|
|
|
870 |
|
|
pages = data->args.pages;
|
871 |
|
|
count = 0;
|
872 |
|
|
while (!list_empty(head)) {
|
873 |
|
|
struct nfs_page *req = nfs_list_entry(head->next);
|
874 |
|
|
nfs_list_remove_request(req);
|
875 |
|
|
nfs_list_add_request(req, &data->pages);
|
876 |
|
|
*pages++ = req->wb_page;
|
877 |
|
|
count += req->wb_bytes;
|
878 |
|
|
}
|
879 |
|
|
req = nfs_list_entry(data->pages.next);
|
880 |
|
|
data->inode = req->wb_inode;
|
881 |
|
|
data->cred = req->wb_cred;
|
882 |
|
|
data->args.fh = NFS_FH(req->wb_inode);
|
883 |
|
|
data->args.offset = page_offset(req->wb_page) + req->wb_offset;
|
884 |
|
|
data->args.pgbase = req->wb_offset;
|
885 |
|
|
data->args.count = count;
|
886 |
|
|
data->res.fattr = &data->fattr;
|
887 |
|
|
data->res.count = count;
|
888 |
|
|
data->res.verf = &data->verf;
|
889 |
|
|
}
|
890 |
|
|
|
891 |
|
|
|
892 |
|
|
/*
|
893 |
|
|
* Create an RPC task for the given write request and kick it.
|
894 |
|
|
* The page must have been locked by the caller.
|
895 |
|
|
*
|
896 |
|
|
* It may happen that the page we're passed is not marked dirty.
|
897 |
|
|
* This is the case if nfs_updatepage detects a conflicting request
|
898 |
|
|
* that has been written but not committed.
|
899 |
|
|
*/
|
900 |
|
|
static int
|
901 |
|
|
nfs_flush_one(struct list_head *head, struct inode *inode, int how)
|
902 |
|
|
{
|
903 |
|
|
struct rpc_clnt *clnt = NFS_CLIENT(inode);
|
904 |
|
|
struct nfs_write_data *data;
|
905 |
|
|
struct rpc_task *task;
|
906 |
|
|
struct rpc_message msg;
|
907 |
|
|
int flags,
|
908 |
|
|
nfsvers = NFS_PROTO(inode)->version,
|
909 |
|
|
async = !(how & FLUSH_SYNC),
|
910 |
|
|
stable = (how & FLUSH_STABLE);
|
911 |
|
|
sigset_t oldset;
|
912 |
|
|
|
913 |
|
|
|
914 |
|
|
data = nfs_writedata_alloc();
|
915 |
|
|
if (!data)
|
916 |
|
|
goto out_bad;
|
917 |
|
|
task = &data->task;
|
918 |
|
|
|
919 |
|
|
/* Set the initial flags for the task. */
|
920 |
|
|
flags = (async) ? RPC_TASK_ASYNC : 0;
|
921 |
|
|
|
922 |
|
|
/* Set up the argument struct */
|
923 |
|
|
nfs_write_rpcsetup(head, data);
|
924 |
|
|
if (nfsvers < 3)
|
925 |
|
|
data->args.stable = NFS_FILE_SYNC;
|
926 |
|
|
else if (stable) {
|
927 |
|
|
if (!inode->u.nfs_i.ncommit)
|
928 |
|
|
data->args.stable = NFS_FILE_SYNC;
|
929 |
|
|
else
|
930 |
|
|
data->args.stable = NFS_DATA_SYNC;
|
931 |
|
|
} else
|
932 |
|
|
data->args.stable = NFS_UNSTABLE;
|
933 |
|
|
|
934 |
|
|
/* Finalize the task. */
|
935 |
|
|
rpc_init_task(task, clnt, nfs_writeback_done, flags);
|
936 |
|
|
task->tk_calldata = data;
|
937 |
|
|
/* Release requests */
|
938 |
|
|
task->tk_release = nfs_writedata_release;
|
939 |
|
|
|
940 |
|
|
#ifdef CONFIG_NFS_V3
|
941 |
|
|
msg.rpc_proc = (nfsvers == 3) ? NFS3PROC_WRITE : NFSPROC_WRITE;
|
942 |
|
|
#else
|
943 |
|
|
msg.rpc_proc = NFSPROC_WRITE;
|
944 |
|
|
#endif
|
945 |
|
|
msg.rpc_argp = &data->args;
|
946 |
|
|
msg.rpc_resp = &data->res;
|
947 |
|
|
msg.rpc_cred = data->cred;
|
948 |
|
|
|
949 |
|
|
dprintk("NFS: %4d initiated write call (req %x/%Ld count %u)\n",
|
950 |
|
|
task->tk_pid,
|
951 |
|
|
inode->i_dev,
|
952 |
|
|
(long long)NFS_FILEID(inode),
|
953 |
|
|
data->args.count);
|
954 |
|
|
|
955 |
|
|
rpc_clnt_sigmask(clnt, &oldset);
|
956 |
|
|
rpc_call_setup(task, &msg, 0);
|
957 |
|
|
lock_kernel();
|
958 |
|
|
rpc_execute(task);
|
959 |
|
|
unlock_kernel();
|
960 |
|
|
rpc_clnt_sigunmask(clnt, &oldset);
|
961 |
|
|
return 0;
|
962 |
|
|
out_bad:
|
963 |
|
|
while (!list_empty(head)) {
|
964 |
|
|
struct nfs_page *req = nfs_list_entry(head->next);
|
965 |
|
|
nfs_list_remove_request(req);
|
966 |
|
|
nfs_mark_request_dirty(req);
|
967 |
|
|
nfs_unlock_request(req);
|
968 |
|
|
}
|
969 |
|
|
return -ENOMEM;
|
970 |
|
|
}
|
971 |
|
|
|
972 |
|
|
int
|
973 |
|
|
nfs_flush_list(struct list_head *head, int wpages, int how)
|
974 |
|
|
{
|
975 |
|
|
LIST_HEAD(one_request);
|
976 |
|
|
struct nfs_page *req;
|
977 |
|
|
int error = 0;
|
978 |
|
|
unsigned int pages = 0;
|
979 |
|
|
|
980 |
|
|
while (!list_empty(head)) {
|
981 |
|
|
pages += nfs_coalesce_requests(head, &one_request, wpages);
|
982 |
|
|
req = nfs_list_entry(one_request.next);
|
983 |
|
|
error = nfs_flush_one(&one_request, req->wb_inode, how);
|
984 |
|
|
if (error < 0)
|
985 |
|
|
break;
|
986 |
|
|
}
|
987 |
|
|
if (error >= 0)
|
988 |
|
|
return pages;
|
989 |
|
|
|
990 |
|
|
while (!list_empty(head)) {
|
991 |
|
|
req = nfs_list_entry(head->next);
|
992 |
|
|
nfs_list_remove_request(req);
|
993 |
|
|
nfs_mark_request_dirty(req);
|
994 |
|
|
nfs_unlock_request(req);
|
995 |
|
|
}
|
996 |
|
|
return error;
|
997 |
|
|
}
|
998 |
|
|
|
999 |
|
|
|
1000 |
|
|
/*
|
1001 |
|
|
* This function is called when the WRITE call is complete.
|
1002 |
|
|
*/
|
1003 |
|
|
static void
|
1004 |
|
|
nfs_writeback_done(struct rpc_task *task)
|
1005 |
|
|
{
|
1006 |
|
|
struct nfs_write_data *data = (struct nfs_write_data *) task->tk_calldata;
|
1007 |
|
|
struct nfs_writeargs *argp = &data->args;
|
1008 |
|
|
struct nfs_writeres *resp = &data->res;
|
1009 |
|
|
struct inode *inode = data->inode;
|
1010 |
|
|
struct nfs_page *req;
|
1011 |
|
|
struct page *page;
|
1012 |
|
|
|
1013 |
|
|
dprintk("NFS: %4d nfs_writeback_done (status %d)\n",
|
1014 |
|
|
task->tk_pid, task->tk_status);
|
1015 |
|
|
|
1016 |
|
|
if (nfs_async_handle_jukebox(task))
|
1017 |
|
|
return;
|
1018 |
|
|
|
1019 |
|
|
/* We can't handle that yet but we check for it nevertheless */
|
1020 |
|
|
if (resp->count < argp->count && task->tk_status >= 0) {
|
1021 |
|
|
static unsigned long complain;
|
1022 |
|
|
if (time_before(complain, jiffies)) {
|
1023 |
|
|
printk(KERN_WARNING
|
1024 |
|
|
"NFS: Server wrote less than requested.\n");
|
1025 |
|
|
complain = jiffies + 300 * HZ;
|
1026 |
|
|
}
|
1027 |
|
|
/* Can't do anything about it right now except throw
|
1028 |
|
|
* an error. */
|
1029 |
|
|
task->tk_status = -EIO;
|
1030 |
|
|
}
|
1031 |
|
|
#ifdef CONFIG_NFS_V3
|
1032 |
|
|
if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
|
1033 |
|
|
/* We tried a write call, but the server did not
|
1034 |
|
|
* commit data to stable storage even though we
|
1035 |
|
|
* requested it.
|
1036 |
|
|
* Note: There is a known bug in Tru64 < 5.0 in which
|
1037 |
|
|
* the server reports NFS_DATA_SYNC, but performs
|
1038 |
|
|
* NFS_FILE_SYNC. We therefore implement this checking
|
1039 |
|
|
* as a dprintk() in order to avoid filling syslog.
|
1040 |
|
|
*/
|
1041 |
|
|
static unsigned long complain;
|
1042 |
|
|
|
1043 |
|
|
if (time_before(complain, jiffies)) {
|
1044 |
|
|
dprintk("NFS: faulty NFSv3 server %s:"
|
1045 |
|
|
" (committed = %d) != (stable = %d)\n",
|
1046 |
|
|
NFS_SERVER(inode)->hostname,
|
1047 |
|
|
resp->verf->committed, argp->stable);
|
1048 |
|
|
complain = jiffies + 300 * HZ;
|
1049 |
|
|
}
|
1050 |
|
|
}
|
1051 |
|
|
#endif
|
1052 |
|
|
|
1053 |
|
|
/*
|
1054 |
|
|
* Update attributes as result of writeback.
|
1055 |
|
|
* FIXME: There is an inherent race with invalidate_inode_pages and
|
1056 |
|
|
* writebacks since the page->count is kept > 1 for as long
|
1057 |
|
|
* as the page has a write request pending.
|
1058 |
|
|
*/
|
1059 |
|
|
nfs_write_attributes(inode, resp->fattr);
|
1060 |
|
|
while (!list_empty(&data->pages)) {
|
1061 |
|
|
req = nfs_list_entry(data->pages.next);
|
1062 |
|
|
nfs_list_remove_request(req);
|
1063 |
|
|
page = req->wb_page;
|
1064 |
|
|
|
1065 |
|
|
dprintk("NFS: write (%x/%Ld %d@%Ld)",
|
1066 |
|
|
req->wb_inode->i_dev,
|
1067 |
|
|
(long long)NFS_FILEID(req->wb_inode),
|
1068 |
|
|
req->wb_bytes,
|
1069 |
|
|
(long long)(page_offset(page) + req->wb_offset));
|
1070 |
|
|
|
1071 |
|
|
if (task->tk_status < 0) {
|
1072 |
|
|
ClearPageUptodate(page);
|
1073 |
|
|
SetPageError(page);
|
1074 |
|
|
if (req->wb_file)
|
1075 |
|
|
req->wb_file->f_error = task->tk_status;
|
1076 |
|
|
nfs_inode_remove_request(req);
|
1077 |
|
|
dprintk(", error = %d\n", task->tk_status);
|
1078 |
|
|
goto next;
|
1079 |
|
|
}
|
1080 |
|
|
|
1081 |
|
|
#ifdef CONFIG_NFS_V3
|
1082 |
|
|
if (argp->stable != NFS_UNSTABLE || resp->verf->committed == NFS_FILE_SYNC) {
|
1083 |
|
|
nfs_inode_remove_request(req);
|
1084 |
|
|
dprintk(" OK\n");
|
1085 |
|
|
goto next;
|
1086 |
|
|
}
|
1087 |
|
|
memcpy(&req->wb_verf, resp->verf, sizeof(req->wb_verf));
|
1088 |
|
|
req->wb_timeout = jiffies + NFS_COMMIT_DELAY;
|
1089 |
|
|
nfs_mark_request_commit(req);
|
1090 |
|
|
dprintk(" marked for commit\n");
|
1091 |
|
|
#else
|
1092 |
|
|
nfs_inode_remove_request(req);
|
1093 |
|
|
#endif
|
1094 |
|
|
next:
|
1095 |
|
|
nfs_unlock_request(req);
|
1096 |
|
|
}
|
1097 |
|
|
}
|
1098 |
|
|
|
1099 |
|
|
|
1100 |
|
|
#ifdef CONFIG_NFS_V3
|
1101 |
|
|
/*
|
1102 |
|
|
* Set up the argument/result storage required for the RPC call.
|
1103 |
|
|
*/
|
1104 |
|
|
static void
|
1105 |
|
|
nfs_commit_rpcsetup(struct list_head *head, struct nfs_write_data *data)
|
1106 |
|
|
{
|
1107 |
|
|
struct nfs_page *first, *last;
|
1108 |
|
|
struct inode *inode;
|
1109 |
|
|
loff_t start, end, len;
|
1110 |
|
|
|
1111 |
|
|
/* Set up the RPC argument and reply structs
|
1112 |
|
|
* NB: take care not to mess about with data->commit et al. */
|
1113 |
|
|
|
1114 |
|
|
list_splice(head, &data->pages);
|
1115 |
|
|
INIT_LIST_HEAD(head);
|
1116 |
|
|
first = nfs_list_entry(data->pages.next);
|
1117 |
|
|
last = nfs_list_entry(data->pages.prev);
|
1118 |
|
|
inode = first->wb_inode;
|
1119 |
|
|
|
1120 |
|
|
/*
|
1121 |
|
|
* Determine the offset range of requests in the COMMIT call.
|
1122 |
|
|
* We rely on the fact that data->pages is an ordered list...
|
1123 |
|
|
*/
|
1124 |
|
|
start = page_offset(first->wb_page) + first->wb_offset;
|
1125 |
|
|
end = page_offset(last->wb_page) + (last->wb_offset + last->wb_bytes);
|
1126 |
|
|
len = end - start;
|
1127 |
|
|
/* If 'len' is not a 32-bit quantity, pass '0' in the COMMIT call */
|
1128 |
|
|
if (end >= inode->i_size || len < 0 || len > (~((u32)0) >> 1))
|
1129 |
|
|
len = 0;
|
1130 |
|
|
|
1131 |
|
|
data->inode = inode;
|
1132 |
|
|
data->cred = first->wb_cred;
|
1133 |
|
|
data->args.fh = NFS_FH(inode);
|
1134 |
|
|
data->args.offset = start;
|
1135 |
|
|
data->res.count = data->args.count = (u32)len;
|
1136 |
|
|
data->res.fattr = &data->fattr;
|
1137 |
|
|
data->res.verf = &data->verf;
|
1138 |
|
|
}
|
1139 |
|
|
|
1140 |
|
|
/*
|
1141 |
|
|
* Commit dirty pages
|
1142 |
|
|
*/
|
1143 |
|
|
int
|
1144 |
|
|
nfs_commit_list(struct list_head *head, int how)
|
1145 |
|
|
{
|
1146 |
|
|
struct rpc_message msg;
|
1147 |
|
|
struct rpc_clnt *clnt;
|
1148 |
|
|
struct nfs_write_data *data;
|
1149 |
|
|
struct rpc_task *task;
|
1150 |
|
|
struct nfs_page *req;
|
1151 |
|
|
int flags,
|
1152 |
|
|
async = !(how & FLUSH_SYNC);
|
1153 |
|
|
sigset_t oldset;
|
1154 |
|
|
|
1155 |
|
|
data = nfs_writedata_alloc();
|
1156 |
|
|
|
1157 |
|
|
if (!data)
|
1158 |
|
|
goto out_bad;
|
1159 |
|
|
task = &data->task;
|
1160 |
|
|
|
1161 |
|
|
flags = (async) ? RPC_TASK_ASYNC : 0;
|
1162 |
|
|
|
1163 |
|
|
/* Set up the argument struct */
|
1164 |
|
|
nfs_commit_rpcsetup(head, data);
|
1165 |
|
|
req = nfs_list_entry(data->pages.next);
|
1166 |
|
|
clnt = NFS_CLIENT(req->wb_inode);
|
1167 |
|
|
|
1168 |
|
|
rpc_init_task(task, clnt, nfs_commit_done, flags);
|
1169 |
|
|
task->tk_calldata = data;
|
1170 |
|
|
/* Release requests */
|
1171 |
|
|
task->tk_release = nfs_writedata_release;
|
1172 |
|
|
|
1173 |
|
|
msg.rpc_proc = NFS3PROC_COMMIT;
|
1174 |
|
|
msg.rpc_argp = &data->args;
|
1175 |
|
|
msg.rpc_resp = &data->res;
|
1176 |
|
|
msg.rpc_cred = data->cred;
|
1177 |
|
|
|
1178 |
|
|
dprintk("NFS: %4d initiated commit call\n", task->tk_pid);
|
1179 |
|
|
rpc_clnt_sigmask(clnt, &oldset);
|
1180 |
|
|
rpc_call_setup(task, &msg, 0);
|
1181 |
|
|
lock_kernel();
|
1182 |
|
|
rpc_execute(task);
|
1183 |
|
|
unlock_kernel();
|
1184 |
|
|
rpc_clnt_sigunmask(clnt, &oldset);
|
1185 |
|
|
return 0;
|
1186 |
|
|
out_bad:
|
1187 |
|
|
while (!list_empty(head)) {
|
1188 |
|
|
req = nfs_list_entry(head->next);
|
1189 |
|
|
nfs_list_remove_request(req);
|
1190 |
|
|
nfs_mark_request_commit(req);
|
1191 |
|
|
nfs_unlock_request(req);
|
1192 |
|
|
}
|
1193 |
|
|
return -ENOMEM;
|
1194 |
|
|
}
|
1195 |
|
|
|
1196 |
|
|
/*
|
1197 |
|
|
* COMMIT call returned
|
1198 |
|
|
*/
|
1199 |
|
|
static void
|
1200 |
|
|
nfs_commit_done(struct rpc_task *task)
|
1201 |
|
|
{
|
1202 |
|
|
struct nfs_write_data *data = (struct nfs_write_data *)task->tk_calldata;
|
1203 |
|
|
struct nfs_writeres *resp = &data->res;
|
1204 |
|
|
struct nfs_page *req;
|
1205 |
|
|
struct inode *inode = data->inode;
|
1206 |
|
|
|
1207 |
|
|
dprintk("NFS: %4d nfs_commit_done (status %d)\n",
|
1208 |
|
|
task->tk_pid, task->tk_status);
|
1209 |
|
|
|
1210 |
|
|
if (nfs_async_handle_jukebox(task))
|
1211 |
|
|
return;
|
1212 |
|
|
|
1213 |
|
|
nfs_write_attributes(inode, resp->fattr);
|
1214 |
|
|
while (!list_empty(&data->pages)) {
|
1215 |
|
|
req = nfs_list_entry(data->pages.next);
|
1216 |
|
|
nfs_list_remove_request(req);
|
1217 |
|
|
|
1218 |
|
|
dprintk("NFS: commit (%x/%Ld %d@%Ld)",
|
1219 |
|
|
req->wb_inode->i_dev,
|
1220 |
|
|
(long long)NFS_FILEID(req->wb_inode),
|
1221 |
|
|
req->wb_bytes,
|
1222 |
|
|
(long long)(page_offset(req->wb_page) + req->wb_offset));
|
1223 |
|
|
if (task->tk_status < 0) {
|
1224 |
|
|
if (req->wb_file)
|
1225 |
|
|
req->wb_file->f_error = task->tk_status;
|
1226 |
|
|
nfs_inode_remove_request(req);
|
1227 |
|
|
dprintk(", error = %d\n", task->tk_status);
|
1228 |
|
|
goto next;
|
1229 |
|
|
}
|
1230 |
|
|
|
1231 |
|
|
/* Okay, COMMIT succeeded, apparently. Check the verifier
|
1232 |
|
|
* returned by the server against all stored verfs. */
|
1233 |
|
|
if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
|
1234 |
|
|
/* We have a match */
|
1235 |
|
|
nfs_inode_remove_request(req);
|
1236 |
|
|
dprintk(" OK\n");
|
1237 |
|
|
goto next;
|
1238 |
|
|
}
|
1239 |
|
|
/* We have a mismatch. Write the page again */
|
1240 |
|
|
dprintk(" mismatch\n");
|
1241 |
|
|
nfs_mark_request_dirty(req);
|
1242 |
|
|
next:
|
1243 |
|
|
nfs_unlock_request(req);
|
1244 |
|
|
}
|
1245 |
|
|
}
|
1246 |
|
|
#endif
|
1247 |
|
|
|
1248 |
|
|
int nfs_flush_file(struct inode *inode, unsigned long idx_start,
|
1249 |
|
|
unsigned int npages, int how)
|
1250 |
|
|
{
|
1251 |
|
|
LIST_HEAD(head);
|
1252 |
|
|
int res,
|
1253 |
|
|
error = 0;
|
1254 |
|
|
|
1255 |
|
|
spin_lock(&nfs_wreq_lock);
|
1256 |
|
|
res = nfs_scan_dirty(inode, &head, idx_start, npages);
|
1257 |
|
|
spin_unlock(&nfs_wreq_lock);
|
1258 |
|
|
if (res)
|
1259 |
|
|
error = nfs_flush_list(&head, NFS_SERVER(inode)->wpages, how);
|
1260 |
|
|
if (error < 0)
|
1261 |
|
|
return error;
|
1262 |
|
|
return res;
|
1263 |
|
|
}
|
1264 |
|
|
|
1265 |
|
|
#ifdef CONFIG_NFS_V3
|
1266 |
|
|
int nfs_commit_file(struct inode *inode, int how)
|
1267 |
|
|
{
|
1268 |
|
|
LIST_HEAD(head);
|
1269 |
|
|
int res,
|
1270 |
|
|
error = 0;
|
1271 |
|
|
|
1272 |
|
|
spin_lock(&nfs_wreq_lock);
|
1273 |
|
|
res = nfs_scan_commit(inode, &head, 0, 0);
|
1274 |
|
|
spin_unlock(&nfs_wreq_lock);
|
1275 |
|
|
if (res)
|
1276 |
|
|
error = nfs_commit_list(&head, how);
|
1277 |
|
|
if (error < 0)
|
1278 |
|
|
return error;
|
1279 |
|
|
return res;
|
1280 |
|
|
}
|
1281 |
|
|
#endif
|
1282 |
|
|
|
1283 |
|
|
int nfs_sync_file(struct inode *inode, unsigned long idx_start,
|
1284 |
|
|
unsigned int npages, int how)
|
1285 |
|
|
{
|
1286 |
|
|
int error,
|
1287 |
|
|
wait;
|
1288 |
|
|
|
1289 |
|
|
wait = how & FLUSH_WAIT;
|
1290 |
|
|
how &= ~FLUSH_WAIT;
|
1291 |
|
|
|
1292 |
|
|
do {
|
1293 |
|
|
error = 0;
|
1294 |
|
|
if (wait)
|
1295 |
|
|
error = nfs_wait_on_requests(inode, idx_start, npages);
|
1296 |
|
|
if (error == 0)
|
1297 |
|
|
error = nfs_flush_file(inode, idx_start, npages, how);
|
1298 |
|
|
#ifdef CONFIG_NFS_V3
|
1299 |
|
|
if (error == 0)
|
1300 |
|
|
error = nfs_commit_file(inode, how);
|
1301 |
|
|
#endif
|
1302 |
|
|
} while (error > 0);
|
1303 |
|
|
return error;
|
1304 |
|
|
}
|
1305 |
|
|
|
1306 |
|
|
int nfs_init_writepagecache(void)
|
1307 |
|
|
{
|
1308 |
|
|
nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
|
1309 |
|
|
sizeof(struct nfs_write_data),
|
1310 |
|
|
0, SLAB_HWCACHE_ALIGN,
|
1311 |
|
|
NULL, NULL);
|
1312 |
|
|
if (nfs_wdata_cachep == NULL)
|
1313 |
|
|
return -ENOMEM;
|
1314 |
|
|
|
1315 |
|
|
return 0;
|
1316 |
|
|
}
|
1317 |
|
|
|
1318 |
|
|
void nfs_destroy_writepagecache(void)
|
1319 |
|
|
{
|
1320 |
|
|
if (kmem_cache_destroy(nfs_wdata_cachep))
|
1321 |
|
|
printk(KERN_INFO "nfs_write_data: not all structures were freed\n");
|
1322 |
|
|
}
|
1323 |
|
|
|