]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/nfs/write.c
NFSv4: Decode getdevicelist should use nfs4_verifier
[mirror_ubuntu-bionic-kernel.git] / fs / nfs / write.c
CommitLineData
1da177e4
LT
1/*
2 * linux/fs/nfs/write.c
3 *
7c85d900 4 * Write file data over NFS.
1da177e4
LT
5 *
6 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
7 */
8
1da177e4
LT
9#include <linux/types.h>
10#include <linux/slab.h>
11#include <linux/mm.h>
12#include <linux/pagemap.h>
13#include <linux/file.h>
1da177e4 14#include <linux/writeback.h>
89a09141 15#include <linux/swap.h>
074cc1de 16#include <linux/migrate.h>
1da177e4
LT
17
18#include <linux/sunrpc/clnt.h>
19#include <linux/nfs_fs.h>
20#include <linux/nfs_mount.h>
21#include <linux/nfs_page.h>
3fcfab16 22#include <linux/backing-dev.h>
afeacc8c 23#include <linux/export.h>
3fcfab16 24
1da177e4 25#include <asm/uaccess.h>
1da177e4
LT
26
27#include "delegation.h"
49a70f27 28#include "internal.h"
91d5b470 29#include "iostat.h"
def6ed7e 30#include "nfs4_fs.h"
074cc1de 31#include "fscache.h"
94ad1c80 32#include "pnfs.h"
1da177e4
LT
33
34#define NFSDBG_FACILITY NFSDBG_PAGECACHE
35
36#define MIN_POOL_WRITE (32)
37#define MIN_POOL_COMMIT (4)
38
39/*
40 * Local function declarations
41 */
f8512ad0 42static void nfs_redirty_request(struct nfs_page *req);
6c75dc0d 43static const struct rpc_call_ops nfs_write_common_ops;
788e7a89 44static const struct rpc_call_ops nfs_commit_ops;
061ae2ed 45static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
f453a54a 46static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
1da177e4 47
e18b890b 48static struct kmem_cache *nfs_wdata_cachep;
3feb2d49 49static mempool_t *nfs_wdata_mempool;
0b7c0153 50static struct kmem_cache *nfs_cdata_cachep;
1da177e4
LT
51static mempool_t *nfs_commit_mempool;
52
0b7c0153 53struct nfs_commit_data *nfs_commitdata_alloc(void)
1da177e4 54{
0b7c0153 55 struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
40859d7e 56
1da177e4
LT
57 if (p) {
58 memset(p, 0, sizeof(*p));
59 INIT_LIST_HEAD(&p->pages);
60 }
61 return p;
62}
e0c2b380 63EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
1da177e4 64
0b7c0153 65void nfs_commit_free(struct nfs_commit_data *p)
1da177e4
LT
66{
67 mempool_free(p, nfs_commit_mempool);
68}
e0c2b380 69EXPORT_SYMBOL_GPL(nfs_commit_free);
1da177e4 70
6c75dc0d 71struct nfs_write_header *nfs_writehdr_alloc(void)
3feb2d49 72{
cd841605 73 struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
3feb2d49
TM
74
75 if (p) {
cd841605 76 struct nfs_pgio_header *hdr = &p->header;
cd841605 77
3feb2d49 78 memset(p, 0, sizeof(*p));
cd841605 79 INIT_LIST_HEAD(&hdr->pages);
6c75dc0d
FI
80 INIT_LIST_HEAD(&hdr->rpc_list);
81 spin_lock_init(&hdr->lock);
82 atomic_set(&hdr->refcnt, 0);
9bce008b 83 hdr->verf = &p->verf;
3feb2d49
TM
84 }
85 return p;
86}
87
1763da12
FI
88static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr,
89 unsigned int pagecount)
6c75dc0d
FI
90{
91 struct nfs_write_data *data, *prealloc;
92
93 prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data;
94 if (prealloc->header == NULL)
95 data = prealloc;
96 else
97 data = kzalloc(sizeof(*data), GFP_KERNEL);
98 if (!data)
99 goto out;
100
101 if (nfs_pgarray_set(&data->pages, pagecount)) {
102 data->header = hdr;
103 atomic_inc(&hdr->refcnt);
104 } else {
105 if (data != prealloc)
106 kfree(data);
107 data = NULL;
108 }
109out:
110 return data;
111}
112
cd841605 113void nfs_writehdr_free(struct nfs_pgio_header *hdr)
3feb2d49 114{
cd841605
FI
115 struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header);
116 mempool_free(whdr, nfs_wdata_mempool);
3feb2d49
TM
117}
118
dce81290 119void nfs_writedata_release(struct nfs_write_data *wdata)
1da177e4 120{
6c75dc0d
FI
121 struct nfs_pgio_header *hdr = wdata->header;
122 struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header);
123
383ba719 124 put_nfs_open_context(wdata->args.context);
30dd374f
FI
125 if (wdata->pages.pagevec != wdata->pages.page_array)
126 kfree(wdata->pages.pagevec);
6c75dc0d
FI
127 if (wdata != &write_header->rpc_data)
128 kfree(wdata);
129 else
130 wdata->header = NULL;
131 if (atomic_dec_and_test(&hdr->refcnt))
061ae2ed 132 hdr->completion_ops->completion(hdr);
1da177e4
LT
133}
134
7b159fc1
TM
135static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
136{
137 ctx->error = error;
138 smp_wmb();
139 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
140}
141
277459d2
TM
142static struct nfs_page *nfs_page_find_request_locked(struct page *page)
143{
144 struct nfs_page *req = NULL;
145
146 if (PagePrivate(page)) {
147 req = (struct nfs_page *)page_private(page);
148 if (req != NULL)
c03b4024 149 kref_get(&req->wb_kref);
277459d2
TM
150 }
151 return req;
152}
153
154static struct nfs_page *nfs_page_find_request(struct page *page)
155{
587142f8 156 struct inode *inode = page->mapping->host;
277459d2 157 struct nfs_page *req = NULL;
277459d2 158
587142f8 159 spin_lock(&inode->i_lock);
277459d2 160 req = nfs_page_find_request_locked(page);
587142f8 161 spin_unlock(&inode->i_lock);
277459d2
TM
162 return req;
163}
164
1da177e4
LT
165/* Adjust the file length if we're writing beyond the end */
166static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
167{
168 struct inode *inode = page->mapping->host;
a3d01454
TM
169 loff_t end, i_size;
170 pgoff_t end_index;
1da177e4 171
a3d01454
TM
172 spin_lock(&inode->i_lock);
173 i_size = i_size_read(inode);
174 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
1da177e4 175 if (i_size > 0 && page->index < end_index)
a3d01454 176 goto out;
1da177e4
LT
177 end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
178 if (i_size >= end)
a3d01454 179 goto out;
1da177e4 180 i_size_write(inode, end);
a3d01454
TM
181 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
182out:
183 spin_unlock(&inode->i_lock);
1da177e4
LT
184}
185
a301b777
TM
186/* A writeback failed: mark the page as bad, and invalidate the page cache */
187static void nfs_set_pageerror(struct page *page)
188{
189 SetPageError(page);
190 nfs_zap_mapping(page->mapping->host, page->mapping);
191}
192
1da177e4
LT
193/* We can set the PG_uptodate flag if we see that a write request
194 * covers the full page.
195 */
196static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
197{
1da177e4
LT
198 if (PageUptodate(page))
199 return;
200 if (base != 0)
201 return;
49a70f27 202 if (count != nfs_page_length(page))
1da177e4 203 return;
49a70f27 204 SetPageUptodate(page);
1da177e4
LT
205}
206
1da177e4
LT
207static int wb_priority(struct writeback_control *wbc)
208{
209 if (wbc->for_reclaim)
c63c7b05 210 return FLUSH_HIGHPRI | FLUSH_STABLE;
b17621fe 211 if (wbc->for_kupdate || wbc->for_background)
b31268ac
TM
212 return FLUSH_LOWPRI | FLUSH_COND_STABLE;
213 return FLUSH_COND_STABLE;
1da177e4
LT
214}
215
89a09141
PZ
216/*
217 * NFS congestion control
218 */
219
220int nfs_congestion_kb;
221
222#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
223#define NFS_CONGESTION_OFF_THRESH \
224 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
225
5a6d41b3 226static int nfs_set_page_writeback(struct page *page)
89a09141 227{
5a6d41b3
TM
228 int ret = test_set_page_writeback(page);
229
230 if (!ret) {
89a09141
PZ
231 struct inode *inode = page->mapping->host;
232 struct nfs_server *nfss = NFS_SERVER(inode);
233
277866a0 234 if (atomic_long_inc_return(&nfss->writeback) >
8aa7e847
JA
235 NFS_CONGESTION_ON_THRESH) {
236 set_bdi_congested(&nfss->backing_dev_info,
237 BLK_RW_ASYNC);
238 }
89a09141 239 }
5a6d41b3 240 return ret;
89a09141
PZ
241}
242
243static void nfs_end_page_writeback(struct page *page)
244{
245 struct inode *inode = page->mapping->host;
246 struct nfs_server *nfss = NFS_SERVER(inode);
247
248 end_page_writeback(page);
c4dc4bee 249 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
8aa7e847 250 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
89a09141
PZ
251}
252
cfb506e1 253static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
e261f51f 254{
587142f8 255 struct inode *inode = page->mapping->host;
e261f51f 256 struct nfs_page *req;
e261f51f
TM
257 int ret;
258
587142f8 259 spin_lock(&inode->i_lock);
074cc1de 260 for (;;) {
e261f51f 261 req = nfs_page_find_request_locked(page);
074cc1de
TM
262 if (req == NULL)
263 break;
7ad84aa9 264 if (nfs_lock_request(req))
e261f51f
TM
265 break;
266 /* Note: If we hold the page lock, as is the case in nfs_writepage,
7ad84aa9 267 * then the call to nfs_lock_request() will always
e261f51f
TM
268 * succeed provided that someone hasn't already marked the
269 * request as dirty (in which case we don't care).
270 */
587142f8 271 spin_unlock(&inode->i_lock);
cfb506e1
TM
272 if (!nonblock)
273 ret = nfs_wait_on_request(req);
274 else
275 ret = -EAGAIN;
e261f51f
TM
276 nfs_release_request(req);
277 if (ret != 0)
074cc1de 278 return ERR_PTR(ret);
587142f8 279 spin_lock(&inode->i_lock);
e261f51f 280 }
587142f8 281 spin_unlock(&inode->i_lock);
074cc1de
TM
282 return req;
283}
284
285/*
286 * Find an associated nfs write request, and prepare to flush it out
287 * May return an error if the user signalled nfs_wait_on_request().
288 */
289static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
cfb506e1 290 struct page *page, bool nonblock)
074cc1de
TM
291{
292 struct nfs_page *req;
293 int ret = 0;
294
cfb506e1 295 req = nfs_find_and_lock_request(page, nonblock);
074cc1de
TM
296 if (!req)
297 goto out;
298 ret = PTR_ERR(req);
299 if (IS_ERR(req))
300 goto out;
301
302 ret = nfs_set_page_writeback(page);
303 BUG_ON(ret != 0);
304 BUG_ON(test_bit(PG_CLEAN, &req->wb_flags));
305
f8512ad0
FI
306 if (!nfs_pageio_add_request(pgio, req)) {
307 nfs_redirty_request(req);
074cc1de 308 ret = pgio->pg_error;
f8512ad0 309 }
074cc1de
TM
310out:
311 return ret;
e261f51f
TM
312}
313
f758c885 314static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
1da177e4 315{
1da177e4 316 struct inode *inode = page->mapping->host;
cfb506e1 317 int ret;
1da177e4 318
91d5b470
CL
319 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
320 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
321
7fe7f848 322 nfs_pageio_cond_complete(pgio, page->index);
1b430bee 323 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
cfb506e1
TM
324 if (ret == -EAGAIN) {
325 redirty_page_for_writepage(wbc, page);
326 ret = 0;
327 }
328 return ret;
f758c885 329}
7fe7f848 330
f758c885
TM
331/*
332 * Write an mmapped page to the server.
333 */
334static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
335{
336 struct nfs_pageio_descriptor pgio;
337 int err;
49a70f27 338
061ae2ed
FI
339 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc),
340 &nfs_async_write_completion_ops);
f758c885
TM
341 err = nfs_do_writepage(page, wbc, &pgio);
342 nfs_pageio_complete(&pgio);
343 if (err < 0)
344 return err;
345 if (pgio.pg_error < 0)
346 return pgio.pg_error;
347 return 0;
4d770ccf
TM
348}
349
350int nfs_writepage(struct page *page, struct writeback_control *wbc)
351{
f758c885 352 int ret;
4d770ccf 353
f758c885 354 ret = nfs_writepage_locked(page, wbc);
1da177e4 355 unlock_page(page);
f758c885
TM
356 return ret;
357}
358
359static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
360{
361 int ret;
362
363 ret = nfs_do_writepage(page, wbc, data);
364 unlock_page(page);
365 return ret;
1da177e4
LT
366}
367
1da177e4
LT
368int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
369{
1da177e4 370 struct inode *inode = mapping->host;
72cb77f4 371 unsigned long *bitlock = &NFS_I(inode)->flags;
c63c7b05 372 struct nfs_pageio_descriptor pgio;
1da177e4
LT
373 int err;
374
72cb77f4
TM
375 /* Stop dirtying of new pages while we sync */
376 err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
377 nfs_wait_bit_killable, TASK_KILLABLE);
378 if (err)
379 goto out_err;
380
91d5b470
CL
381 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
382
061ae2ed
FI
383 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc),
384 &nfs_async_write_completion_ops);
f758c885 385 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
c63c7b05 386 nfs_pageio_complete(&pgio);
72cb77f4
TM
387
388 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
389 smp_mb__after_clear_bit();
390 wake_up_bit(bitlock, NFS_INO_FLUSHING);
391
f758c885 392 if (err < 0)
72cb77f4
TM
393 goto out_err;
394 err = pgio.pg_error;
395 if (err < 0)
396 goto out_err;
c63c7b05 397 return 0;
72cb77f4
TM
398out_err:
399 return err;
1da177e4
LT
400}
401
402/*
403 * Insert a write request into an inode
404 */
d6d6dc7c 405static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
1da177e4
LT
406{
407 struct nfs_inode *nfsi = NFS_I(inode);
e7d39069
TM
408
409 /* Lock the request! */
7ad84aa9 410 nfs_lock_request(req);
e7d39069
TM
411
412 spin_lock(&inode->i_lock);
4d65c520 413 if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE))
a9a4a87a 414 inode->i_version++;
2df485a7 415 set_bit(PG_MAPPED, &req->wb_flags);
deb7d638 416 SetPagePrivate(req->wb_page);
277459d2 417 set_page_private(req->wb_page, (unsigned long)req);
1da177e4 418 nfsi->npages++;
c03b4024 419 kref_get(&req->wb_kref);
e7d39069 420 spin_unlock(&inode->i_lock);
1da177e4
LT
421}
422
423/*
89a09141 424 * Remove a write request from an inode
1da177e4
LT
425 */
426static void nfs_inode_remove_request(struct nfs_page *req)
427{
3d4ff43d 428 struct inode *inode = req->wb_context->dentry->d_inode;
1da177e4
LT
429 struct nfs_inode *nfsi = NFS_I(inode);
430
431 BUG_ON (!NFS_WBACK_BUSY(req));
432
587142f8 433 spin_lock(&inode->i_lock);
277459d2 434 set_page_private(req->wb_page, 0);
deb7d638 435 ClearPagePrivate(req->wb_page);
2df485a7 436 clear_bit(PG_MAPPED, &req->wb_flags);
1da177e4 437 nfsi->npages--;
4d65c520 438 spin_unlock(&inode->i_lock);
1da177e4
LT
439 nfs_release_request(req);
440}
441
61822ab5 442static void
6d884e8f 443nfs_mark_request_dirty(struct nfs_page *req)
61822ab5 444{
61822ab5
TM
445 __set_page_dirty_nobuffers(req->wb_page);
446}
447
1da177e4 448#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
8dd37758
TM
449/**
450 * nfs_request_add_commit_list - add request to a commit list
451 * @req: pointer to a struct nfs_page
ea2cf228
FI
452 * @dst: commit list head
453 * @cinfo: holds list lock and accounting info
8dd37758 454 *
ea2cf228 455 * This sets the PG_CLEAN bit, updates the cinfo count of
8dd37758
TM
456 * number of outstanding requests requiring a commit as well as
457 * the MM page stats.
458 *
ea2cf228 459 * The caller must _not_ hold the cinfo->lock, but must be
8dd37758 460 * holding the nfs_page lock.
1da177e4 461 */
8dd37758 462void
ea2cf228
FI
463nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
464 struct nfs_commit_info *cinfo)
1da177e4 465{
e468bae9 466 set_bit(PG_CLEAN, &(req)->wb_flags);
ea2cf228
FI
467 spin_lock(cinfo->lock);
468 nfs_list_add_request(req, dst);
469 cinfo->mds->ncommit++;
470 spin_unlock(cinfo->lock);
56f9cd68
FI
471 if (!cinfo->dreq) {
472 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
473 inc_bdi_stat(req->wb_page->mapping->backing_dev_info,
474 BDI_RECLAIMABLE);
475 __mark_inode_dirty(req->wb_context->dentry->d_inode,
476 I_DIRTY_DATASYNC);
477 }
1da177e4 478}
8dd37758
TM
479EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
480
481/**
482 * nfs_request_remove_commit_list - Remove request from a commit list
483 * @req: pointer to a nfs_page
ea2cf228 484 * @cinfo: holds list lock and accounting info
8dd37758 485 *
ea2cf228 486 * This clears the PG_CLEAN bit, and updates the cinfo's count of
8dd37758
TM
487 * number of outstanding requests requiring a commit
488 * It does not update the MM page stats.
489 *
ea2cf228 490 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
8dd37758
TM
491 */
492void
ea2cf228
FI
493nfs_request_remove_commit_list(struct nfs_page *req,
494 struct nfs_commit_info *cinfo)
8dd37758 495{
8dd37758
TM
496 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
497 return;
498 nfs_list_remove_request(req);
ea2cf228 499 cinfo->mds->ncommit--;
8dd37758
TM
500}
501EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
502
ea2cf228
FI
503static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
504 struct inode *inode)
505{
506 cinfo->lock = &inode->i_lock;
507 cinfo->mds = &NFS_I(inode)->commit_info;
508 cinfo->ds = pnfs_get_ds_info(inode);
b359f9d0 509 cinfo->dreq = NULL;
f453a54a 510 cinfo->completion_ops = &nfs_commit_completion_ops;
ea2cf228
FI
511}
512
513void nfs_init_cinfo(struct nfs_commit_info *cinfo,
514 struct inode *inode,
515 struct nfs_direct_req *dreq)
516{
1763da12
FI
517 if (dreq)
518 nfs_init_cinfo_from_dreq(cinfo, dreq);
519 else
520 nfs_init_cinfo_from_inode(cinfo, inode);
ea2cf228
FI
521}
522EXPORT_SYMBOL_GPL(nfs_init_cinfo);
8dd37758
TM
523
524/*
525 * Add a request to the inode's commit list.
526 */
1763da12 527void
ea2cf228
FI
528nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
529 struct nfs_commit_info *cinfo)
8dd37758 530{
ea2cf228 531 if (pnfs_mark_request_commit(req, lseg, cinfo))
8dd37758 532 return;
ea2cf228 533 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
8dd37758 534}
8e821cad 535
d6d6dc7c
FI
536static void
537nfs_clear_page_commit(struct page *page)
538{
539 dec_zone_page_state(page, NR_UNSTABLE_NFS);
540 dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
541}
542
8dd37758 543static void
e468bae9
TM
544nfs_clear_request_commit(struct nfs_page *req)
545{
8dd37758
TM
546 if (test_bit(PG_CLEAN, &req->wb_flags)) {
547 struct inode *inode = req->wb_context->dentry->d_inode;
ea2cf228 548 struct nfs_commit_info cinfo;
e468bae9 549
ea2cf228
FI
550 nfs_init_cinfo_from_inode(&cinfo, inode);
551 if (!pnfs_clear_request_commit(req, &cinfo)) {
552 spin_lock(cinfo.lock);
553 nfs_request_remove_commit_list(req, &cinfo);
554 spin_unlock(cinfo.lock);
8dd37758 555 }
d6d6dc7c 556 nfs_clear_page_commit(req->wb_page);
e468bae9 557 }
e468bae9
TM
558}
559
8e821cad
TM
560static inline
561int nfs_write_need_commit(struct nfs_write_data *data)
562{
465d5243 563 if (data->verf.committed == NFS_DATA_SYNC)
cd841605
FI
564 return data->header->lseg == NULL;
565 return data->verf.committed != NFS_FILE_SYNC;
8e821cad
TM
566}
567
8e821cad 568#else
68cd6fa4
BS
569static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
570 struct inode *inode)
571{
572}
573
574void nfs_init_cinfo(struct nfs_commit_info *cinfo,
575 struct inode *inode,
576 struct nfs_direct_req *dreq)
577{
578}
579
1763da12 580void
ea2cf228
FI
581nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
582 struct nfs_commit_info *cinfo)
8e821cad
TM
583{
584}
585
8dd37758 586static void
e468bae9
TM
587nfs_clear_request_commit(struct nfs_page *req)
588{
e468bae9
TM
589}
590
8e821cad
TM
591static inline
592int nfs_write_need_commit(struct nfs_write_data *data)
593{
594 return 0;
595}
596
6c75dc0d
FI
597#endif
598
061ae2ed 599static void nfs_write_completion(struct nfs_pgio_header *hdr)
8e821cad 600{
ea2cf228 601 struct nfs_commit_info cinfo;
6c75dc0d
FI
602 unsigned long bytes = 0;
603
604 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
605 goto out;
ea2cf228 606 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
6c75dc0d
FI
607 while (!list_empty(&hdr->pages)) {
608 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
6c75dc0d
FI
609
610 bytes += req->wb_bytes;
611 nfs_list_remove_request(req);
612 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
613 (hdr->good_bytes < bytes)) {
d1182b33 614 nfs_set_pageerror(req->wb_page);
6c75dc0d
FI
615 nfs_context_set_write_error(req->wb_context, hdr->error);
616 goto remove_req;
617 }
618 if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) {
619 nfs_mark_request_dirty(req);
620 goto next;
621 }
622 if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) {
9bce008b 623 memcpy(&req->wb_verf, hdr->verf, sizeof(req->wb_verf));
ea2cf228 624 nfs_mark_request_commit(req, hdr->lseg, &cinfo);
6c75dc0d
FI
625 goto next;
626 }
627remove_req:
628 nfs_inode_remove_request(req);
629next:
1d1afcbc 630 nfs_unlock_request(req);
d1182b33 631 nfs_end_page_writeback(req->wb_page);
3aff4ebb 632 nfs_release_request(req);
6c75dc0d
FI
633 }
634out:
635 hdr->release(hdr);
8e821cad 636}
1da177e4 637
47c62564 638#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
ea2cf228
FI
639static unsigned long
640nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
fb8a1f11 641{
ea2cf228 642 return cinfo->mds->ncommit;
d6d6dc7c
FI
643}
644
ea2cf228 645/* cinfo->lock held by caller */
1763da12 646int
ea2cf228
FI
647nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
648 struct nfs_commit_info *cinfo, int max)
d6d6dc7c
FI
649{
650 struct nfs_page *req, *tmp;
651 int ret = 0;
652
653 list_for_each_entry_safe(req, tmp, src, wb_list) {
8dd37758
TM
654 if (!nfs_lock_request(req))
655 continue;
7ad84aa9 656 kref_get(&req->wb_kref);
ea2cf228 657 if (cond_resched_lock(cinfo->lock))
3b3be88d 658 list_safe_reset_next(req, tmp, wb_list);
ea2cf228 659 nfs_request_remove_commit_list(req, cinfo);
8dd37758
TM
660 nfs_list_add_request(req, dst);
661 ret++;
1763da12 662 if ((ret == max) && !cinfo->dreq)
8dd37758 663 break;
d6d6dc7c
FI
664 }
665 return ret;
fb8a1f11
TM
666}
667
1da177e4
LT
668/*
669 * nfs_scan_commit - Scan an inode for commit requests
670 * @inode: NFS inode to scan
ea2cf228
FI
671 * @dst: mds destination list
672 * @cinfo: mds and ds lists of reqs ready to commit
1da177e4
LT
673 *
674 * Moves requests from the inode's 'commit' request list.
675 * The requests are *not* checked to ensure that they form a contiguous set.
676 */
1763da12 677int
ea2cf228
FI
678nfs_scan_commit(struct inode *inode, struct list_head *dst,
679 struct nfs_commit_info *cinfo)
1da177e4 680{
d6d6dc7c 681 int ret = 0;
fb8a1f11 682
ea2cf228
FI
683 spin_lock(cinfo->lock);
684 if (cinfo->mds->ncommit > 0) {
8dd37758 685 const int max = INT_MAX;
d6d6dc7c 686
ea2cf228
FI
687 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
688 cinfo, max);
689 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
d6d6dc7c 690 }
ea2cf228 691 spin_unlock(cinfo->lock);
ff778d02 692 return ret;
1da177e4 693}
d6d6dc7c 694
c42de9dd 695#else
ea2cf228 696static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
fb8a1f11
TM
697{
698 return 0;
699}
700
1763da12
FI
701int nfs_scan_commit(struct inode *inode, struct list_head *dst,
702 struct nfs_commit_info *cinfo)
c42de9dd
TM
703{
704 return 0;
705}
1da177e4
LT
706#endif
707
1da177e4 708/*
e7d39069
TM
709 * Search for an existing write request, and attempt to update
710 * it to reflect a new dirty region on a given page.
1da177e4 711 *
e7d39069
TM
712 * If the attempt fails, then the existing request is flushed out
713 * to disk.
1da177e4 714 */
e7d39069
TM
715static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
716 struct page *page,
717 unsigned int offset,
718 unsigned int bytes)
1da177e4 719{
e7d39069
TM
720 struct nfs_page *req;
721 unsigned int rqend;
722 unsigned int end;
723 int error;
724
725 if (!PagePrivate(page))
726 return NULL;
1da177e4
LT
727
728 end = offset + bytes;
e7d39069 729 spin_lock(&inode->i_lock);
1da177e4 730
1da177e4 731 for (;;) {
277459d2 732 req = nfs_page_find_request_locked(page);
e7d39069
TM
733 if (req == NULL)
734 goto out_unlock;
735
736 rqend = req->wb_offset + req->wb_bytes;
737 /*
738 * Tell the caller to flush out the request if
739 * the offsets are non-contiguous.
740 * Note: nfs_flush_incompatible() will already
741 * have flushed out requests having wrong owners.
742 */
e468bae9 743 if (offset > rqend
e7d39069
TM
744 || end < req->wb_offset)
745 goto out_flushme;
746
7ad84aa9 747 if (nfs_lock_request(req))
1da177e4 748 break;
1da177e4 749
e7d39069 750 /* The request is locked, so wait and then retry */
587142f8 751 spin_unlock(&inode->i_lock);
e7d39069
TM
752 error = nfs_wait_on_request(req);
753 nfs_release_request(req);
754 if (error != 0)
755 goto out_err;
756 spin_lock(&inode->i_lock);
1da177e4
LT
757 }
758
759 /* Okay, the request matches. Update the region */
760 if (offset < req->wb_offset) {
761 req->wb_offset = offset;
762 req->wb_pgbase = offset;
1da177e4 763 }
1da177e4
LT
764 if (end > rqend)
765 req->wb_bytes = end - req->wb_offset;
e7d39069
TM
766 else
767 req->wb_bytes = rqend - req->wb_offset;
768out_unlock:
769 spin_unlock(&inode->i_lock);
ca138f36
FI
770 if (req)
771 nfs_clear_request_commit(req);
e7d39069
TM
772 return req;
773out_flushme:
774 spin_unlock(&inode->i_lock);
775 nfs_release_request(req);
776 error = nfs_wb_page(inode, page);
777out_err:
778 return ERR_PTR(error);
779}
780
781/*
782 * Try to update an existing write request, or create one if there is none.
783 *
784 * Note: Should always be called with the Page Lock held to prevent races
785 * if we have to add a new request. Also assumes that the caller has
786 * already called nfs_flush_incompatible() if necessary.
787 */
788static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
789 struct page *page, unsigned int offset, unsigned int bytes)
790{
791 struct inode *inode = page->mapping->host;
792 struct nfs_page *req;
1da177e4 793
e7d39069
TM
794 req = nfs_try_to_update_request(inode, page, offset, bytes);
795 if (req != NULL)
796 goto out;
797 req = nfs_create_request(ctx, inode, page, offset, bytes);
798 if (IS_ERR(req))
799 goto out;
d6d6dc7c 800 nfs_inode_add_request(inode, req);
efc91ed0 801out:
61e930a9 802 return req;
1da177e4
LT
803}
804
e7d39069
TM
805static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
806 unsigned int offset, unsigned int count)
807{
808 struct nfs_page *req;
809
810 req = nfs_setup_write_request(ctx, page, offset, count);
811 if (IS_ERR(req))
812 return PTR_ERR(req);
813 /* Update file length */
814 nfs_grow_file(page, offset, count);
815 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
a6305ddb 816 nfs_mark_request_dirty(req);
1d1afcbc 817 nfs_unlock_and_release_request(req);
e7d39069
TM
818 return 0;
819}
820
1da177e4
LT
821int nfs_flush_incompatible(struct file *file, struct page *page)
822{
cd3758e3 823 struct nfs_open_context *ctx = nfs_file_open_context(file);
1da177e4 824 struct nfs_page *req;
1a54533e 825 int do_flush, status;
1da177e4
LT
826 /*
827 * Look for a request corresponding to this page. If there
828 * is one, and it belongs to another file, we flush it out
829 * before we try to copy anything into the page. Do this
830 * due to the lack of an ACCESS-type call in NFSv2.
831 * Also do the same if we find a request from an existing
832 * dropped page.
833 */
1a54533e
TM
834 do {
835 req = nfs_page_find_request(page);
836 if (req == NULL)
837 return 0;
f11ac8db
TM
838 do_flush = req->wb_page != page || req->wb_context != ctx ||
839 req->wb_lock_context->lockowner != current->files ||
840 req->wb_lock_context->pid != current->tgid;
1da177e4 841 nfs_release_request(req);
1a54533e
TM
842 if (!do_flush)
843 return 0;
844 status = nfs_wb_page(page->mapping->host, page);
845 } while (status == 0);
846 return status;
1da177e4
LT
847}
848
5d47a356
TM
849/*
850 * If the page cache is marked as unsafe or invalid, then we can't rely on
851 * the PageUptodate() flag. In this case, we will need to turn off
852 * write optimisations that depend on the page contents being correct.
853 */
8d197a56 854static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
5d47a356 855{
8d197a56
TM
856 if (nfs_have_delegated_attributes(inode))
857 goto out;
858 if (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
859 return false;
860out:
861 return PageUptodate(page) != 0;
5d47a356
TM
862}
863
1da177e4
LT
864/*
865 * Update and possibly write a cached page of an NFS file.
866 *
867 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
868 * things with a page scheduled for an RPC call (e.g. invalidate it).
869 */
870int nfs_updatepage(struct file *file, struct page *page,
871 unsigned int offset, unsigned int count)
872{
cd3758e3 873 struct nfs_open_context *ctx = nfs_file_open_context(file);
1da177e4 874 struct inode *inode = page->mapping->host;
1da177e4
LT
875 int status = 0;
876
91d5b470
CL
877 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
878
48186c7d 879 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
01cce933
JJS
880 file->f_path.dentry->d_parent->d_name.name,
881 file->f_path.dentry->d_name.name, count,
48186c7d 882 (long long)(page_offset(page) + offset));
1da177e4 883
1da177e4 884 /* If we're not using byte range locks, and we know the page
5d47a356
TM
885 * is up to date, it may be more efficient to extend the write
886 * to cover the entire page in order to avoid fragmentation
887 * inefficiencies.
1da177e4 888 */
5d47a356
TM
889 if (nfs_write_pageuptodate(page, inode) &&
890 inode->i_flock == NULL &&
6b2f3d1f 891 !(file->f_flags & O_DSYNC)) {
49a70f27 892 count = max(count + offset, nfs_page_length(page));
1da177e4 893 offset = 0;
1da177e4
LT
894 }
895
e21195a7 896 status = nfs_writepage_setup(ctx, page, offset, count);
03fa9e84
TM
897 if (status < 0)
898 nfs_set_pageerror(page);
59b7c05f
TM
899 else
900 __set_page_dirty_nobuffers(page);
1da177e4 901
48186c7d 902 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
1da177e4 903 status, (long long)i_size_read(inode));
1da177e4
LT
904 return status;
905}
906
3ff7576d 907static int flush_task_priority(int how)
1da177e4
LT
908{
909 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
910 case FLUSH_HIGHPRI:
911 return RPC_PRIORITY_HIGH;
912 case FLUSH_LOWPRI:
913 return RPC_PRIORITY_LOW;
914 }
915 return RPC_PRIORITY_NORMAL;
916}
917
c5996c4e
FI
918int nfs_initiate_write(struct rpc_clnt *clnt,
919 struct nfs_write_data *data,
d138d5d1 920 const struct rpc_call_ops *call_ops,
9f0ec176 921 int how, int flags)
1da177e4 922{
cd841605 923 struct inode *inode = data->header->inode;
3ff7576d 924 int priority = flush_task_priority(how);
07737691 925 struct rpc_task *task;
bdc7f021
TM
926 struct rpc_message msg = {
927 .rpc_argp = &data->args,
928 .rpc_resp = &data->res,
cd841605 929 .rpc_cred = data->header->cred,
bdc7f021 930 };
84115e1c 931 struct rpc_task_setup task_setup_data = {
d138d5d1 932 .rpc_client = clnt,
07737691 933 .task = &data->task,
bdc7f021 934 .rpc_message = &msg,
84115e1c
TM
935 .callback_ops = call_ops,
936 .callback_data = data,
101070ca 937 .workqueue = nfsiod_workqueue,
9f0ec176 938 .flags = RPC_TASK_ASYNC | flags,
3ff7576d 939 .priority = priority,
84115e1c 940 };
2c61be0a 941 int ret = 0;
1da177e4 942
d138d5d1
AA
943 /* Set up the initial task struct. */
944 NFS_PROTO(inode)->write_setup(data, &msg);
945
946 dprintk("NFS: %5u initiated write call "
947 "(req %s/%lld, %u bytes @ offset %llu)\n",
948 data->task.tk_pid,
949 inode->i_sb->s_id,
950 (long long)NFS_FILEID(inode),
951 data->args.count,
952 (unsigned long long)data->args.offset);
953
954 task = rpc_run_task(&task_setup_data);
955 if (IS_ERR(task)) {
956 ret = PTR_ERR(task);
957 goto out;
958 }
959 if (how & FLUSH_SYNC) {
960 ret = rpc_wait_for_completion_task(task);
961 if (ret == 0)
962 ret = task->tk_status;
963 }
964 rpc_put_task(task);
965out:
966 return ret;
967}
a69aef14 968EXPORT_SYMBOL_GPL(nfs_initiate_write);
d138d5d1
AA
969
970/*
971 * Set up the argument/result storage required for the RPC call.
972 */
6c75dc0d 973static void nfs_write_rpcsetup(struct nfs_write_data *data,
d138d5d1 974 unsigned int count, unsigned int offset,
ea2cf228 975 int how, struct nfs_commit_info *cinfo)
d138d5d1 976{
6c75dc0d 977 struct nfs_page *req = data->header->req;
d138d5d1 978
1da177e4
LT
979 /* Set up the RPC argument and reply structs
980 * NB: take care not to mess about with data->commit et al. */
981
6c75dc0d 982 data->args.fh = NFS_FH(data->header->inode);
1da177e4 983 data->args.offset = req_offset(req) + offset;
2bea038c
BH
984 /* pnfs_set_layoutcommit needs this */
985 data->mds_offset = data->args.offset;
1da177e4 986 data->args.pgbase = req->wb_pgbase + offset;
30dd374f 987 data->args.pages = data->pages.pagevec;
1da177e4 988 data->args.count = count;
383ba719 989 data->args.context = get_nfs_open_context(req->wb_context);
f11ac8db 990 data->args.lock_context = req->wb_lock_context;
bdc7f021 991 data->args.stable = NFS_UNSTABLE;
87ed5eb4
TM
992 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
993 case 0:
994 break;
995 case FLUSH_COND_STABLE:
ea2cf228 996 if (nfs_reqs_to_commit(cinfo))
87ed5eb4
TM
997 break;
998 default:
999 data->args.stable = NFS_FILE_SYNC;
bdc7f021 1000 }
1da177e4
LT
1001
1002 data->res.fattr = &data->fattr;
1003 data->res.count = count;
1004 data->res.verf = &data->verf;
0e574af1 1005 nfs_fattr_init(&data->fattr);
6e4efd56 1006}
1da177e4 1007
6e4efd56
TM
1008static int nfs_do_write(struct nfs_write_data *data,
1009 const struct rpc_call_ops *call_ops,
6e4efd56
TM
1010 int how)
1011{
cd841605 1012 struct inode *inode = data->header->inode;
0382b744 1013
9f0ec176 1014 return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how, 0);
1da177e4
LT
1015}
1016
275acaaf
TM
1017static int nfs_do_multiple_writes(struct list_head *head,
1018 const struct rpc_call_ops *call_ops,
275acaaf
TM
1019 int how)
1020{
1021 struct nfs_write_data *data;
1022 int ret = 0;
1023
1024 while (!list_empty(head)) {
1025 int ret2;
1026
6c75dc0d 1027 data = list_first_entry(head, struct nfs_write_data, list);
275acaaf
TM
1028 list_del_init(&data->list);
1029
dce81290 1030 ret2 = nfs_do_write(data, call_ops, how);
275acaaf
TM
1031 if (ret == 0)
1032 ret = ret2;
1033 }
1034 return ret;
1035}
1036
6d884e8f
F
1037/* If a nfs_flush_* function fails, it should remove reqs from @head and
1038 * call this on each, which will prepare them to be retried on next
1039 * writeback using standard nfs.
1040 */
1041static void nfs_redirty_request(struct nfs_page *req)
1042{
1043 nfs_mark_request_dirty(req);
1d1afcbc 1044 nfs_unlock_request(req);
d1182b33 1045 nfs_end_page_writeback(req->wb_page);
3aff4ebb 1046 nfs_release_request(req);
6d884e8f
F
1047}
1048
061ae2ed 1049static void nfs_async_write_error(struct list_head *head)
6c75dc0d
FI
1050{
1051 struct nfs_page *req;
1052
1053 while (!list_empty(head)) {
1054 req = nfs_list_entry(head->next);
1055 nfs_list_remove_request(req);
1056 nfs_redirty_request(req);
1057 }
1058}
1059
061ae2ed
FI
1060static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1061 .error_cleanup = nfs_async_write_error,
1062 .completion = nfs_write_completion,
1063};
1064
25b11dcd
TM
1065static void nfs_flush_error(struct nfs_pageio_descriptor *desc,
1066 struct nfs_pgio_header *hdr)
1067{
1068 set_bit(NFS_IOHDR_REDO, &hdr->flags);
1069 while (!list_empty(&hdr->rpc_list)) {
1070 struct nfs_write_data *data = list_first_entry(&hdr->rpc_list,
1071 struct nfs_write_data, list);
1072 list_del(&data->list);
1073 nfs_writedata_release(data);
1074 }
1075 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1076}
1077
1da177e4
LT
1078/*
1079 * Generate multiple small requests to write out a single
1080 * contiguous dirty area on one page.
1081 */
6c75dc0d
FI
1082static int nfs_flush_multi(struct nfs_pageio_descriptor *desc,
1083 struct nfs_pgio_header *hdr)
1da177e4 1084{
6c75dc0d 1085 struct nfs_page *req = hdr->req;
1da177e4
LT
1086 struct page *page = req->wb_page;
1087 struct nfs_write_data *data;
d097971d 1088 size_t wsize = desc->pg_bsize, nbytes;
e9f7bee1 1089 unsigned int offset;
1da177e4 1090 int requests = 0;
ea2cf228 1091 struct nfs_commit_info cinfo;
1da177e4 1092
ea2cf228 1093 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
1da177e4 1094
b31268ac 1095 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
ea2cf228 1096 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) ||
b31268ac
TM
1097 desc->pg_count > wsize))
1098 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1099
1100
275acaaf 1101 offset = 0;
c76069bd 1102 nbytes = desc->pg_count;
e9f7bee1
TM
1103 do {
1104 size_t len = min(nbytes, wsize);
1105
6c75dc0d 1106 data = nfs_writedata_alloc(hdr, 1);
25b11dcd
TM
1107 if (!data) {
1108 nfs_flush_error(desc, hdr);
1109 return -ENOMEM;
1110 }
30dd374f 1111 data->pages.pagevec[0] = page;
ea2cf228 1112 nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo);
6c75dc0d 1113 list_add(&data->list, &hdr->rpc_list);
1da177e4 1114 requests++;
e9f7bee1 1115 nbytes -= len;
275acaaf 1116 offset += len;
e9f7bee1 1117 } while (nbytes != 0);
25b11dcd
TM
1118 nfs_list_remove_request(req);
1119 nfs_list_add_request(req, &hdr->pages);
6c75dc0d 1120 desc->pg_rpc_callops = &nfs_write_common_ops;
25b11dcd 1121 return 0;
1da177e4
LT
1122}
1123
1124/*
1125 * Create an RPC task for the given write request and kick it.
1126 * The page must have been locked by the caller.
1127 *
1128 * It may happen that the page we're passed is not marked dirty.
1129 * This is the case if nfs_updatepage detects a conflicting request
1130 * that has been written but not committed.
1131 */
6c75dc0d
FI
1132static int nfs_flush_one(struct nfs_pageio_descriptor *desc,
1133 struct nfs_pgio_header *hdr)
1da177e4
LT
1134{
1135 struct nfs_page *req;
1136 struct page **pages;
1137 struct nfs_write_data *data;
c76069bd 1138 struct list_head *head = &desc->pg_list;
ea2cf228 1139 struct nfs_commit_info cinfo;
1da177e4 1140
6c75dc0d
FI
1141 data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base,
1142 desc->pg_count));
1143 if (!data) {
25b11dcd
TM
1144 nfs_flush_error(desc, hdr);
1145 return -ENOMEM;
44b83799 1146 }
6c75dc0d 1147
ea2cf228 1148 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
30dd374f 1149 pages = data->pages.pagevec;
1da177e4
LT
1150 while (!list_empty(head)) {
1151 req = nfs_list_entry(head->next);
1152 nfs_list_remove_request(req);
6c75dc0d 1153 nfs_list_add_request(req, &hdr->pages);
1da177e4 1154 *pages++ = req->wb_page;
1da177e4 1155 }
1da177e4 1156
b31268ac 1157 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
ea2cf228 1158 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
b31268ac
TM
1159 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1160
1da177e4 1161 /* Set up the argument struct */
ea2cf228 1162 nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
6c75dc0d
FI
1163 list_add(&data->list, &hdr->rpc_list);
1164 desc->pg_rpc_callops = &nfs_write_common_ops;
25b11dcd 1165 return 0;
1da177e4
LT
1166}
1167
6c75dc0d
FI
1168int nfs_generic_flush(struct nfs_pageio_descriptor *desc,
1169 struct nfs_pgio_header *hdr)
dce81290
TM
1170{
1171 if (desc->pg_bsize < PAGE_CACHE_SIZE)
6c75dc0d
FI
1172 return nfs_flush_multi(desc, hdr);
1173 return nfs_flush_one(desc, hdr);
dce81290
TM
1174}
1175
1176static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
1751c363 1177{
6c75dc0d
FI
1178 struct nfs_write_header *whdr;
1179 struct nfs_pgio_header *hdr;
275acaaf
TM
1180 int ret;
1181
6c75dc0d
FI
1182 whdr = nfs_writehdr_alloc();
1183 if (!whdr) {
9b5415b5 1184 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
6c75dc0d
FI
1185 return -ENOMEM;
1186 }
1187 hdr = &whdr->header;
1188 nfs_pgheader_init(desc, hdr, nfs_writehdr_free);
1189 atomic_inc(&hdr->refcnt);
1190 ret = nfs_generic_flush(desc, hdr);
50828d7e 1191 if (ret == 0)
6c75dc0d
FI
1192 ret = nfs_do_multiple_writes(&hdr->rpc_list,
1193 desc->pg_rpc_callops,
1194 desc->pg_ioflags);
6c75dc0d 1195 if (atomic_dec_and_test(&hdr->refcnt))
061ae2ed 1196 hdr->completion_ops->completion(hdr);
275acaaf 1197 return ret;
1751c363 1198}
1751c363
TM
1199
1200static const struct nfs_pageio_ops nfs_pageio_write_ops = {
1201 .pg_test = nfs_generic_pg_test,
1202 .pg_doio = nfs_generic_pg_writepages,
1203};
1204
e2fecb21 1205void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
061ae2ed
FI
1206 struct inode *inode, int ioflags,
1207 const struct nfs_pgio_completion_ops *compl_ops)
1da177e4 1208{
061ae2ed 1209 nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, compl_ops,
1751c363
TM
1210 NFS_SERVER(inode)->wsize, ioflags);
1211}
1da177e4 1212
dce81290
TM
1213void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1214{
1215 pgio->pg_ops = &nfs_pageio_write_ops;
1216 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1217}
1f945357 1218EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
dce81290 1219
1763da12
FI
1220void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1221 struct inode *inode, int ioflags,
1222 const struct nfs_pgio_completion_ops *compl_ops)
1751c363 1223{
061ae2ed
FI
1224 if (!pnfs_pageio_init_write(pgio, inode, ioflags, compl_ops))
1225 nfs_pageio_init_write_mds(pgio, inode, ioflags, compl_ops);
1da177e4
LT
1226}
1227
def6ed7e
AA
1228void nfs_write_prepare(struct rpc_task *task, void *calldata)
1229{
1230 struct nfs_write_data *data = calldata;
cd841605 1231 NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data);
def6ed7e 1232}
def6ed7e 1233
0b7c0153
FI
1234void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1235{
1236 struct nfs_commit_data *data = calldata;
1237
1238 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1239}
1240
1da177e4
LT
1241/*
1242 * Handle a write reply that flushes a whole page.
1243 *
1244 * FIXME: There is an inherent race with invalidate_inode_pages and
1245 * writebacks since the page->count is kept > 1 for as long
1246 * as the page has a write request pending.
1247 */
6c75dc0d 1248static void nfs_writeback_done_common(struct rpc_task *task, void *calldata)
1da177e4 1249{
788e7a89 1250 struct nfs_write_data *data = calldata;
1da177e4 1251
c9d8f89d
TM
1252 nfs_writeback_done(task, data);
1253}
1254
6c75dc0d 1255static void nfs_writeback_release_common(void *calldata)
c9d8f89d
TM
1256{
1257 struct nfs_write_data *data = calldata;
cd841605 1258 struct nfs_pgio_header *hdr = data->header;
e2fecb21 1259 int status = data->task.tk_status;
788e7a89 1260
6c75dc0d
FI
1261 if ((status >= 0) && nfs_write_need_commit(data)) {
1262 spin_lock(&hdr->lock);
1263 if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags))
1264 ; /* Do nothing */
1265 else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags))
9bce008b
TM
1266 memcpy(hdr->verf, &data->verf, sizeof(*hdr->verf));
1267 else if (memcmp(hdr->verf, &data->verf, sizeof(*hdr->verf)))
6c75dc0d
FI
1268 set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags);
1269 spin_unlock(&hdr->lock);
1da177e4 1270 }
cd841605 1271 nfs_writedata_release(data);
1da177e4
LT
1272}
1273
6c75dc0d 1274static const struct rpc_call_ops nfs_write_common_ops = {
def6ed7e 1275 .rpc_call_prepare = nfs_write_prepare,
6c75dc0d
FI
1276 .rpc_call_done = nfs_writeback_done_common,
1277 .rpc_release = nfs_writeback_release_common,
788e7a89
TM
1278};
1279
1280
1da177e4
LT
1281/*
1282 * This function is called when the WRITE call is complete.
1283 */
13602896 1284void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
1da177e4 1285{
1da177e4
LT
1286 struct nfs_writeargs *argp = &data->args;
1287 struct nfs_writeres *resp = &data->res;
cd841605 1288 struct inode *inode = data->header->inode;
788e7a89 1289 int status;
1da177e4 1290
a3f565b1 1291 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
1da177e4
LT
1292 task->tk_pid, task->tk_status);
1293
f551e44f
CL
1294 /*
1295 * ->write_done will attempt to use post-op attributes to detect
1296 * conflicting writes by other clients. A strict interpretation
1297 * of close-to-open would allow us to continue caching even if
1298 * another writer had changed the file, but some applications
1299 * depend on tighter cache coherency when writing.
1300 */
cd841605 1301 status = NFS_PROTO(inode)->write_done(task, data);
788e7a89 1302 if (status != 0)
13602896 1303 return;
cd841605 1304 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
91d5b470 1305
1da177e4
LT
1306#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1307 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1308 /* We tried a write call, but the server did not
1309 * commit data to stable storage even though we
1310 * requested it.
1311 * Note: There is a known bug in Tru64 < 5.0 in which
1312 * the server reports NFS_DATA_SYNC, but performs
1313 * NFS_FILE_SYNC. We therefore implement this checking
1314 * as a dprintk() in order to avoid filling syslog.
1315 */
1316 static unsigned long complain;
1317
a69aef14 1318 /* Note this will print the MDS for a DS write */
1da177e4 1319 if (time_before(complain, jiffies)) {
48186c7d 1320 dprintk("NFS: faulty NFS server %s:"
1da177e4 1321 " (committed = %d) != (stable = %d)\n",
cd841605 1322 NFS_SERVER(inode)->nfs_client->cl_hostname,
1da177e4
LT
1323 resp->verf->committed, argp->stable);
1324 complain = jiffies + 300 * HZ;
1325 }
1326 }
1327#endif
6c75dc0d
FI
1328 if (task->tk_status < 0)
1329 nfs_set_pgio_error(data->header, task->tk_status, argp->offset);
1330 else if (resp->count < argp->count) {
1da177e4
LT
1331 static unsigned long complain;
1332
6c75dc0d 1333 /* This a short write! */
cd841605 1334 nfs_inc_stats(inode, NFSIOS_SHORTWRITE);
91d5b470 1335
1da177e4 1336 /* Has the server at least made some progress? */
6c75dc0d
FI
1337 if (resp->count == 0) {
1338 if (time_before(complain, jiffies)) {
1339 printk(KERN_WARNING
1340 "NFS: Server wrote zero bytes, expected %u.\n",
1341 argp->count);
1342 complain = jiffies + 300 * HZ;
1da177e4 1343 }
6c75dc0d
FI
1344 nfs_set_pgio_error(data->header, -EIO, argp->offset);
1345 task->tk_status = -EIO;
13602896 1346 return;
1da177e4 1347 }
6c75dc0d
FI
1348 /* Was this an NFSv2 write or an NFSv3 stable write? */
1349 if (resp->verf->committed != NFS_UNSTABLE) {
1350 /* Resend from where the server left off */
1351 data->mds_offset += resp->count;
1352 argp->offset += resp->count;
1353 argp->pgbase += resp->count;
1354 argp->count -= resp->count;
1355 } else {
1356 /* Resend as a stable write in order to avoid
1357 * headaches in the case of a server crash.
1358 */
1359 argp->stable = NFS_FILE_SYNC;
1da177e4 1360 }
6c75dc0d 1361 rpc_restart_call_prepare(task);
1da177e4 1362 }
1da177e4
LT
1363}
1364
1365
1366#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
71d0a611
TM
1367static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1368{
b8413f98
TM
1369 int ret;
1370
71d0a611
TM
1371 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1372 return 1;
b8413f98
TM
1373 if (!may_wait)
1374 return 0;
1375 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1376 NFS_INO_COMMIT,
1377 nfs_wait_bit_killable,
1378 TASK_KILLABLE);
1379 return (ret < 0) ? ret : 1;
71d0a611
TM
1380}
1381
f453a54a 1382static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
71d0a611
TM
1383{
1384 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1385 smp_mb__after_clear_bit();
1386 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1387}
1388
0b7c0153 1389void nfs_commitdata_release(struct nfs_commit_data *data)
1da177e4 1390{
0b7c0153
FI
1391 put_nfs_open_context(data->context);
1392 nfs_commit_free(data);
1da177e4 1393}
e0c2b380 1394EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1da177e4 1395
0b7c0153 1396int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
9ace33cd 1397 const struct rpc_call_ops *call_ops,
9f0ec176 1398 int how, int flags)
1da177e4 1399{
07737691 1400 struct rpc_task *task;
9ace33cd 1401 int priority = flush_task_priority(how);
bdc7f021
TM
1402 struct rpc_message msg = {
1403 .rpc_argp = &data->args,
1404 .rpc_resp = &data->res,
9ace33cd 1405 .rpc_cred = data->cred,
bdc7f021 1406 };
84115e1c 1407 struct rpc_task_setup task_setup_data = {
07737691 1408 .task = &data->task,
9ace33cd 1409 .rpc_client = clnt,
bdc7f021 1410 .rpc_message = &msg,
9ace33cd 1411 .callback_ops = call_ops,
84115e1c 1412 .callback_data = data,
101070ca 1413 .workqueue = nfsiod_workqueue,
9f0ec176 1414 .flags = RPC_TASK_ASYNC | flags,
3ff7576d 1415 .priority = priority,
84115e1c 1416 };
9ace33cd
FI
1417 /* Set up the initial task struct. */
1418 NFS_PROTO(data->inode)->commit_setup(data, &msg);
1419
1420 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1421
1422 task = rpc_run_task(&task_setup_data);
1423 if (IS_ERR(task))
1424 return PTR_ERR(task);
1425 if (how & FLUSH_SYNC)
1426 rpc_wait_for_completion_task(task);
1427 rpc_put_task(task);
1428 return 0;
1429}
e0c2b380 1430EXPORT_SYMBOL_GPL(nfs_initiate_commit);
9ace33cd
FI
1431
1432/*
1433 * Set up the argument/result storage required for the RPC call.
1434 */
0b7c0153 1435void nfs_init_commit(struct nfs_commit_data *data,
f453a54a
FI
1436 struct list_head *head,
1437 struct pnfs_layout_segment *lseg,
1438 struct nfs_commit_info *cinfo)
9ace33cd
FI
1439{
1440 struct nfs_page *first = nfs_list_entry(head->next);
3d4ff43d 1441 struct inode *inode = first->wb_context->dentry->d_inode;
1da177e4
LT
1442
1443 /* Set up the RPC argument and reply structs
1444 * NB: take care not to mess about with data->commit et al. */
1445
1446 list_splice_init(head, &data->pages);
1da177e4 1447
1da177e4 1448 data->inode = inode;
9ace33cd 1449 data->cred = first->wb_context->cred;
988b6dce 1450 data->lseg = lseg; /* reference transferred */
9ace33cd 1451 data->mds_ops = &nfs_commit_ops;
f453a54a 1452 data->completion_ops = cinfo->completion_ops;
b359f9d0 1453 data->dreq = cinfo->dreq;
1da177e4
LT
1454
1455 data->args.fh = NFS_FH(data->inode);
3da28eb1
TM
1456 /* Note: we always request a commit of the entire inode */
1457 data->args.offset = 0;
1458 data->args.count = 0;
0b7c0153 1459 data->context = get_nfs_open_context(first->wb_context);
1da177e4
LT
1460 data->res.fattr = &data->fattr;
1461 data->res.verf = &data->verf;
0e574af1 1462 nfs_fattr_init(&data->fattr);
1da177e4 1463}
e0c2b380 1464EXPORT_SYMBOL_GPL(nfs_init_commit);
1da177e4 1465
e0c2b380 1466void nfs_retry_commit(struct list_head *page_list,
ea2cf228
FI
1467 struct pnfs_layout_segment *lseg,
1468 struct nfs_commit_info *cinfo)
64bfeb49
FI
1469{
1470 struct nfs_page *req;
1471
1472 while (!list_empty(page_list)) {
1473 req = nfs_list_entry(page_list->next);
1474 nfs_list_remove_request(req);
ea2cf228 1475 nfs_mark_request_commit(req, lseg, cinfo);
56f9cd68
FI
1476 if (!cinfo->dreq) {
1477 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1478 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1479 BDI_RECLAIMABLE);
1480 }
1d1afcbc 1481 nfs_unlock_and_release_request(req);
64bfeb49
FI
1482 }
1483}
e0c2b380 1484EXPORT_SYMBOL_GPL(nfs_retry_commit);
64bfeb49 1485
1da177e4
LT
1486/*
1487 * Commit dirty pages
1488 */
1489static int
ea2cf228
FI
1490nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1491 struct nfs_commit_info *cinfo)
1da177e4 1492{
0b7c0153 1493 struct nfs_commit_data *data;
1da177e4 1494
c9d8f89d 1495 data = nfs_commitdata_alloc();
1da177e4
LT
1496
1497 if (!data)
1498 goto out_bad;
1499
1500 /* Set up the argument struct */
f453a54a
FI
1501 nfs_init_commit(data, head, NULL, cinfo);
1502 atomic_inc(&cinfo->mds->rpcs_out);
9f0ec176
AA
1503 return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops,
1504 how, 0);
1da177e4 1505 out_bad:
ea2cf228 1506 nfs_retry_commit(head, NULL, cinfo);
f453a54a 1507 cinfo->completion_ops->error_cleanup(NFS_I(inode));
1da177e4
LT
1508 return -ENOMEM;
1509}
1510
1511/*
1512 * COMMIT call returned
1513 */
788e7a89 1514static void nfs_commit_done(struct rpc_task *task, void *calldata)
1da177e4 1515{
0b7c0153 1516 struct nfs_commit_data *data = calldata;
1da177e4 1517
a3f565b1 1518 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1da177e4
LT
1519 task->tk_pid, task->tk_status);
1520
788e7a89 1521 /* Call the NFS version-specific code */
c0d0e96b 1522 NFS_PROTO(data->inode)->commit_done(task, data);
c9d8f89d
TM
1523}
1524
f453a54a 1525static void nfs_commit_release_pages(struct nfs_commit_data *data)
c9d8f89d 1526{
5917ce84 1527 struct nfs_page *req;
c9d8f89d 1528 int status = data->task.tk_status;
f453a54a 1529 struct nfs_commit_info cinfo;
788e7a89 1530
1da177e4
LT
1531 while (!list_empty(&data->pages)) {
1532 req = nfs_list_entry(data->pages.next);
1533 nfs_list_remove_request(req);
d6d6dc7c 1534 nfs_clear_page_commit(req->wb_page);
1da177e4 1535
48186c7d 1536 dprintk("NFS: commit (%s/%lld %d@%lld)",
3d4ff43d
AV
1537 req->wb_context->dentry->d_sb->s_id,
1538 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
1da177e4
LT
1539 req->wb_bytes,
1540 (long long)req_offset(req));
c9d8f89d
TM
1541 if (status < 0) {
1542 nfs_context_set_write_error(req->wb_context, status);
1da177e4 1543 nfs_inode_remove_request(req);
c9d8f89d 1544 dprintk(", error = %d\n", status);
1da177e4
LT
1545 goto next;
1546 }
1547
1548 /* Okay, COMMIT succeeded, apparently. Check the verifier
1549 * returned by the server against all stored verfs. */
1550 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1551 /* We have a match */
1552 nfs_inode_remove_request(req);
1553 dprintk(" OK\n");
1554 goto next;
1555 }
1556 /* We have a mismatch. Write the page again */
1557 dprintk(" mismatch\n");
6d884e8f 1558 nfs_mark_request_dirty(req);
1da177e4 1559 next:
1d1afcbc 1560 nfs_unlock_and_release_request(req);
1da177e4 1561 }
f453a54a
FI
1562 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1563 if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
1564 nfs_commit_clear_lock(NFS_I(data->inode));
5917ce84
FI
1565}
1566
1567static void nfs_commit_release(void *calldata)
1568{
0b7c0153 1569 struct nfs_commit_data *data = calldata;
5917ce84 1570
f453a54a 1571 data->completion_ops->completion(data);
c9d8f89d 1572 nfs_commitdata_release(calldata);
1da177e4 1573}
788e7a89
TM
1574
1575static const struct rpc_call_ops nfs_commit_ops = {
0b7c0153 1576 .rpc_call_prepare = nfs_commit_prepare,
788e7a89
TM
1577 .rpc_call_done = nfs_commit_done,
1578 .rpc_release = nfs_commit_release,
1579};
1da177e4 1580
f453a54a
FI
1581static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1582 .completion = nfs_commit_release_pages,
1583 .error_cleanup = nfs_commit_clear_lock,
1584};
1585
1763da12
FI
1586int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1587 int how, struct nfs_commit_info *cinfo)
84c53ab5
FI
1588{
1589 int status;
1590
ea2cf228 1591 status = pnfs_commit_list(inode, head, how, cinfo);
84c53ab5 1592 if (status == PNFS_NOT_ATTEMPTED)
ea2cf228 1593 status = nfs_commit_list(inode, head, how, cinfo);
84c53ab5
FI
1594 return status;
1595}
1596
b608b283 1597int nfs_commit_inode(struct inode *inode, int how)
1da177e4 1598{
1da177e4 1599 LIST_HEAD(head);
ea2cf228 1600 struct nfs_commit_info cinfo;
71d0a611 1601 int may_wait = how & FLUSH_SYNC;
b8413f98 1602 int res;
1da177e4 1603
b8413f98
TM
1604 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1605 if (res <= 0)
c5efa5fc 1606 goto out_mark_dirty;
ea2cf228
FI
1607 nfs_init_cinfo_from_inode(&cinfo, inode);
1608 res = nfs_scan_commit(inode, &head, &cinfo);
1da177e4 1609 if (res) {
a861a1e1
FI
1610 int error;
1611
ea2cf228 1612 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
3da28eb1
TM
1613 if (error < 0)
1614 return error;
b8413f98 1615 if (!may_wait)
c5efa5fc 1616 goto out_mark_dirty;
b8413f98
TM
1617 error = wait_on_bit(&NFS_I(inode)->flags,
1618 NFS_INO_COMMIT,
1619 nfs_wait_bit_killable,
1620 TASK_KILLABLE);
1621 if (error < 0)
1622 return error;
71d0a611
TM
1623 } else
1624 nfs_commit_clear_lock(NFS_I(inode));
c5efa5fc
TM
1625 return res;
1626 /* Note: If we exit without ensuring that the commit is complete,
1627 * we must mark the inode as dirty. Otherwise, future calls to
1628 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1629 * that the data is on the disk.
1630 */
1631out_mark_dirty:
1632 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1da177e4
LT
1633 return res;
1634}
8fc795f7
TM
1635
1636static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1637{
420e3646
TM
1638 struct nfs_inode *nfsi = NFS_I(inode);
1639 int flags = FLUSH_SYNC;
1640 int ret = 0;
8fc795f7 1641
3236c3e1 1642 /* no commits means nothing needs to be done */
ea2cf228 1643 if (!nfsi->commit_info.ncommit)
3236c3e1
JL
1644 return ret;
1645
a00dd6c0
JL
1646 if (wbc->sync_mode == WB_SYNC_NONE) {
1647 /* Don't commit yet if this is a non-blocking flush and there
1648 * are a lot of outstanding writes for this mapping.
1649 */
ea2cf228 1650 if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1))
a00dd6c0 1651 goto out_mark_dirty;
420e3646 1652
a00dd6c0 1653 /* don't wait for the COMMIT response */
420e3646 1654 flags = 0;
a00dd6c0
JL
1655 }
1656
420e3646
TM
1657 ret = nfs_commit_inode(inode, flags);
1658 if (ret >= 0) {
1659 if (wbc->sync_mode == WB_SYNC_NONE) {
1660 if (ret < wbc->nr_to_write)
1661 wbc->nr_to_write -= ret;
1662 else
1663 wbc->nr_to_write = 0;
1664 }
8fc795f7 1665 return 0;
420e3646
TM
1666 }
1667out_mark_dirty:
8fc795f7
TM
1668 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1669 return ret;
1670}
c63c7b05 1671#else
8fc795f7
TM
1672static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1673{
1674 return 0;
1675}
1da177e4
LT
1676#endif
1677
8fc795f7
TM
1678int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1679{
863a3c6c
AA
1680 int ret;
1681
1682 ret = nfs_commit_unstable_pages(inode, wbc);
1683 if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) {
ef311537
AA
1684 int status;
1685 bool sync = true;
863a3c6c 1686
846d5a09 1687 if (wbc->sync_mode == WB_SYNC_NONE)
ef311537 1688 sync = false;
863a3c6c
AA
1689
1690 status = pnfs_layoutcommit_inode(inode, sync);
1691 if (status < 0)
1692 return status;
1693 }
1694 return ret;
8fc795f7
TM
1695}
1696
acdc53b2
TM
1697/*
1698 * flush the inode to disk.
1699 */
1700int nfs_wb_all(struct inode *inode)
34901f70
TM
1701{
1702 struct writeback_control wbc = {
72cb77f4 1703 .sync_mode = WB_SYNC_ALL,
34901f70 1704 .nr_to_write = LONG_MAX,
d7fb1207
TM
1705 .range_start = 0,
1706 .range_end = LLONG_MAX,
34901f70 1707 };
34901f70 1708
acdc53b2 1709 return sync_inode(inode, &wbc);
1c75950b
TM
1710}
1711
1b3b4a1a
TM
1712int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1713{
1714 struct nfs_page *req;
1b3b4a1a
TM
1715 int ret = 0;
1716
1717 BUG_ON(!PageLocked(page));
1718 for (;;) {
ba8b06e6 1719 wait_on_page_writeback(page);
1b3b4a1a
TM
1720 req = nfs_page_find_request(page);
1721 if (req == NULL)
1b3b4a1a 1722 break;
7ad84aa9 1723 if (nfs_lock_request(req)) {
8dd37758 1724 nfs_clear_request_commit(req);
1b3b4a1a
TM
1725 nfs_inode_remove_request(req);
1726 /*
1727 * In case nfs_inode_remove_request has marked the
1728 * page as being dirty
1729 */
1730 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1d1afcbc 1731 nfs_unlock_and_release_request(req);
1b3b4a1a
TM
1732 break;
1733 }
1734 ret = nfs_wait_on_request(req);
c9edda71 1735 nfs_release_request(req);
1b3b4a1a 1736 if (ret < 0)
c988950e 1737 break;
1b3b4a1a 1738 }
1b3b4a1a
TM
1739 return ret;
1740}
1741
7f2f12d9
TM
1742/*
1743 * Write back all requests on one page - we do this before reading it.
1744 */
1745int nfs_wb_page(struct inode *inode, struct page *page)
1c75950b
TM
1746{
1747 loff_t range_start = page_offset(page);
1748 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
4d770ccf 1749 struct writeback_control wbc = {
4d770ccf 1750 .sync_mode = WB_SYNC_ALL,
7f2f12d9 1751 .nr_to_write = 0,
4d770ccf
TM
1752 .range_start = range_start,
1753 .range_end = range_end,
1754 };
1755 int ret;
1c75950b 1756
0522f6ad 1757 for (;;) {
ba8b06e6 1758 wait_on_page_writeback(page);
73e3302f
TM
1759 if (clear_page_dirty_for_io(page)) {
1760 ret = nfs_writepage_locked(page, &wbc);
1761 if (ret < 0)
1762 goto out_error;
0522f6ad 1763 continue;
7f2f12d9 1764 }
0522f6ad
TM
1765 if (!PagePrivate(page))
1766 break;
1767 ret = nfs_commit_inode(inode, FLUSH_SYNC);
ba8b06e6 1768 if (ret < 0)
73e3302f 1769 goto out_error;
7f2f12d9 1770 }
73e3302f
TM
1771 return 0;
1772out_error:
4d770ccf 1773 return ret;
1c75950b
TM
1774}
1775
074cc1de
TM
1776#ifdef CONFIG_MIGRATION
1777int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
a6bc32b8 1778 struct page *page, enum migrate_mode mode)
074cc1de 1779{
2da95652
JL
1780 /*
1781 * If PagePrivate is set, then the page is currently associated with
1782 * an in-progress read or write request. Don't try to migrate it.
1783 *
1784 * FIXME: we could do this in principle, but we'll need a way to ensure
1785 * that we can safely release the inode reference while holding
1786 * the page lock.
1787 */
1788 if (PagePrivate(page))
1789 return -EBUSY;
074cc1de 1790
7549ad5f 1791 nfs_fscache_release_page(page, GFP_KERNEL);
074cc1de 1792
a6bc32b8 1793 return migrate_page(mapping, newpage, page, mode);
074cc1de
TM
1794}
1795#endif
1796
f7b422b1 1797int __init nfs_init_writepagecache(void)
1da177e4
LT
1798{
1799 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
cd841605 1800 sizeof(struct nfs_write_header),
1da177e4 1801 0, SLAB_HWCACHE_ALIGN,
20c2df83 1802 NULL);
1da177e4
LT
1803 if (nfs_wdata_cachep == NULL)
1804 return -ENOMEM;
1805
93d2341c
MD
1806 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1807 nfs_wdata_cachep);
1da177e4
LT
1808 if (nfs_wdata_mempool == NULL)
1809 return -ENOMEM;
1810
0b7c0153
FI
1811 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
1812 sizeof(struct nfs_commit_data),
1813 0, SLAB_HWCACHE_ALIGN,
1814 NULL);
1815 if (nfs_cdata_cachep == NULL)
1816 return -ENOMEM;
1817
93d2341c
MD
1818 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1819 nfs_wdata_cachep);
1da177e4
LT
1820 if (nfs_commit_mempool == NULL)
1821 return -ENOMEM;
1822
89a09141
PZ
1823 /*
1824 * NFS congestion size, scale with available memory.
1825 *
1826 * 64MB: 8192k
1827 * 128MB: 11585k
1828 * 256MB: 16384k
1829 * 512MB: 23170k
1830 * 1GB: 32768k
1831 * 2GB: 46340k
1832 * 4GB: 65536k
1833 * 8GB: 92681k
1834 * 16GB: 131072k
1835 *
1836 * This allows larger machines to have larger/more transfers.
1837 * Limit the default to 256M
1838 */
1839 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1840 if (nfs_congestion_kb > 256*1024)
1841 nfs_congestion_kb = 256*1024;
1842
1da177e4
LT
1843 return 0;
1844}
1845
266bee88 1846void nfs_destroy_writepagecache(void)
1da177e4
LT
1847{
1848 mempool_destroy(nfs_commit_mempool);
1849 mempool_destroy(nfs_wdata_mempool);
1a1d92c1 1850 kmem_cache_destroy(nfs_wdata_cachep);
1da177e4
LT
1851}
1852