Commit | Line | Data |
---|---|---|
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 | 42 | static void nfs_redirty_request(struct nfs_page *req); |
6c75dc0d | 43 | static const struct rpc_call_ops nfs_write_common_ops; |
788e7a89 | 44 | static const struct rpc_call_ops nfs_commit_ops; |
061ae2ed | 45 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; |
f453a54a | 46 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops; |
1da177e4 | 47 | |
e18b890b | 48 | static struct kmem_cache *nfs_wdata_cachep; |
3feb2d49 | 49 | static mempool_t *nfs_wdata_mempool; |
0b7c0153 | 50 | static struct kmem_cache *nfs_cdata_cachep; |
1da177e4 LT |
51 | static mempool_t *nfs_commit_mempool; |
52 | ||
0b7c0153 | 53 | struct nfs_commit_data *nfs_commitdata_alloc(void) |
1da177e4 | 54 | { |
192e501b | 55 | struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); |
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 | 63 | EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); |
1da177e4 | 64 | |
0b7c0153 | 65 | void nfs_commit_free(struct nfs_commit_data *p) |
1da177e4 LT |
66 | { |
67 | mempool_free(p, nfs_commit_mempool); | |
68 | } | |
e0c2b380 | 69 | EXPORT_SYMBOL_GPL(nfs_commit_free); |
1da177e4 | 70 | |
6c75dc0d | 71 | struct nfs_write_header *nfs_writehdr_alloc(void) |
3feb2d49 | 72 | { |
192e501b | 73 | struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); |
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 | } | |
89d77c8f | 87 | EXPORT_SYMBOL_GPL(nfs_writehdr_alloc); |
3feb2d49 | 88 | |
1763da12 FI |
89 | static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr, |
90 | unsigned int pagecount) | |
6c75dc0d FI |
91 | { |
92 | struct nfs_write_data *data, *prealloc; | |
93 | ||
94 | prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data; | |
95 | if (prealloc->header == NULL) | |
96 | data = prealloc; | |
97 | else | |
98 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
99 | if (!data) | |
100 | goto out; | |
101 | ||
102 | if (nfs_pgarray_set(&data->pages, pagecount)) { | |
103 | data->header = hdr; | |
104 | atomic_inc(&hdr->refcnt); | |
105 | } else { | |
106 | if (data != prealloc) | |
107 | kfree(data); | |
108 | data = NULL; | |
109 | } | |
110 | out: | |
111 | return data; | |
112 | } | |
113 | ||
cd841605 | 114 | void nfs_writehdr_free(struct nfs_pgio_header *hdr) |
3feb2d49 | 115 | { |
cd841605 FI |
116 | struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header); |
117 | mempool_free(whdr, nfs_wdata_mempool); | |
3feb2d49 | 118 | } |
89d77c8f | 119 | EXPORT_SYMBOL_GPL(nfs_writehdr_free); |
3feb2d49 | 120 | |
dce81290 | 121 | void nfs_writedata_release(struct nfs_write_data *wdata) |
1da177e4 | 122 | { |
6c75dc0d FI |
123 | struct nfs_pgio_header *hdr = wdata->header; |
124 | struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header); | |
125 | ||
383ba719 | 126 | put_nfs_open_context(wdata->args.context); |
30dd374f FI |
127 | if (wdata->pages.pagevec != wdata->pages.page_array) |
128 | kfree(wdata->pages.pagevec); | |
6db6dd7d | 129 | if (wdata == &write_header->rpc_data) { |
6c75dc0d | 130 | wdata->header = NULL; |
6db6dd7d TM |
131 | wdata = NULL; |
132 | } | |
6c75dc0d | 133 | if (atomic_dec_and_test(&hdr->refcnt)) |
061ae2ed | 134 | hdr->completion_ops->completion(hdr); |
6db6dd7d TM |
135 | /* Note: we only free the rpc_task after callbacks are done. |
136 | * See the comment in rpc_free_task() for why | |
137 | */ | |
138 | kfree(wdata); | |
1da177e4 | 139 | } |
89d77c8f | 140 | EXPORT_SYMBOL_GPL(nfs_writedata_release); |
1da177e4 | 141 | |
7b159fc1 TM |
142 | static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error) |
143 | { | |
144 | ctx->error = error; | |
145 | smp_wmb(); | |
146 | set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
147 | } | |
148 | ||
29418aa4 MG |
149 | static struct nfs_page * |
150 | nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page) | |
277459d2 TM |
151 | { |
152 | struct nfs_page *req = NULL; | |
153 | ||
29418aa4 | 154 | if (PagePrivate(page)) |
277459d2 | 155 | req = (struct nfs_page *)page_private(page); |
29418aa4 MG |
156 | else if (unlikely(PageSwapCache(page))) { |
157 | struct nfs_page *freq, *t; | |
158 | ||
159 | /* Linearly search the commit list for the correct req */ | |
160 | list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) { | |
161 | if (freq->wb_page == page) { | |
162 | req = freq; | |
163 | break; | |
164 | } | |
165 | } | |
277459d2 | 166 | } |
29418aa4 MG |
167 | |
168 | if (req) | |
169 | kref_get(&req->wb_kref); | |
170 | ||
277459d2 TM |
171 | return req; |
172 | } | |
173 | ||
174 | static struct nfs_page *nfs_page_find_request(struct page *page) | |
175 | { | |
d56b4ddf | 176 | struct inode *inode = page_file_mapping(page)->host; |
277459d2 | 177 | struct nfs_page *req = NULL; |
277459d2 | 178 | |
587142f8 | 179 | spin_lock(&inode->i_lock); |
29418aa4 | 180 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
587142f8 | 181 | spin_unlock(&inode->i_lock); |
277459d2 TM |
182 | return req; |
183 | } | |
184 | ||
1da177e4 LT |
185 | /* Adjust the file length if we're writing beyond the end */ |
186 | static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) | |
187 | { | |
d56b4ddf | 188 | struct inode *inode = page_file_mapping(page)->host; |
a3d01454 TM |
189 | loff_t end, i_size; |
190 | pgoff_t end_index; | |
1da177e4 | 191 | |
a3d01454 TM |
192 | spin_lock(&inode->i_lock); |
193 | i_size = i_size_read(inode); | |
194 | end_index = (i_size - 1) >> PAGE_CACHE_SHIFT; | |
d56b4ddf | 195 | if (i_size > 0 && page_file_index(page) < end_index) |
a3d01454 | 196 | goto out; |
d56b4ddf | 197 | end = page_file_offset(page) + ((loff_t)offset+count); |
1da177e4 | 198 | if (i_size >= end) |
a3d01454 | 199 | goto out; |
1da177e4 | 200 | i_size_write(inode, end); |
a3d01454 TM |
201 | nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); |
202 | out: | |
203 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
204 | } |
205 | ||
a301b777 TM |
206 | /* A writeback failed: mark the page as bad, and invalidate the page cache */ |
207 | static void nfs_set_pageerror(struct page *page) | |
208 | { | |
d56b4ddf | 209 | nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); |
a301b777 TM |
210 | } |
211 | ||
1da177e4 LT |
212 | /* We can set the PG_uptodate flag if we see that a write request |
213 | * covers the full page. | |
214 | */ | |
215 | static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count) | |
216 | { | |
1da177e4 LT |
217 | if (PageUptodate(page)) |
218 | return; | |
219 | if (base != 0) | |
220 | return; | |
49a70f27 | 221 | if (count != nfs_page_length(page)) |
1da177e4 | 222 | return; |
49a70f27 | 223 | SetPageUptodate(page); |
1da177e4 LT |
224 | } |
225 | ||
1da177e4 LT |
226 | static int wb_priority(struct writeback_control *wbc) |
227 | { | |
228 | if (wbc->for_reclaim) | |
c63c7b05 | 229 | return FLUSH_HIGHPRI | FLUSH_STABLE; |
b17621fe | 230 | if (wbc->for_kupdate || wbc->for_background) |
b31268ac TM |
231 | return FLUSH_LOWPRI | FLUSH_COND_STABLE; |
232 | return FLUSH_COND_STABLE; | |
1da177e4 LT |
233 | } |
234 | ||
89a09141 PZ |
235 | /* |
236 | * NFS congestion control | |
237 | */ | |
238 | ||
239 | int nfs_congestion_kb; | |
240 | ||
241 | #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) | |
242 | #define NFS_CONGESTION_OFF_THRESH \ | |
243 | (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) | |
244 | ||
deed85e7 | 245 | static void nfs_set_page_writeback(struct page *page) |
89a09141 | 246 | { |
deed85e7 | 247 | struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host); |
5a6d41b3 TM |
248 | int ret = test_set_page_writeback(page); |
249 | ||
deed85e7 | 250 | WARN_ON_ONCE(ret != 0); |
89a09141 | 251 | |
deed85e7 TM |
252 | if (atomic_long_inc_return(&nfss->writeback) > |
253 | NFS_CONGESTION_ON_THRESH) { | |
254 | set_bdi_congested(&nfss->backing_dev_info, | |
255 | BLK_RW_ASYNC); | |
89a09141 PZ |
256 | } |
257 | } | |
258 | ||
259 | static void nfs_end_page_writeback(struct page *page) | |
260 | { | |
d56b4ddf | 261 | struct inode *inode = page_file_mapping(page)->host; |
89a09141 PZ |
262 | struct nfs_server *nfss = NFS_SERVER(inode); |
263 | ||
264 | end_page_writeback(page); | |
c4dc4bee | 265 | if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) |
8aa7e847 | 266 | clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC); |
89a09141 PZ |
267 | } |
268 | ||
cfb506e1 | 269 | static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock) |
e261f51f | 270 | { |
d56b4ddf | 271 | struct inode *inode = page_file_mapping(page)->host; |
e261f51f | 272 | struct nfs_page *req; |
e261f51f TM |
273 | int ret; |
274 | ||
587142f8 | 275 | spin_lock(&inode->i_lock); |
074cc1de | 276 | for (;;) { |
29418aa4 | 277 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
074cc1de TM |
278 | if (req == NULL) |
279 | break; | |
7ad84aa9 | 280 | if (nfs_lock_request(req)) |
e261f51f TM |
281 | break; |
282 | /* Note: If we hold the page lock, as is the case in nfs_writepage, | |
7ad84aa9 | 283 | * then the call to nfs_lock_request() will always |
e261f51f TM |
284 | * succeed provided that someone hasn't already marked the |
285 | * request as dirty (in which case we don't care). | |
286 | */ | |
587142f8 | 287 | spin_unlock(&inode->i_lock); |
cfb506e1 TM |
288 | if (!nonblock) |
289 | ret = nfs_wait_on_request(req); | |
290 | else | |
291 | ret = -EAGAIN; | |
e261f51f TM |
292 | nfs_release_request(req); |
293 | if (ret != 0) | |
074cc1de | 294 | return ERR_PTR(ret); |
587142f8 | 295 | spin_lock(&inode->i_lock); |
e261f51f | 296 | } |
587142f8 | 297 | spin_unlock(&inode->i_lock); |
074cc1de TM |
298 | return req; |
299 | } | |
300 | ||
301 | /* | |
302 | * Find an associated nfs write request, and prepare to flush it out | |
303 | * May return an error if the user signalled nfs_wait_on_request(). | |
304 | */ | |
305 | static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, | |
cfb506e1 | 306 | struct page *page, bool nonblock) |
074cc1de TM |
307 | { |
308 | struct nfs_page *req; | |
309 | int ret = 0; | |
310 | ||
cfb506e1 | 311 | req = nfs_find_and_lock_request(page, nonblock); |
074cc1de TM |
312 | if (!req) |
313 | goto out; | |
314 | ret = PTR_ERR(req); | |
315 | if (IS_ERR(req)) | |
316 | goto out; | |
317 | ||
deed85e7 TM |
318 | nfs_set_page_writeback(page); |
319 | WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); | |
074cc1de | 320 | |
deed85e7 | 321 | ret = 0; |
f8512ad0 FI |
322 | if (!nfs_pageio_add_request(pgio, req)) { |
323 | nfs_redirty_request(req); | |
074cc1de | 324 | ret = pgio->pg_error; |
f8512ad0 | 325 | } |
074cc1de TM |
326 | out: |
327 | return ret; | |
e261f51f TM |
328 | } |
329 | ||
f758c885 | 330 | static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio) |
1da177e4 | 331 | { |
d56b4ddf | 332 | struct inode *inode = page_file_mapping(page)->host; |
cfb506e1 | 333 | int ret; |
1da177e4 | 334 | |
91d5b470 CL |
335 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); |
336 | nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1); | |
337 | ||
d56b4ddf | 338 | nfs_pageio_cond_complete(pgio, page_file_index(page)); |
1b430bee | 339 | ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE); |
cfb506e1 TM |
340 | if (ret == -EAGAIN) { |
341 | redirty_page_for_writepage(wbc, page); | |
342 | ret = 0; | |
343 | } | |
344 | return ret; | |
f758c885 | 345 | } |
7fe7f848 | 346 | |
f758c885 TM |
347 | /* |
348 | * Write an mmapped page to the server. | |
349 | */ | |
350 | static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc) | |
351 | { | |
352 | struct nfs_pageio_descriptor pgio; | |
353 | int err; | |
49a70f27 | 354 | |
d56b4ddf | 355 | NFS_PROTO(page_file_mapping(page)->host)->write_pageio_init(&pgio, |
57208fa7 BS |
356 | page->mapping->host, |
357 | wb_priority(wbc), | |
358 | &nfs_async_write_completion_ops); | |
f758c885 TM |
359 | err = nfs_do_writepage(page, wbc, &pgio); |
360 | nfs_pageio_complete(&pgio); | |
361 | if (err < 0) | |
362 | return err; | |
363 | if (pgio.pg_error < 0) | |
364 | return pgio.pg_error; | |
365 | return 0; | |
4d770ccf TM |
366 | } |
367 | ||
368 | int nfs_writepage(struct page *page, struct writeback_control *wbc) | |
369 | { | |
f758c885 | 370 | int ret; |
4d770ccf | 371 | |
f758c885 | 372 | ret = nfs_writepage_locked(page, wbc); |
1da177e4 | 373 | unlock_page(page); |
f758c885 TM |
374 | return ret; |
375 | } | |
376 | ||
377 | static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) | |
378 | { | |
379 | int ret; | |
380 | ||
381 | ret = nfs_do_writepage(page, wbc, data); | |
382 | unlock_page(page); | |
383 | return ret; | |
1da177e4 LT |
384 | } |
385 | ||
1da177e4 LT |
386 | int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) |
387 | { | |
1da177e4 | 388 | struct inode *inode = mapping->host; |
72cb77f4 | 389 | unsigned long *bitlock = &NFS_I(inode)->flags; |
c63c7b05 | 390 | struct nfs_pageio_descriptor pgio; |
1da177e4 LT |
391 | int err; |
392 | ||
72cb77f4 TM |
393 | /* Stop dirtying of new pages while we sync */ |
394 | err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING, | |
395 | nfs_wait_bit_killable, TASK_KILLABLE); | |
396 | if (err) | |
397 | goto out_err; | |
398 | ||
91d5b470 CL |
399 | nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); |
400 | ||
57208fa7 | 401 | NFS_PROTO(inode)->write_pageio_init(&pgio, inode, wb_priority(wbc), &nfs_async_write_completion_ops); |
f758c885 | 402 | err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); |
c63c7b05 | 403 | nfs_pageio_complete(&pgio); |
72cb77f4 TM |
404 | |
405 | clear_bit_unlock(NFS_INO_FLUSHING, bitlock); | |
406 | smp_mb__after_clear_bit(); | |
407 | wake_up_bit(bitlock, NFS_INO_FLUSHING); | |
408 | ||
f758c885 | 409 | if (err < 0) |
72cb77f4 TM |
410 | goto out_err; |
411 | err = pgio.pg_error; | |
412 | if (err < 0) | |
413 | goto out_err; | |
c63c7b05 | 414 | return 0; |
72cb77f4 TM |
415 | out_err: |
416 | return err; | |
1da177e4 LT |
417 | } |
418 | ||
419 | /* | |
420 | * Insert a write request into an inode | |
421 | */ | |
d6d6dc7c | 422 | static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) |
1da177e4 LT |
423 | { |
424 | struct nfs_inode *nfsi = NFS_I(inode); | |
e7d39069 TM |
425 | |
426 | /* Lock the request! */ | |
7ad84aa9 | 427 | nfs_lock_request(req); |
e7d39069 TM |
428 | |
429 | spin_lock(&inode->i_lock); | |
011e2a7f | 430 | if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
a9a4a87a | 431 | inode->i_version++; |
29418aa4 MG |
432 | /* |
433 | * Swap-space should not get truncated. Hence no need to plug the race | |
434 | * with invalidate/truncate. | |
435 | */ | |
436 | if (likely(!PageSwapCache(req->wb_page))) { | |
437 | set_bit(PG_MAPPED, &req->wb_flags); | |
438 | SetPagePrivate(req->wb_page); | |
439 | set_page_private(req->wb_page, (unsigned long)req); | |
440 | } | |
1da177e4 | 441 | nfsi->npages++; |
c03b4024 | 442 | kref_get(&req->wb_kref); |
e7d39069 | 443 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
444 | } |
445 | ||
446 | /* | |
89a09141 | 447 | * Remove a write request from an inode |
1da177e4 LT |
448 | */ |
449 | static void nfs_inode_remove_request(struct nfs_page *req) | |
450 | { | |
3d4ff43d | 451 | struct inode *inode = req->wb_context->dentry->d_inode; |
1da177e4 LT |
452 | struct nfs_inode *nfsi = NFS_I(inode); |
453 | ||
587142f8 | 454 | spin_lock(&inode->i_lock); |
29418aa4 MG |
455 | if (likely(!PageSwapCache(req->wb_page))) { |
456 | set_page_private(req->wb_page, 0); | |
457 | ClearPagePrivate(req->wb_page); | |
458 | clear_bit(PG_MAPPED, &req->wb_flags); | |
459 | } | |
1da177e4 | 460 | nfsi->npages--; |
4d65c520 | 461 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
462 | nfs_release_request(req); |
463 | } | |
464 | ||
61822ab5 | 465 | static void |
6d884e8f | 466 | nfs_mark_request_dirty(struct nfs_page *req) |
61822ab5 | 467 | { |
61822ab5 TM |
468 | __set_page_dirty_nobuffers(req->wb_page); |
469 | } | |
470 | ||
89d77c8f | 471 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
8dd37758 TM |
472 | /** |
473 | * nfs_request_add_commit_list - add request to a commit list | |
474 | * @req: pointer to a struct nfs_page | |
ea2cf228 FI |
475 | * @dst: commit list head |
476 | * @cinfo: holds list lock and accounting info | |
8dd37758 | 477 | * |
ea2cf228 | 478 | * This sets the PG_CLEAN bit, updates the cinfo count of |
8dd37758 TM |
479 | * number of outstanding requests requiring a commit as well as |
480 | * the MM page stats. | |
481 | * | |
ea2cf228 | 482 | * The caller must _not_ hold the cinfo->lock, but must be |
8dd37758 | 483 | * holding the nfs_page lock. |
1da177e4 | 484 | */ |
8dd37758 | 485 | void |
ea2cf228 FI |
486 | nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst, |
487 | struct nfs_commit_info *cinfo) | |
1da177e4 | 488 | { |
e468bae9 | 489 | set_bit(PG_CLEAN, &(req)->wb_flags); |
ea2cf228 FI |
490 | spin_lock(cinfo->lock); |
491 | nfs_list_add_request(req, dst); | |
492 | cinfo->mds->ncommit++; | |
493 | spin_unlock(cinfo->lock); | |
56f9cd68 FI |
494 | if (!cinfo->dreq) { |
495 | inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 496 | inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
497 | BDI_RECLAIMABLE); |
498 | __mark_inode_dirty(req->wb_context->dentry->d_inode, | |
499 | I_DIRTY_DATASYNC); | |
500 | } | |
1da177e4 | 501 | } |
8dd37758 TM |
502 | EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); |
503 | ||
504 | /** | |
505 | * nfs_request_remove_commit_list - Remove request from a commit list | |
506 | * @req: pointer to a nfs_page | |
ea2cf228 | 507 | * @cinfo: holds list lock and accounting info |
8dd37758 | 508 | * |
ea2cf228 | 509 | * This clears the PG_CLEAN bit, and updates the cinfo's count of |
8dd37758 TM |
510 | * number of outstanding requests requiring a commit |
511 | * It does not update the MM page stats. | |
512 | * | |
ea2cf228 | 513 | * The caller _must_ hold the cinfo->lock and the nfs_page lock. |
8dd37758 TM |
514 | */ |
515 | void | |
ea2cf228 FI |
516 | nfs_request_remove_commit_list(struct nfs_page *req, |
517 | struct nfs_commit_info *cinfo) | |
8dd37758 | 518 | { |
8dd37758 TM |
519 | if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) |
520 | return; | |
521 | nfs_list_remove_request(req); | |
ea2cf228 | 522 | cinfo->mds->ncommit--; |
8dd37758 TM |
523 | } |
524 | EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); | |
525 | ||
ea2cf228 FI |
526 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
527 | struct inode *inode) | |
528 | { | |
529 | cinfo->lock = &inode->i_lock; | |
530 | cinfo->mds = &NFS_I(inode)->commit_info; | |
531 | cinfo->ds = pnfs_get_ds_info(inode); | |
b359f9d0 | 532 | cinfo->dreq = NULL; |
f453a54a | 533 | cinfo->completion_ops = &nfs_commit_completion_ops; |
ea2cf228 FI |
534 | } |
535 | ||
536 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
537 | struct inode *inode, | |
538 | struct nfs_direct_req *dreq) | |
539 | { | |
1763da12 FI |
540 | if (dreq) |
541 | nfs_init_cinfo_from_dreq(cinfo, dreq); | |
542 | else | |
543 | nfs_init_cinfo_from_inode(cinfo, inode); | |
ea2cf228 FI |
544 | } |
545 | EXPORT_SYMBOL_GPL(nfs_init_cinfo); | |
8dd37758 TM |
546 | |
547 | /* | |
548 | * Add a request to the inode's commit list. | |
549 | */ | |
1763da12 | 550 | void |
ea2cf228 FI |
551 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
552 | struct nfs_commit_info *cinfo) | |
8dd37758 | 553 | { |
ea2cf228 | 554 | if (pnfs_mark_request_commit(req, lseg, cinfo)) |
8dd37758 | 555 | return; |
ea2cf228 | 556 | nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo); |
8dd37758 | 557 | } |
8e821cad | 558 | |
d6d6dc7c FI |
559 | static void |
560 | nfs_clear_page_commit(struct page *page) | |
561 | { | |
562 | dec_zone_page_state(page, NR_UNSTABLE_NFS); | |
d56b4ddf | 563 | dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE); |
d6d6dc7c FI |
564 | } |
565 | ||
8dd37758 | 566 | static void |
e468bae9 TM |
567 | nfs_clear_request_commit(struct nfs_page *req) |
568 | { | |
8dd37758 TM |
569 | if (test_bit(PG_CLEAN, &req->wb_flags)) { |
570 | struct inode *inode = req->wb_context->dentry->d_inode; | |
ea2cf228 | 571 | struct nfs_commit_info cinfo; |
e468bae9 | 572 | |
ea2cf228 FI |
573 | nfs_init_cinfo_from_inode(&cinfo, inode); |
574 | if (!pnfs_clear_request_commit(req, &cinfo)) { | |
575 | spin_lock(cinfo.lock); | |
576 | nfs_request_remove_commit_list(req, &cinfo); | |
577 | spin_unlock(cinfo.lock); | |
8dd37758 | 578 | } |
d6d6dc7c | 579 | nfs_clear_page_commit(req->wb_page); |
e468bae9 | 580 | } |
e468bae9 TM |
581 | } |
582 | ||
8e821cad TM |
583 | static inline |
584 | int nfs_write_need_commit(struct nfs_write_data *data) | |
585 | { | |
465d5243 | 586 | if (data->verf.committed == NFS_DATA_SYNC) |
cd841605 FI |
587 | return data->header->lseg == NULL; |
588 | return data->verf.committed != NFS_FILE_SYNC; | |
8e821cad TM |
589 | } |
590 | ||
8e821cad | 591 | #else |
68cd6fa4 BS |
592 | static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, |
593 | struct inode *inode) | |
594 | { | |
595 | } | |
596 | ||
597 | void nfs_init_cinfo(struct nfs_commit_info *cinfo, | |
598 | struct inode *inode, | |
599 | struct nfs_direct_req *dreq) | |
600 | { | |
601 | } | |
602 | ||
1763da12 | 603 | void |
ea2cf228 FI |
604 | nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, |
605 | struct nfs_commit_info *cinfo) | |
8e821cad TM |
606 | { |
607 | } | |
608 | ||
8dd37758 | 609 | static void |
e468bae9 TM |
610 | nfs_clear_request_commit(struct nfs_page *req) |
611 | { | |
e468bae9 TM |
612 | } |
613 | ||
8e821cad TM |
614 | static inline |
615 | int nfs_write_need_commit(struct nfs_write_data *data) | |
616 | { | |
617 | return 0; | |
618 | } | |
619 | ||
6c75dc0d FI |
620 | #endif |
621 | ||
061ae2ed | 622 | static void nfs_write_completion(struct nfs_pgio_header *hdr) |
8e821cad | 623 | { |
ea2cf228 | 624 | struct nfs_commit_info cinfo; |
6c75dc0d FI |
625 | unsigned long bytes = 0; |
626 | ||
627 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
628 | goto out; | |
ea2cf228 | 629 | nfs_init_cinfo_from_inode(&cinfo, hdr->inode); |
6c75dc0d FI |
630 | while (!list_empty(&hdr->pages)) { |
631 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
6c75dc0d FI |
632 | |
633 | bytes += req->wb_bytes; | |
634 | nfs_list_remove_request(req); | |
635 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && | |
636 | (hdr->good_bytes < bytes)) { | |
d1182b33 | 637 | nfs_set_pageerror(req->wb_page); |
6c75dc0d FI |
638 | nfs_context_set_write_error(req->wb_context, hdr->error); |
639 | goto remove_req; | |
640 | } | |
641 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { | |
642 | nfs_mark_request_dirty(req); | |
643 | goto next; | |
644 | } | |
645 | if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { | |
2f2c63bc | 646 | memcpy(&req->wb_verf, &hdr->verf->verifier, sizeof(req->wb_verf)); |
ea2cf228 | 647 | nfs_mark_request_commit(req, hdr->lseg, &cinfo); |
6c75dc0d FI |
648 | goto next; |
649 | } | |
650 | remove_req: | |
651 | nfs_inode_remove_request(req); | |
652 | next: | |
1d1afcbc | 653 | nfs_unlock_request(req); |
d1182b33 | 654 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 655 | nfs_release_request(req); |
6c75dc0d FI |
656 | } |
657 | out: | |
658 | hdr->release(hdr); | |
8e821cad | 659 | } |
1da177e4 | 660 | |
89d77c8f | 661 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
ea2cf228 FI |
662 | static unsigned long |
663 | nfs_reqs_to_commit(struct nfs_commit_info *cinfo) | |
fb8a1f11 | 664 | { |
ea2cf228 | 665 | return cinfo->mds->ncommit; |
d6d6dc7c FI |
666 | } |
667 | ||
ea2cf228 | 668 | /* cinfo->lock held by caller */ |
1763da12 | 669 | int |
ea2cf228 FI |
670 | nfs_scan_commit_list(struct list_head *src, struct list_head *dst, |
671 | struct nfs_commit_info *cinfo, int max) | |
d6d6dc7c FI |
672 | { |
673 | struct nfs_page *req, *tmp; | |
674 | int ret = 0; | |
675 | ||
676 | list_for_each_entry_safe(req, tmp, src, wb_list) { | |
8dd37758 TM |
677 | if (!nfs_lock_request(req)) |
678 | continue; | |
7ad84aa9 | 679 | kref_get(&req->wb_kref); |
ea2cf228 | 680 | if (cond_resched_lock(cinfo->lock)) |
3b3be88d | 681 | list_safe_reset_next(req, tmp, wb_list); |
ea2cf228 | 682 | nfs_request_remove_commit_list(req, cinfo); |
8dd37758 TM |
683 | nfs_list_add_request(req, dst); |
684 | ret++; | |
1763da12 | 685 | if ((ret == max) && !cinfo->dreq) |
8dd37758 | 686 | break; |
d6d6dc7c FI |
687 | } |
688 | return ret; | |
fb8a1f11 TM |
689 | } |
690 | ||
1da177e4 LT |
691 | /* |
692 | * nfs_scan_commit - Scan an inode for commit requests | |
693 | * @inode: NFS inode to scan | |
ea2cf228 FI |
694 | * @dst: mds destination list |
695 | * @cinfo: mds and ds lists of reqs ready to commit | |
1da177e4 LT |
696 | * |
697 | * Moves requests from the inode's 'commit' request list. | |
698 | * The requests are *not* checked to ensure that they form a contiguous set. | |
699 | */ | |
1763da12 | 700 | int |
ea2cf228 FI |
701 | nfs_scan_commit(struct inode *inode, struct list_head *dst, |
702 | struct nfs_commit_info *cinfo) | |
1da177e4 | 703 | { |
d6d6dc7c | 704 | int ret = 0; |
fb8a1f11 | 705 | |
ea2cf228 FI |
706 | spin_lock(cinfo->lock); |
707 | if (cinfo->mds->ncommit > 0) { | |
8dd37758 | 708 | const int max = INT_MAX; |
d6d6dc7c | 709 | |
ea2cf228 FI |
710 | ret = nfs_scan_commit_list(&cinfo->mds->list, dst, |
711 | cinfo, max); | |
712 | ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); | |
d6d6dc7c | 713 | } |
ea2cf228 | 714 | spin_unlock(cinfo->lock); |
ff778d02 | 715 | return ret; |
1da177e4 | 716 | } |
d6d6dc7c | 717 | |
c42de9dd | 718 | #else |
ea2cf228 | 719 | static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo) |
fb8a1f11 TM |
720 | { |
721 | return 0; | |
722 | } | |
723 | ||
1763da12 FI |
724 | int nfs_scan_commit(struct inode *inode, struct list_head *dst, |
725 | struct nfs_commit_info *cinfo) | |
c42de9dd TM |
726 | { |
727 | return 0; | |
728 | } | |
1da177e4 LT |
729 | #endif |
730 | ||
1da177e4 | 731 | /* |
e7d39069 TM |
732 | * Search for an existing write request, and attempt to update |
733 | * it to reflect a new dirty region on a given page. | |
1da177e4 | 734 | * |
e7d39069 TM |
735 | * If the attempt fails, then the existing request is flushed out |
736 | * to disk. | |
1da177e4 | 737 | */ |
e7d39069 TM |
738 | static struct nfs_page *nfs_try_to_update_request(struct inode *inode, |
739 | struct page *page, | |
740 | unsigned int offset, | |
741 | unsigned int bytes) | |
1da177e4 | 742 | { |
e7d39069 TM |
743 | struct nfs_page *req; |
744 | unsigned int rqend; | |
745 | unsigned int end; | |
746 | int error; | |
747 | ||
748 | if (!PagePrivate(page)) | |
749 | return NULL; | |
1da177e4 LT |
750 | |
751 | end = offset + bytes; | |
e7d39069 | 752 | spin_lock(&inode->i_lock); |
1da177e4 | 753 | |
1da177e4 | 754 | for (;;) { |
29418aa4 | 755 | req = nfs_page_find_request_locked(NFS_I(inode), page); |
e7d39069 TM |
756 | if (req == NULL) |
757 | goto out_unlock; | |
758 | ||
759 | rqend = req->wb_offset + req->wb_bytes; | |
760 | /* | |
761 | * Tell the caller to flush out the request if | |
762 | * the offsets are non-contiguous. | |
763 | * Note: nfs_flush_incompatible() will already | |
764 | * have flushed out requests having wrong owners. | |
765 | */ | |
e468bae9 | 766 | if (offset > rqend |
e7d39069 TM |
767 | || end < req->wb_offset) |
768 | goto out_flushme; | |
769 | ||
7ad84aa9 | 770 | if (nfs_lock_request(req)) |
1da177e4 | 771 | break; |
1da177e4 | 772 | |
e7d39069 | 773 | /* The request is locked, so wait and then retry */ |
587142f8 | 774 | spin_unlock(&inode->i_lock); |
e7d39069 TM |
775 | error = nfs_wait_on_request(req); |
776 | nfs_release_request(req); | |
777 | if (error != 0) | |
778 | goto out_err; | |
779 | spin_lock(&inode->i_lock); | |
1da177e4 LT |
780 | } |
781 | ||
782 | /* Okay, the request matches. Update the region */ | |
783 | if (offset < req->wb_offset) { | |
784 | req->wb_offset = offset; | |
785 | req->wb_pgbase = offset; | |
1da177e4 | 786 | } |
1da177e4 LT |
787 | if (end > rqend) |
788 | req->wb_bytes = end - req->wb_offset; | |
e7d39069 TM |
789 | else |
790 | req->wb_bytes = rqend - req->wb_offset; | |
791 | out_unlock: | |
792 | spin_unlock(&inode->i_lock); | |
ca138f36 FI |
793 | if (req) |
794 | nfs_clear_request_commit(req); | |
e7d39069 TM |
795 | return req; |
796 | out_flushme: | |
797 | spin_unlock(&inode->i_lock); | |
798 | nfs_release_request(req); | |
799 | error = nfs_wb_page(inode, page); | |
800 | out_err: | |
801 | return ERR_PTR(error); | |
802 | } | |
803 | ||
804 | /* | |
805 | * Try to update an existing write request, or create one if there is none. | |
806 | * | |
807 | * Note: Should always be called with the Page Lock held to prevent races | |
808 | * if we have to add a new request. Also assumes that the caller has | |
809 | * already called nfs_flush_incompatible() if necessary. | |
810 | */ | |
811 | static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, | |
812 | struct page *page, unsigned int offset, unsigned int bytes) | |
813 | { | |
d56b4ddf | 814 | struct inode *inode = page_file_mapping(page)->host; |
e7d39069 | 815 | struct nfs_page *req; |
1da177e4 | 816 | |
e7d39069 TM |
817 | req = nfs_try_to_update_request(inode, page, offset, bytes); |
818 | if (req != NULL) | |
819 | goto out; | |
820 | req = nfs_create_request(ctx, inode, page, offset, bytes); | |
821 | if (IS_ERR(req)) | |
822 | goto out; | |
d6d6dc7c | 823 | nfs_inode_add_request(inode, req); |
efc91ed0 | 824 | out: |
61e930a9 | 825 | return req; |
1da177e4 LT |
826 | } |
827 | ||
e7d39069 TM |
828 | static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, |
829 | unsigned int offset, unsigned int count) | |
830 | { | |
831 | struct nfs_page *req; | |
832 | ||
833 | req = nfs_setup_write_request(ctx, page, offset, count); | |
834 | if (IS_ERR(req)) | |
835 | return PTR_ERR(req); | |
836 | /* Update file length */ | |
837 | nfs_grow_file(page, offset, count); | |
838 | nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes); | |
a6305ddb | 839 | nfs_mark_request_dirty(req); |
1d1afcbc | 840 | nfs_unlock_and_release_request(req); |
e7d39069 TM |
841 | return 0; |
842 | } | |
843 | ||
1da177e4 LT |
844 | int nfs_flush_incompatible(struct file *file, struct page *page) |
845 | { | |
cd3758e3 | 846 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
2a369153 | 847 | struct nfs_lock_context *l_ctx; |
1da177e4 | 848 | struct nfs_page *req; |
1a54533e | 849 | int do_flush, status; |
1da177e4 LT |
850 | /* |
851 | * Look for a request corresponding to this page. If there | |
852 | * is one, and it belongs to another file, we flush it out | |
853 | * before we try to copy anything into the page. Do this | |
854 | * due to the lack of an ACCESS-type call in NFSv2. | |
855 | * Also do the same if we find a request from an existing | |
856 | * dropped page. | |
857 | */ | |
1a54533e TM |
858 | do { |
859 | req = nfs_page_find_request(page); | |
860 | if (req == NULL) | |
861 | return 0; | |
2a369153 TM |
862 | l_ctx = req->wb_lock_context; |
863 | do_flush = req->wb_page != page || req->wb_context != ctx; | |
864 | if (l_ctx) { | |
865 | do_flush |= l_ctx->lockowner.l_owner != current->files | |
866 | || l_ctx->lockowner.l_pid != current->tgid; | |
867 | } | |
1da177e4 | 868 | nfs_release_request(req); |
1a54533e TM |
869 | if (!do_flush) |
870 | return 0; | |
d56b4ddf | 871 | status = nfs_wb_page(page_file_mapping(page)->host, page); |
1a54533e TM |
872 | } while (status == 0); |
873 | return status; | |
1da177e4 LT |
874 | } |
875 | ||
5d47a356 TM |
876 | /* |
877 | * If the page cache is marked as unsafe or invalid, then we can't rely on | |
878 | * the PageUptodate() flag. In this case, we will need to turn off | |
879 | * write optimisations that depend on the page contents being correct. | |
880 | */ | |
8d197a56 | 881 | static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) |
5d47a356 | 882 | { |
8d197a56 TM |
883 | if (nfs_have_delegated_attributes(inode)) |
884 | goto out; | |
81d9bce5 | 885 | if (NFS_I(inode)->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE)) |
8d197a56 TM |
886 | return false; |
887 | out: | |
888 | return PageUptodate(page) != 0; | |
5d47a356 TM |
889 | } |
890 | ||
c7559663 SM |
891 | /* If we know the page is up to date, and we're not using byte range locks (or |
892 | * if we have the whole file locked for writing), it may be more efficient to | |
893 | * extend the write to cover the entire page in order to avoid fragmentation | |
894 | * inefficiencies. | |
895 | * | |
896 | * If the file is opened for synchronous writes or if we have a write delegation | |
897 | * from the server then we can just skip the rest of the checks. | |
898 | */ | |
899 | static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) | |
900 | { | |
901 | if (file->f_flags & O_DSYNC) | |
902 | return 0; | |
903 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) | |
904 | return 1; | |
905 | if (nfs_write_pageuptodate(page, inode) && (inode->i_flock == NULL || | |
906 | (inode->i_flock->fl_start == 0 && | |
907 | inode->i_flock->fl_end == OFFSET_MAX && | |
908 | inode->i_flock->fl_type != F_RDLCK))) | |
909 | return 1; | |
910 | return 0; | |
911 | } | |
912 | ||
1da177e4 LT |
913 | /* |
914 | * Update and possibly write a cached page of an NFS file. | |
915 | * | |
916 | * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad | |
917 | * things with a page scheduled for an RPC call (e.g. invalidate it). | |
918 | */ | |
919 | int nfs_updatepage(struct file *file, struct page *page, | |
920 | unsigned int offset, unsigned int count) | |
921 | { | |
cd3758e3 | 922 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
d56b4ddf | 923 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
924 | int status = 0; |
925 | ||
91d5b470 CL |
926 | nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); |
927 | ||
48186c7d | 928 | dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n", |
01cce933 JJS |
929 | file->f_path.dentry->d_parent->d_name.name, |
930 | file->f_path.dentry->d_name.name, count, | |
d56b4ddf | 931 | (long long)(page_file_offset(page) + offset)); |
1da177e4 | 932 | |
c7559663 | 933 | if (nfs_can_extend_write(file, page, inode)) { |
49a70f27 | 934 | count = max(count + offset, nfs_page_length(page)); |
1da177e4 | 935 | offset = 0; |
1da177e4 LT |
936 | } |
937 | ||
e21195a7 | 938 | status = nfs_writepage_setup(ctx, page, offset, count); |
03fa9e84 TM |
939 | if (status < 0) |
940 | nfs_set_pageerror(page); | |
59b7c05f TM |
941 | else |
942 | __set_page_dirty_nobuffers(page); | |
1da177e4 | 943 | |
48186c7d | 944 | dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", |
1da177e4 | 945 | status, (long long)i_size_read(inode)); |
1da177e4 LT |
946 | return status; |
947 | } | |
948 | ||
3ff7576d | 949 | static int flush_task_priority(int how) |
1da177e4 LT |
950 | { |
951 | switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { | |
952 | case FLUSH_HIGHPRI: | |
953 | return RPC_PRIORITY_HIGH; | |
954 | case FLUSH_LOWPRI: | |
955 | return RPC_PRIORITY_LOW; | |
956 | } | |
957 | return RPC_PRIORITY_NORMAL; | |
958 | } | |
959 | ||
c5996c4e FI |
960 | int nfs_initiate_write(struct rpc_clnt *clnt, |
961 | struct nfs_write_data *data, | |
d138d5d1 | 962 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 963 | int how, int flags) |
1da177e4 | 964 | { |
cd841605 | 965 | struct inode *inode = data->header->inode; |
3ff7576d | 966 | int priority = flush_task_priority(how); |
07737691 | 967 | struct rpc_task *task; |
bdc7f021 TM |
968 | struct rpc_message msg = { |
969 | .rpc_argp = &data->args, | |
970 | .rpc_resp = &data->res, | |
cd841605 | 971 | .rpc_cred = data->header->cred, |
bdc7f021 | 972 | }; |
84115e1c | 973 | struct rpc_task_setup task_setup_data = { |
d138d5d1 | 974 | .rpc_client = clnt, |
07737691 | 975 | .task = &data->task, |
bdc7f021 | 976 | .rpc_message = &msg, |
84115e1c TM |
977 | .callback_ops = call_ops, |
978 | .callback_data = data, | |
101070ca | 979 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 980 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 981 | .priority = priority, |
84115e1c | 982 | }; |
2c61be0a | 983 | int ret = 0; |
1da177e4 | 984 | |
d138d5d1 AA |
985 | /* Set up the initial task struct. */ |
986 | NFS_PROTO(inode)->write_setup(data, &msg); | |
987 | ||
988 | dprintk("NFS: %5u initiated write call " | |
989 | "(req %s/%lld, %u bytes @ offset %llu)\n", | |
990 | data->task.tk_pid, | |
991 | inode->i_sb->s_id, | |
992 | (long long)NFS_FILEID(inode), | |
993 | data->args.count, | |
994 | (unsigned long long)data->args.offset); | |
995 | ||
996 | task = rpc_run_task(&task_setup_data); | |
997 | if (IS_ERR(task)) { | |
998 | ret = PTR_ERR(task); | |
999 | goto out; | |
1000 | } | |
1001 | if (how & FLUSH_SYNC) { | |
1002 | ret = rpc_wait_for_completion_task(task); | |
1003 | if (ret == 0) | |
1004 | ret = task->tk_status; | |
1005 | } | |
1006 | rpc_put_task(task); | |
1007 | out: | |
1008 | return ret; | |
1009 | } | |
a69aef14 | 1010 | EXPORT_SYMBOL_GPL(nfs_initiate_write); |
d138d5d1 AA |
1011 | |
1012 | /* | |
1013 | * Set up the argument/result storage required for the RPC call. | |
1014 | */ | |
6c75dc0d | 1015 | static void nfs_write_rpcsetup(struct nfs_write_data *data, |
d138d5d1 | 1016 | unsigned int count, unsigned int offset, |
ea2cf228 | 1017 | int how, struct nfs_commit_info *cinfo) |
d138d5d1 | 1018 | { |
6c75dc0d | 1019 | struct nfs_page *req = data->header->req; |
d138d5d1 | 1020 | |
1da177e4 LT |
1021 | /* Set up the RPC argument and reply structs |
1022 | * NB: take care not to mess about with data->commit et al. */ | |
1023 | ||
6c75dc0d | 1024 | data->args.fh = NFS_FH(data->header->inode); |
1da177e4 | 1025 | data->args.offset = req_offset(req) + offset; |
2bea038c BH |
1026 | /* pnfs_set_layoutcommit needs this */ |
1027 | data->mds_offset = data->args.offset; | |
1da177e4 | 1028 | data->args.pgbase = req->wb_pgbase + offset; |
30dd374f | 1029 | data->args.pages = data->pages.pagevec; |
1da177e4 | 1030 | data->args.count = count; |
383ba719 | 1031 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 1032 | data->args.lock_context = req->wb_lock_context; |
bdc7f021 | 1033 | data->args.stable = NFS_UNSTABLE; |
87ed5eb4 TM |
1034 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { |
1035 | case 0: | |
1036 | break; | |
1037 | case FLUSH_COND_STABLE: | |
ea2cf228 | 1038 | if (nfs_reqs_to_commit(cinfo)) |
87ed5eb4 TM |
1039 | break; |
1040 | default: | |
1041 | data->args.stable = NFS_FILE_SYNC; | |
bdc7f021 | 1042 | } |
1da177e4 LT |
1043 | |
1044 | data->res.fattr = &data->fattr; | |
1045 | data->res.count = count; | |
1046 | data->res.verf = &data->verf; | |
0e574af1 | 1047 | nfs_fattr_init(&data->fattr); |
6e4efd56 | 1048 | } |
1da177e4 | 1049 | |
6e4efd56 TM |
1050 | static int nfs_do_write(struct nfs_write_data *data, |
1051 | const struct rpc_call_ops *call_ops, | |
6e4efd56 TM |
1052 | int how) |
1053 | { | |
cd841605 | 1054 | struct inode *inode = data->header->inode; |
0382b744 | 1055 | |
9f0ec176 | 1056 | return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how, 0); |
1da177e4 LT |
1057 | } |
1058 | ||
275acaaf TM |
1059 | static int nfs_do_multiple_writes(struct list_head *head, |
1060 | const struct rpc_call_ops *call_ops, | |
275acaaf TM |
1061 | int how) |
1062 | { | |
1063 | struct nfs_write_data *data; | |
1064 | int ret = 0; | |
1065 | ||
1066 | while (!list_empty(head)) { | |
1067 | int ret2; | |
1068 | ||
6c75dc0d | 1069 | data = list_first_entry(head, struct nfs_write_data, list); |
275acaaf TM |
1070 | list_del_init(&data->list); |
1071 | ||
dce81290 | 1072 | ret2 = nfs_do_write(data, call_ops, how); |
275acaaf TM |
1073 | if (ret == 0) |
1074 | ret = ret2; | |
1075 | } | |
1076 | return ret; | |
1077 | } | |
1078 | ||
6d884e8f F |
1079 | /* If a nfs_flush_* function fails, it should remove reqs from @head and |
1080 | * call this on each, which will prepare them to be retried on next | |
1081 | * writeback using standard nfs. | |
1082 | */ | |
1083 | static void nfs_redirty_request(struct nfs_page *req) | |
1084 | { | |
1085 | nfs_mark_request_dirty(req); | |
1d1afcbc | 1086 | nfs_unlock_request(req); |
d1182b33 | 1087 | nfs_end_page_writeback(req->wb_page); |
3aff4ebb | 1088 | nfs_release_request(req); |
6d884e8f F |
1089 | } |
1090 | ||
061ae2ed | 1091 | static void nfs_async_write_error(struct list_head *head) |
6c75dc0d FI |
1092 | { |
1093 | struct nfs_page *req; | |
1094 | ||
1095 | while (!list_empty(head)) { | |
1096 | req = nfs_list_entry(head->next); | |
1097 | nfs_list_remove_request(req); | |
1098 | nfs_redirty_request(req); | |
1099 | } | |
1100 | } | |
1101 | ||
061ae2ed FI |
1102 | static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { |
1103 | .error_cleanup = nfs_async_write_error, | |
1104 | .completion = nfs_write_completion, | |
1105 | }; | |
1106 | ||
25b11dcd TM |
1107 | static void nfs_flush_error(struct nfs_pageio_descriptor *desc, |
1108 | struct nfs_pgio_header *hdr) | |
1109 | { | |
1110 | set_bit(NFS_IOHDR_REDO, &hdr->flags); | |
1111 | while (!list_empty(&hdr->rpc_list)) { | |
1112 | struct nfs_write_data *data = list_first_entry(&hdr->rpc_list, | |
1113 | struct nfs_write_data, list); | |
1114 | list_del(&data->list); | |
1115 | nfs_writedata_release(data); | |
1116 | } | |
1117 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); | |
1118 | } | |
1119 | ||
1da177e4 LT |
1120 | /* |
1121 | * Generate multiple small requests to write out a single | |
1122 | * contiguous dirty area on one page. | |
1123 | */ | |
6c75dc0d FI |
1124 | static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, |
1125 | struct nfs_pgio_header *hdr) | |
1da177e4 | 1126 | { |
6c75dc0d | 1127 | struct nfs_page *req = hdr->req; |
1da177e4 LT |
1128 | struct page *page = req->wb_page; |
1129 | struct nfs_write_data *data; | |
d097971d | 1130 | size_t wsize = desc->pg_bsize, nbytes; |
e9f7bee1 | 1131 | unsigned int offset; |
1da177e4 | 1132 | int requests = 0; |
ea2cf228 | 1133 | struct nfs_commit_info cinfo; |
1da177e4 | 1134 | |
ea2cf228 | 1135 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
1da177e4 | 1136 | |
b31268ac | 1137 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1138 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || |
b31268ac TM |
1139 | desc->pg_count > wsize)) |
1140 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
1141 | ||
1142 | ||
275acaaf | 1143 | offset = 0; |
c76069bd | 1144 | nbytes = desc->pg_count; |
e9f7bee1 TM |
1145 | do { |
1146 | size_t len = min(nbytes, wsize); | |
1147 | ||
6c75dc0d | 1148 | data = nfs_writedata_alloc(hdr, 1); |
25b11dcd TM |
1149 | if (!data) { |
1150 | nfs_flush_error(desc, hdr); | |
1151 | return -ENOMEM; | |
1152 | } | |
30dd374f | 1153 | data->pages.pagevec[0] = page; |
ea2cf228 | 1154 | nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); |
6c75dc0d | 1155 | list_add(&data->list, &hdr->rpc_list); |
1da177e4 | 1156 | requests++; |
e9f7bee1 | 1157 | nbytes -= len; |
275acaaf | 1158 | offset += len; |
e9f7bee1 | 1159 | } while (nbytes != 0); |
25b11dcd TM |
1160 | nfs_list_remove_request(req); |
1161 | nfs_list_add_request(req, &hdr->pages); | |
6c75dc0d | 1162 | desc->pg_rpc_callops = &nfs_write_common_ops; |
25b11dcd | 1163 | return 0; |
1da177e4 LT |
1164 | } |
1165 | ||
1166 | /* | |
1167 | * Create an RPC task for the given write request and kick it. | |
1168 | * The page must have been locked by the caller. | |
1169 | * | |
1170 | * It may happen that the page we're passed is not marked dirty. | |
1171 | * This is the case if nfs_updatepage detects a conflicting request | |
1172 | * that has been written but not committed. | |
1173 | */ | |
6c75dc0d FI |
1174 | static int nfs_flush_one(struct nfs_pageio_descriptor *desc, |
1175 | struct nfs_pgio_header *hdr) | |
1da177e4 LT |
1176 | { |
1177 | struct nfs_page *req; | |
1178 | struct page **pages; | |
1179 | struct nfs_write_data *data; | |
c76069bd | 1180 | struct list_head *head = &desc->pg_list; |
ea2cf228 | 1181 | struct nfs_commit_info cinfo; |
1da177e4 | 1182 | |
6c75dc0d FI |
1183 | data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base, |
1184 | desc->pg_count)); | |
1185 | if (!data) { | |
25b11dcd TM |
1186 | nfs_flush_error(desc, hdr); |
1187 | return -ENOMEM; | |
44b83799 | 1188 | } |
6c75dc0d | 1189 | |
ea2cf228 | 1190 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); |
30dd374f | 1191 | pages = data->pages.pagevec; |
1da177e4 LT |
1192 | while (!list_empty(head)) { |
1193 | req = nfs_list_entry(head->next); | |
1194 | nfs_list_remove_request(req); | |
6c75dc0d | 1195 | nfs_list_add_request(req, &hdr->pages); |
1da177e4 | 1196 | *pages++ = req->wb_page; |
1da177e4 | 1197 | } |
1da177e4 | 1198 | |
b31268ac | 1199 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && |
ea2cf228 | 1200 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) |
b31268ac TM |
1201 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; |
1202 | ||
1da177e4 | 1203 | /* Set up the argument struct */ |
ea2cf228 | 1204 | nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); |
6c75dc0d FI |
1205 | list_add(&data->list, &hdr->rpc_list); |
1206 | desc->pg_rpc_callops = &nfs_write_common_ops; | |
25b11dcd | 1207 | return 0; |
1da177e4 LT |
1208 | } |
1209 | ||
6c75dc0d FI |
1210 | int nfs_generic_flush(struct nfs_pageio_descriptor *desc, |
1211 | struct nfs_pgio_header *hdr) | |
dce81290 TM |
1212 | { |
1213 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
6c75dc0d FI |
1214 | return nfs_flush_multi(desc, hdr); |
1215 | return nfs_flush_one(desc, hdr); | |
dce81290 | 1216 | } |
89d77c8f | 1217 | EXPORT_SYMBOL_GPL(nfs_generic_flush); |
dce81290 TM |
1218 | |
1219 | static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc) | |
1751c363 | 1220 | { |
6c75dc0d FI |
1221 | struct nfs_write_header *whdr; |
1222 | struct nfs_pgio_header *hdr; | |
275acaaf TM |
1223 | int ret; |
1224 | ||
6c75dc0d FI |
1225 | whdr = nfs_writehdr_alloc(); |
1226 | if (!whdr) { | |
9b5415b5 | 1227 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); |
6c75dc0d FI |
1228 | return -ENOMEM; |
1229 | } | |
1230 | hdr = &whdr->header; | |
1231 | nfs_pgheader_init(desc, hdr, nfs_writehdr_free); | |
1232 | atomic_inc(&hdr->refcnt); | |
1233 | ret = nfs_generic_flush(desc, hdr); | |
50828d7e | 1234 | if (ret == 0) |
6c75dc0d FI |
1235 | ret = nfs_do_multiple_writes(&hdr->rpc_list, |
1236 | desc->pg_rpc_callops, | |
1237 | desc->pg_ioflags); | |
6c75dc0d | 1238 | if (atomic_dec_and_test(&hdr->refcnt)) |
061ae2ed | 1239 | hdr->completion_ops->completion(hdr); |
275acaaf | 1240 | return ret; |
1751c363 | 1241 | } |
1751c363 TM |
1242 | |
1243 | static const struct nfs_pageio_ops nfs_pageio_write_ops = { | |
1244 | .pg_test = nfs_generic_pg_test, | |
1245 | .pg_doio = nfs_generic_pg_writepages, | |
1246 | }; | |
1247 | ||
57208fa7 | 1248 | void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, |
061ae2ed FI |
1249 | struct inode *inode, int ioflags, |
1250 | const struct nfs_pgio_completion_ops *compl_ops) | |
1da177e4 | 1251 | { |
061ae2ed | 1252 | nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops, compl_ops, |
1751c363 TM |
1253 | NFS_SERVER(inode)->wsize, ioflags); |
1254 | } | |
ddda8e0a | 1255 | EXPORT_SYMBOL_GPL(nfs_pageio_init_write); |
1da177e4 | 1256 | |
dce81290 TM |
1257 | void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) |
1258 | { | |
1259 | pgio->pg_ops = &nfs_pageio_write_ops; | |
1260 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; | |
1261 | } | |
1f945357 | 1262 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); |
dce81290 | 1263 | |
1da177e4 | 1264 | |
def6ed7e AA |
1265 | void nfs_write_prepare(struct rpc_task *task, void *calldata) |
1266 | { | |
1267 | struct nfs_write_data *data = calldata; | |
cd841605 | 1268 | NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data); |
c58c8441 TM |
1269 | if (unlikely(test_bit(NFS_CONTEXT_BAD, &data->args.context->flags))) |
1270 | rpc_exit(task, -EIO); | |
def6ed7e | 1271 | } |
def6ed7e | 1272 | |
0b7c0153 FI |
1273 | void nfs_commit_prepare(struct rpc_task *task, void *calldata) |
1274 | { | |
1275 | struct nfs_commit_data *data = calldata; | |
1276 | ||
1277 | NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); | |
1278 | } | |
1279 | ||
1da177e4 LT |
1280 | /* |
1281 | * Handle a write reply that flushes a whole page. | |
1282 | * | |
1283 | * FIXME: There is an inherent race with invalidate_inode_pages and | |
1284 | * writebacks since the page->count is kept > 1 for as long | |
1285 | * as the page has a write request pending. | |
1286 | */ | |
6c75dc0d | 1287 | static void nfs_writeback_done_common(struct rpc_task *task, void *calldata) |
1da177e4 | 1288 | { |
788e7a89 | 1289 | struct nfs_write_data *data = calldata; |
1da177e4 | 1290 | |
c9d8f89d TM |
1291 | nfs_writeback_done(task, data); |
1292 | } | |
1293 | ||
6c75dc0d | 1294 | static void nfs_writeback_release_common(void *calldata) |
c9d8f89d TM |
1295 | { |
1296 | struct nfs_write_data *data = calldata; | |
cd841605 | 1297 | struct nfs_pgio_header *hdr = data->header; |
e2fecb21 | 1298 | int status = data->task.tk_status; |
788e7a89 | 1299 | |
6c75dc0d FI |
1300 | if ((status >= 0) && nfs_write_need_commit(data)) { |
1301 | spin_lock(&hdr->lock); | |
1302 | if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) | |
1303 | ; /* Do nothing */ | |
1304 | else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) | |
9bce008b TM |
1305 | memcpy(hdr->verf, &data->verf, sizeof(*hdr->verf)); |
1306 | else if (memcmp(hdr->verf, &data->verf, sizeof(*hdr->verf))) | |
6c75dc0d FI |
1307 | set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags); |
1308 | spin_unlock(&hdr->lock); | |
1da177e4 | 1309 | } |
cd841605 | 1310 | nfs_writedata_release(data); |
1da177e4 LT |
1311 | } |
1312 | ||
6c75dc0d | 1313 | static const struct rpc_call_ops nfs_write_common_ops = { |
def6ed7e | 1314 | .rpc_call_prepare = nfs_write_prepare, |
6c75dc0d FI |
1315 | .rpc_call_done = nfs_writeback_done_common, |
1316 | .rpc_release = nfs_writeback_release_common, | |
788e7a89 TM |
1317 | }; |
1318 | ||
1319 | ||
1da177e4 LT |
1320 | /* |
1321 | * This function is called when the WRITE call is complete. | |
1322 | */ | |
13602896 | 1323 | void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data) |
1da177e4 | 1324 | { |
1da177e4 LT |
1325 | struct nfs_writeargs *argp = &data->args; |
1326 | struct nfs_writeres *resp = &data->res; | |
cd841605 | 1327 | struct inode *inode = data->header->inode; |
788e7a89 | 1328 | int status; |
1da177e4 | 1329 | |
a3f565b1 | 1330 | dprintk("NFS: %5u nfs_writeback_done (status %d)\n", |
1da177e4 LT |
1331 | task->tk_pid, task->tk_status); |
1332 | ||
f551e44f CL |
1333 | /* |
1334 | * ->write_done will attempt to use post-op attributes to detect | |
1335 | * conflicting writes by other clients. A strict interpretation | |
1336 | * of close-to-open would allow us to continue caching even if | |
1337 | * another writer had changed the file, but some applications | |
1338 | * depend on tighter cache coherency when writing. | |
1339 | */ | |
cd841605 | 1340 | status = NFS_PROTO(inode)->write_done(task, data); |
788e7a89 | 1341 | if (status != 0) |
13602896 | 1342 | return; |
cd841605 | 1343 | nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count); |
91d5b470 | 1344 | |
89d77c8f | 1345 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
1da177e4 LT |
1346 | if (resp->verf->committed < argp->stable && task->tk_status >= 0) { |
1347 | /* We tried a write call, but the server did not | |
1348 | * commit data to stable storage even though we | |
1349 | * requested it. | |
1350 | * Note: There is a known bug in Tru64 < 5.0 in which | |
1351 | * the server reports NFS_DATA_SYNC, but performs | |
1352 | * NFS_FILE_SYNC. We therefore implement this checking | |
1353 | * as a dprintk() in order to avoid filling syslog. | |
1354 | */ | |
1355 | static unsigned long complain; | |
1356 | ||
a69aef14 | 1357 | /* Note this will print the MDS for a DS write */ |
1da177e4 | 1358 | if (time_before(complain, jiffies)) { |
48186c7d | 1359 | dprintk("NFS: faulty NFS server %s:" |
1da177e4 | 1360 | " (committed = %d) != (stable = %d)\n", |
cd841605 | 1361 | NFS_SERVER(inode)->nfs_client->cl_hostname, |
1da177e4 LT |
1362 | resp->verf->committed, argp->stable); |
1363 | complain = jiffies + 300 * HZ; | |
1364 | } | |
1365 | } | |
1366 | #endif | |
6c75dc0d FI |
1367 | if (task->tk_status < 0) |
1368 | nfs_set_pgio_error(data->header, task->tk_status, argp->offset); | |
1369 | else if (resp->count < argp->count) { | |
1da177e4 LT |
1370 | static unsigned long complain; |
1371 | ||
6c75dc0d | 1372 | /* This a short write! */ |
cd841605 | 1373 | nfs_inc_stats(inode, NFSIOS_SHORTWRITE); |
91d5b470 | 1374 | |
1da177e4 | 1375 | /* Has the server at least made some progress? */ |
6c75dc0d FI |
1376 | if (resp->count == 0) { |
1377 | if (time_before(complain, jiffies)) { | |
1378 | printk(KERN_WARNING | |
1379 | "NFS: Server wrote zero bytes, expected %u.\n", | |
1380 | argp->count); | |
1381 | complain = jiffies + 300 * HZ; | |
1da177e4 | 1382 | } |
6c75dc0d FI |
1383 | nfs_set_pgio_error(data->header, -EIO, argp->offset); |
1384 | task->tk_status = -EIO; | |
13602896 | 1385 | return; |
1da177e4 | 1386 | } |
6c75dc0d FI |
1387 | /* Was this an NFSv2 write or an NFSv3 stable write? */ |
1388 | if (resp->verf->committed != NFS_UNSTABLE) { | |
1389 | /* Resend from where the server left off */ | |
1390 | data->mds_offset += resp->count; | |
1391 | argp->offset += resp->count; | |
1392 | argp->pgbase += resp->count; | |
1393 | argp->count -= resp->count; | |
1394 | } else { | |
1395 | /* Resend as a stable write in order to avoid | |
1396 | * headaches in the case of a server crash. | |
1397 | */ | |
1398 | argp->stable = NFS_FILE_SYNC; | |
1da177e4 | 1399 | } |
6c75dc0d | 1400 | rpc_restart_call_prepare(task); |
1da177e4 | 1401 | } |
1da177e4 LT |
1402 | } |
1403 | ||
1404 | ||
89d77c8f | 1405 | #if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4) |
71d0a611 TM |
1406 | static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait) |
1407 | { | |
b8413f98 TM |
1408 | int ret; |
1409 | ||
71d0a611 TM |
1410 | if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags)) |
1411 | return 1; | |
b8413f98 TM |
1412 | if (!may_wait) |
1413 | return 0; | |
1414 | ret = out_of_line_wait_on_bit_lock(&nfsi->flags, | |
1415 | NFS_INO_COMMIT, | |
1416 | nfs_wait_bit_killable, | |
1417 | TASK_KILLABLE); | |
1418 | return (ret < 0) ? ret : 1; | |
71d0a611 TM |
1419 | } |
1420 | ||
f453a54a | 1421 | static void nfs_commit_clear_lock(struct nfs_inode *nfsi) |
71d0a611 TM |
1422 | { |
1423 | clear_bit(NFS_INO_COMMIT, &nfsi->flags); | |
1424 | smp_mb__after_clear_bit(); | |
1425 | wake_up_bit(&nfsi->flags, NFS_INO_COMMIT); | |
1426 | } | |
1427 | ||
0b7c0153 | 1428 | void nfs_commitdata_release(struct nfs_commit_data *data) |
1da177e4 | 1429 | { |
0b7c0153 FI |
1430 | put_nfs_open_context(data->context); |
1431 | nfs_commit_free(data); | |
1da177e4 | 1432 | } |
e0c2b380 | 1433 | EXPORT_SYMBOL_GPL(nfs_commitdata_release); |
1da177e4 | 1434 | |
0b7c0153 | 1435 | int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, |
9ace33cd | 1436 | const struct rpc_call_ops *call_ops, |
9f0ec176 | 1437 | int how, int flags) |
1da177e4 | 1438 | { |
07737691 | 1439 | struct rpc_task *task; |
9ace33cd | 1440 | int priority = flush_task_priority(how); |
bdc7f021 TM |
1441 | struct rpc_message msg = { |
1442 | .rpc_argp = &data->args, | |
1443 | .rpc_resp = &data->res, | |
9ace33cd | 1444 | .rpc_cred = data->cred, |
bdc7f021 | 1445 | }; |
84115e1c | 1446 | struct rpc_task_setup task_setup_data = { |
07737691 | 1447 | .task = &data->task, |
9ace33cd | 1448 | .rpc_client = clnt, |
bdc7f021 | 1449 | .rpc_message = &msg, |
9ace33cd | 1450 | .callback_ops = call_ops, |
84115e1c | 1451 | .callback_data = data, |
101070ca | 1452 | .workqueue = nfsiod_workqueue, |
9f0ec176 | 1453 | .flags = RPC_TASK_ASYNC | flags, |
3ff7576d | 1454 | .priority = priority, |
84115e1c | 1455 | }; |
9ace33cd FI |
1456 | /* Set up the initial task struct. */ |
1457 | NFS_PROTO(data->inode)->commit_setup(data, &msg); | |
1458 | ||
1459 | dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid); | |
1460 | ||
1461 | task = rpc_run_task(&task_setup_data); | |
1462 | if (IS_ERR(task)) | |
1463 | return PTR_ERR(task); | |
1464 | if (how & FLUSH_SYNC) | |
1465 | rpc_wait_for_completion_task(task); | |
1466 | rpc_put_task(task); | |
1467 | return 0; | |
1468 | } | |
e0c2b380 | 1469 | EXPORT_SYMBOL_GPL(nfs_initiate_commit); |
9ace33cd FI |
1470 | |
1471 | /* | |
1472 | * Set up the argument/result storage required for the RPC call. | |
1473 | */ | |
0b7c0153 | 1474 | void nfs_init_commit(struct nfs_commit_data *data, |
f453a54a FI |
1475 | struct list_head *head, |
1476 | struct pnfs_layout_segment *lseg, | |
1477 | struct nfs_commit_info *cinfo) | |
9ace33cd FI |
1478 | { |
1479 | struct nfs_page *first = nfs_list_entry(head->next); | |
3d4ff43d | 1480 | struct inode *inode = first->wb_context->dentry->d_inode; |
1da177e4 LT |
1481 | |
1482 | /* Set up the RPC argument and reply structs | |
1483 | * NB: take care not to mess about with data->commit et al. */ | |
1484 | ||
1485 | list_splice_init(head, &data->pages); | |
1da177e4 | 1486 | |
1da177e4 | 1487 | data->inode = inode; |
9ace33cd | 1488 | data->cred = first->wb_context->cred; |
988b6dce | 1489 | data->lseg = lseg; /* reference transferred */ |
9ace33cd | 1490 | data->mds_ops = &nfs_commit_ops; |
f453a54a | 1491 | data->completion_ops = cinfo->completion_ops; |
b359f9d0 | 1492 | data->dreq = cinfo->dreq; |
1da177e4 LT |
1493 | |
1494 | data->args.fh = NFS_FH(data->inode); | |
3da28eb1 TM |
1495 | /* Note: we always request a commit of the entire inode */ |
1496 | data->args.offset = 0; | |
1497 | data->args.count = 0; | |
0b7c0153 | 1498 | data->context = get_nfs_open_context(first->wb_context); |
1da177e4 LT |
1499 | data->res.fattr = &data->fattr; |
1500 | data->res.verf = &data->verf; | |
0e574af1 | 1501 | nfs_fattr_init(&data->fattr); |
1da177e4 | 1502 | } |
e0c2b380 | 1503 | EXPORT_SYMBOL_GPL(nfs_init_commit); |
1da177e4 | 1504 | |
e0c2b380 | 1505 | void nfs_retry_commit(struct list_head *page_list, |
ea2cf228 FI |
1506 | struct pnfs_layout_segment *lseg, |
1507 | struct nfs_commit_info *cinfo) | |
64bfeb49 FI |
1508 | { |
1509 | struct nfs_page *req; | |
1510 | ||
1511 | while (!list_empty(page_list)) { | |
1512 | req = nfs_list_entry(page_list->next); | |
1513 | nfs_list_remove_request(req); | |
ea2cf228 | 1514 | nfs_mark_request_commit(req, lseg, cinfo); |
56f9cd68 FI |
1515 | if (!cinfo->dreq) { |
1516 | dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS); | |
d56b4ddf | 1517 | dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info, |
56f9cd68 FI |
1518 | BDI_RECLAIMABLE); |
1519 | } | |
1d1afcbc | 1520 | nfs_unlock_and_release_request(req); |
64bfeb49 FI |
1521 | } |
1522 | } | |
e0c2b380 | 1523 | EXPORT_SYMBOL_GPL(nfs_retry_commit); |
64bfeb49 | 1524 | |
1da177e4 LT |
1525 | /* |
1526 | * Commit dirty pages | |
1527 | */ | |
1528 | static int | |
ea2cf228 FI |
1529 | nfs_commit_list(struct inode *inode, struct list_head *head, int how, |
1530 | struct nfs_commit_info *cinfo) | |
1da177e4 | 1531 | { |
0b7c0153 | 1532 | struct nfs_commit_data *data; |
1da177e4 | 1533 | |
c9d8f89d | 1534 | data = nfs_commitdata_alloc(); |
1da177e4 LT |
1535 | |
1536 | if (!data) | |
1537 | goto out_bad; | |
1538 | ||
1539 | /* Set up the argument struct */ | |
f453a54a FI |
1540 | nfs_init_commit(data, head, NULL, cinfo); |
1541 | atomic_inc(&cinfo->mds->rpcs_out); | |
9f0ec176 AA |
1542 | return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops, |
1543 | how, 0); | |
1da177e4 | 1544 | out_bad: |
ea2cf228 | 1545 | nfs_retry_commit(head, NULL, cinfo); |
f453a54a | 1546 | cinfo->completion_ops->error_cleanup(NFS_I(inode)); |
1da177e4 LT |
1547 | return -ENOMEM; |
1548 | } | |
1549 | ||
1550 | /* | |
1551 | * COMMIT call returned | |
1552 | */ | |
788e7a89 | 1553 | static void nfs_commit_done(struct rpc_task *task, void *calldata) |
1da177e4 | 1554 | { |
0b7c0153 | 1555 | struct nfs_commit_data *data = calldata; |
1da177e4 | 1556 | |
a3f565b1 | 1557 | dprintk("NFS: %5u nfs_commit_done (status %d)\n", |
1da177e4 LT |
1558 | task->tk_pid, task->tk_status); |
1559 | ||
788e7a89 | 1560 | /* Call the NFS version-specific code */ |
c0d0e96b | 1561 | NFS_PROTO(data->inode)->commit_done(task, data); |
c9d8f89d TM |
1562 | } |
1563 | ||
f453a54a | 1564 | static void nfs_commit_release_pages(struct nfs_commit_data *data) |
c9d8f89d | 1565 | { |
5917ce84 | 1566 | struct nfs_page *req; |
c9d8f89d | 1567 | int status = data->task.tk_status; |
f453a54a | 1568 | struct nfs_commit_info cinfo; |
788e7a89 | 1569 | |
1da177e4 LT |
1570 | while (!list_empty(&data->pages)) { |
1571 | req = nfs_list_entry(data->pages.next); | |
1572 | nfs_list_remove_request(req); | |
d6d6dc7c | 1573 | nfs_clear_page_commit(req->wb_page); |
1da177e4 | 1574 | |
48186c7d | 1575 | dprintk("NFS: commit (%s/%lld %d@%lld)", |
3d4ff43d AV |
1576 | req->wb_context->dentry->d_sb->s_id, |
1577 | (long long)NFS_FILEID(req->wb_context->dentry->d_inode), | |
1da177e4 LT |
1578 | req->wb_bytes, |
1579 | (long long)req_offset(req)); | |
c9d8f89d TM |
1580 | if (status < 0) { |
1581 | nfs_context_set_write_error(req->wb_context, status); | |
1da177e4 | 1582 | nfs_inode_remove_request(req); |
c9d8f89d | 1583 | dprintk(", error = %d\n", status); |
1da177e4 LT |
1584 | goto next; |
1585 | } | |
1586 | ||
1587 | /* Okay, COMMIT succeeded, apparently. Check the verifier | |
1588 | * returned by the server against all stored verfs. */ | |
2f2c63bc | 1589 | if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) { |
1da177e4 LT |
1590 | /* We have a match */ |
1591 | nfs_inode_remove_request(req); | |
1592 | dprintk(" OK\n"); | |
1593 | goto next; | |
1594 | } | |
1595 | /* We have a mismatch. Write the page again */ | |
1596 | dprintk(" mismatch\n"); | |
6d884e8f | 1597 | nfs_mark_request_dirty(req); |
05990d1b | 1598 | set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); |
1da177e4 | 1599 | next: |
1d1afcbc | 1600 | nfs_unlock_and_release_request(req); |
1da177e4 | 1601 | } |
f453a54a FI |
1602 | nfs_init_cinfo(&cinfo, data->inode, data->dreq); |
1603 | if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) | |
1604 | nfs_commit_clear_lock(NFS_I(data->inode)); | |
5917ce84 FI |
1605 | } |
1606 | ||
1607 | static void nfs_commit_release(void *calldata) | |
1608 | { | |
0b7c0153 | 1609 | struct nfs_commit_data *data = calldata; |
5917ce84 | 1610 | |
f453a54a | 1611 | data->completion_ops->completion(data); |
c9d8f89d | 1612 | nfs_commitdata_release(calldata); |
1da177e4 | 1613 | } |
788e7a89 TM |
1614 | |
1615 | static const struct rpc_call_ops nfs_commit_ops = { | |
0b7c0153 | 1616 | .rpc_call_prepare = nfs_commit_prepare, |
788e7a89 TM |
1617 | .rpc_call_done = nfs_commit_done, |
1618 | .rpc_release = nfs_commit_release, | |
1619 | }; | |
1da177e4 | 1620 | |
f453a54a FI |
1621 | static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { |
1622 | .completion = nfs_commit_release_pages, | |
1623 | .error_cleanup = nfs_commit_clear_lock, | |
1624 | }; | |
1625 | ||
1763da12 FI |
1626 | int nfs_generic_commit_list(struct inode *inode, struct list_head *head, |
1627 | int how, struct nfs_commit_info *cinfo) | |
84c53ab5 FI |
1628 | { |
1629 | int status; | |
1630 | ||
ea2cf228 | 1631 | status = pnfs_commit_list(inode, head, how, cinfo); |
84c53ab5 | 1632 | if (status == PNFS_NOT_ATTEMPTED) |
ea2cf228 | 1633 | status = nfs_commit_list(inode, head, how, cinfo); |
84c53ab5 FI |
1634 | return status; |
1635 | } | |
1636 | ||
b608b283 | 1637 | int nfs_commit_inode(struct inode *inode, int how) |
1da177e4 | 1638 | { |
1da177e4 | 1639 | LIST_HEAD(head); |
ea2cf228 | 1640 | struct nfs_commit_info cinfo; |
71d0a611 | 1641 | int may_wait = how & FLUSH_SYNC; |
b8413f98 | 1642 | int res; |
1da177e4 | 1643 | |
b8413f98 TM |
1644 | res = nfs_commit_set_lock(NFS_I(inode), may_wait); |
1645 | if (res <= 0) | |
c5efa5fc | 1646 | goto out_mark_dirty; |
ea2cf228 FI |
1647 | nfs_init_cinfo_from_inode(&cinfo, inode); |
1648 | res = nfs_scan_commit(inode, &head, &cinfo); | |
1da177e4 | 1649 | if (res) { |
a861a1e1 FI |
1650 | int error; |
1651 | ||
ea2cf228 | 1652 | error = nfs_generic_commit_list(inode, &head, how, &cinfo); |
3da28eb1 TM |
1653 | if (error < 0) |
1654 | return error; | |
b8413f98 | 1655 | if (!may_wait) |
c5efa5fc | 1656 | goto out_mark_dirty; |
b8413f98 TM |
1657 | error = wait_on_bit(&NFS_I(inode)->flags, |
1658 | NFS_INO_COMMIT, | |
1659 | nfs_wait_bit_killable, | |
1660 | TASK_KILLABLE); | |
1661 | if (error < 0) | |
1662 | return error; | |
71d0a611 TM |
1663 | } else |
1664 | nfs_commit_clear_lock(NFS_I(inode)); | |
c5efa5fc TM |
1665 | return res; |
1666 | /* Note: If we exit without ensuring that the commit is complete, | |
1667 | * we must mark the inode as dirty. Otherwise, future calls to | |
1668 | * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure | |
1669 | * that the data is on the disk. | |
1670 | */ | |
1671 | out_mark_dirty: | |
1672 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
1da177e4 LT |
1673 | return res; |
1674 | } | |
8fc795f7 TM |
1675 | |
1676 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) | |
1677 | { | |
420e3646 TM |
1678 | struct nfs_inode *nfsi = NFS_I(inode); |
1679 | int flags = FLUSH_SYNC; | |
1680 | int ret = 0; | |
8fc795f7 | 1681 | |
3236c3e1 | 1682 | /* no commits means nothing needs to be done */ |
ea2cf228 | 1683 | if (!nfsi->commit_info.ncommit) |
3236c3e1 JL |
1684 | return ret; |
1685 | ||
a00dd6c0 JL |
1686 | if (wbc->sync_mode == WB_SYNC_NONE) { |
1687 | /* Don't commit yet if this is a non-blocking flush and there | |
1688 | * are a lot of outstanding writes for this mapping. | |
1689 | */ | |
ea2cf228 | 1690 | if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1)) |
a00dd6c0 | 1691 | goto out_mark_dirty; |
420e3646 | 1692 | |
a00dd6c0 | 1693 | /* don't wait for the COMMIT response */ |
420e3646 | 1694 | flags = 0; |
a00dd6c0 JL |
1695 | } |
1696 | ||
420e3646 TM |
1697 | ret = nfs_commit_inode(inode, flags); |
1698 | if (ret >= 0) { | |
1699 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
1700 | if (ret < wbc->nr_to_write) | |
1701 | wbc->nr_to_write -= ret; | |
1702 | else | |
1703 | wbc->nr_to_write = 0; | |
1704 | } | |
8fc795f7 | 1705 | return 0; |
420e3646 TM |
1706 | } |
1707 | out_mark_dirty: | |
8fc795f7 TM |
1708 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); |
1709 | return ret; | |
1710 | } | |
c63c7b05 | 1711 | #else |
8fc795f7 TM |
1712 | static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc) |
1713 | { | |
1714 | return 0; | |
1715 | } | |
1da177e4 LT |
1716 | #endif |
1717 | ||
8fc795f7 TM |
1718 | int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
1719 | { | |
a8d8f02c BS |
1720 | return nfs_commit_unstable_pages(inode, wbc); |
1721 | } | |
89d77c8f | 1722 | EXPORT_SYMBOL_GPL(nfs_write_inode); |
a8d8f02c | 1723 | |
acdc53b2 TM |
1724 | /* |
1725 | * flush the inode to disk. | |
1726 | */ | |
1727 | int nfs_wb_all(struct inode *inode) | |
34901f70 TM |
1728 | { |
1729 | struct writeback_control wbc = { | |
72cb77f4 | 1730 | .sync_mode = WB_SYNC_ALL, |
34901f70 | 1731 | .nr_to_write = LONG_MAX, |
d7fb1207 TM |
1732 | .range_start = 0, |
1733 | .range_end = LLONG_MAX, | |
34901f70 | 1734 | }; |
34901f70 | 1735 | |
acdc53b2 | 1736 | return sync_inode(inode, &wbc); |
1c75950b | 1737 | } |
ddda8e0a | 1738 | EXPORT_SYMBOL_GPL(nfs_wb_all); |
1c75950b | 1739 | |
1b3b4a1a TM |
1740 | int nfs_wb_page_cancel(struct inode *inode, struct page *page) |
1741 | { | |
1742 | struct nfs_page *req; | |
1b3b4a1a TM |
1743 | int ret = 0; |
1744 | ||
1b3b4a1a | 1745 | for (;;) { |
ba8b06e6 | 1746 | wait_on_page_writeback(page); |
1b3b4a1a TM |
1747 | req = nfs_page_find_request(page); |
1748 | if (req == NULL) | |
1b3b4a1a | 1749 | break; |
7ad84aa9 | 1750 | if (nfs_lock_request(req)) { |
8dd37758 | 1751 | nfs_clear_request_commit(req); |
1b3b4a1a TM |
1752 | nfs_inode_remove_request(req); |
1753 | /* | |
1754 | * In case nfs_inode_remove_request has marked the | |
1755 | * page as being dirty | |
1756 | */ | |
1757 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
1d1afcbc | 1758 | nfs_unlock_and_release_request(req); |
1b3b4a1a TM |
1759 | break; |
1760 | } | |
1761 | ret = nfs_wait_on_request(req); | |
c9edda71 | 1762 | nfs_release_request(req); |
1b3b4a1a | 1763 | if (ret < 0) |
c988950e | 1764 | break; |
1b3b4a1a | 1765 | } |
1b3b4a1a TM |
1766 | return ret; |
1767 | } | |
1768 | ||
7f2f12d9 TM |
1769 | /* |
1770 | * Write back all requests on one page - we do this before reading it. | |
1771 | */ | |
1772 | int nfs_wb_page(struct inode *inode, struct page *page) | |
1c75950b | 1773 | { |
29418aa4 | 1774 | loff_t range_start = page_file_offset(page); |
1c75950b | 1775 | loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1); |
4d770ccf | 1776 | struct writeback_control wbc = { |
4d770ccf | 1777 | .sync_mode = WB_SYNC_ALL, |
7f2f12d9 | 1778 | .nr_to_write = 0, |
4d770ccf TM |
1779 | .range_start = range_start, |
1780 | .range_end = range_end, | |
1781 | }; | |
1782 | int ret; | |
1c75950b | 1783 | |
0522f6ad | 1784 | for (;;) { |
ba8b06e6 | 1785 | wait_on_page_writeback(page); |
73e3302f TM |
1786 | if (clear_page_dirty_for_io(page)) { |
1787 | ret = nfs_writepage_locked(page, &wbc); | |
1788 | if (ret < 0) | |
1789 | goto out_error; | |
0522f6ad | 1790 | continue; |
7f2f12d9 | 1791 | } |
0522f6ad TM |
1792 | if (!PagePrivate(page)) |
1793 | break; | |
1794 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
ba8b06e6 | 1795 | if (ret < 0) |
73e3302f | 1796 | goto out_error; |
7f2f12d9 | 1797 | } |
73e3302f TM |
1798 | return 0; |
1799 | out_error: | |
4d770ccf | 1800 | return ret; |
1c75950b TM |
1801 | } |
1802 | ||
074cc1de TM |
1803 | #ifdef CONFIG_MIGRATION |
1804 | int nfs_migrate_page(struct address_space *mapping, struct page *newpage, | |
a6bc32b8 | 1805 | struct page *page, enum migrate_mode mode) |
074cc1de | 1806 | { |
2da95652 JL |
1807 | /* |
1808 | * If PagePrivate is set, then the page is currently associated with | |
1809 | * an in-progress read or write request. Don't try to migrate it. | |
1810 | * | |
1811 | * FIXME: we could do this in principle, but we'll need a way to ensure | |
1812 | * that we can safely release the inode reference while holding | |
1813 | * the page lock. | |
1814 | */ | |
1815 | if (PagePrivate(page)) | |
1816 | return -EBUSY; | |
074cc1de | 1817 | |
8c209ce7 DH |
1818 | if (!nfs_fscache_release_page(page, GFP_KERNEL)) |
1819 | return -EBUSY; | |
074cc1de | 1820 | |
a6bc32b8 | 1821 | return migrate_page(mapping, newpage, page, mode); |
074cc1de TM |
1822 | } |
1823 | #endif | |
1824 | ||
f7b422b1 | 1825 | int __init nfs_init_writepagecache(void) |
1da177e4 LT |
1826 | { |
1827 | nfs_wdata_cachep = kmem_cache_create("nfs_write_data", | |
cd841605 | 1828 | sizeof(struct nfs_write_header), |
1da177e4 | 1829 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 1830 | NULL); |
1da177e4 LT |
1831 | if (nfs_wdata_cachep == NULL) |
1832 | return -ENOMEM; | |
1833 | ||
93d2341c MD |
1834 | nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, |
1835 | nfs_wdata_cachep); | |
1da177e4 | 1836 | if (nfs_wdata_mempool == NULL) |
3dd4765f | 1837 | goto out_destroy_write_cache; |
1da177e4 | 1838 | |
0b7c0153 FI |
1839 | nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", |
1840 | sizeof(struct nfs_commit_data), | |
1841 | 0, SLAB_HWCACHE_ALIGN, | |
1842 | NULL); | |
1843 | if (nfs_cdata_cachep == NULL) | |
3dd4765f | 1844 | goto out_destroy_write_mempool; |
0b7c0153 | 1845 | |
93d2341c | 1846 | nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, |
4c100210 | 1847 | nfs_cdata_cachep); |
1da177e4 | 1848 | if (nfs_commit_mempool == NULL) |
3dd4765f | 1849 | goto out_destroy_commit_cache; |
1da177e4 | 1850 | |
89a09141 PZ |
1851 | /* |
1852 | * NFS congestion size, scale with available memory. | |
1853 | * | |
1854 | * 64MB: 8192k | |
1855 | * 128MB: 11585k | |
1856 | * 256MB: 16384k | |
1857 | * 512MB: 23170k | |
1858 | * 1GB: 32768k | |
1859 | * 2GB: 46340k | |
1860 | * 4GB: 65536k | |
1861 | * 8GB: 92681k | |
1862 | * 16GB: 131072k | |
1863 | * | |
1864 | * This allows larger machines to have larger/more transfers. | |
1865 | * Limit the default to 256M | |
1866 | */ | |
1867 | nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); | |
1868 | if (nfs_congestion_kb > 256*1024) | |
1869 | nfs_congestion_kb = 256*1024; | |
1870 | ||
1da177e4 | 1871 | return 0; |
3dd4765f JL |
1872 | |
1873 | out_destroy_commit_cache: | |
1874 | kmem_cache_destroy(nfs_cdata_cachep); | |
1875 | out_destroy_write_mempool: | |
1876 | mempool_destroy(nfs_wdata_mempool); | |
1877 | out_destroy_write_cache: | |
1878 | kmem_cache_destroy(nfs_wdata_cachep); | |
1879 | return -ENOMEM; | |
1da177e4 LT |
1880 | } |
1881 | ||
266bee88 | 1882 | void nfs_destroy_writepagecache(void) |
1da177e4 LT |
1883 | { |
1884 | mempool_destroy(nfs_commit_mempool); | |
3dd4765f | 1885 | kmem_cache_destroy(nfs_cdata_cachep); |
1da177e4 | 1886 | mempool_destroy(nfs_wdata_mempool); |
1a1d92c1 | 1887 | kmem_cache_destroy(nfs_wdata_cachep); |
1da177e4 LT |
1888 | } |
1889 |