]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/read.c | |
3 | * | |
4 | * Block I/O for NFS | |
5 | * | |
6 | * Partial copy of Linus' read cache modifications to fs/nfs/file.c | |
7 | * modified for async RPC by okir@monad.swb.de | |
1da177e4 LT |
8 | */ |
9 | ||
1da177e4 LT |
10 | #include <linux/time.h> |
11 | #include <linux/kernel.h> | |
12 | #include <linux/errno.h> | |
13 | #include <linux/fcntl.h> | |
14 | #include <linux/stat.h> | |
15 | #include <linux/mm.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_page.h> | |
64419a9b | 21 | #include <linux/module.h> |
1da177e4 | 22 | |
f11c88af | 23 | #include "nfs4_fs.h" |
49a70f27 | 24 | #include "internal.h" |
91d5b470 | 25 | #include "iostat.h" |
9a9fc1c0 | 26 | #include "fscache.h" |
fab5fc25 | 27 | #include "pnfs.h" |
91d5b470 | 28 | |
1da177e4 LT |
29 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
30 | ||
061ae2ed | 31 | static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops; |
4a0de55c | 32 | static const struct nfs_rw_ops nfs_rw_read_ops; |
1da177e4 | 33 | |
e18b890b | 34 | static struct kmem_cache *nfs_rdata_cachep; |
1da177e4 | 35 | |
1e7f3a48 | 36 | static struct nfs_pgio_header *nfs_readhdr_alloc(void) |
3feb2d49 | 37 | { |
4a0de55c | 38 | return kmem_cache_zalloc(nfs_rdata_cachep, GFP_KERNEL); |
4db6e0b7 FI |
39 | } |
40 | ||
1e7f3a48 | 41 | static void nfs_readhdr_free(struct nfs_pgio_header *rhdr) |
3feb2d49 | 42 | { |
cd841605 | 43 | kmem_cache_free(nfs_rdata_cachep, rhdr); |
3feb2d49 TM |
44 | } |
45 | ||
1da177e4 LT |
46 | static |
47 | int nfs_return_empty_page(struct page *page) | |
48 | { | |
eebd2aa3 | 49 | zero_user(page, 0, PAGE_CACHE_SIZE); |
1da177e4 LT |
50 | SetPageUptodate(page); |
51 | unlock_page(page); | |
52 | return 0; | |
53 | } | |
54 | ||
1abb5088 | 55 | void nfs_pageio_init_read(struct nfs_pageio_descriptor *pgio, |
fab5fc25 | 56 | struct inode *inode, bool force_mds, |
061ae2ed | 57 | const struct nfs_pgio_completion_ops *compl_ops) |
1751c363 | 58 | { |
fab5fc25 | 59 | struct nfs_server *server = NFS_SERVER(inode); |
41d8d5b7 | 60 | const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; |
fab5fc25 CH |
61 | |
62 | #ifdef CONFIG_NFS_V4_1 | |
63 | if (server->pnfs_curr_ld && !force_mds) | |
64 | pg_ops = server->pnfs_curr_ld->pg_read_ops; | |
65 | #endif | |
4a0de55c AS |
66 | nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_read_ops, |
67 | server->rsize, 0); | |
1751c363 | 68 | } |
ddda8e0a | 69 | EXPORT_SYMBOL_GPL(nfs_pageio_init_read); |
1751c363 | 70 | |
493292dd TM |
71 | void nfs_pageio_reset_read_mds(struct nfs_pageio_descriptor *pgio) |
72 | { | |
41d8d5b7 | 73 | pgio->pg_ops = &nfs_pgio_rw_ops; |
493292dd TM |
74 | pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->rsize; |
75 | } | |
1f945357 | 76 | EXPORT_SYMBOL_GPL(nfs_pageio_reset_read_mds); |
493292dd | 77 | |
f42b293d DH |
78 | int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, |
79 | struct page *page) | |
1da177e4 | 80 | { |
1da177e4 LT |
81 | struct nfs_page *new; |
82 | unsigned int len; | |
c76069bd | 83 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 84 | |
49a70f27 | 85 | len = nfs_page_length(page); |
1da177e4 LT |
86 | if (len == 0) |
87 | return nfs_return_empty_page(page); | |
2bfc6e56 | 88 | new = nfs_create_request(ctx, page, NULL, 0, len); |
1da177e4 LT |
89 | if (IS_ERR(new)) { |
90 | unlock_page(page); | |
91 | return PTR_ERR(new); | |
92 | } | |
93 | if (len < PAGE_CACHE_SIZE) | |
eebd2aa3 | 94 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
1da177e4 | 95 | |
fab5fc25 CH |
96 | nfs_pageio_init_read(&pgio, inode, false, |
97 | &nfs_async_read_completion_ops); | |
d8007d4d | 98 | nfs_pageio_add_request(&pgio, new); |
1751c363 | 99 | nfs_pageio_complete(&pgio); |
2701d086 | 100 | NFS_I(inode)->read_io += pgio.pg_bytes_written; |
1da177e4 LT |
101 | return 0; |
102 | } | |
103 | ||
104 | static void nfs_readpage_release(struct nfs_page *req) | |
105 | { | |
3d4ff43d | 106 | struct inode *d_inode = req->wb_context->dentry->d_inode; |
7f8e05f6 | 107 | |
67d0338e WAA |
108 | dprintk("NFS: read done (%s/%llu %d@%lld)\n", d_inode->i_sb->s_id, |
109 | (unsigned long long)NFS_FILEID(d_inode), req->wb_bytes, | |
110 | (long long)req_offset(req)); | |
7f8e05f6 | 111 | |
67d0338e WAA |
112 | if (nfs_page_group_sync_on_bit(req, PG_UNLOCKPAGE)) { |
113 | if (PageUptodate(req->wb_page)) | |
114 | nfs_readpage_to_fscache(d_inode, req->wb_page, 0); | |
115 | ||
116 | unlock_page(req->wb_page); | |
117 | } | |
1da177e4 | 118 | |
1e8968c5 | 119 | dprintk("NFS: read done (%s/%Lu %d@%Ld)\n", |
3d4ff43d | 120 | req->wb_context->dentry->d_inode->i_sb->s_id, |
1e8968c5 | 121 | (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode), |
1da177e4 LT |
122 | req->wb_bytes, |
123 | (long long)req_offset(req)); | |
10d2c46f | 124 | nfs_release_request(req); |
1da177e4 LT |
125 | } |
126 | ||
67d0338e WAA |
127 | static void nfs_page_group_set_uptodate(struct nfs_page *req) |
128 | { | |
129 | if (nfs_page_group_sync_on_bit(req, PG_UPTODATE)) | |
130 | SetPageUptodate(req->wb_page); | |
131 | } | |
132 | ||
061ae2ed | 133 | static void nfs_read_completion(struct nfs_pgio_header *hdr) |
4db6e0b7 FI |
134 | { |
135 | unsigned long bytes = 0; | |
136 | ||
137 | if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) | |
138 | goto out; | |
4bd8b010 TM |
139 | while (!list_empty(&hdr->pages)) { |
140 | struct nfs_page *req = nfs_list_entry(hdr->pages.next); | |
141 | struct page *page = req->wb_page; | |
68072992 WAA |
142 | unsigned long start = req->wb_pgbase; |
143 | unsigned long end = req->wb_pgbase + req->wb_bytes; | |
4bd8b010 TM |
144 | |
145 | if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) { | |
68072992 WAA |
146 | /* note: regions of the page not covered by a |
147 | * request are zeroed in nfs_readpage_async / | |
148 | * readpage_async_filler */ | |
149 | if (bytes > hdr->good_bytes) { | |
150 | /* nothing in this request was good, so zero | |
151 | * the full extent of the request */ | |
152 | zero_user_segment(page, start, end); | |
153 | ||
154 | } else if (hdr->good_bytes - bytes < req->wb_bytes) { | |
155 | /* part of this request has good bytes, but | |
156 | * not all. zero the bad bytes */ | |
157 | start += hdr->good_bytes - bytes; | |
158 | WARN_ON(start < req->wb_pgbase); | |
159 | zero_user_segment(page, start, end); | |
160 | } | |
4db6e0b7 | 161 | } |
4bd8b010 TM |
162 | bytes += req->wb_bytes; |
163 | if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { | |
4db6e0b7 | 164 | if (bytes <= hdr->good_bytes) |
67d0338e | 165 | nfs_page_group_set_uptodate(req); |
4bd8b010 | 166 | } else |
67d0338e | 167 | nfs_page_group_set_uptodate(req); |
4bd8b010 TM |
168 | nfs_list_remove_request(req); |
169 | nfs_readpage_release(req); | |
4db6e0b7 FI |
170 | } |
171 | out: | |
172 | hdr->release(hdr); | |
173 | } | |
174 | ||
d45f60c6 WAA |
175 | static void nfs_initiate_read(struct nfs_pgio_header *hdr, |
176 | struct rpc_message *msg, | |
1ed26f33 | 177 | struct rpc_task_setup *task_setup_data, int how) |
1da177e4 | 178 | { |
d45f60c6 | 179 | struct inode *inode = hdr->inode; |
84115e1c | 180 | int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; |
1da177e4 | 181 | |
1ed26f33 | 182 | task_setup_data->flags |= swap_flags; |
d45f60c6 | 183 | NFS_PROTO(inode)->read_setup(hdr, msg); |
64419a9b AA |
184 | } |
185 | ||
061ae2ed | 186 | static void |
1da177e4 LT |
187 | nfs_async_read_error(struct list_head *head) |
188 | { | |
189 | struct nfs_page *req; | |
190 | ||
191 | while (!list_empty(head)) { | |
192 | req = nfs_list_entry(head->next); | |
193 | nfs_list_remove_request(req); | |
1da177e4 LT |
194 | nfs_readpage_release(req); |
195 | } | |
196 | } | |
197 | ||
061ae2ed FI |
198 | static const struct nfs_pgio_completion_ops nfs_async_read_completion_ops = { |
199 | .error_cleanup = nfs_async_read_error, | |
200 | .completion = nfs_read_completion, | |
201 | }; | |
202 | ||
0b671301 TM |
203 | /* |
204 | * This is the callback from RPC telling us whether a reply was | |
205 | * received or some error occurred (timeout or socket shutdown). | |
206 | */ | |
d45f60c6 WAA |
207 | static int nfs_readpage_done(struct rpc_task *task, |
208 | struct nfs_pgio_header *hdr, | |
0eecb214 | 209 | struct inode *inode) |
0b671301 | 210 | { |
d45f60c6 | 211 | int status = NFS_PROTO(inode)->read_done(task, hdr); |
0b671301 TM |
212 | if (status != 0) |
213 | return status; | |
214 | ||
d45f60c6 | 215 | nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, hdr->res.count); |
0b671301 TM |
216 | |
217 | if (task->tk_status == -ESTALE) { | |
cd841605 FI |
218 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); |
219 | nfs_mark_for_revalidate(inode); | |
0b671301 | 220 | } |
0b671301 TM |
221 | return 0; |
222 | } | |
223 | ||
d45f60c6 WAA |
224 | static void nfs_readpage_retry(struct rpc_task *task, |
225 | struct nfs_pgio_header *hdr) | |
0b671301 | 226 | { |
d45f60c6 WAA |
227 | struct nfs_pgio_args *argp = &hdr->args; |
228 | struct nfs_pgio_res *resp = &hdr->res; | |
0b671301 | 229 | |
0b671301 | 230 | /* This is a short read! */ |
d45f60c6 | 231 | nfs_inc_stats(hdr->inode, NFSIOS_SHORTREAD); |
0b671301 | 232 | /* Has the server at least made some progress? */ |
4db6e0b7 | 233 | if (resp->count == 0) { |
d45f60c6 | 234 | nfs_set_pgio_error(hdr, -EIO, argp->offset); |
d61e612a | 235 | return; |
4db6e0b7 | 236 | } |
d45f60c6 WAA |
237 | /* Yes, so retry the read at the end of the hdr */ |
238 | hdr->mds_offset += resp->count; | |
0b671301 TM |
239 | argp->offset += resp->count; |
240 | argp->pgbase += resp->count; | |
241 | argp->count -= resp->count; | |
d00c5d43 | 242 | rpc_restart_call_prepare(task); |
0b671301 TM |
243 | } |
244 | ||
d45f60c6 WAA |
245 | static void nfs_readpage_result(struct rpc_task *task, |
246 | struct nfs_pgio_header *hdr) | |
1da177e4 | 247 | { |
d45f60c6 | 248 | if (hdr->res.eof) { |
4db6e0b7 FI |
249 | loff_t bound; |
250 | ||
d45f60c6 | 251 | bound = hdr->args.offset + hdr->res.count; |
4db6e0b7 FI |
252 | spin_lock(&hdr->lock); |
253 | if (bound < hdr->io_start + hdr->good_bytes) { | |
254 | set_bit(NFS_IOHDR_EOF, &hdr->flags); | |
255 | clear_bit(NFS_IOHDR_ERROR, &hdr->flags); | |
256 | hdr->good_bytes = bound - hdr->io_start; | |
257 | } | |
258 | spin_unlock(&hdr->lock); | |
d45f60c6 WAA |
259 | } else if (hdr->res.count != hdr->args.count) |
260 | nfs_readpage_retry(task, hdr); | |
fdd1e74c TM |
261 | } |
262 | ||
1da177e4 LT |
263 | /* |
264 | * Read a page over NFS. | |
265 | * We read the page synchronously in the following case: | |
266 | * - The error flag is set for this page. This happens only when a | |
267 | * previous async read operation failed. | |
268 | */ | |
269 | int nfs_readpage(struct file *file, struct page *page) | |
270 | { | |
271 | struct nfs_open_context *ctx; | |
d56b4ddf | 272 | struct inode *inode = page_file_mapping(page)->host; |
1da177e4 LT |
273 | int error; |
274 | ||
275 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
d56b4ddf | 276 | page, PAGE_CACHE_SIZE, page_file_index(page)); |
91d5b470 CL |
277 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
278 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
279 | ||
1da177e4 LT |
280 | /* |
281 | * Try to flush any pending writes to the file.. | |
282 | * | |
283 | * NOTE! Because we own the page lock, there cannot | |
284 | * be any new pending writes generated at this point | |
285 | * for this page (other pages can be written to). | |
286 | */ | |
287 | error = nfs_wb_page(inode, page); | |
288 | if (error) | |
de05a0cc TM |
289 | goto out_unlock; |
290 | if (PageUptodate(page)) | |
291 | goto out_unlock; | |
1da177e4 | 292 | |
5f004cf2 TM |
293 | error = -ESTALE; |
294 | if (NFS_STALE(inode)) | |
de05a0cc | 295 | goto out_unlock; |
5f004cf2 | 296 | |
1da177e4 | 297 | if (file == NULL) { |
cf1308ff | 298 | error = -EBADF; |
d530838b | 299 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 | 300 | if (ctx == NULL) |
de05a0cc | 301 | goto out_unlock; |
1da177e4 | 302 | } else |
cd3758e3 | 303 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 304 | |
9a9fc1c0 DH |
305 | if (!IS_SYNC(inode)) { |
306 | error = nfs_readpage_from_fscache(ctx, inode, page); | |
307 | if (error == 0) | |
308 | goto out; | |
309 | } | |
310 | ||
8e0969f0 TM |
311 | error = nfs_readpage_async(ctx, inode, page); |
312 | ||
9a9fc1c0 | 313 | out: |
1da177e4 LT |
314 | put_nfs_open_context(ctx); |
315 | return error; | |
de05a0cc | 316 | out_unlock: |
1da177e4 LT |
317 | unlock_page(page); |
318 | return error; | |
319 | } | |
320 | ||
321 | struct nfs_readdesc { | |
8b09bee3 | 322 | struct nfs_pageio_descriptor *pgio; |
1da177e4 LT |
323 | struct nfs_open_context *ctx; |
324 | }; | |
325 | ||
326 | static int | |
327 | readpage_async_filler(void *data, struct page *page) | |
328 | { | |
329 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
1da177e4 LT |
330 | struct nfs_page *new; |
331 | unsigned int len; | |
de05a0cc TM |
332 | int error; |
333 | ||
49a70f27 | 334 | len = nfs_page_length(page); |
1da177e4 LT |
335 | if (len == 0) |
336 | return nfs_return_empty_page(page); | |
de05a0cc | 337 | |
2bfc6e56 | 338 | new = nfs_create_request(desc->ctx, page, NULL, 0, len); |
de05a0cc TM |
339 | if (IS_ERR(new)) |
340 | goto out_error; | |
341 | ||
1da177e4 | 342 | if (len < PAGE_CACHE_SIZE) |
eebd2aa3 | 343 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
f8512ad0 FI |
344 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
345 | error = desc->pgio->pg_error; | |
346 | goto out_unlock; | |
347 | } | |
1da177e4 | 348 | return 0; |
de05a0cc TM |
349 | out_error: |
350 | error = PTR_ERR(new); | |
de05a0cc TM |
351 | out_unlock: |
352 | unlock_page(page); | |
353 | return error; | |
1da177e4 LT |
354 | } |
355 | ||
356 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
357 | struct list_head *pages, unsigned nr_pages) | |
358 | { | |
8b09bee3 | 359 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 360 | struct nfs_readdesc desc = { |
8b09bee3 | 361 | .pgio = &pgio, |
1da177e4 LT |
362 | }; |
363 | struct inode *inode = mapping->host; | |
8b09bee3 | 364 | unsigned long npages; |
5f004cf2 | 365 | int ret = -ESTALE; |
1da177e4 | 366 | |
1e8968c5 | 367 | dprintk("NFS: nfs_readpages (%s/%Lu %d)\n", |
1da177e4 | 368 | inode->i_sb->s_id, |
1e8968c5 | 369 | (unsigned long long)NFS_FILEID(inode), |
1da177e4 | 370 | nr_pages); |
91d5b470 | 371 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 372 | |
5f004cf2 TM |
373 | if (NFS_STALE(inode)) |
374 | goto out; | |
375 | ||
1da177e4 | 376 | if (filp == NULL) { |
d530838b | 377 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
378 | if (desc.ctx == NULL) |
379 | return -EBADF; | |
380 | } else | |
cd3758e3 | 381 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
9a9fc1c0 DH |
382 | |
383 | /* attempt to read as many of the pages as possible from the cache | |
384 | * - this returns -ENOBUFS immediately if the cookie is negative | |
385 | */ | |
386 | ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping, | |
387 | pages, &nr_pages); | |
388 | if (ret == 0) | |
389 | goto read_complete; /* all pages were read */ | |
390 | ||
fab5fc25 CH |
391 | nfs_pageio_init_read(&pgio, inode, false, |
392 | &nfs_async_read_completion_ops); | |
8b09bee3 | 393 | |
1da177e4 | 394 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
8b09bee3 TM |
395 | |
396 | nfs_pageio_complete(&pgio); | |
2701d086 | 397 | NFS_I(inode)->read_io += pgio.pg_bytes_written; |
8b09bee3 TM |
398 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
399 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); | |
9a9fc1c0 | 400 | read_complete: |
1da177e4 | 401 | put_nfs_open_context(desc.ctx); |
5f004cf2 | 402 | out: |
1da177e4 LT |
403 | return ret; |
404 | } | |
405 | ||
f7b422b1 | 406 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
407 | { |
408 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
1e7f3a48 | 409 | sizeof(struct nfs_pgio_header), |
1da177e4 | 410 | 0, SLAB_HWCACHE_ALIGN, |
20c2df83 | 411 | NULL); |
1da177e4 LT |
412 | if (nfs_rdata_cachep == NULL) |
413 | return -ENOMEM; | |
414 | ||
1da177e4 LT |
415 | return 0; |
416 | } | |
417 | ||
266bee88 | 418 | void nfs_destroy_readpagecache(void) |
1da177e4 | 419 | { |
1a1d92c1 | 420 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 421 | } |
4a0de55c AS |
422 | |
423 | static const struct nfs_rw_ops nfs_rw_read_ops = { | |
a4cdda59 | 424 | .rw_mode = FMODE_READ, |
4a0de55c AS |
425 | .rw_alloc_header = nfs_readhdr_alloc, |
426 | .rw_free_header = nfs_readhdr_free, | |
0eecb214 AS |
427 | .rw_done = nfs_readpage_done, |
428 | .rw_result = nfs_readpage_result, | |
1ed26f33 | 429 | .rw_initiate = nfs_initiate_read, |
4a0de55c | 430 | }; |