Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/pagelist.c | |
3 | * | |
4 | * A set of helper functions for managing NFS read and write requests. | |
5 | * The main purpose of these routines is to provide support for the | |
6 | * coalescing of several requests into a single RPC call. | |
7 | * | |
8 | * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no> | |
9 | * | |
10 | */ | |
11 | ||
1da177e4 LT |
12 | #include <linux/slab.h> |
13 | #include <linux/file.h> | |
e8edc6e0 | 14 | #include <linux/sched.h> |
1da177e4 | 15 | #include <linux/sunrpc/clnt.h> |
1313e603 | 16 | #include <linux/nfs.h> |
1da177e4 LT |
17 | #include <linux/nfs3.h> |
18 | #include <linux/nfs4.h> | |
19 | #include <linux/nfs_page.h> | |
20 | #include <linux/nfs_fs.h> | |
21 | #include <linux/nfs_mount.h> | |
afeacc8c | 22 | #include <linux/export.h> |
1da177e4 | 23 | |
8d5658c9 | 24 | #include "internal.h" |
bae724ef | 25 | #include "pnfs.h" |
8d5658c9 | 26 | |
e18b890b | 27 | static struct kmem_cache *nfs_page_cachep; |
1da177e4 LT |
28 | |
29 | static inline struct nfs_page * | |
30 | nfs_page_alloc(void) | |
31 | { | |
72895b1a JJ |
32 | struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_KERNEL); |
33 | if (p) | |
1da177e4 | 34 | INIT_LIST_HEAD(&p->wb_list); |
1da177e4 LT |
35 | return p; |
36 | } | |
37 | ||
38 | static inline void | |
39 | nfs_page_free(struct nfs_page *p) | |
40 | { | |
41 | kmem_cache_free(nfs_page_cachep, p); | |
42 | } | |
43 | ||
44 | /** | |
45 | * nfs_create_request - Create an NFS read/write request. | |
c02f557d | 46 | * @ctx: open context to use |
1da177e4 LT |
47 | * @inode: inode to which the request is attached |
48 | * @page: page to write | |
49 | * @offset: starting offset within the page for the write | |
50 | * @count: number of bytes to read/write | |
51 | * | |
52 | * The page must be locked by the caller. This makes sure we never | |
a19b89ca | 53 | * create two different requests for the same page. |
1da177e4 LT |
54 | * User should ensure it is safe to sleep in this function. |
55 | */ | |
56 | struct nfs_page * | |
57 | nfs_create_request(struct nfs_open_context *ctx, struct inode *inode, | |
58 | struct page *page, | |
59 | unsigned int offset, unsigned int count) | |
60 | { | |
1da177e4 LT |
61 | struct nfs_page *req; |
62 | ||
18eb8842 TM |
63 | /* try to allocate the request struct */ |
64 | req = nfs_page_alloc(); | |
65 | if (req == NULL) | |
66 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 67 | |
015f0212 JL |
68 | /* get lock context early so we can deal with alloc failures */ |
69 | req->wb_lock_context = nfs_get_lock_context(ctx); | |
70 | if (req->wb_lock_context == NULL) { | |
71 | nfs_page_free(req); | |
72 | return ERR_PTR(-ENOMEM); | |
73 | } | |
74 | ||
1da177e4 LT |
75 | /* Initialize the request struct. Initially, we assume a |
76 | * long write-back delay. This will be adjusted in | |
77 | * update_nfs_request below if the region is not locked. */ | |
78 | req->wb_page = page; | |
79 | atomic_set(&req->wb_complete, 0); | |
80 | req->wb_index = page->index; | |
81 | page_cache_get(page); | |
cd52ed35 TM |
82 | BUG_ON(PagePrivate(page)); |
83 | BUG_ON(!PageLocked(page)); | |
84 | BUG_ON(page->mapping->host != inode); | |
1da177e4 LT |
85 | req->wb_offset = offset; |
86 | req->wb_pgbase = offset; | |
87 | req->wb_bytes = count; | |
1da177e4 | 88 | req->wb_context = get_nfs_open_context(ctx); |
c03b4024 | 89 | kref_init(&req->wb_kref); |
1da177e4 LT |
90 | return req; |
91 | } | |
92 | ||
93 | /** | |
94 | * nfs_unlock_request - Unlock request and wake up sleepers. | |
95 | * @req: | |
96 | */ | |
97 | void nfs_unlock_request(struct nfs_page *req) | |
98 | { | |
99 | if (!NFS_WBACK_BUSY(req)) { | |
100 | printk(KERN_ERR "NFS: Invalid unlock attempted\n"); | |
101 | BUG(); | |
102 | } | |
103 | smp_mb__before_clear_bit(); | |
104 | clear_bit(PG_BUSY, &req->wb_flags); | |
105 | smp_mb__after_clear_bit(); | |
464a98bd | 106 | wake_up_bit(&req->wb_flags, PG_BUSY); |
1da177e4 LT |
107 | nfs_release_request(req); |
108 | } | |
109 | ||
4d65c520 | 110 | /* |
1da177e4 LT |
111 | * nfs_clear_request - Free up all resources allocated to the request |
112 | * @req: | |
113 | * | |
bb6fbc45 TM |
114 | * Release page and open context resources associated with a read/write |
115 | * request after it has completed. | |
1da177e4 | 116 | */ |
4d65c520 | 117 | static void nfs_clear_request(struct nfs_page *req) |
1da177e4 | 118 | { |
cd52ed35 | 119 | struct page *page = req->wb_page; |
bb6fbc45 | 120 | struct nfs_open_context *ctx = req->wb_context; |
f11ac8db | 121 | struct nfs_lock_context *l_ctx = req->wb_lock_context; |
bb6fbc45 | 122 | |
cd52ed35 | 123 | if (page != NULL) { |
cd52ed35 | 124 | page_cache_release(page); |
1da177e4 LT |
125 | req->wb_page = NULL; |
126 | } | |
f11ac8db TM |
127 | if (l_ctx != NULL) { |
128 | nfs_put_lock_context(l_ctx); | |
129 | req->wb_lock_context = NULL; | |
130 | } | |
bb6fbc45 TM |
131 | if (ctx != NULL) { |
132 | put_nfs_open_context(ctx); | |
133 | req->wb_context = NULL; | |
134 | } | |
1da177e4 LT |
135 | } |
136 | ||
137 | ||
138 | /** | |
139 | * nfs_release_request - Release the count on an NFS read/write request | |
140 | * @req: request to release | |
141 | * | |
142 | * Note: Should never be called with the spinlock held! | |
143 | */ | |
c03b4024 | 144 | static void nfs_free_request(struct kref *kref) |
1da177e4 | 145 | { |
c03b4024 | 146 | struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref); |
1da177e4 | 147 | |
bb6fbc45 | 148 | /* Release struct file and open context */ |
1da177e4 | 149 | nfs_clear_request(req); |
1da177e4 LT |
150 | nfs_page_free(req); |
151 | } | |
152 | ||
c03b4024 TM |
153 | void nfs_release_request(struct nfs_page *req) |
154 | { | |
155 | kref_put(&req->wb_kref, nfs_free_request); | |
156 | } | |
157 | ||
9f557cd8 TM |
158 | static int nfs_wait_bit_uninterruptible(void *word) |
159 | { | |
160 | io_schedule(); | |
161 | return 0; | |
162 | } | |
163 | ||
1da177e4 LT |
164 | /** |
165 | * nfs_wait_on_request - Wait for a request to complete. | |
166 | * @req: request to wait upon. | |
167 | * | |
150030b7 | 168 | * Interruptible by fatal signals only. |
1da177e4 LT |
169 | * The user is responsible for holding a count on the request. |
170 | */ | |
171 | int | |
172 | nfs_wait_on_request(struct nfs_page *req) | |
173 | { | |
9f557cd8 TM |
174 | return wait_on_bit(&req->wb_flags, PG_BUSY, |
175 | nfs_wait_bit_uninterruptible, | |
176 | TASK_UNINTERRUPTIBLE); | |
1da177e4 LT |
177 | } |
178 | ||
19345cb2 | 179 | bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req) |
5b36c7dc BH |
180 | { |
181 | /* | |
182 | * FIXME: ideally we should be able to coalesce all requests | |
183 | * that are not block boundary aligned, but currently this | |
184 | * is problematic for the case of bsize < PAGE_CACHE_SIZE, | |
185 | * since nfs_flush_multi and nfs_pagein_multi assume you | |
186 | * can have only one struct nfs_page. | |
187 | */ | |
188 | if (desc->pg_bsize < PAGE_SIZE) | |
189 | return 0; | |
190 | ||
191 | return desc->pg_count + req->wb_bytes <= desc->pg_bsize; | |
192 | } | |
19345cb2 | 193 | EXPORT_SYMBOL_GPL(nfs_generic_pg_test); |
5b36c7dc | 194 | |
1da177e4 | 195 | /** |
d8a5ad75 TM |
196 | * nfs_pageio_init - initialise a page io descriptor |
197 | * @desc: pointer to descriptor | |
bcb71bba TM |
198 | * @inode: pointer to inode |
199 | * @doio: pointer to io function | |
200 | * @bsize: io block size | |
201 | * @io_flags: extra parameters for the io function | |
d8a5ad75 | 202 | */ |
bcb71bba TM |
203 | void nfs_pageio_init(struct nfs_pageio_descriptor *desc, |
204 | struct inode *inode, | |
1751c363 | 205 | const struct nfs_pageio_ops *pg_ops, |
84dde76c | 206 | size_t bsize, |
bcb71bba | 207 | int io_flags) |
d8a5ad75 TM |
208 | { |
209 | INIT_LIST_HEAD(&desc->pg_list); | |
bcb71bba | 210 | desc->pg_bytes_written = 0; |
d8a5ad75 TM |
211 | desc->pg_count = 0; |
212 | desc->pg_bsize = bsize; | |
213 | desc->pg_base = 0; | |
b31268ac | 214 | desc->pg_moreio = 0; |
d9156f9f | 215 | desc->pg_recoalesce = 0; |
bcb71bba | 216 | desc->pg_inode = inode; |
1751c363 | 217 | desc->pg_ops = pg_ops; |
bcb71bba TM |
218 | desc->pg_ioflags = io_flags; |
219 | desc->pg_error = 0; | |
94ad1c80 | 220 | desc->pg_lseg = NULL; |
d8a5ad75 TM |
221 | } |
222 | ||
223 | /** | |
224 | * nfs_can_coalesce_requests - test two requests for compatibility | |
225 | * @prev: pointer to nfs_page | |
226 | * @req: pointer to nfs_page | |
227 | * | |
228 | * The nfs_page structures 'prev' and 'req' are compared to ensure that the | |
229 | * page data area they describe is contiguous, and that their RPC | |
230 | * credentials, NFSv4 open state, and lockowners are the same. | |
231 | * | |
232 | * Return 'true' if this is the case, else return 'false'. | |
233 | */ | |
18ad0a9f BH |
234 | static bool nfs_can_coalesce_requests(struct nfs_page *prev, |
235 | struct nfs_page *req, | |
236 | struct nfs_pageio_descriptor *pgio) | |
d8a5ad75 TM |
237 | { |
238 | if (req->wb_context->cred != prev->wb_context->cred) | |
18ad0a9f | 239 | return false; |
f11ac8db | 240 | if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner) |
18ad0a9f | 241 | return false; |
d8a5ad75 | 242 | if (req->wb_context->state != prev->wb_context->state) |
18ad0a9f | 243 | return false; |
d8a5ad75 | 244 | if (req->wb_index != (prev->wb_index + 1)) |
18ad0a9f | 245 | return false; |
d8a5ad75 | 246 | if (req->wb_pgbase != 0) |
18ad0a9f | 247 | return false; |
d8a5ad75 | 248 | if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE) |
18ad0a9f | 249 | return false; |
1751c363 | 250 | return pgio->pg_ops->pg_test(pgio, prev, req); |
d8a5ad75 TM |
251 | } |
252 | ||
253 | /** | |
bcb71bba | 254 | * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list. |
d8a5ad75 TM |
255 | * @desc: destination io descriptor |
256 | * @req: request | |
257 | * | |
258 | * Returns true if the request 'req' was successfully coalesced into the | |
259 | * existing list of pages 'desc'. | |
260 | */ | |
bcb71bba TM |
261 | static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc, |
262 | struct nfs_page *req) | |
d8a5ad75 | 263 | { |
d8a5ad75 TM |
264 | if (desc->pg_count != 0) { |
265 | struct nfs_page *prev; | |
266 | ||
d8a5ad75 | 267 | prev = nfs_list_entry(desc->pg_list.prev); |
94ad1c80 | 268 | if (!nfs_can_coalesce_requests(prev, req, desc)) |
d8a5ad75 | 269 | return 0; |
5b36c7dc | 270 | } else { |
d8007d4d TM |
271 | if (desc->pg_ops->pg_init) |
272 | desc->pg_ops->pg_init(desc, req); | |
d8a5ad75 | 273 | desc->pg_base = req->wb_pgbase; |
5b36c7dc | 274 | } |
d8a5ad75 TM |
275 | nfs_list_remove_request(req); |
276 | nfs_list_add_request(req, &desc->pg_list); | |
5b36c7dc | 277 | desc->pg_count += req->wb_bytes; |
d8a5ad75 TM |
278 | return 1; |
279 | } | |
280 | ||
bcb71bba TM |
281 | /* |
282 | * Helper for nfs_pageio_add_request and nfs_pageio_complete | |
283 | */ | |
284 | static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc) | |
285 | { | |
286 | if (!list_empty(&desc->pg_list)) { | |
1751c363 | 287 | int error = desc->pg_ops->pg_doio(desc); |
bcb71bba TM |
288 | if (error < 0) |
289 | desc->pg_error = error; | |
290 | else | |
291 | desc->pg_bytes_written += desc->pg_count; | |
292 | } | |
293 | if (list_empty(&desc->pg_list)) { | |
294 | desc->pg_count = 0; | |
295 | desc->pg_base = 0; | |
296 | } | |
297 | } | |
298 | ||
299 | /** | |
300 | * nfs_pageio_add_request - Attempt to coalesce a request into a page list. | |
301 | * @desc: destination io descriptor | |
302 | * @req: request | |
303 | * | |
304 | * Returns true if the request 'req' was successfully coalesced into the | |
305 | * existing list of pages 'desc'. | |
306 | */ | |
d9156f9f | 307 | static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, |
8b09bee3 | 308 | struct nfs_page *req) |
bcb71bba TM |
309 | { |
310 | while (!nfs_pageio_do_add_request(desc, req)) { | |
b31268ac | 311 | desc->pg_moreio = 1; |
bcb71bba TM |
312 | nfs_pageio_doio(desc); |
313 | if (desc->pg_error < 0) | |
314 | return 0; | |
b31268ac | 315 | desc->pg_moreio = 0; |
d9156f9f TM |
316 | if (desc->pg_recoalesce) |
317 | return 0; | |
bcb71bba TM |
318 | } |
319 | return 1; | |
320 | } | |
321 | ||
d9156f9f TM |
322 | static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc) |
323 | { | |
324 | LIST_HEAD(head); | |
325 | ||
326 | do { | |
327 | list_splice_init(&desc->pg_list, &head); | |
328 | desc->pg_bytes_written -= desc->pg_count; | |
329 | desc->pg_count = 0; | |
330 | desc->pg_base = 0; | |
331 | desc->pg_recoalesce = 0; | |
332 | ||
333 | while (!list_empty(&head)) { | |
334 | struct nfs_page *req; | |
335 | ||
336 | req = list_first_entry(&head, struct nfs_page, wb_list); | |
337 | nfs_list_remove_request(req); | |
338 | if (__nfs_pageio_add_request(desc, req)) | |
339 | continue; | |
340 | if (desc->pg_error < 0) | |
341 | return 0; | |
342 | break; | |
343 | } | |
344 | } while (desc->pg_recoalesce); | |
345 | return 1; | |
346 | } | |
347 | ||
348 | int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, | |
349 | struct nfs_page *req) | |
350 | { | |
351 | int ret; | |
352 | ||
353 | do { | |
354 | ret = __nfs_pageio_add_request(desc, req); | |
355 | if (ret) | |
356 | break; | |
357 | if (desc->pg_error < 0) | |
358 | break; | |
359 | ret = nfs_do_recoalesce(desc); | |
360 | } while (ret); | |
361 | return ret; | |
362 | } | |
363 | ||
bcb71bba TM |
364 | /** |
365 | * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor | |
366 | * @desc: pointer to io descriptor | |
367 | */ | |
368 | void nfs_pageio_complete(struct nfs_pageio_descriptor *desc) | |
369 | { | |
d9156f9f TM |
370 | for (;;) { |
371 | nfs_pageio_doio(desc); | |
372 | if (!desc->pg_recoalesce) | |
373 | break; | |
374 | if (!nfs_do_recoalesce(desc)) | |
375 | break; | |
376 | } | |
bcb71bba TM |
377 | } |
378 | ||
7fe7f848 TM |
379 | /** |
380 | * nfs_pageio_cond_complete - Conditional I/O completion | |
381 | * @desc: pointer to io descriptor | |
382 | * @index: page index | |
383 | * | |
384 | * It is important to ensure that processes don't try to take locks | |
385 | * on non-contiguous ranges of pages as that might deadlock. This | |
386 | * function should be called before attempting to wait on a locked | |
387 | * nfs_page. It will complete the I/O if the page index 'index' | |
388 | * is not contiguous with the existing list of pages in 'desc'. | |
389 | */ | |
390 | void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index) | |
391 | { | |
392 | if (!list_empty(&desc->pg_list)) { | |
393 | struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev); | |
394 | if (index != prev->wb_index + 1) | |
d9156f9f | 395 | nfs_pageio_complete(desc); |
7fe7f848 TM |
396 | } |
397 | } | |
398 | ||
f7b422b1 | 399 | int __init nfs_init_nfspagecache(void) |
1da177e4 LT |
400 | { |
401 | nfs_page_cachep = kmem_cache_create("nfs_page", | |
402 | sizeof(struct nfs_page), | |
403 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 404 | NULL); |
1da177e4 LT |
405 | if (nfs_page_cachep == NULL) |
406 | return -ENOMEM; | |
407 | ||
408 | return 0; | |
409 | } | |
410 | ||
266bee88 | 411 | void nfs_destroy_nfspagecache(void) |
1da177e4 | 412 | { |
1a1d92c1 | 413 | kmem_cache_destroy(nfs_page_cachep); |
1da177e4 LT |
414 | } |
415 |