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 LT |
15 | #include <linux/sunrpc/clnt.h> |
16 | #include <linux/nfs3.h> | |
17 | #include <linux/nfs4.h> | |
18 | #include <linux/nfs_page.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_mount.h> | |
21 | ||
8d5658c9 | 22 | #include "internal.h" |
bae724ef | 23 | #include "pnfs.h" |
8d5658c9 | 24 | |
e18b890b | 25 | static struct kmem_cache *nfs_page_cachep; |
1da177e4 LT |
26 | |
27 | static inline struct nfs_page * | |
28 | nfs_page_alloc(void) | |
29 | { | |
72895b1a JJ |
30 | struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_KERNEL); |
31 | if (p) | |
1da177e4 | 32 | INIT_LIST_HEAD(&p->wb_list); |
1da177e4 LT |
33 | return p; |
34 | } | |
35 | ||
36 | static inline void | |
37 | nfs_page_free(struct nfs_page *p) | |
38 | { | |
39 | kmem_cache_free(nfs_page_cachep, p); | |
40 | } | |
41 | ||
42 | /** | |
43 | * nfs_create_request - Create an NFS read/write request. | |
44 | * @file: file descriptor to use | |
45 | * @inode: inode to which the request is attached | |
46 | * @page: page to write | |
47 | * @offset: starting offset within the page for the write | |
48 | * @count: number of bytes to read/write | |
49 | * | |
50 | * The page must be locked by the caller. This makes sure we never | |
a19b89ca | 51 | * create two different requests for the same page. |
1da177e4 LT |
52 | * User should ensure it is safe to sleep in this function. |
53 | */ | |
54 | struct nfs_page * | |
55 | nfs_create_request(struct nfs_open_context *ctx, struct inode *inode, | |
56 | struct page *page, | |
57 | unsigned int offset, unsigned int count) | |
58 | { | |
1da177e4 LT |
59 | struct nfs_page *req; |
60 | ||
18eb8842 TM |
61 | /* try to allocate the request struct */ |
62 | req = nfs_page_alloc(); | |
63 | if (req == NULL) | |
64 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 65 | |
015f0212 JL |
66 | /* get lock context early so we can deal with alloc failures */ |
67 | req->wb_lock_context = nfs_get_lock_context(ctx); | |
68 | if (req->wb_lock_context == NULL) { | |
69 | nfs_page_free(req); | |
70 | return ERR_PTR(-ENOMEM); | |
71 | } | |
72 | ||
1da177e4 LT |
73 | /* Initialize the request struct. Initially, we assume a |
74 | * long write-back delay. This will be adjusted in | |
75 | * update_nfs_request below if the region is not locked. */ | |
76 | req->wb_page = page; | |
77 | atomic_set(&req->wb_complete, 0); | |
78 | req->wb_index = page->index; | |
79 | page_cache_get(page); | |
cd52ed35 TM |
80 | BUG_ON(PagePrivate(page)); |
81 | BUG_ON(!PageLocked(page)); | |
82 | BUG_ON(page->mapping->host != inode); | |
1da177e4 LT |
83 | req->wb_offset = offset; |
84 | req->wb_pgbase = offset; | |
85 | req->wb_bytes = count; | |
1da177e4 | 86 | req->wb_context = get_nfs_open_context(ctx); |
c03b4024 | 87 | kref_init(&req->wb_kref); |
1da177e4 LT |
88 | return req; |
89 | } | |
90 | ||
91 | /** | |
92 | * nfs_unlock_request - Unlock request and wake up sleepers. | |
93 | * @req: | |
94 | */ | |
95 | void nfs_unlock_request(struct nfs_page *req) | |
96 | { | |
97 | if (!NFS_WBACK_BUSY(req)) { | |
98 | printk(KERN_ERR "NFS: Invalid unlock attempted\n"); | |
99 | BUG(); | |
100 | } | |
101 | smp_mb__before_clear_bit(); | |
102 | clear_bit(PG_BUSY, &req->wb_flags); | |
103 | smp_mb__after_clear_bit(); | |
464a98bd | 104 | wake_up_bit(&req->wb_flags, PG_BUSY); |
1da177e4 LT |
105 | nfs_release_request(req); |
106 | } | |
107 | ||
c6a556b8 | 108 | /** |
9fd367f0 | 109 | * nfs_set_page_tag_locked - Tag a request as locked |
c6a556b8 TM |
110 | * @req: |
111 | */ | |
acee478a | 112 | int nfs_set_page_tag_locked(struct nfs_page *req) |
c6a556b8 | 113 | { |
acee478a | 114 | if (!nfs_lock_request_dontget(req)) |
c6a556b8 | 115 | return 0; |
2df485a7 | 116 | if (test_bit(PG_MAPPED, &req->wb_flags)) |
bb6fbc45 | 117 | radix_tree_tag_set(&NFS_I(req->wb_context->path.dentry->d_inode)->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED); |
c6a556b8 TM |
118 | return 1; |
119 | } | |
120 | ||
121 | /** | |
9fd367f0 | 122 | * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers |
c6a556b8 | 123 | */ |
9fd367f0 | 124 | void nfs_clear_page_tag_locked(struct nfs_page *req) |
c6a556b8 | 125 | { |
2df485a7 | 126 | if (test_bit(PG_MAPPED, &req->wb_flags)) { |
bb6fbc45 TM |
127 | struct inode *inode = req->wb_context->path.dentry->d_inode; |
128 | struct nfs_inode *nfsi = NFS_I(inode); | |
129 | ||
587142f8 | 130 | spin_lock(&inode->i_lock); |
9fd367f0 | 131 | radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED); |
acee478a | 132 | nfs_unlock_request(req); |
587142f8 | 133 | spin_unlock(&inode->i_lock); |
acee478a TM |
134 | } else |
135 | nfs_unlock_request(req); | |
c6a556b8 TM |
136 | } |
137 | ||
1da177e4 LT |
138 | /** |
139 | * nfs_clear_request - Free up all resources allocated to the request | |
140 | * @req: | |
141 | * | |
bb6fbc45 TM |
142 | * Release page and open context resources associated with a read/write |
143 | * request after it has completed. | |
1da177e4 LT |
144 | */ |
145 | void nfs_clear_request(struct nfs_page *req) | |
146 | { | |
cd52ed35 | 147 | struct page *page = req->wb_page; |
bb6fbc45 | 148 | struct nfs_open_context *ctx = req->wb_context; |
f11ac8db | 149 | struct nfs_lock_context *l_ctx = req->wb_lock_context; |
bb6fbc45 | 150 | |
cd52ed35 | 151 | if (page != NULL) { |
cd52ed35 | 152 | page_cache_release(page); |
1da177e4 LT |
153 | req->wb_page = NULL; |
154 | } | |
f11ac8db TM |
155 | if (l_ctx != NULL) { |
156 | nfs_put_lock_context(l_ctx); | |
157 | req->wb_lock_context = NULL; | |
158 | } | |
bb6fbc45 TM |
159 | if (ctx != NULL) { |
160 | put_nfs_open_context(ctx); | |
161 | req->wb_context = NULL; | |
162 | } | |
1da177e4 LT |
163 | } |
164 | ||
165 | ||
166 | /** | |
167 | * nfs_release_request - Release the count on an NFS read/write request | |
168 | * @req: request to release | |
169 | * | |
170 | * Note: Should never be called with the spinlock held! | |
171 | */ | |
c03b4024 | 172 | static void nfs_free_request(struct kref *kref) |
1da177e4 | 173 | { |
c03b4024 | 174 | struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref); |
1da177e4 | 175 | |
bb6fbc45 | 176 | /* Release struct file and open context */ |
1da177e4 | 177 | nfs_clear_request(req); |
1da177e4 LT |
178 | nfs_page_free(req); |
179 | } | |
180 | ||
c03b4024 TM |
181 | void nfs_release_request(struct nfs_page *req) |
182 | { | |
183 | kref_put(&req->wb_kref, nfs_free_request); | |
184 | } | |
185 | ||
9f557cd8 TM |
186 | static int nfs_wait_bit_uninterruptible(void *word) |
187 | { | |
188 | io_schedule(); | |
189 | return 0; | |
190 | } | |
191 | ||
1da177e4 LT |
192 | /** |
193 | * nfs_wait_on_request - Wait for a request to complete. | |
194 | * @req: request to wait upon. | |
195 | * | |
150030b7 | 196 | * Interruptible by fatal signals only. |
1da177e4 LT |
197 | * The user is responsible for holding a count on the request. |
198 | */ | |
199 | int | |
200 | nfs_wait_on_request(struct nfs_page *req) | |
201 | { | |
9f557cd8 TM |
202 | return wait_on_bit(&req->wb_flags, PG_BUSY, |
203 | nfs_wait_bit_uninterruptible, | |
204 | TASK_UNINTERRUPTIBLE); | |
1da177e4 LT |
205 | } |
206 | ||
207 | /** | |
d8a5ad75 TM |
208 | * nfs_pageio_init - initialise a page io descriptor |
209 | * @desc: pointer to descriptor | |
bcb71bba TM |
210 | * @inode: pointer to inode |
211 | * @doio: pointer to io function | |
212 | * @bsize: io block size | |
213 | * @io_flags: extra parameters for the io function | |
d8a5ad75 | 214 | */ |
bcb71bba TM |
215 | void nfs_pageio_init(struct nfs_pageio_descriptor *desc, |
216 | struct inode *inode, | |
c76069bd | 217 | int (*doio)(struct nfs_pageio_descriptor *), |
84dde76c | 218 | size_t bsize, |
bcb71bba | 219 | int io_flags) |
d8a5ad75 TM |
220 | { |
221 | INIT_LIST_HEAD(&desc->pg_list); | |
bcb71bba | 222 | desc->pg_bytes_written = 0; |
d8a5ad75 TM |
223 | desc->pg_count = 0; |
224 | desc->pg_bsize = bsize; | |
225 | desc->pg_base = 0; | |
bcb71bba TM |
226 | desc->pg_inode = inode; |
227 | desc->pg_doio = doio; | |
228 | desc->pg_ioflags = io_flags; | |
229 | desc->pg_error = 0; | |
94ad1c80 | 230 | desc->pg_lseg = NULL; |
d8a5ad75 TM |
231 | } |
232 | ||
233 | /** | |
234 | * nfs_can_coalesce_requests - test two requests for compatibility | |
235 | * @prev: pointer to nfs_page | |
236 | * @req: pointer to nfs_page | |
237 | * | |
238 | * The nfs_page structures 'prev' and 'req' are compared to ensure that the | |
239 | * page data area they describe is contiguous, and that their RPC | |
240 | * credentials, NFSv4 open state, and lockowners are the same. | |
241 | * | |
242 | * Return 'true' if this is the case, else return 'false'. | |
243 | */ | |
244 | static int nfs_can_coalesce_requests(struct nfs_page *prev, | |
94ad1c80 FI |
245 | struct nfs_page *req, |
246 | struct nfs_pageio_descriptor *pgio) | |
d8a5ad75 TM |
247 | { |
248 | if (req->wb_context->cred != prev->wb_context->cred) | |
249 | return 0; | |
f11ac8db | 250 | if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner) |
d8a5ad75 TM |
251 | return 0; |
252 | if (req->wb_context->state != prev->wb_context->state) | |
253 | return 0; | |
254 | if (req->wb_index != (prev->wb_index + 1)) | |
255 | return 0; | |
256 | if (req->wb_pgbase != 0) | |
257 | return 0; | |
258 | if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE) | |
259 | return 0; | |
94ad1c80 FI |
260 | /* |
261 | * Non-whole file layouts need to check that req is inside of | |
262 | * pgio->pg_lseg. | |
263 | */ | |
264 | if (pgio->pg_test && !pgio->pg_test(pgio, prev, req)) | |
265 | return 0; | |
d8a5ad75 TM |
266 | return 1; |
267 | } | |
268 | ||
269 | /** | |
bcb71bba | 270 | * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list. |
d8a5ad75 TM |
271 | * @desc: destination io descriptor |
272 | * @req: request | |
273 | * | |
274 | * Returns true if the request 'req' was successfully coalesced into the | |
275 | * existing list of pages 'desc'. | |
276 | */ | |
bcb71bba TM |
277 | static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc, |
278 | struct nfs_page *req) | |
d8a5ad75 TM |
279 | { |
280 | size_t newlen = req->wb_bytes; | |
281 | ||
282 | if (desc->pg_count != 0) { | |
283 | struct nfs_page *prev; | |
284 | ||
285 | /* | |
286 | * FIXME: ideally we should be able to coalesce all requests | |
287 | * that are not block boundary aligned, but currently this | |
288 | * is problematic for the case of bsize < PAGE_CACHE_SIZE, | |
289 | * since nfs_flush_multi and nfs_pagein_multi assume you | |
290 | * can have only one struct nfs_page. | |
291 | */ | |
8d5658c9 TM |
292 | if (desc->pg_bsize < PAGE_SIZE) |
293 | return 0; | |
d8a5ad75 | 294 | newlen += desc->pg_count; |
8d5658c9 | 295 | if (newlen > desc->pg_bsize) |
d8a5ad75 TM |
296 | return 0; |
297 | prev = nfs_list_entry(desc->pg_list.prev); | |
94ad1c80 | 298 | if (!nfs_can_coalesce_requests(prev, req, desc)) |
d8a5ad75 TM |
299 | return 0; |
300 | } else | |
301 | desc->pg_base = req->wb_pgbase; | |
302 | nfs_list_remove_request(req); | |
303 | nfs_list_add_request(req, &desc->pg_list); | |
304 | desc->pg_count = newlen; | |
305 | return 1; | |
306 | } | |
307 | ||
bcb71bba TM |
308 | /* |
309 | * Helper for nfs_pageio_add_request and nfs_pageio_complete | |
310 | */ | |
311 | static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc) | |
312 | { | |
313 | if (!list_empty(&desc->pg_list)) { | |
c76069bd | 314 | int error = desc->pg_doio(desc); |
bcb71bba TM |
315 | if (error < 0) |
316 | desc->pg_error = error; | |
317 | else | |
318 | desc->pg_bytes_written += desc->pg_count; | |
319 | } | |
320 | if (list_empty(&desc->pg_list)) { | |
321 | desc->pg_count = 0; | |
322 | desc->pg_base = 0; | |
323 | } | |
324 | } | |
325 | ||
326 | /** | |
327 | * nfs_pageio_add_request - Attempt to coalesce a request into a page list. | |
328 | * @desc: destination io descriptor | |
329 | * @req: request | |
330 | * | |
331 | * Returns true if the request 'req' was successfully coalesced into the | |
332 | * existing list of pages 'desc'. | |
333 | */ | |
8b09bee3 TM |
334 | int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, |
335 | struct nfs_page *req) | |
bcb71bba TM |
336 | { |
337 | while (!nfs_pageio_do_add_request(desc, req)) { | |
338 | nfs_pageio_doio(desc); | |
339 | if (desc->pg_error < 0) | |
340 | return 0; | |
341 | } | |
342 | return 1; | |
343 | } | |
344 | ||
bcb71bba TM |
345 | /** |
346 | * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor | |
347 | * @desc: pointer to io descriptor | |
348 | */ | |
349 | void nfs_pageio_complete(struct nfs_pageio_descriptor *desc) | |
350 | { | |
351 | nfs_pageio_doio(desc); | |
352 | } | |
353 | ||
7fe7f848 TM |
354 | /** |
355 | * nfs_pageio_cond_complete - Conditional I/O completion | |
356 | * @desc: pointer to io descriptor | |
357 | * @index: page index | |
358 | * | |
359 | * It is important to ensure that processes don't try to take locks | |
360 | * on non-contiguous ranges of pages as that might deadlock. This | |
361 | * function should be called before attempting to wait on a locked | |
362 | * nfs_page. It will complete the I/O if the page index 'index' | |
363 | * is not contiguous with the existing list of pages in 'desc'. | |
364 | */ | |
365 | void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index) | |
366 | { | |
367 | if (!list_empty(&desc->pg_list)) { | |
368 | struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev); | |
369 | if (index != prev->wb_index + 1) | |
370 | nfs_pageio_doio(desc); | |
371 | } | |
372 | } | |
373 | ||
3da28eb1 | 374 | #define NFS_SCAN_MAXENTRIES 16 |
1da177e4 LT |
375 | /** |
376 | * nfs_scan_list - Scan a list for matching requests | |
d2ccddf0 | 377 | * @nfsi: NFS inode |
1da177e4 LT |
378 | * @dst: Destination list |
379 | * @idx_start: lower bound of page->index to scan | |
380 | * @npages: idx_start + npages sets the upper bound to scan. | |
5c369683 | 381 | * @tag: tag to scan for |
1da177e4 LT |
382 | * |
383 | * Moves elements from one of the inode request lists. | |
384 | * If the number of requests is set to 0, the entire address_space | |
385 | * starting at index idx_start, is scanned. | |
386 | * The requests are *not* checked to ensure that they form a contiguous set. | |
587142f8 | 387 | * You must be holding the inode's i_lock when calling this function |
1da177e4 | 388 | */ |
5c369683 | 389 | int nfs_scan_list(struct nfs_inode *nfsi, |
ca52fec1 | 390 | struct list_head *dst, pgoff_t idx_start, |
5c369683 | 391 | unsigned int npages, int tag) |
1da177e4 | 392 | { |
d2ccddf0 TM |
393 | struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES]; |
394 | struct nfs_page *req; | |
ca52fec1 | 395 | pgoff_t idx_end; |
d2ccddf0 TM |
396 | int found, i; |
397 | int res; | |
1da177e4 LT |
398 | |
399 | res = 0; | |
400 | if (npages == 0) | |
401 | idx_end = ~0; | |
402 | else | |
403 | idx_end = idx_start + npages - 1; | |
404 | ||
d2ccddf0 | 405 | for (;;) { |
5c369683 | 406 | found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, |
d2ccddf0 | 407 | (void **)&pgvec[0], idx_start, |
5c369683 | 408 | NFS_SCAN_MAXENTRIES, tag); |
d2ccddf0 | 409 | if (found <= 0) |
1da177e4 | 410 | break; |
d2ccddf0 TM |
411 | for (i = 0; i < found; i++) { |
412 | req = pgvec[i]; | |
413 | if (req->wb_index > idx_end) | |
414 | goto out; | |
415 | idx_start = req->wb_index + 1; | |
9fd367f0 | 416 | if (nfs_set_page_tag_locked(req)) { |
acee478a | 417 | kref_get(&req->wb_kref); |
d2ccddf0 | 418 | nfs_list_remove_request(req); |
5c369683 TM |
419 | radix_tree_tag_clear(&nfsi->nfs_page_tree, |
420 | req->wb_index, tag); | |
d2ccddf0 TM |
421 | nfs_list_add_request(req, dst); |
422 | res++; | |
dce34ce2 TM |
423 | if (res == INT_MAX) |
424 | goto out; | |
d2ccddf0 TM |
425 | } |
426 | } | |
edc05fc1 | 427 | /* for latency reduction */ |
587142f8 | 428 | cond_resched_lock(&nfsi->vfs_inode.i_lock); |
1da177e4 | 429 | } |
d2ccddf0 | 430 | out: |
1da177e4 LT |
431 | return res; |
432 | } | |
433 | ||
f7b422b1 | 434 | int __init nfs_init_nfspagecache(void) |
1da177e4 LT |
435 | { |
436 | nfs_page_cachep = kmem_cache_create("nfs_page", | |
437 | sizeof(struct nfs_page), | |
438 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 439 | NULL); |
1da177e4 LT |
440 | if (nfs_page_cachep == NULL) |
441 | return -ENOMEM; | |
442 | ||
443 | return 0; | |
444 | } | |
445 | ||
266bee88 | 446 | void nfs_destroy_nfspagecache(void) |
1da177e4 | 447 | { |
1a1d92c1 | 448 | kmem_cache_destroy(nfs_page_cachep); |
1da177e4 LT |
449 | } |
450 |