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 | ||
12 | #include <linux/config.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/file.h> | |
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 | ||
22 | #define NFS_PARANOIA 1 | |
23 | ||
24 | static kmem_cache_t *nfs_page_cachep; | |
25 | ||
26 | static inline struct nfs_page * | |
27 | nfs_page_alloc(void) | |
28 | { | |
29 | struct nfs_page *p; | |
30 | p = kmem_cache_alloc(nfs_page_cachep, SLAB_KERNEL); | |
31 | if (p) { | |
32 | memset(p, 0, sizeof(*p)); | |
33 | INIT_LIST_HEAD(&p->wb_list); | |
34 | } | |
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. | |
46 | * @file: file descriptor to use | |
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 | |
53 | * create two different requests for the same page, and avoids | |
54 | * a possible deadlock when we reach the hard limit on the number | |
55 | * of dirty pages. | |
56 | * User should ensure it is safe to sleep in this function. | |
57 | */ | |
58 | struct nfs_page * | |
59 | nfs_create_request(struct nfs_open_context *ctx, struct inode *inode, | |
60 | struct page *page, | |
61 | unsigned int offset, unsigned int count) | |
62 | { | |
63 | struct nfs_server *server = NFS_SERVER(inode); | |
64 | struct nfs_page *req; | |
65 | ||
66 | /* Deal with hard limits. */ | |
67 | for (;;) { | |
68 | /* try to allocate the request struct */ | |
69 | req = nfs_page_alloc(); | |
70 | if (req != NULL) | |
71 | break; | |
72 | ||
73 | /* Try to free up at least one request in order to stay | |
74 | * below the hard limit | |
75 | */ | |
76 | if (signalled() && (server->flags & NFS_MOUNT_INTR)) | |
77 | return ERR_PTR(-ERESTARTSYS); | |
78 | yield(); | |
79 | } | |
80 | ||
81 | /* Initialize the request struct. Initially, we assume a | |
82 | * long write-back delay. This will be adjusted in | |
83 | * update_nfs_request below if the region is not locked. */ | |
84 | req->wb_page = page; | |
85 | atomic_set(&req->wb_complete, 0); | |
86 | req->wb_index = page->index; | |
87 | page_cache_get(page); | |
88 | req->wb_offset = offset; | |
89 | req->wb_pgbase = offset; | |
90 | req->wb_bytes = count; | |
91 | atomic_set(&req->wb_count, 1); | |
92 | req->wb_context = get_nfs_open_context(ctx); | |
93 | ||
94 | return req; | |
95 | } | |
96 | ||
97 | /** | |
98 | * nfs_unlock_request - Unlock request and wake up sleepers. | |
99 | * @req: | |
100 | */ | |
101 | void nfs_unlock_request(struct nfs_page *req) | |
102 | { | |
103 | if (!NFS_WBACK_BUSY(req)) { | |
104 | printk(KERN_ERR "NFS: Invalid unlock attempted\n"); | |
105 | BUG(); | |
106 | } | |
107 | smp_mb__before_clear_bit(); | |
108 | clear_bit(PG_BUSY, &req->wb_flags); | |
109 | smp_mb__after_clear_bit(); | |
110 | wake_up_all(&req->wb_context->waitq); | |
111 | nfs_release_request(req); | |
112 | } | |
113 | ||
114 | /** | |
115 | * nfs_clear_request - Free up all resources allocated to the request | |
116 | * @req: | |
117 | * | |
118 | * Release page resources associated with a write request after it | |
119 | * has completed. | |
120 | */ | |
121 | void nfs_clear_request(struct nfs_page *req) | |
122 | { | |
123 | if (req->wb_page) { | |
124 | page_cache_release(req->wb_page); | |
125 | req->wb_page = NULL; | |
126 | } | |
127 | } | |
128 | ||
129 | ||
130 | /** | |
131 | * nfs_release_request - Release the count on an NFS read/write request | |
132 | * @req: request to release | |
133 | * | |
134 | * Note: Should never be called with the spinlock held! | |
135 | */ | |
136 | void | |
137 | nfs_release_request(struct nfs_page *req) | |
138 | { | |
139 | if (!atomic_dec_and_test(&req->wb_count)) | |
140 | return; | |
141 | ||
142 | #ifdef NFS_PARANOIA | |
143 | BUG_ON (!list_empty(&req->wb_list)); | |
144 | BUG_ON (NFS_WBACK_BUSY(req)); | |
145 | #endif | |
146 | ||
147 | /* Release struct file or cached credential */ | |
148 | nfs_clear_request(req); | |
149 | put_nfs_open_context(req->wb_context); | |
150 | nfs_page_free(req); | |
151 | } | |
152 | ||
153 | /** | |
154 | * nfs_list_add_request - Insert a request into a sorted list | |
155 | * @req: request | |
156 | * @head: head of list into which to insert the request. | |
157 | * | |
158 | * Note that the wb_list is sorted by page index in order to facilitate | |
159 | * coalescing of requests. | |
160 | * We use an insertion sort that is optimized for the case of appended | |
161 | * writes. | |
162 | */ | |
163 | void | |
164 | nfs_list_add_request(struct nfs_page *req, struct list_head *head) | |
165 | { | |
166 | struct list_head *pos; | |
167 | ||
168 | #ifdef NFS_PARANOIA | |
169 | if (!list_empty(&req->wb_list)) { | |
170 | printk(KERN_ERR "NFS: Add to list failed!\n"); | |
171 | BUG(); | |
172 | } | |
173 | #endif | |
174 | list_for_each_prev(pos, head) { | |
175 | struct nfs_page *p = nfs_list_entry(pos); | |
176 | if (p->wb_index < req->wb_index) | |
177 | break; | |
178 | } | |
179 | list_add(&req->wb_list, pos); | |
180 | req->wb_list_head = head; | |
181 | } | |
182 | ||
183 | /** | |
184 | * nfs_wait_on_request - Wait for a request to complete. | |
185 | * @req: request to wait upon. | |
186 | * | |
187 | * Interruptible by signals only if mounted with intr flag. | |
188 | * The user is responsible for holding a count on the request. | |
189 | */ | |
190 | int | |
191 | nfs_wait_on_request(struct nfs_page *req) | |
192 | { | |
193 | struct inode *inode = req->wb_context->dentry->d_inode; | |
194 | struct rpc_clnt *clnt = NFS_CLIENT(inode); | |
195 | ||
196 | if (!NFS_WBACK_BUSY(req)) | |
197 | return 0; | |
198 | return nfs_wait_event(clnt, req->wb_context->waitq, !NFS_WBACK_BUSY(req)); | |
199 | } | |
200 | ||
201 | /** | |
202 | * nfs_coalesce_requests - Split coalesced requests out from a list. | |
203 | * @head: source list | |
204 | * @dst: destination list | |
205 | * @nmax: maximum number of requests to coalesce | |
206 | * | |
207 | * Moves a maximum of 'nmax' elements from one list to another. | |
208 | * The elements are checked to ensure that they form a contiguous set | |
209 | * of pages, and that the RPC credentials are the same. | |
210 | */ | |
211 | int | |
212 | nfs_coalesce_requests(struct list_head *head, struct list_head *dst, | |
213 | unsigned int nmax) | |
214 | { | |
215 | struct nfs_page *req = NULL; | |
216 | unsigned int npages = 0; | |
217 | ||
218 | while (!list_empty(head)) { | |
219 | struct nfs_page *prev = req; | |
220 | ||
221 | req = nfs_list_entry(head->next); | |
222 | if (prev) { | |
223 | if (req->wb_context->cred != prev->wb_context->cred) | |
224 | break; | |
225 | if (req->wb_context->lockowner != prev->wb_context->lockowner) | |
226 | break; | |
227 | if (req->wb_context->state != prev->wb_context->state) | |
228 | break; | |
229 | if (req->wb_index != (prev->wb_index + 1)) | |
230 | break; | |
231 | ||
232 | if (req->wb_pgbase != 0) | |
233 | break; | |
234 | } | |
235 | nfs_list_remove_request(req); | |
236 | nfs_list_add_request(req, dst); | |
237 | npages++; | |
238 | if (req->wb_pgbase + req->wb_bytes != PAGE_CACHE_SIZE) | |
239 | break; | |
240 | if (npages >= nmax) | |
241 | break; | |
242 | } | |
243 | return npages; | |
244 | } | |
245 | ||
246 | /** | |
247 | * nfs_scan_list - Scan a list for matching requests | |
248 | * @head: One of the NFS inode request lists | |
249 | * @dst: Destination list | |
250 | * @idx_start: lower bound of page->index to scan | |
251 | * @npages: idx_start + npages sets the upper bound to scan. | |
252 | * | |
253 | * Moves elements from one of the inode request lists. | |
254 | * If the number of requests is set to 0, the entire address_space | |
255 | * starting at index idx_start, is scanned. | |
256 | * The requests are *not* checked to ensure that they form a contiguous set. | |
257 | * You must be holding the inode's req_lock when calling this function | |
258 | */ | |
259 | int | |
260 | nfs_scan_list(struct list_head *head, struct list_head *dst, | |
261 | unsigned long idx_start, unsigned int npages) | |
262 | { | |
263 | struct list_head *pos, *tmp; | |
264 | struct nfs_page *req; | |
265 | unsigned long idx_end; | |
266 | int res; | |
267 | ||
268 | res = 0; | |
269 | if (npages == 0) | |
270 | idx_end = ~0; | |
271 | else | |
272 | idx_end = idx_start + npages - 1; | |
273 | ||
274 | list_for_each_safe(pos, tmp, head) { | |
275 | ||
276 | req = nfs_list_entry(pos); | |
277 | ||
278 | if (req->wb_index < idx_start) | |
279 | continue; | |
280 | if (req->wb_index > idx_end) | |
281 | break; | |
282 | ||
283 | if (!nfs_lock_request(req)) | |
284 | continue; | |
285 | nfs_list_remove_request(req); | |
286 | nfs_list_add_request(req, dst); | |
287 | res++; | |
288 | } | |
289 | return res; | |
290 | } | |
291 | ||
292 | int nfs_init_nfspagecache(void) | |
293 | { | |
294 | nfs_page_cachep = kmem_cache_create("nfs_page", | |
295 | sizeof(struct nfs_page), | |
296 | 0, SLAB_HWCACHE_ALIGN, | |
297 | NULL, NULL); | |
298 | if (nfs_page_cachep == NULL) | |
299 | return -ENOMEM; | |
300 | ||
301 | return 0; | |
302 | } | |
303 | ||
304 | void nfs_destroy_nfspagecache(void) | |
305 | { | |
306 | if (kmem_cache_destroy(nfs_page_cachep)) | |
307 | printk(KERN_INFO "nfs_page: not all structures were freed\n"); | |
308 | } | |
309 |