2 * linux/fs/nfs/pagelist.c
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.
8 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
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>
22 #define NFS_PARANOIA 1
24 static kmem_cache_t
*nfs_page_cachep
;
26 static inline struct nfs_page
*
30 p
= kmem_cache_alloc(nfs_page_cachep
, SLAB_KERNEL
);
32 memset(p
, 0, sizeof(*p
));
33 INIT_LIST_HEAD(&p
->wb_list
);
39 nfs_page_free(struct nfs_page
*p
)
41 kmem_cache_free(nfs_page_cachep
, p
);
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
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
56 * User should ensure it is safe to sleep in this function.
59 nfs_create_request(struct nfs_open_context
*ctx
, struct inode
*inode
,
61 unsigned int offset
, unsigned int count
)
63 struct nfs_server
*server
= NFS_SERVER(inode
);
66 /* Deal with hard limits. */
68 /* try to allocate the request struct */
69 req
= nfs_page_alloc();
73 /* Try to free up at least one request in order to stay
74 * below the hard limit
76 if (signalled() && (server
->flags
& NFS_MOUNT_INTR
))
77 return ERR_PTR(-ERESTARTSYS
);
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. */
85 atomic_set(&req
->wb_complete
, 0);
86 req
->wb_index
= page
->index
;
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
);
98 * nfs_unlock_request - Unlock request and wake up sleepers.
101 void nfs_unlock_request(struct nfs_page
*req
)
103 if (!NFS_WBACK_BUSY(req
)) {
104 printk(KERN_ERR
"NFS: Invalid unlock attempted\n");
107 smp_mb__before_clear_bit();
108 clear_bit(PG_BUSY
, &req
->wb_flags
);
109 smp_mb__after_clear_bit();
110 wake_up_bit(&req
->wb_flags
, PG_BUSY
);
111 nfs_release_request(req
);
115 * nfs_clear_request - Free up all resources allocated to the request
118 * Release page resources associated with a write request after it
121 void nfs_clear_request(struct nfs_page
*req
)
124 page_cache_release(req
->wb_page
);
131 * nfs_release_request - Release the count on an NFS read/write request
132 * @req: request to release
134 * Note: Should never be called with the spinlock held!
137 nfs_release_request(struct nfs_page
*req
)
139 if (!atomic_dec_and_test(&req
->wb_count
))
143 BUG_ON (!list_empty(&req
->wb_list
));
144 BUG_ON (NFS_WBACK_BUSY(req
));
147 /* Release struct file or cached credential */
148 nfs_clear_request(req
);
149 put_nfs_open_context(req
->wb_context
);
154 * nfs_list_add_request - Insert a request into a sorted list
156 * @head: head of list into which to insert the request.
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
164 nfs_list_add_request(struct nfs_page
*req
, struct list_head
*head
)
166 struct list_head
*pos
;
169 if (!list_empty(&req
->wb_list
)) {
170 printk(KERN_ERR
"NFS: Add to list failed!\n");
174 list_for_each_prev(pos
, head
) {
175 struct nfs_page
*p
= nfs_list_entry(pos
);
176 if (p
->wb_index
< req
->wb_index
)
179 list_add(&req
->wb_list
, pos
);
180 req
->wb_list_head
= head
;
183 static int nfs_wait_bit_interruptible(void *word
)
187 if (signal_pending(current
))
195 * nfs_wait_on_request - Wait for a request to complete.
196 * @req: request to wait upon.
198 * Interruptible by signals only if mounted with intr flag.
199 * The user is responsible for holding a count on the request.
202 nfs_wait_on_request(struct nfs_page
*req
)
204 struct rpc_clnt
*clnt
= NFS_CLIENT(req
->wb_context
->dentry
->d_inode
);
208 if (!test_bit(PG_BUSY
, &req
->wb_flags
))
211 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
212 * are not interrupted if intr flag is not set
214 rpc_clnt_sigmask(clnt
, &oldmask
);
215 ret
= out_of_line_wait_on_bit(&req
->wb_flags
, PG_BUSY
,
216 nfs_wait_bit_interruptible
, TASK_INTERRUPTIBLE
);
217 rpc_clnt_sigunmask(clnt
, &oldmask
);
223 * nfs_coalesce_requests - Split coalesced requests out from a list.
225 * @dst: destination list
226 * @nmax: maximum number of requests to coalesce
228 * Moves a maximum of 'nmax' elements from one list to another.
229 * The elements are checked to ensure that they form a contiguous set
230 * of pages, and that the RPC credentials are the same.
233 nfs_coalesce_requests(struct list_head
*head
, struct list_head
*dst
,
236 struct nfs_page
*req
= NULL
;
237 unsigned int npages
= 0;
239 while (!list_empty(head
)) {
240 struct nfs_page
*prev
= req
;
242 req
= nfs_list_entry(head
->next
);
244 if (req
->wb_context
->cred
!= prev
->wb_context
->cred
)
246 if (req
->wb_context
->lockowner
!= prev
->wb_context
->lockowner
)
248 if (req
->wb_context
->state
!= prev
->wb_context
->state
)
250 if (req
->wb_index
!= (prev
->wb_index
+ 1))
253 if (req
->wb_pgbase
!= 0)
256 nfs_list_remove_request(req
);
257 nfs_list_add_request(req
, dst
);
259 if (req
->wb_pgbase
+ req
->wb_bytes
!= PAGE_CACHE_SIZE
)
268 * nfs_scan_list - Scan a list for matching requests
269 * @head: One of the NFS inode request lists
270 * @dst: Destination list
271 * @idx_start: lower bound of page->index to scan
272 * @npages: idx_start + npages sets the upper bound to scan.
274 * Moves elements from one of the inode request lists.
275 * If the number of requests is set to 0, the entire address_space
276 * starting at index idx_start, is scanned.
277 * The requests are *not* checked to ensure that they form a contiguous set.
278 * You must be holding the inode's req_lock when calling this function
281 nfs_scan_list(struct list_head
*head
, struct list_head
*dst
,
282 unsigned long idx_start
, unsigned int npages
)
284 struct list_head
*pos
, *tmp
;
285 struct nfs_page
*req
;
286 unsigned long idx_end
;
293 idx_end
= idx_start
+ npages
- 1;
295 list_for_each_safe(pos
, tmp
, head
) {
297 req
= nfs_list_entry(pos
);
299 if (req
->wb_index
< idx_start
)
301 if (req
->wb_index
> idx_end
)
304 if (!nfs_lock_request(req
))
306 nfs_list_remove_request(req
);
307 nfs_list_add_request(req
, dst
);
313 int nfs_init_nfspagecache(void)
315 nfs_page_cachep
= kmem_cache_create("nfs_page",
316 sizeof(struct nfs_page
),
317 0, SLAB_HWCACHE_ALIGN
,
319 if (nfs_page_cachep
== NULL
)
325 void nfs_destroy_nfspagecache(void)
327 if (kmem_cache_destroy(nfs_page_cachep
))
328 printk(KERN_INFO
"nfs_page: not all structures were freed\n");