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/slab.h>
13 #include <linux/file.h>
14 #include <linux/sched.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>
24 static struct kmem_cache
*nfs_page_cachep
;
26 static inline struct nfs_page
*
30 p
= kmem_cache_alloc(nfs_page_cachep
, GFP_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.
54 * User should ensure it is safe to sleep in this function.
57 nfs_create_request(struct nfs_open_context
*ctx
, struct inode
*inode
,
59 unsigned int offset
, unsigned int count
)
63 /* try to allocate the request struct */
64 req
= nfs_page_alloc();
66 return ERR_PTR(-ENOMEM
);
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
) {
72 return ERR_PTR(-ENOMEM
);
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. */
79 atomic_set(&req
->wb_complete
, 0);
80 req
->wb_index
= page
->index
;
82 BUG_ON(PagePrivate(page
));
83 BUG_ON(!PageLocked(page
));
84 BUG_ON(page
->mapping
->host
!= inode
);
85 req
->wb_offset
= offset
;
86 req
->wb_pgbase
= offset
;
87 req
->wb_bytes
= count
;
88 req
->wb_context
= get_nfs_open_context(ctx
);
89 kref_init(&req
->wb_kref
);
94 * nfs_unlock_request - Unlock request and wake up sleepers.
97 void nfs_unlock_request(struct nfs_page
*req
)
99 if (!NFS_WBACK_BUSY(req
)) {
100 printk(KERN_ERR
"NFS: Invalid unlock attempted\n");
103 smp_mb__before_clear_bit();
104 clear_bit(PG_BUSY
, &req
->wb_flags
);
105 smp_mb__after_clear_bit();
106 wake_up_bit(&req
->wb_flags
, PG_BUSY
);
107 nfs_release_request(req
);
111 * nfs_set_page_tag_locked - Tag a request as locked
114 int nfs_set_page_tag_locked(struct nfs_page
*req
)
116 if (!nfs_lock_request_dontget(req
))
118 if (req
->wb_page
!= NULL
)
119 radix_tree_tag_set(&NFS_I(req
->wb_context
->path
.dentry
->d_inode
)->nfs_page_tree
, req
->wb_index
, NFS_PAGE_TAG_LOCKED
);
124 * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers
126 void nfs_clear_page_tag_locked(struct nfs_page
*req
)
128 if (req
->wb_page
!= NULL
) {
129 struct inode
*inode
= req
->wb_context
->path
.dentry
->d_inode
;
130 struct nfs_inode
*nfsi
= NFS_I(inode
);
132 spin_lock(&inode
->i_lock
);
133 radix_tree_tag_clear(&nfsi
->nfs_page_tree
, req
->wb_index
, NFS_PAGE_TAG_LOCKED
);
134 nfs_unlock_request(req
);
135 spin_unlock(&inode
->i_lock
);
137 nfs_unlock_request(req
);
141 * nfs_clear_request - Free up all resources allocated to the request
144 * Release page and open context resources associated with a read/write
145 * request after it has completed.
147 void nfs_clear_request(struct nfs_page
*req
)
149 struct page
*page
= req
->wb_page
;
150 struct nfs_open_context
*ctx
= req
->wb_context
;
151 struct nfs_lock_context
*l_ctx
= req
->wb_lock_context
;
154 page_cache_release(page
);
158 nfs_put_lock_context(l_ctx
);
159 req
->wb_lock_context
= NULL
;
162 put_nfs_open_context(ctx
);
163 req
->wb_context
= NULL
;
169 * nfs_release_request - Release the count on an NFS read/write request
170 * @req: request to release
172 * Note: Should never be called with the spinlock held!
174 static void nfs_free_request(struct kref
*kref
)
176 struct nfs_page
*req
= container_of(kref
, struct nfs_page
, wb_kref
);
178 /* Release struct file and open context */
179 nfs_clear_request(req
);
183 void nfs_release_request(struct nfs_page
*req
)
185 kref_put(&req
->wb_kref
, nfs_free_request
);
188 static int nfs_wait_bit_uninterruptible(void *word
)
195 * nfs_wait_on_request - Wait for a request to complete.
196 * @req: request to wait upon.
198 * Interruptible by fatal signals only.
199 * The user is responsible for holding a count on the request.
202 nfs_wait_on_request(struct nfs_page
*req
)
204 return wait_on_bit(&req
->wb_flags
, PG_BUSY
,
205 nfs_wait_bit_uninterruptible
,
206 TASK_UNINTERRUPTIBLE
);
210 * nfs_pageio_init - initialise a page io descriptor
211 * @desc: pointer to descriptor
212 * @inode: pointer to inode
213 * @doio: pointer to io function
214 * @bsize: io block size
215 * @io_flags: extra parameters for the io function
217 void nfs_pageio_init(struct nfs_pageio_descriptor
*desc
,
219 int (*doio
)(struct inode
*, struct list_head
*, unsigned int, size_t, int),
223 INIT_LIST_HEAD(&desc
->pg_list
);
224 desc
->pg_bytes_written
= 0;
226 desc
->pg_bsize
= bsize
;
228 desc
->pg_inode
= inode
;
229 desc
->pg_doio
= doio
;
230 desc
->pg_ioflags
= io_flags
;
235 * nfs_can_coalesce_requests - test two requests for compatibility
236 * @prev: pointer to nfs_page
237 * @req: pointer to nfs_page
239 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
240 * page data area they describe is contiguous, and that their RPC
241 * credentials, NFSv4 open state, and lockowners are the same.
243 * Return 'true' if this is the case, else return 'false'.
245 static int nfs_can_coalesce_requests(struct nfs_page
*prev
,
246 struct nfs_page
*req
)
248 if (req
->wb_context
->cred
!= prev
->wb_context
->cred
)
250 if (req
->wb_lock_context
->lockowner
!= prev
->wb_lock_context
->lockowner
)
252 if (req
->wb_context
->state
!= prev
->wb_context
->state
)
254 if (req
->wb_index
!= (prev
->wb_index
+ 1))
256 if (req
->wb_pgbase
!= 0)
258 if (prev
->wb_pgbase
+ prev
->wb_bytes
!= PAGE_CACHE_SIZE
)
264 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
265 * @desc: destination io descriptor
268 * Returns true if the request 'req' was successfully coalesced into the
269 * existing list of pages 'desc'.
271 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor
*desc
,
272 struct nfs_page
*req
)
274 size_t newlen
= req
->wb_bytes
;
276 if (desc
->pg_count
!= 0) {
277 struct nfs_page
*prev
;
279 if (desc
->pg_bsize
< PAGE_SIZE
)
281 newlen
+= desc
->pg_count
;
282 if (newlen
> desc
->pg_bsize
)
284 prev
= nfs_list_entry(desc
->pg_list
.prev
);
285 if (!nfs_can_coalesce_requests(prev
, req
))
288 desc
->pg_base
= req
->wb_pgbase
;
289 nfs_list_remove_request(req
);
290 nfs_list_add_request(req
, &desc
->pg_list
);
291 desc
->pg_count
= newlen
;
296 * Helper for nfs_pageio_add_request and nfs_pageio_complete
298 static void nfs_pageio_doio(struct nfs_pageio_descriptor
*desc
)
300 if (!list_empty(&desc
->pg_list
)) {
301 int error
= desc
->pg_doio(desc
->pg_inode
,
303 nfs_page_array_len(desc
->pg_base
,
308 desc
->pg_error
= error
;
310 desc
->pg_bytes_written
+= desc
->pg_count
;
312 if (list_empty(&desc
->pg_list
)) {
319 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
320 * @desc: destination io descriptor
323 * Returns true if the request 'req' was successfully coalesced into the
324 * existing list of pages 'desc'.
326 int nfs_pageio_add_request(struct nfs_pageio_descriptor
*desc
,
327 struct nfs_page
*req
)
329 while (!nfs_pageio_do_add_request(desc
, req
)) {
330 nfs_pageio_doio(desc
);
331 if (desc
->pg_error
< 0)
338 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
339 * @desc: pointer to io descriptor
341 void nfs_pageio_complete(struct nfs_pageio_descriptor
*desc
)
343 nfs_pageio_doio(desc
);
347 * nfs_pageio_cond_complete - Conditional I/O completion
348 * @desc: pointer to io descriptor
351 * It is important to ensure that processes don't try to take locks
352 * on non-contiguous ranges of pages as that might deadlock. This
353 * function should be called before attempting to wait on a locked
354 * nfs_page. It will complete the I/O if the page index 'index'
355 * is not contiguous with the existing list of pages in 'desc'.
357 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor
*desc
, pgoff_t index
)
359 if (!list_empty(&desc
->pg_list
)) {
360 struct nfs_page
*prev
= nfs_list_entry(desc
->pg_list
.prev
);
361 if (index
!= prev
->wb_index
+ 1)
362 nfs_pageio_doio(desc
);
366 #define NFS_SCAN_MAXENTRIES 16
368 * nfs_scan_list - Scan a list for matching requests
370 * @dst: Destination list
371 * @idx_start: lower bound of page->index to scan
372 * @npages: idx_start + npages sets the upper bound to scan.
373 * @tag: tag to scan for
375 * Moves elements from one of the inode request lists.
376 * If the number of requests is set to 0, the entire address_space
377 * starting at index idx_start, is scanned.
378 * The requests are *not* checked to ensure that they form a contiguous set.
379 * You must be holding the inode's i_lock when calling this function
381 int nfs_scan_list(struct nfs_inode
*nfsi
,
382 struct list_head
*dst
, pgoff_t idx_start
,
383 unsigned int npages
, int tag
)
385 struct nfs_page
*pgvec
[NFS_SCAN_MAXENTRIES
];
386 struct nfs_page
*req
;
395 idx_end
= idx_start
+ npages
- 1;
398 found
= radix_tree_gang_lookup_tag(&nfsi
->nfs_page_tree
,
399 (void **)&pgvec
[0], idx_start
,
400 NFS_SCAN_MAXENTRIES
, tag
);
403 for (i
= 0; i
< found
; i
++) {
405 if (req
->wb_index
> idx_end
)
407 idx_start
= req
->wb_index
+ 1;
408 if (nfs_set_page_tag_locked(req
)) {
409 kref_get(&req
->wb_kref
);
410 nfs_list_remove_request(req
);
411 radix_tree_tag_clear(&nfsi
->nfs_page_tree
,
413 nfs_list_add_request(req
, dst
);
419 /* for latency reduction */
420 cond_resched_lock(&nfsi
->vfs_inode
.i_lock
);
426 int __init
nfs_init_nfspagecache(void)
428 nfs_page_cachep
= kmem_cache_create("nfs_page",
429 sizeof(struct nfs_page
),
430 0, SLAB_HWCACHE_ALIGN
,
432 if (nfs_page_cachep
== NULL
)
438 void nfs_destroy_nfspagecache(void)
440 kmem_cache_destroy(nfs_page_cachep
);