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
)
64 /* try to allocate the request struct */
65 req
= nfs_page_alloc();
69 if (fatal_signal_pending(current
))
70 return ERR_PTR(-ERESTARTSYS
);
74 /* Initialize the request struct. Initially, we assume a
75 * long write-back delay. This will be adjusted in
76 * update_nfs_request below if the region is not locked. */
78 atomic_set(&req
->wb_complete
, 0);
79 req
->wb_index
= page
->index
;
81 BUG_ON(PagePrivate(page
));
82 BUG_ON(!PageLocked(page
));
83 BUG_ON(page
->mapping
->host
!= inode
);
84 req
->wb_offset
= offset
;
85 req
->wb_pgbase
= offset
;
86 req
->wb_bytes
= count
;
87 req
->wb_context
= get_nfs_open_context(ctx
);
88 kref_init(&req
->wb_kref
);
93 * nfs_unlock_request - Unlock request and wake up sleepers.
96 void nfs_unlock_request(struct nfs_page
*req
)
98 if (!NFS_WBACK_BUSY(req
)) {
99 printk(KERN_ERR
"NFS: Invalid unlock attempted\n");
102 smp_mb__before_clear_bit();
103 clear_bit(PG_BUSY
, &req
->wb_flags
);
104 smp_mb__after_clear_bit();
105 wake_up_bit(&req
->wb_flags
, PG_BUSY
);
106 nfs_release_request(req
);
110 * nfs_set_page_tag_locked - Tag a request as locked
113 int nfs_set_page_tag_locked(struct nfs_page
*req
)
115 if (!nfs_lock_request_dontget(req
))
117 if (req
->wb_page
!= NULL
)
118 radix_tree_tag_set(&NFS_I(req
->wb_context
->path
.dentry
->d_inode
)->nfs_page_tree
, req
->wb_index
, NFS_PAGE_TAG_LOCKED
);
123 * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers
125 void nfs_clear_page_tag_locked(struct nfs_page
*req
)
127 if (req
->wb_page
!= NULL
) {
128 struct inode
*inode
= req
->wb_context
->path
.dentry
->d_inode
;
129 struct nfs_inode
*nfsi
= NFS_I(inode
);
131 spin_lock(&inode
->i_lock
);
132 radix_tree_tag_clear(&nfsi
->nfs_page_tree
, req
->wb_index
, NFS_PAGE_TAG_LOCKED
);
133 nfs_unlock_request(req
);
134 spin_unlock(&inode
->i_lock
);
136 nfs_unlock_request(req
);
140 * nfs_clear_request - Free up all resources allocated to the request
143 * Release page and open context resources associated with a read/write
144 * request after it has completed.
146 void nfs_clear_request(struct nfs_page
*req
)
148 struct page
*page
= req
->wb_page
;
149 struct nfs_open_context
*ctx
= req
->wb_context
;
152 page_cache_release(page
);
156 put_nfs_open_context(ctx
);
157 req
->wb_context
= NULL
;
163 * nfs_release_request - Release the count on an NFS read/write request
164 * @req: request to release
166 * Note: Should never be called with the spinlock held!
168 static void nfs_free_request(struct kref
*kref
)
170 struct nfs_page
*req
= container_of(kref
, struct nfs_page
, wb_kref
);
172 /* Release struct file and open context */
173 nfs_clear_request(req
);
177 void nfs_release_request(struct nfs_page
*req
)
179 kref_put(&req
->wb_kref
, nfs_free_request
);
182 static int nfs_wait_bit_uninterruptible(void *word
)
189 * nfs_wait_on_request - Wait for a request to complete.
190 * @req: request to wait upon.
192 * Interruptible by fatal signals only.
193 * The user is responsible for holding a count on the request.
196 nfs_wait_on_request(struct nfs_page
*req
)
198 return wait_on_bit(&req
->wb_flags
, PG_BUSY
,
199 nfs_wait_bit_uninterruptible
,
200 TASK_UNINTERRUPTIBLE
);
204 * nfs_pageio_init - initialise a page io descriptor
205 * @desc: pointer to descriptor
206 * @inode: pointer to inode
207 * @doio: pointer to io function
208 * @bsize: io block size
209 * @io_flags: extra parameters for the io function
211 void nfs_pageio_init(struct nfs_pageio_descriptor
*desc
,
213 int (*doio
)(struct inode
*, struct list_head
*, unsigned int, size_t, int),
217 INIT_LIST_HEAD(&desc
->pg_list
);
218 desc
->pg_bytes_written
= 0;
220 desc
->pg_bsize
= bsize
;
222 desc
->pg_inode
= inode
;
223 desc
->pg_doio
= doio
;
224 desc
->pg_ioflags
= io_flags
;
229 * nfs_can_coalesce_requests - test two requests for compatibility
230 * @prev: pointer to nfs_page
231 * @req: pointer to nfs_page
233 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
234 * page data area they describe is contiguous, and that their RPC
235 * credentials, NFSv4 open state, and lockowners are the same.
237 * Return 'true' if this is the case, else return 'false'.
239 static int nfs_can_coalesce_requests(struct nfs_page
*prev
,
240 struct nfs_page
*req
)
242 if (req
->wb_context
->cred
!= prev
->wb_context
->cred
)
244 if (req
->wb_context
->lockowner
!= prev
->wb_context
->lockowner
)
246 if (req
->wb_context
->state
!= prev
->wb_context
->state
)
248 if (req
->wb_index
!= (prev
->wb_index
+ 1))
250 if (req
->wb_pgbase
!= 0)
252 if (prev
->wb_pgbase
+ prev
->wb_bytes
!= PAGE_CACHE_SIZE
)
258 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
259 * @desc: destination io descriptor
262 * Returns true if the request 'req' was successfully coalesced into the
263 * existing list of pages 'desc'.
265 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor
*desc
,
266 struct nfs_page
*req
)
268 size_t newlen
= req
->wb_bytes
;
270 if (desc
->pg_count
!= 0) {
271 struct nfs_page
*prev
;
274 * FIXME: ideally we should be able to coalesce all requests
275 * that are not block boundary aligned, but currently this
276 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
277 * since nfs_flush_multi and nfs_pagein_multi assume you
278 * can have only one struct nfs_page.
280 if (desc
->pg_bsize
< PAGE_SIZE
)
282 newlen
+= desc
->pg_count
;
283 if (newlen
> desc
->pg_bsize
)
285 prev
= nfs_list_entry(desc
->pg_list
.prev
);
286 if (!nfs_can_coalesce_requests(prev
, req
))
289 desc
->pg_base
= req
->wb_pgbase
;
290 nfs_list_remove_request(req
);
291 nfs_list_add_request(req
, &desc
->pg_list
);
292 desc
->pg_count
= newlen
;
297 * Helper for nfs_pageio_add_request and nfs_pageio_complete
299 static void nfs_pageio_doio(struct nfs_pageio_descriptor
*desc
)
301 if (!list_empty(&desc
->pg_list
)) {
302 int error
= desc
->pg_doio(desc
->pg_inode
,
304 nfs_page_array_len(desc
->pg_base
,
309 desc
->pg_error
= error
;
311 desc
->pg_bytes_written
+= desc
->pg_count
;
313 if (list_empty(&desc
->pg_list
)) {
320 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
321 * @desc: destination io descriptor
324 * Returns true if the request 'req' was successfully coalesced into the
325 * existing list of pages 'desc'.
327 int nfs_pageio_add_request(struct nfs_pageio_descriptor
*desc
,
328 struct nfs_page
*req
)
330 while (!nfs_pageio_do_add_request(desc
, req
)) {
331 nfs_pageio_doio(desc
);
332 if (desc
->pg_error
< 0)
339 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
340 * @desc: pointer to io descriptor
342 void nfs_pageio_complete(struct nfs_pageio_descriptor
*desc
)
344 nfs_pageio_doio(desc
);
348 * nfs_pageio_cond_complete - Conditional I/O completion
349 * @desc: pointer to io descriptor
352 * It is important to ensure that processes don't try to take locks
353 * on non-contiguous ranges of pages as that might deadlock. This
354 * function should be called before attempting to wait on a locked
355 * nfs_page. It will complete the I/O if the page index 'index'
356 * is not contiguous with the existing list of pages in 'desc'.
358 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor
*desc
, pgoff_t index
)
360 if (!list_empty(&desc
->pg_list
)) {
361 struct nfs_page
*prev
= nfs_list_entry(desc
->pg_list
.prev
);
362 if (index
!= prev
->wb_index
+ 1)
363 nfs_pageio_doio(desc
);
367 #define NFS_SCAN_MAXENTRIES 16
369 * nfs_scan_list - Scan a list for matching requests
371 * @dst: Destination list
372 * @idx_start: lower bound of page->index to scan
373 * @npages: idx_start + npages sets the upper bound to scan.
374 * @tag: tag to scan for
376 * Moves elements from one of the inode request lists.
377 * If the number of requests is set to 0, the entire address_space
378 * starting at index idx_start, is scanned.
379 * The requests are *not* checked to ensure that they form a contiguous set.
380 * You must be holding the inode's i_lock when calling this function
382 int nfs_scan_list(struct nfs_inode
*nfsi
,
383 struct list_head
*dst
, pgoff_t idx_start
,
384 unsigned int npages
, int tag
)
386 struct nfs_page
*pgvec
[NFS_SCAN_MAXENTRIES
];
387 struct nfs_page
*req
;
396 idx_end
= idx_start
+ npages
- 1;
399 found
= radix_tree_gang_lookup_tag(&nfsi
->nfs_page_tree
,
400 (void **)&pgvec
[0], idx_start
,
401 NFS_SCAN_MAXENTRIES
, tag
);
404 for (i
= 0; i
< found
; i
++) {
406 if (req
->wb_index
> idx_end
)
408 idx_start
= req
->wb_index
+ 1;
409 if (nfs_set_page_tag_locked(req
)) {
410 kref_get(&req
->wb_kref
);
411 nfs_list_remove_request(req
);
412 radix_tree_tag_clear(&nfsi
->nfs_page_tree
,
414 nfs_list_add_request(req
, dst
);
420 /* for latency reduction */
421 cond_resched_lock(&nfsi
->vfs_inode
.i_lock
);
427 int __init
nfs_init_nfspagecache(void)
429 nfs_page_cachep
= kmem_cache_create("nfs_page",
430 sizeof(struct nfs_page
),
431 0, SLAB_HWCACHE_ALIGN
,
433 if (nfs_page_cachep
== NULL
)
439 void nfs_destroy_nfspagecache(void)
441 kmem_cache_destroy(nfs_page_cachep
);