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/nfs.h>
17 #include <linux/nfs3.h>
18 #include <linux/nfs4.h>
19 #include <linux/nfs_page.h>
20 #include <linux/nfs_fs.h>
21 #include <linux/nfs_mount.h>
22 #include <linux/export.h>
27 static struct kmem_cache
*nfs_page_cachep
;
29 bool nfs_pgarray_set(struct nfs_page_array
*p
, unsigned int pagecount
)
31 p
->npages
= pagecount
;
32 if (pagecount
<= ARRAY_SIZE(p
->page_array
))
33 p
->pagevec
= p
->page_array
;
35 p
->pagevec
= kcalloc(pagecount
, sizeof(struct page
*), GFP_KERNEL
);
39 return p
->pagevec
!= NULL
;
42 void nfs_pgheader_init(struct nfs_pageio_descriptor
*desc
,
43 struct nfs_pgio_header
*hdr
,
44 void (*release
)(struct nfs_pgio_header
*hdr
))
46 hdr
->req
= nfs_list_entry(desc
->pg_list
.next
);
47 hdr
->inode
= desc
->pg_inode
;
48 hdr
->cred
= hdr
->req
->wb_context
->cred
;
49 hdr
->io_start
= req_offset(hdr
->req
);
50 hdr
->good_bytes
= desc
->pg_count
;
51 hdr
->dreq
= desc
->pg_dreq
;
52 hdr
->release
= release
;
53 hdr
->completion_ops
= desc
->pg_completion_ops
;
54 if (hdr
->completion_ops
->init_hdr
)
55 hdr
->completion_ops
->init_hdr(hdr
);
58 void nfs_set_pgio_error(struct nfs_pgio_header
*hdr
, int error
, loff_t pos
)
60 spin_lock(&hdr
->lock
);
61 if (pos
< hdr
->io_start
+ hdr
->good_bytes
) {
62 set_bit(NFS_IOHDR_ERROR
, &hdr
->flags
);
63 clear_bit(NFS_IOHDR_EOF
, &hdr
->flags
);
64 hdr
->good_bytes
= pos
- hdr
->io_start
;
67 spin_unlock(&hdr
->lock
);
70 static inline struct nfs_page
*
73 struct nfs_page
*p
= kmem_cache_zalloc(nfs_page_cachep
, GFP_KERNEL
);
75 INIT_LIST_HEAD(&p
->wb_list
);
80 nfs_page_free(struct nfs_page
*p
)
82 kmem_cache_free(nfs_page_cachep
, p
);
86 * nfs_create_request - Create an NFS read/write request.
87 * @ctx: open context to use
88 * @inode: inode to which the request is attached
89 * @page: page to write
90 * @offset: starting offset within the page for the write
91 * @count: number of bytes to read/write
93 * The page must be locked by the caller. This makes sure we never
94 * create two different requests for the same page.
95 * User should ensure it is safe to sleep in this function.
98 nfs_create_request(struct nfs_open_context
*ctx
, struct inode
*inode
,
100 unsigned int offset
, unsigned int count
)
102 struct nfs_page
*req
;
104 /* try to allocate the request struct */
105 req
= nfs_page_alloc();
107 return ERR_PTR(-ENOMEM
);
109 /* get lock context early so we can deal with alloc failures */
110 req
->wb_lock_context
= nfs_get_lock_context(ctx
);
111 if (req
->wb_lock_context
== NULL
) {
113 return ERR_PTR(-ENOMEM
);
116 /* Initialize the request struct. Initially, we assume a
117 * long write-back delay. This will be adjusted in
118 * update_nfs_request below if the region is not locked. */
120 req
->wb_index
= page
->index
;
121 page_cache_get(page
);
122 req
->wb_offset
= offset
;
123 req
->wb_pgbase
= offset
;
124 req
->wb_bytes
= count
;
125 req
->wb_context
= get_nfs_open_context(ctx
);
126 kref_init(&req
->wb_kref
);
131 * nfs_unlock_request - Unlock request and wake up sleepers.
134 void nfs_unlock_request(struct nfs_page
*req
)
136 if (!NFS_WBACK_BUSY(req
)) {
137 printk(KERN_ERR
"NFS: Invalid unlock attempted\n");
140 smp_mb__before_clear_bit();
141 clear_bit(PG_BUSY
, &req
->wb_flags
);
142 smp_mb__after_clear_bit();
143 wake_up_bit(&req
->wb_flags
, PG_BUSY
);
147 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
150 void nfs_unlock_and_release_request(struct nfs_page
*req
)
152 nfs_unlock_request(req
);
153 nfs_release_request(req
);
157 * nfs_clear_request - Free up all resources allocated to the request
160 * Release page and open context resources associated with a read/write
161 * request after it has completed.
163 static void nfs_clear_request(struct nfs_page
*req
)
165 struct page
*page
= req
->wb_page
;
166 struct nfs_open_context
*ctx
= req
->wb_context
;
167 struct nfs_lock_context
*l_ctx
= req
->wb_lock_context
;
170 page_cache_release(page
);
174 nfs_put_lock_context(l_ctx
);
175 req
->wb_lock_context
= NULL
;
178 put_nfs_open_context(ctx
);
179 req
->wb_context
= NULL
;
185 * nfs_release_request - Release the count on an NFS read/write request
186 * @req: request to release
188 * Note: Should never be called with the spinlock held!
190 static void nfs_free_request(struct kref
*kref
)
192 struct nfs_page
*req
= container_of(kref
, struct nfs_page
, wb_kref
);
194 /* Release struct file and open context */
195 nfs_clear_request(req
);
199 void nfs_release_request(struct nfs_page
*req
)
201 kref_put(&req
->wb_kref
, nfs_free_request
);
204 static int nfs_wait_bit_uninterruptible(void *word
)
211 * nfs_wait_on_request - Wait for a request to complete.
212 * @req: request to wait upon.
214 * Interruptible by fatal signals only.
215 * The user is responsible for holding a count on the request.
218 nfs_wait_on_request(struct nfs_page
*req
)
220 return wait_on_bit(&req
->wb_flags
, PG_BUSY
,
221 nfs_wait_bit_uninterruptible
,
222 TASK_UNINTERRUPTIBLE
);
225 bool nfs_generic_pg_test(struct nfs_pageio_descriptor
*desc
, struct nfs_page
*prev
, struct nfs_page
*req
)
228 * FIXME: ideally we should be able to coalesce all requests
229 * that are not block boundary aligned, but currently this
230 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
231 * since nfs_flush_multi and nfs_pagein_multi assume you
232 * can have only one struct nfs_page.
234 if (desc
->pg_bsize
< PAGE_SIZE
)
237 return desc
->pg_count
+ req
->wb_bytes
<= desc
->pg_bsize
;
239 EXPORT_SYMBOL_GPL(nfs_generic_pg_test
);
242 * nfs_pageio_init - initialise a page io descriptor
243 * @desc: pointer to descriptor
244 * @inode: pointer to inode
245 * @doio: pointer to io function
246 * @bsize: io block size
247 * @io_flags: extra parameters for the io function
249 void nfs_pageio_init(struct nfs_pageio_descriptor
*desc
,
251 const struct nfs_pageio_ops
*pg_ops
,
252 const struct nfs_pgio_completion_ops
*compl_ops
,
256 INIT_LIST_HEAD(&desc
->pg_list
);
257 desc
->pg_bytes_written
= 0;
259 desc
->pg_bsize
= bsize
;
262 desc
->pg_recoalesce
= 0;
263 desc
->pg_inode
= inode
;
264 desc
->pg_ops
= pg_ops
;
265 desc
->pg_completion_ops
= compl_ops
;
266 desc
->pg_ioflags
= io_flags
;
268 desc
->pg_lseg
= NULL
;
269 desc
->pg_dreq
= NULL
;
273 * nfs_can_coalesce_requests - test two requests for compatibility
274 * @prev: pointer to nfs_page
275 * @req: pointer to nfs_page
277 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
278 * page data area they describe is contiguous, and that their RPC
279 * credentials, NFSv4 open state, and lockowners are the same.
281 * Return 'true' if this is the case, else return 'false'.
283 static bool nfs_can_coalesce_requests(struct nfs_page
*prev
,
284 struct nfs_page
*req
,
285 struct nfs_pageio_descriptor
*pgio
)
287 if (req
->wb_context
->cred
!= prev
->wb_context
->cred
)
289 if (req
->wb_lock_context
->lockowner
!= prev
->wb_lock_context
->lockowner
)
291 if (req
->wb_context
->state
!= prev
->wb_context
->state
)
293 if (req
->wb_pgbase
!= 0)
295 if (prev
->wb_pgbase
+ prev
->wb_bytes
!= PAGE_CACHE_SIZE
)
297 if (req_offset(req
) != req_offset(prev
) + prev
->wb_bytes
)
299 return pgio
->pg_ops
->pg_test(pgio
, prev
, req
);
303 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
304 * @desc: destination io descriptor
307 * Returns true if the request 'req' was successfully coalesced into the
308 * existing list of pages 'desc'.
310 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor
*desc
,
311 struct nfs_page
*req
)
313 if (desc
->pg_count
!= 0) {
314 struct nfs_page
*prev
;
316 prev
= nfs_list_entry(desc
->pg_list
.prev
);
317 if (!nfs_can_coalesce_requests(prev
, req
, desc
))
320 if (desc
->pg_ops
->pg_init
)
321 desc
->pg_ops
->pg_init(desc
, req
);
322 desc
->pg_base
= req
->wb_pgbase
;
324 nfs_list_remove_request(req
);
325 nfs_list_add_request(req
, &desc
->pg_list
);
326 desc
->pg_count
+= req
->wb_bytes
;
331 * Helper for nfs_pageio_add_request and nfs_pageio_complete
333 static void nfs_pageio_doio(struct nfs_pageio_descriptor
*desc
)
335 if (!list_empty(&desc
->pg_list
)) {
336 int error
= desc
->pg_ops
->pg_doio(desc
);
338 desc
->pg_error
= error
;
340 desc
->pg_bytes_written
+= desc
->pg_count
;
342 if (list_empty(&desc
->pg_list
)) {
349 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
350 * @desc: destination io descriptor
353 * Returns true if the request 'req' was successfully coalesced into the
354 * existing list of pages 'desc'.
356 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor
*desc
,
357 struct nfs_page
*req
)
359 while (!nfs_pageio_do_add_request(desc
, req
)) {
361 nfs_pageio_doio(desc
);
362 if (desc
->pg_error
< 0)
365 if (desc
->pg_recoalesce
)
371 static int nfs_do_recoalesce(struct nfs_pageio_descriptor
*desc
)
376 list_splice_init(&desc
->pg_list
, &head
);
377 desc
->pg_bytes_written
-= desc
->pg_count
;
380 desc
->pg_recoalesce
= 0;
382 while (!list_empty(&head
)) {
383 struct nfs_page
*req
;
385 req
= list_first_entry(&head
, struct nfs_page
, wb_list
);
386 nfs_list_remove_request(req
);
387 if (__nfs_pageio_add_request(desc
, req
))
389 if (desc
->pg_error
< 0)
393 } while (desc
->pg_recoalesce
);
397 int nfs_pageio_add_request(struct nfs_pageio_descriptor
*desc
,
398 struct nfs_page
*req
)
403 ret
= __nfs_pageio_add_request(desc
, req
);
406 if (desc
->pg_error
< 0)
408 ret
= nfs_do_recoalesce(desc
);
414 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
415 * @desc: pointer to io descriptor
417 void nfs_pageio_complete(struct nfs_pageio_descriptor
*desc
)
420 nfs_pageio_doio(desc
);
421 if (!desc
->pg_recoalesce
)
423 if (!nfs_do_recoalesce(desc
))
429 * nfs_pageio_cond_complete - Conditional I/O completion
430 * @desc: pointer to io descriptor
433 * It is important to ensure that processes don't try to take locks
434 * on non-contiguous ranges of pages as that might deadlock. This
435 * function should be called before attempting to wait on a locked
436 * nfs_page. It will complete the I/O if the page index 'index'
437 * is not contiguous with the existing list of pages in 'desc'.
439 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor
*desc
, pgoff_t index
)
441 if (!list_empty(&desc
->pg_list
)) {
442 struct nfs_page
*prev
= nfs_list_entry(desc
->pg_list
.prev
);
443 if (index
!= prev
->wb_index
+ 1)
444 nfs_pageio_complete(desc
);
448 int __init
nfs_init_nfspagecache(void)
450 nfs_page_cachep
= kmem_cache_create("nfs_page",
451 sizeof(struct nfs_page
),
452 0, SLAB_HWCACHE_ALIGN
,
454 if (nfs_page_cachep
== NULL
)
460 void nfs_destroy_nfspagecache(void)
462 kmem_cache_destroy(nfs_page_cachep
);