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>
21 #include <linux/export.h>
26 static struct kmem_cache
*nfs_page_cachep
;
28 static inline struct nfs_page
*
31 struct nfs_page
*p
= kmem_cache_zalloc(nfs_page_cachep
, GFP_KERNEL
);
33 INIT_LIST_HEAD(&p
->wb_list
);
38 nfs_page_free(struct nfs_page
*p
)
40 kmem_cache_free(nfs_page_cachep
, p
);
44 * nfs_create_request - Create an NFS read/write request.
45 * @ctx: open context to use
46 * @inode: inode to which the request is attached
47 * @page: page to write
48 * @offset: starting offset within the page for the write
49 * @count: number of bytes to read/write
51 * The page must be locked by the caller. This makes sure we never
52 * create two different requests for the same page.
53 * User should ensure it is safe to sleep in this function.
56 nfs_create_request(struct nfs_open_context
*ctx
, struct inode
*inode
,
58 unsigned int offset
, unsigned int count
)
62 /* try to allocate the request struct */
63 req
= nfs_page_alloc();
65 return ERR_PTR(-ENOMEM
);
67 /* get lock context early so we can deal with alloc failures */
68 req
->wb_lock_context
= nfs_get_lock_context(ctx
);
69 if (req
->wb_lock_context
== NULL
) {
71 return ERR_PTR(-ENOMEM
);
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 (test_bit(PG_MAPPED
, &req
->wb_flags
))
118 radix_tree_tag_set(&NFS_I(req
->wb_context
->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 (test_bit(PG_MAPPED
, &req
->wb_flags
)) {
128 struct inode
*inode
= req
->wb_context
->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 static void nfs_clear_request(struct nfs_page
*req
)
148 struct page
*page
= req
->wb_page
;
149 struct nfs_open_context
*ctx
= req
->wb_context
;
150 struct nfs_lock_context
*l_ctx
= req
->wb_lock_context
;
153 page_cache_release(page
);
157 nfs_put_lock_context(l_ctx
);
158 req
->wb_lock_context
= NULL
;
161 put_nfs_open_context(ctx
);
162 req
->wb_context
= NULL
;
168 * nfs_release_request - Release the count on an NFS read/write request
169 * @req: request to release
171 * Note: Should never be called with the spinlock held!
173 static void nfs_free_request(struct kref
*kref
)
175 struct nfs_page
*req
= container_of(kref
, struct nfs_page
, wb_kref
);
177 /* Release struct file and open context */
178 nfs_clear_request(req
);
182 void nfs_release_request(struct nfs_page
*req
)
184 kref_put(&req
->wb_kref
, nfs_free_request
);
187 static int nfs_wait_bit_uninterruptible(void *word
)
194 * nfs_wait_on_request - Wait for a request to complete.
195 * @req: request to wait upon.
197 * Interruptible by fatal signals only.
198 * The user is responsible for holding a count on the request.
201 nfs_wait_on_request(struct nfs_page
*req
)
203 return wait_on_bit(&req
->wb_flags
, PG_BUSY
,
204 nfs_wait_bit_uninterruptible
,
205 TASK_UNINTERRUPTIBLE
);
208 bool nfs_generic_pg_test(struct nfs_pageio_descriptor
*desc
, struct nfs_page
*prev
, struct nfs_page
*req
)
211 * FIXME: ideally we should be able to coalesce all requests
212 * that are not block boundary aligned, but currently this
213 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
214 * since nfs_flush_multi and nfs_pagein_multi assume you
215 * can have only one struct nfs_page.
217 if (desc
->pg_bsize
< PAGE_SIZE
)
220 return desc
->pg_count
+ req
->wb_bytes
<= desc
->pg_bsize
;
222 EXPORT_SYMBOL_GPL(nfs_generic_pg_test
);
225 * nfs_pageio_init - initialise a page io descriptor
226 * @desc: pointer to descriptor
227 * @inode: pointer to inode
228 * @doio: pointer to io function
229 * @bsize: io block size
230 * @io_flags: extra parameters for the io function
232 void nfs_pageio_init(struct nfs_pageio_descriptor
*desc
,
234 const struct nfs_pageio_ops
*pg_ops
,
238 INIT_LIST_HEAD(&desc
->pg_list
);
239 desc
->pg_bytes_written
= 0;
241 desc
->pg_bsize
= bsize
;
244 desc
->pg_recoalesce
= 0;
245 desc
->pg_inode
= inode
;
246 desc
->pg_ops
= pg_ops
;
247 desc
->pg_ioflags
= io_flags
;
249 desc
->pg_lseg
= NULL
;
253 * nfs_can_coalesce_requests - test two requests for compatibility
254 * @prev: pointer to nfs_page
255 * @req: pointer to nfs_page
257 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
258 * page data area they describe is contiguous, and that their RPC
259 * credentials, NFSv4 open state, and lockowners are the same.
261 * Return 'true' if this is the case, else return 'false'.
263 static bool nfs_can_coalesce_requests(struct nfs_page
*prev
,
264 struct nfs_page
*req
,
265 struct nfs_pageio_descriptor
*pgio
)
267 if (req
->wb_context
->cred
!= prev
->wb_context
->cred
)
269 if (req
->wb_lock_context
->lockowner
!= prev
->wb_lock_context
->lockowner
)
271 if (req
->wb_context
->state
!= prev
->wb_context
->state
)
273 if (req
->wb_index
!= (prev
->wb_index
+ 1))
275 if (req
->wb_pgbase
!= 0)
277 if (prev
->wb_pgbase
+ prev
->wb_bytes
!= PAGE_CACHE_SIZE
)
279 return pgio
->pg_ops
->pg_test(pgio
, prev
, req
);
283 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
284 * @desc: destination io descriptor
287 * Returns true if the request 'req' was successfully coalesced into the
288 * existing list of pages 'desc'.
290 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor
*desc
,
291 struct nfs_page
*req
)
293 if (desc
->pg_count
!= 0) {
294 struct nfs_page
*prev
;
296 prev
= nfs_list_entry(desc
->pg_list
.prev
);
297 if (!nfs_can_coalesce_requests(prev
, req
, desc
))
300 if (desc
->pg_ops
->pg_init
)
301 desc
->pg_ops
->pg_init(desc
, req
);
302 desc
->pg_base
= req
->wb_pgbase
;
304 nfs_list_remove_request(req
);
305 nfs_list_add_request(req
, &desc
->pg_list
);
306 desc
->pg_count
+= req
->wb_bytes
;
311 * Helper for nfs_pageio_add_request and nfs_pageio_complete
313 static void nfs_pageio_doio(struct nfs_pageio_descriptor
*desc
)
315 if (!list_empty(&desc
->pg_list
)) {
316 int error
= desc
->pg_ops
->pg_doio(desc
);
318 desc
->pg_error
= error
;
320 desc
->pg_bytes_written
+= desc
->pg_count
;
322 if (list_empty(&desc
->pg_list
)) {
329 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
330 * @desc: destination io descriptor
333 * Returns true if the request 'req' was successfully coalesced into the
334 * existing list of pages 'desc'.
336 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor
*desc
,
337 struct nfs_page
*req
)
339 while (!nfs_pageio_do_add_request(desc
, req
)) {
341 nfs_pageio_doio(desc
);
342 if (desc
->pg_error
< 0)
345 if (desc
->pg_recoalesce
)
351 static int nfs_do_recoalesce(struct nfs_pageio_descriptor
*desc
)
356 list_splice_init(&desc
->pg_list
, &head
);
357 desc
->pg_bytes_written
-= desc
->pg_count
;
360 desc
->pg_recoalesce
= 0;
362 while (!list_empty(&head
)) {
363 struct nfs_page
*req
;
365 req
= list_first_entry(&head
, struct nfs_page
, wb_list
);
366 nfs_list_remove_request(req
);
367 if (__nfs_pageio_add_request(desc
, req
))
369 if (desc
->pg_error
< 0)
373 } while (desc
->pg_recoalesce
);
377 int nfs_pageio_add_request(struct nfs_pageio_descriptor
*desc
,
378 struct nfs_page
*req
)
383 ret
= __nfs_pageio_add_request(desc
, req
);
386 if (desc
->pg_error
< 0)
388 ret
= nfs_do_recoalesce(desc
);
394 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
395 * @desc: pointer to io descriptor
397 void nfs_pageio_complete(struct nfs_pageio_descriptor
*desc
)
400 nfs_pageio_doio(desc
);
401 if (!desc
->pg_recoalesce
)
403 if (!nfs_do_recoalesce(desc
))
409 * nfs_pageio_cond_complete - Conditional I/O completion
410 * @desc: pointer to io descriptor
413 * It is important to ensure that processes don't try to take locks
414 * on non-contiguous ranges of pages as that might deadlock. This
415 * function should be called before attempting to wait on a locked
416 * nfs_page. It will complete the I/O if the page index 'index'
417 * is not contiguous with the existing list of pages in 'desc'.
419 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor
*desc
, pgoff_t index
)
421 if (!list_empty(&desc
->pg_list
)) {
422 struct nfs_page
*prev
= nfs_list_entry(desc
->pg_list
.prev
);
423 if (index
!= prev
->wb_index
+ 1)
424 nfs_pageio_complete(desc
);
428 #define NFS_SCAN_MAXENTRIES 16
430 * nfs_scan_list - Scan a list for matching requests
432 * @dst: Destination list
433 * @idx_start: lower bound of page->index to scan
434 * @npages: idx_start + npages sets the upper bound to scan.
435 * @tag: tag to scan for
437 * Moves elements from one of the inode request lists.
438 * If the number of requests is set to 0, the entire address_space
439 * starting at index idx_start, is scanned.
440 * The requests are *not* checked to ensure that they form a contiguous set.
441 * You must be holding the inode's i_lock when calling this function
443 int nfs_scan_list(struct nfs_inode
*nfsi
,
444 struct list_head
*dst
, pgoff_t idx_start
,
445 unsigned int npages
, int tag
)
447 struct nfs_page
*pgvec
[NFS_SCAN_MAXENTRIES
];
448 struct nfs_page
*req
;
452 struct list_head
*list
;
458 idx_end
= idx_start
+ npages
- 1;
461 found
= radix_tree_gang_lookup_tag(&nfsi
->nfs_page_tree
,
462 (void **)&pgvec
[0], idx_start
,
463 NFS_SCAN_MAXENTRIES
, tag
);
466 for (i
= 0; i
< found
; i
++) {
468 if (req
->wb_index
> idx_end
)
470 idx_start
= req
->wb_index
+ 1;
471 if (nfs_set_page_tag_locked(req
)) {
472 kref_get(&req
->wb_kref
);
473 radix_tree_tag_clear(&nfsi
->nfs_page_tree
,
475 list
= pnfs_choose_commit_list(req
, dst
);
476 nfs_list_add_request(req
, list
);
482 /* for latency reduction */
483 cond_resched_lock(&nfsi
->vfs_inode
.i_lock
);
489 int __init
nfs_init_nfspagecache(void)
491 nfs_page_cachep
= kmem_cache_create("nfs_page",
492 sizeof(struct nfs_page
),
493 0, SLAB_HWCACHE_ALIGN
,
495 if (nfs_page_cachep
== NULL
)
501 void nfs_destroy_nfspagecache(void)
503 kmem_cache_destroy(nfs_page_cachep
);