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 struct nfs_inode *nfsi = NFS_I(req->wb_context->path.dentry->d_inode);
117 if (!nfs_lock_request_dontget(req))
119 if (req->wb_page != NULL)
120 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
125 * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers
127 void nfs_clear_page_tag_locked(struct nfs_page *req)
129 struct inode *inode = req->wb_context->path.dentry->d_inode;
130 struct nfs_inode *nfsi = NFS_I(inode);
132 if (req->wb_page != NULL) {
133 spin_lock(&inode->i_lock);
134 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
135 nfs_unlock_request(req);
136 spin_unlock(&inode->i_lock);
138 nfs_unlock_request(req);
142 * nfs_clear_request - Free up all resources allocated to the request
145 * Release page resources associated with a write request after it
148 void nfs_clear_request(struct nfs_page *req)
150 struct page *page = req->wb_page;
152 page_cache_release(page);
159 * nfs_release_request - Release the count on an NFS read/write request
160 * @req: request to release
162 * Note: Should never be called with the spinlock held!
164 static void nfs_free_request(struct kref *kref)
166 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
168 /* Release struct file or cached credential */
169 nfs_clear_request(req);
170 put_nfs_open_context(req->wb_context);
174 void nfs_release_request(struct nfs_page *req)
176 kref_put(&req->wb_kref, nfs_free_request);
179 static int nfs_wait_bit_killable(void *word)
183 if (fatal_signal_pending(current))
191 * nfs_wait_on_request - Wait for a request to complete.
192 * @req: request to wait upon.
194 * Interruptible by fatal signals only.
195 * The user is responsible for holding a count on the request.
198 nfs_wait_on_request(struct nfs_page *req)
202 if (!test_bit(PG_BUSY, &req->wb_flags))
204 ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
205 nfs_wait_bit_killable, TASK_KILLABLE);
211 * nfs_pageio_init - initialise a page io descriptor
212 * @desc: pointer to descriptor
213 * @inode: pointer to inode
214 * @doio: pointer to io function
215 * @bsize: io block size
216 * @io_flags: extra parameters for the io function
218 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
220 int (*doio)(struct inode *, struct list_head *, unsigned int, size_t, int),
224 INIT_LIST_HEAD(&desc->pg_list);
225 desc->pg_bytes_written = 0;
227 desc->pg_bsize = bsize;
229 desc->pg_inode = inode;
230 desc->pg_doio = doio;
231 desc->pg_ioflags = io_flags;
236 * nfs_can_coalesce_requests - test two requests for compatibility
237 * @prev: pointer to nfs_page
238 * @req: pointer to nfs_page
240 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
241 * page data area they describe is contiguous, and that their RPC
242 * credentials, NFSv4 open state, and lockowners are the same.
244 * Return 'true' if this is the case, else return 'false'.
246 static int nfs_can_coalesce_requests(struct nfs_page *prev,
247 struct nfs_page *req)
249 if (req->wb_context->cred != prev->wb_context->cred)
251 if (req->wb_context->lockowner != prev->wb_context->lockowner)
253 if (req->wb_context->state != prev->wb_context->state)
255 if (req->wb_index != (prev->wb_index + 1))
257 if (req->wb_pgbase != 0)
259 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
265 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
266 * @desc: destination io descriptor
269 * Returns true if the request 'req' was successfully coalesced into the
270 * existing list of pages 'desc'.
272 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
273 struct nfs_page *req)
275 size_t newlen = req->wb_bytes;
277 if (desc->pg_count != 0) {
278 struct nfs_page *prev;
281 * FIXME: ideally we should be able to coalesce all requests
282 * that are not block boundary aligned, but currently this
283 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
284 * since nfs_flush_multi and nfs_pagein_multi assume you
285 * can have only one struct nfs_page.
287 if (desc->pg_bsize < PAGE_SIZE)
289 newlen += desc->pg_count;
290 if (newlen > desc->pg_bsize)
292 prev = nfs_list_entry(desc->pg_list.prev);
293 if (!nfs_can_coalesce_requests(prev, req))
296 desc->pg_base = req->wb_pgbase;
297 nfs_list_remove_request(req);
298 nfs_list_add_request(req, &desc->pg_list);
299 desc->pg_count = newlen;
304 * Helper for nfs_pageio_add_request and nfs_pageio_complete
306 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
308 if (!list_empty(&desc->pg_list)) {
309 int error = desc->pg_doio(desc->pg_inode,
311 nfs_page_array_len(desc->pg_base,
316 desc->pg_error = error;
318 desc->pg_bytes_written += desc->pg_count;
320 if (list_empty(&desc->pg_list)) {
327 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
328 * @desc: destination io descriptor
331 * Returns true if the request 'req' was successfully coalesced into the
332 * existing list of pages 'desc'.
334 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
335 struct nfs_page *req)
337 while (!nfs_pageio_do_add_request(desc, req)) {
338 nfs_pageio_doio(desc);
339 if (desc->pg_error < 0)
346 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
347 * @desc: pointer to io descriptor
349 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
351 nfs_pageio_doio(desc);
355 * nfs_pageio_cond_complete - Conditional I/O completion
356 * @desc: pointer to io descriptor
359 * It is important to ensure that processes don't try to take locks
360 * on non-contiguous ranges of pages as that might deadlock. This
361 * function should be called before attempting to wait on a locked
362 * nfs_page. It will complete the I/O if the page index 'index'
363 * is not contiguous with the existing list of pages in 'desc'.
365 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
367 if (!list_empty(&desc->pg_list)) {
368 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
369 if (index != prev->wb_index + 1)
370 nfs_pageio_doio(desc);
374 #define NFS_SCAN_MAXENTRIES 16
376 * nfs_scan_list - Scan a list for matching requests
378 * @dst: Destination list
379 * @idx_start: lower bound of page->index to scan
380 * @npages: idx_start + npages sets the upper bound to scan.
381 * @tag: tag to scan for
383 * Moves elements from one of the inode request lists.
384 * If the number of requests is set to 0, the entire address_space
385 * starting at index idx_start, is scanned.
386 * The requests are *not* checked to ensure that they form a contiguous set.
387 * You must be holding the inode's i_lock when calling this function
389 int nfs_scan_list(struct nfs_inode *nfsi,
390 struct list_head *dst, pgoff_t idx_start,
391 unsigned int npages, int tag)
393 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
394 struct nfs_page *req;
403 idx_end = idx_start + npages - 1;
406 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
407 (void **)&pgvec[0], idx_start,
408 NFS_SCAN_MAXENTRIES, tag);
411 for (i = 0; i < found; i++) {
413 if (req->wb_index > idx_end)
415 idx_start = req->wb_index + 1;
416 if (nfs_set_page_tag_locked(req)) {
417 kref_get(&req->wb_kref);
418 nfs_list_remove_request(req);
419 radix_tree_tag_clear(&nfsi->nfs_page_tree,
421 nfs_list_add_request(req, dst);
427 /* for latency reduction */
428 cond_resched_lock(&nfsi->vfs_inode.i_lock);
434 int __init nfs_init_nfspagecache(void)
436 nfs_page_cachep = kmem_cache_create("nfs_page",
437 sizeof(struct nfs_page),
438 0, SLAB_HWCACHE_ALIGN,
440 if (nfs_page_cachep == NULL)
446 void nfs_destroy_nfspagecache(void)
448 kmem_cache_destroy(nfs_page_cachep);