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)
61 struct nfs_server *server = NFS_SERVER(inode);
65 /* try to allocate the request struct */
66 req = nfs_page_alloc();
70 if (signalled() && (server->flags & NFS_MOUNT_INTR))
71 return ERR_PTR(-ERESTARTSYS);
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 struct nfs_inode *nfsi = NFS_I(req->wb_context->path.dentry->d_inode);
118 if (!nfs_lock_request_dontget(req))
120 if (req->wb_page != NULL)
121 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
126 * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers
128 void nfs_clear_page_tag_locked(struct nfs_page *req)
130 struct inode *inode = req->wb_context->path.dentry->d_inode;
131 struct nfs_inode *nfsi = NFS_I(inode);
133 if (req->wb_page != NULL) {
134 spin_lock(&inode->i_lock);
135 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
136 nfs_unlock_request(req);
137 spin_unlock(&inode->i_lock);
139 nfs_unlock_request(req);
143 * nfs_clear_request - Free up all resources allocated to the request
146 * Release page resources associated with a write request after it
149 void nfs_clear_request(struct nfs_page *req)
151 struct page *page = req->wb_page;
153 page_cache_release(page);
160 * nfs_release_request - Release the count on an NFS read/write request
161 * @req: request to release
163 * Note: Should never be called with the spinlock held!
165 static void nfs_free_request(struct kref *kref)
167 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
169 /* Release struct file or cached credential */
170 nfs_clear_request(req);
171 put_nfs_open_context(req->wb_context);
175 void nfs_release_request(struct nfs_page *req)
177 kref_put(&req->wb_kref, nfs_free_request);
180 static int nfs_wait_bit_interruptible(void *word)
184 if (signal_pending(current))
192 * nfs_wait_on_request - Wait for a request to complete.
193 * @req: request to wait upon.
195 * Interruptible by signals only if mounted with intr flag.
196 * The user is responsible for holding a count on the request.
199 nfs_wait_on_request(struct nfs_page *req)
201 struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->path.dentry->d_inode);
205 if (!test_bit(PG_BUSY, &req->wb_flags))
208 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
209 * are not interrupted if intr flag is not set
211 rpc_clnt_sigmask(clnt, &oldmask);
212 ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
213 nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
214 rpc_clnt_sigunmask(clnt, &oldmask);
220 * nfs_pageio_init - initialise a page io descriptor
221 * @desc: pointer to descriptor
222 * @inode: pointer to inode
223 * @doio: pointer to io function
224 * @bsize: io block size
225 * @io_flags: extra parameters for the io function
227 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
229 int (*doio)(struct inode *, struct list_head *, unsigned int, size_t, int),
233 INIT_LIST_HEAD(&desc->pg_list);
234 desc->pg_bytes_written = 0;
236 desc->pg_bsize = bsize;
238 desc->pg_inode = inode;
239 desc->pg_doio = doio;
240 desc->pg_ioflags = io_flags;
245 * nfs_can_coalesce_requests - test two requests for compatibility
246 * @prev: pointer to nfs_page
247 * @req: pointer to nfs_page
249 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
250 * page data area they describe is contiguous, and that their RPC
251 * credentials, NFSv4 open state, and lockowners are the same.
253 * Return 'true' if this is the case, else return 'false'.
255 static int nfs_can_coalesce_requests(struct nfs_page *prev,
256 struct nfs_page *req)
258 if (req->wb_context->cred != prev->wb_context->cred)
260 if (req->wb_context->lockowner != prev->wb_context->lockowner)
262 if (req->wb_context->state != prev->wb_context->state)
264 if (req->wb_index != (prev->wb_index + 1))
266 if (req->wb_pgbase != 0)
268 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
274 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
275 * @desc: destination io descriptor
278 * Returns true if the request 'req' was successfully coalesced into the
279 * existing list of pages 'desc'.
281 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
282 struct nfs_page *req)
284 size_t newlen = req->wb_bytes;
286 if (desc->pg_count != 0) {
287 struct nfs_page *prev;
290 * FIXME: ideally we should be able to coalesce all requests
291 * that are not block boundary aligned, but currently this
292 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
293 * since nfs_flush_multi and nfs_pagein_multi assume you
294 * can have only one struct nfs_page.
296 if (desc->pg_bsize < PAGE_SIZE)
298 newlen += desc->pg_count;
299 if (newlen > desc->pg_bsize)
301 prev = nfs_list_entry(desc->pg_list.prev);
302 if (!nfs_can_coalesce_requests(prev, req))
305 desc->pg_base = req->wb_pgbase;
306 nfs_list_remove_request(req);
307 nfs_list_add_request(req, &desc->pg_list);
308 desc->pg_count = newlen;
313 * Helper for nfs_pageio_add_request and nfs_pageio_complete
315 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
317 if (!list_empty(&desc->pg_list)) {
318 int error = desc->pg_doio(desc->pg_inode,
320 nfs_page_array_len(desc->pg_base,
325 desc->pg_error = error;
327 desc->pg_bytes_written += desc->pg_count;
329 if (list_empty(&desc->pg_list)) {
336 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
337 * @desc: destination io descriptor
340 * Returns true if the request 'req' was successfully coalesced into the
341 * existing list of pages 'desc'.
343 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
344 struct nfs_page *req)
346 while (!nfs_pageio_do_add_request(desc, req)) {
347 nfs_pageio_doio(desc);
348 if (desc->pg_error < 0)
355 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
356 * @desc: pointer to io descriptor
358 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
360 nfs_pageio_doio(desc);
364 * nfs_pageio_cond_complete - Conditional I/O completion
365 * @desc: pointer to io descriptor
368 * It is important to ensure that processes don't try to take locks
369 * on non-contiguous ranges of pages as that might deadlock. This
370 * function should be called before attempting to wait on a locked
371 * nfs_page. It will complete the I/O if the page index 'index'
372 * is not contiguous with the existing list of pages in 'desc'.
374 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
376 if (!list_empty(&desc->pg_list)) {
377 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
378 if (index != prev->wb_index + 1)
379 nfs_pageio_doio(desc);
383 #define NFS_SCAN_MAXENTRIES 16
385 * nfs_scan_list - Scan a list for matching requests
387 * @dst: Destination list
388 * @idx_start: lower bound of page->index to scan
389 * @npages: idx_start + npages sets the upper bound to scan.
390 * @tag: tag to scan for
392 * Moves elements from one of the inode request lists.
393 * If the number of requests is set to 0, the entire address_space
394 * starting at index idx_start, is scanned.
395 * The requests are *not* checked to ensure that they form a contiguous set.
396 * You must be holding the inode's i_lock when calling this function
398 int nfs_scan_list(struct nfs_inode *nfsi,
399 struct list_head *dst, pgoff_t idx_start,
400 unsigned int npages, int tag)
402 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
403 struct nfs_page *req;
412 idx_end = idx_start + npages - 1;
415 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
416 (void **)&pgvec[0], idx_start,
417 NFS_SCAN_MAXENTRIES, tag);
420 for (i = 0; i < found; i++) {
422 if (req->wb_index > idx_end)
424 idx_start = req->wb_index + 1;
425 if (nfs_set_page_tag_locked(req)) {
426 kref_get(&req->wb_kref);
427 nfs_list_remove_request(req);
428 radix_tree_tag_clear(&nfsi->nfs_page_tree,
430 nfs_list_add_request(req, dst);
436 /* for latency reduction */
437 cond_resched_lock(&nfsi->vfs_inode.i_lock);
443 int __init nfs_init_nfspagecache(void)
445 nfs_page_cachep = kmem_cache_create("nfs_page",
446 sizeof(struct nfs_page),
447 0, SLAB_HWCACHE_ALIGN,
449 if (nfs_page_cachep == NULL)
455 void nfs_destroy_nfspagecache(void)
457 kmem_cache_destroy(nfs_page_cachep);