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>
25 static struct kmem_cache *nfs_page_cachep;
27 static inline struct nfs_page *
30 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_KERNEL);
32 INIT_LIST_HEAD(&p->wb_list);
37 nfs_page_free(struct nfs_page *p)
39 kmem_cache_free(nfs_page_cachep, p);
43 * nfs_create_request - Create an NFS read/write request.
44 * @file: file descriptor to use
45 * @inode: inode to which the request is attached
46 * @page: page to write
47 * @offset: starting offset within the page for the write
48 * @count: number of bytes to read/write
50 * The page must be locked by the caller. This makes sure we never
51 * create two different requests for the same page.
52 * User should ensure it is safe to sleep in this function.
55 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
57 unsigned int offset, unsigned int count)
61 /* try to allocate the request struct */
62 req = nfs_page_alloc();
64 return ERR_PTR(-ENOMEM);
66 /* get lock context early so we can deal with alloc failures */
67 req->wb_lock_context = nfs_get_lock_context(ctx);
68 if (req->wb_lock_context == NULL) {
70 return ERR_PTR(-ENOMEM);
73 /* Initialize the request struct. Initially, we assume a
74 * long write-back delay. This will be adjusted in
75 * update_nfs_request below if the region is not locked. */
77 atomic_set(&req->wb_complete, 0);
78 req->wb_index = page->index;
80 BUG_ON(PagePrivate(page));
81 BUG_ON(!PageLocked(page));
82 BUG_ON(page->mapping->host != inode);
83 req->wb_offset = offset;
84 req->wb_pgbase = offset;
85 req->wb_bytes = count;
86 req->wb_context = get_nfs_open_context(ctx);
87 kref_init(&req->wb_kref);
92 * nfs_unlock_request - Unlock request and wake up sleepers.
95 void nfs_unlock_request(struct nfs_page *req)
97 if (!NFS_WBACK_BUSY(req)) {
98 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
101 smp_mb__before_clear_bit();
102 clear_bit(PG_BUSY, &req->wb_flags);
103 smp_mb__after_clear_bit();
104 wake_up_bit(&req->wb_flags, PG_BUSY);
105 nfs_release_request(req);
109 * nfs_set_page_tag_locked - Tag a request as locked
112 int nfs_set_page_tag_locked(struct nfs_page *req)
114 if (!nfs_lock_request_dontget(req))
116 if (test_bit(PG_MAPPED, &req->wb_flags))
117 radix_tree_tag_set(&NFS_I(req->wb_context->path.dentry->d_inode)->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
122 * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers
124 void nfs_clear_page_tag_locked(struct nfs_page *req)
126 if (test_bit(PG_MAPPED, &req->wb_flags)) {
127 struct inode *inode = req->wb_context->path.dentry->d_inode;
128 struct nfs_inode *nfsi = NFS_I(inode);
130 spin_lock(&inode->i_lock);
131 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
132 nfs_unlock_request(req);
133 spin_unlock(&inode->i_lock);
135 nfs_unlock_request(req);
139 * nfs_clear_request - Free up all resources allocated to the request
142 * Release page and open context resources associated with a read/write
143 * request after it has completed.
145 static void nfs_clear_request(struct nfs_page *req)
147 struct page *page = req->wb_page;
148 struct nfs_open_context *ctx = req->wb_context;
149 struct nfs_lock_context *l_ctx = req->wb_lock_context;
152 page_cache_release(page);
156 nfs_put_lock_context(l_ctx);
157 req->wb_lock_context = NULL;
160 put_nfs_open_context(ctx);
161 req->wb_context = NULL;
167 * nfs_release_request - Release the count on an NFS read/write request
168 * @req: request to release
170 * Note: Should never be called with the spinlock held!
172 static void nfs_free_request(struct kref *kref)
174 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
176 /* Release struct file and open context */
177 nfs_clear_request(req);
181 void nfs_release_request(struct nfs_page *req)
183 kref_put(&req->wb_kref, nfs_free_request);
186 static int nfs_wait_bit_uninterruptible(void *word)
193 * nfs_wait_on_request - Wait for a request to complete.
194 * @req: request to wait upon.
196 * Interruptible by fatal signals only.
197 * The user is responsible for holding a count on the request.
200 nfs_wait_on_request(struct nfs_page *req)
202 return wait_on_bit(&req->wb_flags, PG_BUSY,
203 nfs_wait_bit_uninterruptible,
204 TASK_UNINTERRUPTIBLE);
207 bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
210 * FIXME: ideally we should be able to coalesce all requests
211 * that are not block boundary aligned, but currently this
212 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
213 * since nfs_flush_multi and nfs_pagein_multi assume you
214 * can have only one struct nfs_page.
216 if (desc->pg_bsize < PAGE_SIZE)
219 return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
221 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
224 * nfs_pageio_init - initialise a page io descriptor
225 * @desc: pointer to descriptor
226 * @inode: pointer to inode
227 * @doio: pointer to io function
228 * @bsize: io block size
229 * @io_flags: extra parameters for the io function
231 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
233 const struct nfs_pageio_ops *pg_ops,
237 INIT_LIST_HEAD(&desc->pg_list);
238 desc->pg_bytes_written = 0;
240 desc->pg_bsize = bsize;
243 desc->pg_inode = inode;
244 desc->pg_ops = pg_ops;
245 desc->pg_ioflags = io_flags;
247 desc->pg_lseg = NULL;
251 * nfs_can_coalesce_requests - test two requests for compatibility
252 * @prev: pointer to nfs_page
253 * @req: pointer to nfs_page
255 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
256 * page data area they describe is contiguous, and that their RPC
257 * credentials, NFSv4 open state, and lockowners are the same.
259 * Return 'true' if this is the case, else return 'false'.
261 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
262 struct nfs_page *req,
263 struct nfs_pageio_descriptor *pgio)
265 if (req->wb_context->cred != prev->wb_context->cred)
267 if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner)
269 if (req->wb_context->state != prev->wb_context->state)
271 if (req->wb_index != (prev->wb_index + 1))
273 if (req->wb_pgbase != 0)
275 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
277 return pgio->pg_ops->pg_test(pgio, prev, req);
281 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
282 * @desc: destination io descriptor
285 * Returns true if the request 'req' was successfully coalesced into the
286 * existing list of pages 'desc'.
288 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
289 struct nfs_page *req)
291 if (desc->pg_count != 0) {
292 struct nfs_page *prev;
294 prev = nfs_list_entry(desc->pg_list.prev);
295 if (!nfs_can_coalesce_requests(prev, req, desc))
298 desc->pg_base = req->wb_pgbase;
300 nfs_list_remove_request(req);
301 nfs_list_add_request(req, &desc->pg_list);
302 desc->pg_count += req->wb_bytes;
307 * Helper for nfs_pageio_add_request and nfs_pageio_complete
309 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
311 if (!list_empty(&desc->pg_list)) {
312 int error = desc->pg_ops->pg_doio(desc);
314 desc->pg_error = error;
316 desc->pg_bytes_written += desc->pg_count;
318 if (list_empty(&desc->pg_list)) {
325 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
326 * @desc: destination io descriptor
329 * Returns true if the request 'req' was successfully coalesced into the
330 * existing list of pages 'desc'.
332 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
333 struct nfs_page *req)
335 while (!nfs_pageio_do_add_request(desc, req)) {
337 nfs_pageio_doio(desc);
338 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;
398 struct list_head *list;
404 idx_end = idx_start + npages - 1;
407 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
408 (void **)&pgvec[0], idx_start,
409 NFS_SCAN_MAXENTRIES, tag);
412 for (i = 0; i < found; i++) {
414 if (req->wb_index > idx_end)
416 idx_start = req->wb_index + 1;
417 if (nfs_set_page_tag_locked(req)) {
418 kref_get(&req->wb_kref);
419 radix_tree_tag_clear(&nfsi->nfs_page_tree,
421 list = pnfs_choose_commit_list(req, dst);
422 nfs_list_add_request(req, list);
428 /* for latency reduction */
429 cond_resched_lock(&nfsi->vfs_inode.i_lock);
435 int __init nfs_init_nfspagecache(void)
437 nfs_page_cachep = kmem_cache_create("nfs_page",
438 sizeof(struct nfs_page),
439 0, SLAB_HWCACHE_ALIGN,
441 if (nfs_page_cachep == NULL)
447 void nfs_destroy_nfspagecache(void)
449 kmem_cache_destroy(nfs_page_cachep);