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/sunrpc/clnt.h>
15 #include <linux/nfs3.h>
16 #include <linux/nfs4.h>
17 #include <linux/nfs_page.h>
18 #include <linux/nfs_fs.h>
19 #include <linux/nfs_mount.h>
20 #include <linux/writeback.h>
22 #define NFS_PARANOIA 1
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, and avoids
54 * a possible deadlock when we reach the hard limit on the number
56 * User should ensure it is safe to sleep in this function.
59 nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
61 unsigned int offset, unsigned int count)
63 struct nfs_server *server = NFS_SERVER(inode);
66 /* Deal with hard limits. */
68 /* try to allocate the request struct */
69 req = nfs_page_alloc();
73 /* Try to free up at least one request in order to stay
74 * below the hard limit
76 if (signalled() && (server->flags & NFS_MOUNT_INTR))
77 return ERR_PTR(-ERESTARTSYS);
81 /* Initialize the request struct. Initially, we assume a
82 * long write-back delay. This will be adjusted in
83 * update_nfs_request below if the region is not locked. */
85 atomic_set(&req->wb_complete, 0);
86 req->wb_index = page->index;
88 BUG_ON(PagePrivate(page));
89 BUG_ON(!PageLocked(page));
90 BUG_ON(page->mapping->host != inode);
91 req->wb_offset = offset;
92 req->wb_pgbase = offset;
93 req->wb_bytes = count;
94 atomic_set(&req->wb_count, 1);
95 req->wb_context = get_nfs_open_context(ctx);
101 * nfs_unlock_request - Unlock request and wake up sleepers.
104 void nfs_unlock_request(struct nfs_page *req)
106 if (!NFS_WBACK_BUSY(req)) {
107 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
110 smp_mb__before_clear_bit();
111 clear_bit(PG_BUSY, &req->wb_flags);
112 smp_mb__after_clear_bit();
113 wake_up_bit(&req->wb_flags, PG_BUSY);
114 nfs_release_request(req);
118 * nfs_set_page_writeback_locked - Lock a request for writeback
121 int nfs_set_page_writeback_locked(struct nfs_page *req)
123 struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
125 if (!nfs_lock_request(req))
127 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
132 * nfs_clear_page_writeback - Unlock request and wake up sleepers
134 void nfs_clear_page_writeback(struct nfs_page *req)
136 struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
138 if (req->wb_page != NULL) {
139 spin_lock(&nfsi->req_lock);
140 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
141 spin_unlock(&nfsi->req_lock);
143 nfs_unlock_request(req);
147 * nfs_clear_request - Free up all resources allocated to the request
150 * Release page resources associated with a write request after it
153 void nfs_clear_request(struct nfs_page *req)
155 struct page *page = req->wb_page;
157 page_cache_release(page);
164 * nfs_release_request - Release the count on an NFS read/write request
165 * @req: request to release
167 * Note: Should never be called with the spinlock held!
170 nfs_release_request(struct nfs_page *req)
172 if (!atomic_dec_and_test(&req->wb_count))
176 BUG_ON (!list_empty(&req->wb_list));
177 BUG_ON (NFS_WBACK_BUSY(req));
180 /* Release struct file or cached credential */
181 nfs_clear_request(req);
182 put_nfs_open_context(req->wb_context);
186 static int nfs_wait_bit_interruptible(void *word)
190 if (signal_pending(current))
198 * nfs_wait_on_request - Wait for a request to complete.
199 * @req: request to wait upon.
201 * Interruptible by signals only if mounted with intr flag.
202 * The user is responsible for holding a count on the request.
205 nfs_wait_on_request(struct nfs_page *req)
207 struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
211 if (!test_bit(PG_BUSY, &req->wb_flags))
214 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
215 * are not interrupted if intr flag is not set
217 rpc_clnt_sigmask(clnt, &oldmask);
218 ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
219 nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
220 rpc_clnt_sigunmask(clnt, &oldmask);
226 * nfs_pageio_init - initialise a page io descriptor
227 * @desc: pointer to descriptor
228 * @iosize: io block size
230 void nfs_pageio_init(struct nfs_pageio_descriptor *desc, unsigned int bsize)
232 INIT_LIST_HEAD(&desc->pg_list);
234 desc->pg_bsize = bsize;
239 * nfs_can_coalesce_requests - test two requests for compatibility
240 * @prev: pointer to nfs_page
241 * @req: pointer to nfs_page
243 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
244 * page data area they describe is contiguous, and that their RPC
245 * credentials, NFSv4 open state, and lockowners are the same.
247 * Return 'true' if this is the case, else return 'false'.
249 static int nfs_can_coalesce_requests(struct nfs_page *prev,
250 struct nfs_page *req)
252 if (req->wb_context->cred != prev->wb_context->cred)
254 if (req->wb_context->lockowner != prev->wb_context->lockowner)
256 if (req->wb_context->state != prev->wb_context->state)
258 if (req->wb_index != (prev->wb_index + 1))
260 if (req->wb_pgbase != 0)
262 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
268 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
269 * @desc: destination io descriptor
272 * Returns true if the request 'req' was successfully coalesced into the
273 * existing list of pages 'desc'.
275 static int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
276 struct nfs_page *req)
278 size_t newlen = req->wb_bytes;
280 if (desc->pg_count != 0) {
281 struct nfs_page *prev;
284 * FIXME: ideally we should be able to coalesce all requests
285 * that are not block boundary aligned, but currently this
286 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
287 * since nfs_flush_multi and nfs_pagein_multi assume you
288 * can have only one struct nfs_page.
290 newlen += desc->pg_count;
291 if (desc->pg_base + newlen > desc->pg_bsize)
293 prev = nfs_list_entry(desc->pg_list.prev);
294 if (!nfs_can_coalesce_requests(prev, req))
297 desc->pg_base = req->wb_pgbase;
298 nfs_list_remove_request(req);
299 nfs_list_add_request(req, &desc->pg_list);
300 desc->pg_count = newlen;
305 * nfs_pageio_add_list - Split coalesced requests out from a list.
306 * @desc: destination io descriptor
309 * Moves a maximum of 'nmax' elements from one list to another.
310 * The elements are checked to ensure that they form a contiguous set
311 * of pages, and that the RPC credentials are the same.
313 void nfs_pageio_add_list(struct nfs_pageio_descriptor *desc,
314 struct list_head *head)
316 while (!list_empty(head)) {
317 struct nfs_page *req = nfs_list_entry(head->next);
318 if (!nfs_pageio_add_request(desc, req))
323 #define NFS_SCAN_MAXENTRIES 16
325 * nfs_scan_dirty - Scan the radix tree for dirty requests
326 * @mapping: pointer to address space
327 * @wbc: writeback_control structure
328 * @dst: Destination list
330 * Moves elements from one of the inode request lists.
331 * If the number of requests is set to 0, the entire address_space
332 * starting at index idx_start, is scanned.
333 * The requests are *not* checked to ensure that they form a contiguous set.
334 * You must be holding the inode's req_lock when calling this function
336 long nfs_scan_dirty(struct address_space *mapping,
337 struct writeback_control *wbc,
338 struct list_head *dst)
340 struct nfs_inode *nfsi = NFS_I(mapping->host);
341 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
342 struct nfs_page *req;
343 pgoff_t idx_start, idx_end;
347 if (nfsi->ndirty == 0)
349 if (wbc->range_cyclic) {
352 } else if (wbc->range_end == 0) {
353 idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
356 idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
357 idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
361 unsigned int toscan = NFS_SCAN_MAXENTRIES;
363 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
364 (void **)&pgvec[0], idx_start, toscan,
367 /* Did we make progress? */
371 for (i = 0; i < found; i++) {
373 if (!wbc->range_cyclic && req->wb_index > idx_end)
376 /* Try to lock request and mark it for writeback */
377 if (!nfs_set_page_writeback_locked(req))
379 radix_tree_tag_clear(&nfsi->nfs_page_tree,
380 req->wb_index, NFS_PAGE_TAG_DIRTY);
382 nfs_list_remove_request(req);
383 nfs_list_add_request(req, dst);
388 idx_start = req->wb_index + 1;
392 WARN_ON ((nfsi->ndirty == 0) != list_empty(&nfsi->dirty));
397 * nfs_scan_list - Scan a list for matching requests
399 * @head: One of the NFS inode request lists
400 * @dst: Destination list
401 * @idx_start: lower bound of page->index to scan
402 * @npages: idx_start + npages sets the upper bound to scan.
404 * Moves elements from one of the inode request lists.
405 * If the number of requests is set to 0, the entire address_space
406 * starting at index idx_start, is scanned.
407 * The requests are *not* checked to ensure that they form a contiguous set.
408 * You must be holding the inode's req_lock when calling this function
410 int nfs_scan_list(struct nfs_inode *nfsi, struct list_head *head,
411 struct list_head *dst, unsigned long idx_start,
414 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
415 struct nfs_page *req;
416 unsigned long idx_end;
424 idx_end = idx_start + npages - 1;
427 found = radix_tree_gang_lookup(&nfsi->nfs_page_tree,
428 (void **)&pgvec[0], idx_start,
429 NFS_SCAN_MAXENTRIES);
432 for (i = 0; i < found; i++) {
434 if (req->wb_index > idx_end)
436 idx_start = req->wb_index + 1;
437 if (req->wb_list_head != head)
439 if (nfs_set_page_writeback_locked(req)) {
440 nfs_list_remove_request(req);
441 nfs_list_add_request(req, dst);
451 int __init nfs_init_nfspagecache(void)
453 nfs_page_cachep = kmem_cache_create("nfs_page",
454 sizeof(struct nfs_page),
455 0, SLAB_HWCACHE_ALIGN,
457 if (nfs_page_cachep == NULL)
463 void nfs_destroy_nfspagecache(void)
465 kmem_cache_destroy(nfs_page_cachep);