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/nfs.h>
17 #include <linux/nfs3.h>
18 #include <linux/nfs4.h>
19 #include <linux/nfs_page.h>
20 #include <linux/nfs_fs.h>
21 #include <linux/nfs_mount.h>
22 #include <linux/export.h>
27 static struct kmem_cache *nfs_page_cachep;
29 static inline struct nfs_page *
32 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_KERNEL);
34 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 * @ctx: open context 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)
63 /* try to allocate the request struct */
64 req = nfs_page_alloc();
66 return ERR_PTR(-ENOMEM);
68 /* get lock context early so we can deal with alloc failures */
69 req->wb_lock_context = nfs_get_lock_context(ctx);
70 if (req->wb_lock_context == NULL) {
72 return ERR_PTR(-ENOMEM);
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 if (!nfs_lock_request_dontget(req))
118 if (test_bit(PG_MAPPED, &req->wb_flags))
119 radix_tree_tag_set(&NFS_I(req->wb_context->dentry->d_inode)->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
124 * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers
126 void nfs_clear_page_tag_locked(struct nfs_page *req)
128 if (test_bit(PG_MAPPED, &req->wb_flags)) {
129 struct inode *inode = req->wb_context->dentry->d_inode;
130 struct nfs_inode *nfsi = NFS_I(inode);
132 spin_lock(&inode->i_lock);
133 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
134 nfs_unlock_request(req);
135 spin_unlock(&inode->i_lock);
137 nfs_unlock_request(req);
141 * nfs_clear_request - Free up all resources allocated to the request
144 * Release page and open context resources associated with a read/write
145 * request after it has completed.
147 static void nfs_clear_request(struct nfs_page *req)
149 struct page *page = req->wb_page;
150 struct nfs_open_context *ctx = req->wb_context;
151 struct nfs_lock_context *l_ctx = req->wb_lock_context;
154 page_cache_release(page);
158 nfs_put_lock_context(l_ctx);
159 req->wb_lock_context = NULL;
162 put_nfs_open_context(ctx);
163 req->wb_context = NULL;
169 * nfs_release_request - Release the count on an NFS read/write request
170 * @req: request to release
172 * Note: Should never be called with the spinlock held!
174 static void nfs_free_request(struct kref *kref)
176 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
178 /* Release struct file and open context */
179 nfs_clear_request(req);
183 void nfs_release_request(struct nfs_page *req)
185 kref_put(&req->wb_kref, nfs_free_request);
188 static int nfs_wait_bit_uninterruptible(void *word)
195 * nfs_wait_on_request - Wait for a request to complete.
196 * @req: request to wait upon.
198 * Interruptible by fatal signals only.
199 * The user is responsible for holding a count on the request.
202 nfs_wait_on_request(struct nfs_page *req)
204 return wait_on_bit(&req->wb_flags, PG_BUSY,
205 nfs_wait_bit_uninterruptible,
206 TASK_UNINTERRUPTIBLE);
209 bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
212 * FIXME: ideally we should be able to coalesce all requests
213 * that are not block boundary aligned, but currently this
214 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
215 * since nfs_flush_multi and nfs_pagein_multi assume you
216 * can have only one struct nfs_page.
218 if (desc->pg_bsize < PAGE_SIZE)
221 return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
223 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
226 * nfs_pageio_init - initialise a page io descriptor
227 * @desc: pointer to descriptor
228 * @inode: pointer to inode
229 * @doio: pointer to io function
230 * @bsize: io block size
231 * @io_flags: extra parameters for the io function
233 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
235 const struct nfs_pageio_ops *pg_ops,
239 INIT_LIST_HEAD(&desc->pg_list);
240 desc->pg_bytes_written = 0;
242 desc->pg_bsize = bsize;
245 desc->pg_recoalesce = 0;
246 desc->pg_inode = inode;
247 desc->pg_ops = pg_ops;
248 desc->pg_ioflags = io_flags;
250 desc->pg_lseg = NULL;
254 * nfs_can_coalesce_requests - test two requests for compatibility
255 * @prev: pointer to nfs_page
256 * @req: pointer to nfs_page
258 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
259 * page data area they describe is contiguous, and that their RPC
260 * credentials, NFSv4 open state, and lockowners are the same.
262 * Return 'true' if this is the case, else return 'false'.
264 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
265 struct nfs_page *req,
266 struct nfs_pageio_descriptor *pgio)
268 if (req->wb_context->cred != prev->wb_context->cred)
270 if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner)
272 if (req->wb_context->state != prev->wb_context->state)
274 if (req->wb_index != (prev->wb_index + 1))
276 if (req->wb_pgbase != 0)
278 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
280 return pgio->pg_ops->pg_test(pgio, prev, req);
284 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
285 * @desc: destination io descriptor
288 * Returns true if the request 'req' was successfully coalesced into the
289 * existing list of pages 'desc'.
291 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
292 struct nfs_page *req)
294 if (desc->pg_count != 0) {
295 struct nfs_page *prev;
297 prev = nfs_list_entry(desc->pg_list.prev);
298 if (!nfs_can_coalesce_requests(prev, req, desc))
301 if (desc->pg_ops->pg_init)
302 desc->pg_ops->pg_init(desc, req);
303 desc->pg_base = req->wb_pgbase;
305 nfs_list_remove_request(req);
306 nfs_list_add_request(req, &desc->pg_list);
307 desc->pg_count += req->wb_bytes;
312 * Helper for nfs_pageio_add_request and nfs_pageio_complete
314 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
316 if (!list_empty(&desc->pg_list)) {
317 int error = desc->pg_ops->pg_doio(desc);
319 desc->pg_error = error;
321 desc->pg_bytes_written += desc->pg_count;
323 if (list_empty(&desc->pg_list)) {
330 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
331 * @desc: destination io descriptor
334 * Returns true if the request 'req' was successfully coalesced into the
335 * existing list of pages 'desc'.
337 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
338 struct nfs_page *req)
340 while (!nfs_pageio_do_add_request(desc, req)) {
342 nfs_pageio_doio(desc);
343 if (desc->pg_error < 0)
346 if (desc->pg_recoalesce)
352 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
357 list_splice_init(&desc->pg_list, &head);
358 desc->pg_bytes_written -= desc->pg_count;
361 desc->pg_recoalesce = 0;
363 while (!list_empty(&head)) {
364 struct nfs_page *req;
366 req = list_first_entry(&head, struct nfs_page, wb_list);
367 nfs_list_remove_request(req);
368 if (__nfs_pageio_add_request(desc, req))
370 if (desc->pg_error < 0)
374 } while (desc->pg_recoalesce);
378 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
379 struct nfs_page *req)
384 ret = __nfs_pageio_add_request(desc, req);
387 if (desc->pg_error < 0)
389 ret = nfs_do_recoalesce(desc);
395 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
396 * @desc: pointer to io descriptor
398 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
401 nfs_pageio_doio(desc);
402 if (!desc->pg_recoalesce)
404 if (!nfs_do_recoalesce(desc))
410 * nfs_pageio_cond_complete - Conditional I/O completion
411 * @desc: pointer to io descriptor
414 * It is important to ensure that processes don't try to take locks
415 * on non-contiguous ranges of pages as that might deadlock. This
416 * function should be called before attempting to wait on a locked
417 * nfs_page. It will complete the I/O if the page index 'index'
418 * is not contiguous with the existing list of pages in 'desc'.
420 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
422 if (!list_empty(&desc->pg_list)) {
423 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
424 if (index != prev->wb_index + 1)
425 nfs_pageio_complete(desc);
429 #define NFS_SCAN_MAXENTRIES 16
431 * nfs_scan_list - Scan a list for matching requests
433 * @dst: Destination list
434 * @idx_start: lower bound of page->index to scan
435 * @npages: idx_start + npages sets the upper bound to scan.
436 * @tag: tag to scan for
438 * Moves elements from one of the inode request lists.
439 * If the number of requests is set to 0, the entire address_space
440 * starting at index idx_start, is scanned.
441 * The requests are *not* checked to ensure that they form a contiguous set.
442 * You must be holding the inode's i_lock when calling this function
444 int nfs_scan_list(struct nfs_inode *nfsi,
445 struct list_head *dst, pgoff_t idx_start,
446 unsigned int npages, int tag)
448 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
449 struct nfs_page *req;
453 struct list_head *list;
459 idx_end = idx_start + npages - 1;
462 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
463 (void **)&pgvec[0], idx_start,
464 NFS_SCAN_MAXENTRIES, tag);
467 for (i = 0; i < found; i++) {
469 if (req->wb_index > idx_end)
471 idx_start = req->wb_index + 1;
472 if (nfs_set_page_tag_locked(req)) {
473 kref_get(&req->wb_kref);
474 radix_tree_tag_clear(&nfsi->nfs_page_tree,
476 list = pnfs_choose_commit_list(req, dst);
477 nfs_list_add_request(req, list);
483 /* for latency reduction */
484 cond_resched_lock(&nfsi->vfs_inode.i_lock);
490 int __init nfs_init_nfspagecache(void)
492 nfs_page_cachep = kmem_cache_create("nfs_page",
493 sizeof(struct nfs_page),
494 0, SLAB_HWCACHE_ALIGN,
496 if (nfs_page_cachep == NULL)
502 void nfs_destroy_nfspagecache(void)
504 kmem_cache_destroy(nfs_page_cachep);