blob: f56dae5216f4dfadd82bc316a439487c8fd2cc72 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/pagelist.c
3 *
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.
7 *
8 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
9 *
10 */
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/slab.h>
13#include <linux/file.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040014#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#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>
21
Trond Myklebust8d5658c2007-04-10 09:26:35 -040022#include "internal.h"
23
Christoph Lametere18b8902006-12-06 20:33:20 -080024static struct kmem_cache *nfs_page_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26static inline struct nfs_page *
27nfs_page_alloc(void)
28{
29 struct nfs_page *p;
Christoph Lametere94b1762006-12-06 20:33:17 -080030 p = kmem_cache_alloc(nfs_page_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 if (p) {
32 memset(p, 0, sizeof(*p));
33 INIT_LIST_HEAD(&p->wb_list);
34 }
35 return p;
36}
37
38static inline void
39nfs_page_free(struct nfs_page *p)
40{
41 kmem_cache_free(nfs_page_cachep, p);
42}
43
44/**
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
51 *
52 * The page must be locked by the caller. This makes sure we never
Jason Uhlenkotta19b89c2007-04-26 17:25:51 -070053 * create two different requests for the same page.
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 * User should ensure it is safe to sleep in this function.
55 */
56struct nfs_page *
57nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
58 struct page *page,
59 unsigned int offset, unsigned int count)
60{
61 struct nfs_server *server = NFS_SERVER(inode);
62 struct nfs_page *req;
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 for (;;) {
65 /* try to allocate the request struct */
66 req = nfs_page_alloc();
67 if (req != NULL)
68 break;
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 if (signalled() && (server->flags & NFS_MOUNT_INTR))
71 return ERR_PTR(-ERESTARTSYS);
72 yield();
73 }
74
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. */
78 req->wb_page = page;
79 atomic_set(&req->wb_complete, 0);
80 req->wb_index = page->index;
81 page_cache_get(page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -050082 BUG_ON(PagePrivate(page));
83 BUG_ON(!PageLocked(page));
84 BUG_ON(page->mapping->host != inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 req->wb_offset = offset;
86 req->wb_pgbase = offset;
87 req->wb_bytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 req->wb_context = get_nfs_open_context(ctx);
Trond Myklebustc03b4022007-06-17 13:26:38 -040089 kref_init(&req->wb_kref);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 return req;
91}
92
93/**
94 * nfs_unlock_request - Unlock request and wake up sleepers.
95 * @req:
96 */
97void nfs_unlock_request(struct nfs_page *req)
98{
99 if (!NFS_WBACK_BUSY(req)) {
100 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
101 BUG();
102 }
103 smp_mb__before_clear_bit();
104 clear_bit(PG_BUSY, &req->wb_flags);
105 smp_mb__after_clear_bit();
Trond Myklebust464a98b2005-06-22 17:16:21 +0000106 wake_up_bit(&req->wb_flags, PG_BUSY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 nfs_release_request(req);
108}
109
110/**
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400111 * nfs_set_page_tag_locked - Tag a request as locked
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000112 * @req:
113 */
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400114static int nfs_set_page_tag_locked(struct nfs_page *req)
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000115{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400116 struct nfs_inode *nfsi = NFS_I(req->wb_context->path.dentry->d_inode);
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000117
118 if (!nfs_lock_request(req))
119 return 0;
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400120 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000121 return 1;
122}
123
124/**
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400125 * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000126 */
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400127void nfs_clear_page_tag_locked(struct nfs_page *req)
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000128{
Trond Myklebust587142f2007-07-02 09:57:54 -0400129 struct inode *inode = req->wb_context->path.dentry->d_inode;
130 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000131
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500132 if (req->wb_page != NULL) {
Trond Myklebust587142f2007-07-02 09:57:54 -0400133 spin_lock(&inode->i_lock);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400134 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
Trond Myklebust587142f2007-07-02 09:57:54 -0400135 spin_unlock(&inode->i_lock);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500136 }
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000137 nfs_unlock_request(req);
138}
139
140/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 * nfs_clear_request - Free up all resources allocated to the request
142 * @req:
143 *
144 * Release page resources associated with a write request after it
145 * has completed.
146 */
147void nfs_clear_request(struct nfs_page *req)
148{
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500149 struct page *page = req->wb_page;
150 if (page != NULL) {
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500151 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 req->wb_page = NULL;
153 }
154}
155
156
157/**
158 * nfs_release_request - Release the count on an NFS read/write request
159 * @req: request to release
160 *
161 * Note: Should never be called with the spinlock held!
162 */
Trond Myklebustc03b4022007-06-17 13:26:38 -0400163static void nfs_free_request(struct kref *kref)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
Trond Myklebustc03b4022007-06-17 13:26:38 -0400165 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 /* Release struct file or cached credential */
168 nfs_clear_request(req);
169 put_nfs_open_context(req->wb_context);
170 nfs_page_free(req);
171}
172
Trond Myklebustc03b4022007-06-17 13:26:38 -0400173void nfs_release_request(struct nfs_page *req)
174{
175 kref_put(&req->wb_kref, nfs_free_request);
176}
177
Trond Myklebust464a98b2005-06-22 17:16:21 +0000178static int nfs_wait_bit_interruptible(void *word)
179{
180 int ret = 0;
181
182 if (signal_pending(current))
183 ret = -ERESTARTSYS;
184 else
185 schedule();
186 return ret;
187}
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/**
190 * nfs_wait_on_request - Wait for a request to complete.
191 * @req: request to wait upon.
192 *
193 * Interruptible by signals only if mounted with intr flag.
194 * The user is responsible for holding a count on the request.
195 */
196int
197nfs_wait_on_request(struct nfs_page *req)
198{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400199 struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->path.dentry->d_inode);
Trond Myklebust464a98b2005-06-22 17:16:21 +0000200 sigset_t oldmask;
201 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Trond Myklebust464a98b2005-06-22 17:16:21 +0000203 if (!test_bit(PG_BUSY, &req->wb_flags))
204 goto out;
205 /*
206 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
207 * are not interrupted if intr flag is not set
208 */
209 rpc_clnt_sigmask(clnt, &oldmask);
210 ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
211 nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
212 rpc_clnt_sigunmask(clnt, &oldmask);
213out:
214 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
217/**
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400218 * nfs_pageio_init - initialise a page io descriptor
219 * @desc: pointer to descriptor
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400220 * @inode: pointer to inode
221 * @doio: pointer to io function
222 * @bsize: io block size
223 * @io_flags: extra parameters for the io function
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400224 */
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400225void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
226 struct inode *inode,
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400227 int (*doio)(struct inode *, struct list_head *, unsigned int, size_t, int),
Trond Myklebust84dde762007-05-04 14:44:06 -0400228 size_t bsize,
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400229 int io_flags)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400230{
231 INIT_LIST_HEAD(&desc->pg_list);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400232 desc->pg_bytes_written = 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400233 desc->pg_count = 0;
234 desc->pg_bsize = bsize;
235 desc->pg_base = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400236 desc->pg_inode = inode;
237 desc->pg_doio = doio;
238 desc->pg_ioflags = io_flags;
239 desc->pg_error = 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400240}
241
242/**
243 * nfs_can_coalesce_requests - test two requests for compatibility
244 * @prev: pointer to nfs_page
245 * @req: pointer to nfs_page
246 *
247 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
248 * page data area they describe is contiguous, and that their RPC
249 * credentials, NFSv4 open state, and lockowners are the same.
250 *
251 * Return 'true' if this is the case, else return 'false'.
252 */
253static int nfs_can_coalesce_requests(struct nfs_page *prev,
254 struct nfs_page *req)
255{
256 if (req->wb_context->cred != prev->wb_context->cred)
257 return 0;
258 if (req->wb_context->lockowner != prev->wb_context->lockowner)
259 return 0;
260 if (req->wb_context->state != prev->wb_context->state)
261 return 0;
262 if (req->wb_index != (prev->wb_index + 1))
263 return 0;
264 if (req->wb_pgbase != 0)
265 return 0;
266 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
267 return 0;
268 return 1;
269}
270
271/**
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400272 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400273 * @desc: destination io descriptor
274 * @req: request
275 *
276 * Returns true if the request 'req' was successfully coalesced into the
277 * existing list of pages 'desc'.
278 */
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400279static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
280 struct nfs_page *req)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400281{
282 size_t newlen = req->wb_bytes;
283
284 if (desc->pg_count != 0) {
285 struct nfs_page *prev;
286
287 /*
288 * FIXME: ideally we should be able to coalesce all requests
289 * that are not block boundary aligned, but currently this
290 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
291 * since nfs_flush_multi and nfs_pagein_multi assume you
292 * can have only one struct nfs_page.
293 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400294 if (desc->pg_bsize < PAGE_SIZE)
295 return 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400296 newlen += desc->pg_count;
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400297 if (newlen > desc->pg_bsize)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400298 return 0;
299 prev = nfs_list_entry(desc->pg_list.prev);
300 if (!nfs_can_coalesce_requests(prev, req))
301 return 0;
302 } else
303 desc->pg_base = req->wb_pgbase;
304 nfs_list_remove_request(req);
305 nfs_list_add_request(req, &desc->pg_list);
306 desc->pg_count = newlen;
307 return 1;
308}
309
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400310/*
311 * Helper for nfs_pageio_add_request and nfs_pageio_complete
312 */
313static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
314{
315 if (!list_empty(&desc->pg_list)) {
316 int error = desc->pg_doio(desc->pg_inode,
317 &desc->pg_list,
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400318 nfs_page_array_len(desc->pg_base,
319 desc->pg_count),
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400320 desc->pg_count,
321 desc->pg_ioflags);
322 if (error < 0)
323 desc->pg_error = error;
324 else
325 desc->pg_bytes_written += desc->pg_count;
326 }
327 if (list_empty(&desc->pg_list)) {
328 desc->pg_count = 0;
329 desc->pg_base = 0;
330 }
331}
332
333/**
334 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
335 * @desc: destination io descriptor
336 * @req: request
337 *
338 * Returns true if the request 'req' was successfully coalesced into the
339 * existing list of pages 'desc'.
340 */
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400341int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
342 struct nfs_page *req)
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400343{
344 while (!nfs_pageio_do_add_request(desc, req)) {
345 nfs_pageio_doio(desc);
346 if (desc->pg_error < 0)
347 return 0;
348 }
349 return 1;
350}
351
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400352/**
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400353 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
354 * @desc: pointer to io descriptor
355 */
356void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
357{
358 nfs_pageio_doio(desc);
359}
360
Trond Myklebust7fe7f842007-05-20 10:18:27 -0400361/**
362 * nfs_pageio_cond_complete - Conditional I/O completion
363 * @desc: pointer to io descriptor
364 * @index: page index
365 *
366 * It is important to ensure that processes don't try to take locks
367 * on non-contiguous ranges of pages as that might deadlock. This
368 * function should be called before attempting to wait on a locked
369 * nfs_page. It will complete the I/O if the page index 'index'
370 * is not contiguous with the existing list of pages in 'desc'.
371 */
372void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
373{
374 if (!list_empty(&desc->pg_list)) {
375 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
376 if (index != prev->wb_index + 1)
377 nfs_pageio_doio(desc);
378 }
379}
380
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000381#define NFS_SCAN_MAXENTRIES 16
382/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 * nfs_scan_list - Scan a list for matching requests
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400384 * @nfsi: NFS inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 * @dst: Destination list
386 * @idx_start: lower bound of page->index to scan
387 * @npages: idx_start + npages sets the upper bound to scan.
Trond Myklebust5c369682007-06-17 15:27:42 -0400388 * @tag: tag to scan for
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 *
390 * Moves elements from one of the inode request lists.
391 * If the number of requests is set to 0, the entire address_space
392 * starting at index idx_start, is scanned.
393 * The requests are *not* checked to ensure that they form a contiguous set.
Trond Myklebust587142f2007-07-02 09:57:54 -0400394 * You must be holding the inode's i_lock when calling this function
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 */
Trond Myklebust5c369682007-06-17 15:27:42 -0400396int nfs_scan_list(struct nfs_inode *nfsi,
Trond Myklebustca52fec2007-04-17 17:22:13 -0400397 struct list_head *dst, pgoff_t idx_start,
Trond Myklebust5c369682007-06-17 15:27:42 -0400398 unsigned int npages, int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400400 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
401 struct nfs_page *req;
Trond Myklebustca52fec2007-04-17 17:22:13 -0400402 pgoff_t idx_end;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400403 int found, i;
404 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
406 res = 0;
407 if (npages == 0)
408 idx_end = ~0;
409 else
410 idx_end = idx_start + npages - 1;
411
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400412 for (;;) {
Trond Myklebust5c369682007-06-17 15:27:42 -0400413 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400414 (void **)&pgvec[0], idx_start,
Trond Myklebust5c369682007-06-17 15:27:42 -0400415 NFS_SCAN_MAXENTRIES, tag);
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400416 if (found <= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400418 for (i = 0; i < found; i++) {
419 req = pgvec[i];
420 if (req->wb_index > idx_end)
421 goto out;
422 idx_start = req->wb_index + 1;
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400423 if (nfs_set_page_tag_locked(req)) {
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400424 nfs_list_remove_request(req);
Trond Myklebust5c369682007-06-17 15:27:42 -0400425 radix_tree_tag_clear(&nfsi->nfs_page_tree,
426 req->wb_index, tag);
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400427 nfs_list_add_request(req, dst);
428 res++;
Trond Myklebustdce34ce2007-06-17 15:47:53 -0400429 if (res == INT_MAX)
430 goto out;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400431 }
432 }
Trond Myklebustedc05fc2007-06-17 16:02:34 -0400433 /* for latency reduction */
Trond Myklebust587142f2007-07-02 09:57:54 -0400434 cond_resched_lock(&nfsi->vfs_inode.i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 }
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400436out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 return res;
438}
439
David Howellsf7b422b2006-06-09 09:34:33 -0400440int __init nfs_init_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
442 nfs_page_cachep = kmem_cache_create("nfs_page",
443 sizeof(struct nfs_page),
444 0, SLAB_HWCACHE_ALIGN,
445 NULL, NULL);
446 if (nfs_page_cachep == NULL)
447 return -ENOMEM;
448
449 return 0;
450}
451
David Brownell266bee82006-06-27 12:59:15 -0700452void nfs_destroy_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700454 kmem_cache_destroy(nfs_page_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456