blob: 919490232e17ea887fd96b50f0b709e12b5f0520 [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{
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 struct nfs_page *req;
62
Trond Myklebust18eb8842010-05-13 12:51:02 -040063 /* try to allocate the request struct */
64 req = nfs_page_alloc();
65 if (req == NULL)
66 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68 /* Initialize the request struct. Initially, we assume a
69 * long write-back delay. This will be adjusted in
70 * update_nfs_request below if the region is not locked. */
71 req->wb_page = page;
72 atomic_set(&req->wb_complete, 0);
73 req->wb_index = page->index;
74 page_cache_get(page);
Trond Myklebustcd52ed32006-03-20 13:44:04 -050075 BUG_ON(PagePrivate(page));
76 BUG_ON(!PageLocked(page));
77 BUG_ON(page->mapping->host != inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 req->wb_offset = offset;
79 req->wb_pgbase = offset;
80 req->wb_bytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 req->wb_context = get_nfs_open_context(ctx);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -040082 req->wb_lock_context = nfs_get_lock_context(ctx);
Trond Myklebustc03b4022007-06-17 13:26:38 -040083 kref_init(&req->wb_kref);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 return req;
85}
86
87/**
88 * nfs_unlock_request - Unlock request and wake up sleepers.
89 * @req:
90 */
91void nfs_unlock_request(struct nfs_page *req)
92{
93 if (!NFS_WBACK_BUSY(req)) {
94 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
95 BUG();
96 }
97 smp_mb__before_clear_bit();
98 clear_bit(PG_BUSY, &req->wb_flags);
99 smp_mb__after_clear_bit();
Trond Myklebust464a98b2005-06-22 17:16:21 +0000100 wake_up_bit(&req->wb_flags, PG_BUSY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 nfs_release_request(req);
102}
103
104/**
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400105 * nfs_set_page_tag_locked - Tag a request as locked
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000106 * @req:
107 */
Trond Myklebustacee4782008-01-22 17:13:07 -0500108int nfs_set_page_tag_locked(struct nfs_page *req)
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000109{
Trond Myklebustacee4782008-01-22 17:13:07 -0500110 if (!nfs_lock_request_dontget(req))
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000111 return 0;
Trond Myklebustacee4782008-01-22 17:13:07 -0500112 if (req->wb_page != NULL)
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500113 radix_tree_tag_set(&NFS_I(req->wb_context->path.dentry->d_inode)->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000114 return 1;
115}
116
117/**
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400118 * nfs_clear_page_tag_locked - Clear request tag and wake up sleepers
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000119 */
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400120void nfs_clear_page_tag_locked(struct nfs_page *req)
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000121{
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500122 if (req->wb_page != NULL) {
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500123 struct inode *inode = req->wb_context->path.dentry->d_inode;
124 struct nfs_inode *nfsi = NFS_I(inode);
125
Trond Myklebust587142f2007-07-02 09:57:54 -0400126 spin_lock(&inode->i_lock);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400127 radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_LOCKED);
Trond Myklebustacee4782008-01-22 17:13:07 -0500128 nfs_unlock_request(req);
Trond Myklebust587142f2007-07-02 09:57:54 -0400129 spin_unlock(&inode->i_lock);
Trond Myklebustacee4782008-01-22 17:13:07 -0500130 } else
131 nfs_unlock_request(req);
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000132}
133
134/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 * nfs_clear_request - Free up all resources allocated to the request
136 * @req:
137 *
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500138 * Release page and open context resources associated with a read/write
139 * request after it has completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 */
141void nfs_clear_request(struct nfs_page *req)
142{
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500143 struct page *page = req->wb_page;
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500144 struct nfs_open_context *ctx = req->wb_context;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400145 struct nfs_lock_context *l_ctx = req->wb_lock_context;
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500146
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500147 if (page != NULL) {
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500148 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 req->wb_page = NULL;
150 }
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400151 if (l_ctx != NULL) {
152 nfs_put_lock_context(l_ctx);
153 req->wb_lock_context = NULL;
154 }
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500155 if (ctx != NULL) {
156 put_nfs_open_context(ctx);
157 req->wb_context = NULL;
158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159}
160
161
162/**
163 * nfs_release_request - Release the count on an NFS read/write request
164 * @req: request to release
165 *
166 * Note: Should never be called with the spinlock held!
167 */
Trond Myklebustc03b4022007-06-17 13:26:38 -0400168static void nfs_free_request(struct kref *kref)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Trond Myklebustc03b4022007-06-17 13:26:38 -0400170 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500172 /* Release struct file and open context */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 nfs_clear_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 nfs_page_free(req);
175}
176
Trond Myklebustc03b4022007-06-17 13:26:38 -0400177void nfs_release_request(struct nfs_page *req)
178{
179 kref_put(&req->wb_kref, nfs_free_request);
180}
181
Trond Myklebust9f557cd2010-02-03 08:27:22 -0500182static int nfs_wait_bit_uninterruptible(void *word)
183{
184 io_schedule();
185 return 0;
186}
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188/**
189 * nfs_wait_on_request - Wait for a request to complete.
190 * @req: request to wait upon.
191 *
Matthew Wilcox150030b2007-12-06 16:24:39 -0500192 * Interruptible by fatal signals only.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 * The user is responsible for holding a count on the request.
194 */
195int
196nfs_wait_on_request(struct nfs_page *req)
197{
Trond Myklebust9f557cd2010-02-03 08:27:22 -0500198 return wait_on_bit(&req->wb_flags, PG_BUSY,
199 nfs_wait_bit_uninterruptible,
200 TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
202
203/**
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400204 * nfs_pageio_init - initialise a page io descriptor
205 * @desc: pointer to descriptor
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400206 * @inode: pointer to inode
207 * @doio: pointer to io function
208 * @bsize: io block size
209 * @io_flags: extra parameters for the io function
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400210 */
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400211void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
212 struct inode *inode,
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400213 int (*doio)(struct inode *, struct list_head *, unsigned int, size_t, int),
Trond Myklebust84dde762007-05-04 14:44:06 -0400214 size_t bsize,
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400215 int io_flags)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400216{
217 INIT_LIST_HEAD(&desc->pg_list);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400218 desc->pg_bytes_written = 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400219 desc->pg_count = 0;
220 desc->pg_bsize = bsize;
221 desc->pg_base = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400222 desc->pg_inode = inode;
223 desc->pg_doio = doio;
224 desc->pg_ioflags = io_flags;
225 desc->pg_error = 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400226}
227
228/**
229 * nfs_can_coalesce_requests - test two requests for compatibility
230 * @prev: pointer to nfs_page
231 * @req: pointer to nfs_page
232 *
233 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
234 * page data area they describe is contiguous, and that their RPC
235 * credentials, NFSv4 open state, and lockowners are the same.
236 *
237 * Return 'true' if this is the case, else return 'false'.
238 */
239static int nfs_can_coalesce_requests(struct nfs_page *prev,
240 struct nfs_page *req)
241{
242 if (req->wb_context->cred != prev->wb_context->cred)
243 return 0;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400244 if (req->wb_lock_context->lockowner != prev->wb_lock_context->lockowner)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400245 return 0;
246 if (req->wb_context->state != prev->wb_context->state)
247 return 0;
248 if (req->wb_index != (prev->wb_index + 1))
249 return 0;
250 if (req->wb_pgbase != 0)
251 return 0;
252 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
253 return 0;
254 return 1;
255}
256
257/**
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400258 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400259 * @desc: destination io descriptor
260 * @req: request
261 *
262 * Returns true if the request 'req' was successfully coalesced into the
263 * existing list of pages 'desc'.
264 */
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400265static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
266 struct nfs_page *req)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400267{
268 size_t newlen = req->wb_bytes;
269
270 if (desc->pg_count != 0) {
271 struct nfs_page *prev;
272
273 /*
274 * FIXME: ideally we should be able to coalesce all requests
275 * that are not block boundary aligned, but currently this
276 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
277 * since nfs_flush_multi and nfs_pagein_multi assume you
278 * can have only one struct nfs_page.
279 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400280 if (desc->pg_bsize < PAGE_SIZE)
281 return 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400282 newlen += desc->pg_count;
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400283 if (newlen > desc->pg_bsize)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400284 return 0;
285 prev = nfs_list_entry(desc->pg_list.prev);
286 if (!nfs_can_coalesce_requests(prev, req))
287 return 0;
288 } else
289 desc->pg_base = req->wb_pgbase;
290 nfs_list_remove_request(req);
291 nfs_list_add_request(req, &desc->pg_list);
292 desc->pg_count = newlen;
293 return 1;
294}
295
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400296/*
297 * Helper for nfs_pageio_add_request and nfs_pageio_complete
298 */
299static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
300{
301 if (!list_empty(&desc->pg_list)) {
302 int error = desc->pg_doio(desc->pg_inode,
303 &desc->pg_list,
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400304 nfs_page_array_len(desc->pg_base,
305 desc->pg_count),
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400306 desc->pg_count,
307 desc->pg_ioflags);
308 if (error < 0)
309 desc->pg_error = error;
310 else
311 desc->pg_bytes_written += desc->pg_count;
312 }
313 if (list_empty(&desc->pg_list)) {
314 desc->pg_count = 0;
315 desc->pg_base = 0;
316 }
317}
318
319/**
320 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
321 * @desc: destination io descriptor
322 * @req: request
323 *
324 * Returns true if the request 'req' was successfully coalesced into the
325 * existing list of pages 'desc'.
326 */
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400327int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
328 struct nfs_page *req)
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400329{
330 while (!nfs_pageio_do_add_request(desc, req)) {
331 nfs_pageio_doio(desc);
332 if (desc->pg_error < 0)
333 return 0;
334 }
335 return 1;
336}
337
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400338/**
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400339 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
340 * @desc: pointer to io descriptor
341 */
342void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
343{
344 nfs_pageio_doio(desc);
345}
346
Trond Myklebust7fe7f842007-05-20 10:18:27 -0400347/**
348 * nfs_pageio_cond_complete - Conditional I/O completion
349 * @desc: pointer to io descriptor
350 * @index: page index
351 *
352 * It is important to ensure that processes don't try to take locks
353 * on non-contiguous ranges of pages as that might deadlock. This
354 * function should be called before attempting to wait on a locked
355 * nfs_page. It will complete the I/O if the page index 'index'
356 * is not contiguous with the existing list of pages in 'desc'.
357 */
358void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
359{
360 if (!list_empty(&desc->pg_list)) {
361 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
362 if (index != prev->wb_index + 1)
363 nfs_pageio_doio(desc);
364 }
365}
366
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000367#define NFS_SCAN_MAXENTRIES 16
368/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 * nfs_scan_list - Scan a list for matching requests
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400370 * @nfsi: NFS inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 * @dst: Destination list
372 * @idx_start: lower bound of page->index to scan
373 * @npages: idx_start + npages sets the upper bound to scan.
Trond Myklebust5c369682007-06-17 15:27:42 -0400374 * @tag: tag to scan for
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 *
376 * Moves elements from one of the inode request lists.
377 * If the number of requests is set to 0, the entire address_space
378 * starting at index idx_start, is scanned.
379 * The requests are *not* checked to ensure that they form a contiguous set.
Trond Myklebust587142f2007-07-02 09:57:54 -0400380 * You must be holding the inode's i_lock when calling this function
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 */
Trond Myklebust5c369682007-06-17 15:27:42 -0400382int nfs_scan_list(struct nfs_inode *nfsi,
Trond Myklebustca52fec2007-04-17 17:22:13 -0400383 struct list_head *dst, pgoff_t idx_start,
Trond Myklebust5c369682007-06-17 15:27:42 -0400384 unsigned int npages, int tag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400386 struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
387 struct nfs_page *req;
Trond Myklebustca52fec2007-04-17 17:22:13 -0400388 pgoff_t idx_end;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400389 int found, i;
390 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 res = 0;
393 if (npages == 0)
394 idx_end = ~0;
395 else
396 idx_end = idx_start + npages - 1;
397
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400398 for (;;) {
Trond Myklebust5c369682007-06-17 15:27:42 -0400399 found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400400 (void **)&pgvec[0], idx_start,
Trond Myklebust5c369682007-06-17 15:27:42 -0400401 NFS_SCAN_MAXENTRIES, tag);
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400402 if (found <= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400404 for (i = 0; i < found; i++) {
405 req = pgvec[i];
406 if (req->wb_index > idx_end)
407 goto out;
408 idx_start = req->wb_index + 1;
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400409 if (nfs_set_page_tag_locked(req)) {
Trond Myklebustacee4782008-01-22 17:13:07 -0500410 kref_get(&req->wb_kref);
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400411 nfs_list_remove_request(req);
Trond Myklebust5c369682007-06-17 15:27:42 -0400412 radix_tree_tag_clear(&nfsi->nfs_page_tree,
413 req->wb_index, tag);
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400414 nfs_list_add_request(req, dst);
415 res++;
Trond Myklebustdce34ce2007-06-17 15:47:53 -0400416 if (res == INT_MAX)
417 goto out;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400418 }
419 }
Trond Myklebustedc05fc2007-06-17 16:02:34 -0400420 /* for latency reduction */
Trond Myklebust587142f2007-07-02 09:57:54 -0400421 cond_resched_lock(&nfsi->vfs_inode.i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 }
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400423out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 return res;
425}
426
David Howellsf7b422b2006-06-09 09:34:33 -0400427int __init nfs_init_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
429 nfs_page_cachep = kmem_cache_create("nfs_page",
430 sizeof(struct nfs_page),
431 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +0900432 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (nfs_page_cachep == NULL)
434 return -ENOMEM;
435
436 return 0;
437}
438
David Brownell266bee82006-06-27 12:59:15 -0700439void nfs_destroy_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700441 kmem_cache_destroy(nfs_page_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442}
443