blob: 03ed984ab4d8016e3e062a7fc95c2b3268e6287c [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>
Trond Myklebust1313e602012-01-17 22:04:24 -050016#include <linux/nfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#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>
Paul Gortmakerafeacc82011-05-26 16:00:52 -040022#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
Trond Myklebust8d5658c2007-04-10 09:26:35 -040024#include "internal.h"
Fred Isamanbae724e2011-03-01 01:34:15 +000025#include "pnfs.h"
Trond Myklebust8d5658c2007-04-10 09:26:35 -040026
Christoph Lametere18b8902006-12-06 20:33:20 -080027static struct kmem_cache *nfs_page_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
Fred Isaman30dd3742012-04-20 14:47:45 -040029bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
30{
31 p->npages = pagecount;
32 if (pagecount <= ARRAY_SIZE(p->page_array))
33 p->pagevec = p->page_array;
34 else {
35 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
36 if (!p->pagevec)
37 p->npages = 0;
38 }
39 return p->pagevec != NULL;
40}
41
Fred Isaman4db6e0b2012-04-20 14:47:46 -040042void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
43 struct nfs_pgio_header *hdr,
44 void (*release)(struct nfs_pgio_header *hdr))
45{
46 hdr->req = nfs_list_entry(desc->pg_list.next);
47 hdr->inode = desc->pg_inode;
48 hdr->cred = hdr->req->wb_context->cred;
49 hdr->io_start = req_offset(hdr->req);
50 hdr->good_bytes = desc->pg_count;
Fred Isaman584aa812012-04-20 14:47:51 -040051 hdr->dreq = desc->pg_dreq;
Peng Taof6166382012-08-02 15:36:09 +030052 hdr->layout_private = desc->pg_layout_private;
Fred Isaman4db6e0b2012-04-20 14:47:46 -040053 hdr->release = release;
Fred Isaman061ae2e2012-04-20 14:47:48 -040054 hdr->completion_ops = desc->pg_completion_ops;
Fred Isaman584aa812012-04-20 14:47:51 -040055 if (hdr->completion_ops->init_hdr)
56 hdr->completion_ops->init_hdr(hdr);
Fred Isaman4db6e0b2012-04-20 14:47:46 -040057}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040058EXPORT_SYMBOL_GPL(nfs_pgheader_init);
Fred Isaman4db6e0b2012-04-20 14:47:46 -040059
60void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
61{
62 spin_lock(&hdr->lock);
63 if (pos < hdr->io_start + hdr->good_bytes) {
64 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
65 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
66 hdr->good_bytes = pos - hdr->io_start;
67 hdr->error = error;
68 }
69 spin_unlock(&hdr->lock);
70}
71
Linus Torvalds1da177e2005-04-16 15:20:36 -070072static inline struct nfs_page *
73nfs_page_alloc(void)
74{
Mel Gorman192e5012012-07-31 16:45:16 -070075 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
Jesper Juhl72895b12010-12-09 23:17:15 +010076 if (p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 INIT_LIST_HEAD(&p->wb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 return p;
79}
80
81static inline void
82nfs_page_free(struct nfs_page *p)
83{
84 kmem_cache_free(nfs_page_cachep, p);
85}
86
Trond Myklebust577b4232013-04-08 21:38:12 -040087static void
88nfs_iocounter_inc(struct nfs_io_counter *c)
89{
90 atomic_inc(&c->io_count);
91}
92
93static void
94nfs_iocounter_dec(struct nfs_io_counter *c)
95{
96 if (atomic_dec_and_test(&c->io_count)) {
97 clear_bit(NFS_IO_INPROGRESS, &c->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +010098 smp_mb__after_atomic();
Trond Myklebust577b4232013-04-08 21:38:12 -040099 wake_up_bit(&c->flags, NFS_IO_INPROGRESS);
100 }
101}
102
103static int
104__nfs_iocounter_wait(struct nfs_io_counter *c)
105{
106 wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS);
107 DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS);
108 int ret = 0;
109
110 do {
111 prepare_to_wait(wq, &q.wait, TASK_KILLABLE);
112 set_bit(NFS_IO_INPROGRESS, &c->flags);
113 if (atomic_read(&c->io_count) == 0)
114 break;
115 ret = nfs_wait_bit_killable(&c->flags);
116 } while (atomic_read(&c->io_count) != 0);
117 finish_wait(wq, &q.wait);
118 return ret;
119}
120
121/**
122 * nfs_iocounter_wait - wait for i/o to complete
123 * @c: nfs_io_counter to use
124 *
125 * returns -ERESTARTSYS if interrupted by a fatal signal.
126 * Otherwise returns 0 once the io_count hits 0.
127 */
128int
129nfs_iocounter_wait(struct nfs_io_counter *c)
130{
131 if (atomic_read(&c->io_count) == 0)
132 return 0;
133 return __nfs_iocounter_wait(c);
134}
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136/**
137 * nfs_create_request - Create an NFS read/write request.
Chuck Leverc02f5572011-10-25 12:17:43 -0400138 * @ctx: open context to use
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 * @inode: inode to which the request is attached
140 * @page: page to write
141 * @offset: starting offset within the page for the write
142 * @count: number of bytes to read/write
143 *
144 * The page must be locked by the caller. This makes sure we never
Jason Uhlenkotta19b89c2007-04-26 17:25:51 -0700145 * create two different requests for the same page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 * User should ensure it is safe to sleep in this function.
147 */
148struct nfs_page *
149nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
150 struct page *page,
151 unsigned int offset, unsigned int count)
152{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 struct nfs_page *req;
Trond Myklebustb3c54de2012-08-13 17:15:50 -0400154 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
Trond Myklebustc58c8442013-03-18 19:45:14 -0400156 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
157 return ERR_PTR(-EBADF);
Trond Myklebust18eb8842010-05-13 12:51:02 -0400158 /* try to allocate the request struct */
159 req = nfs_page_alloc();
160 if (req == NULL)
161 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Jeff Layton015f0212010-10-28 10:10:37 -0400163 /* get lock context early so we can deal with alloc failures */
Trond Myklebustb3c54de2012-08-13 17:15:50 -0400164 l_ctx = nfs_get_lock_context(ctx);
165 if (IS_ERR(l_ctx)) {
Jeff Layton015f0212010-10-28 10:10:37 -0400166 nfs_page_free(req);
Trond Myklebustb3c54de2012-08-13 17:15:50 -0400167 return ERR_CAST(l_ctx);
Jeff Layton015f0212010-10-28 10:10:37 -0400168 }
Trond Myklebustb3c54de2012-08-13 17:15:50 -0400169 req->wb_lock_context = l_ctx;
Trond Myklebust577b4232013-04-08 21:38:12 -0400170 nfs_iocounter_inc(&l_ctx->io_count);
Jeff Layton015f0212010-10-28 10:10:37 -0400171
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 /* Initialize the request struct. Initially, we assume a
173 * long write-back delay. This will be adjusted in
174 * update_nfs_request below if the region is not locked. */
175 req->wb_page = page;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700176 req->wb_index = page_file_index(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 page_cache_get(page);
178 req->wb_offset = offset;
179 req->wb_pgbase = offset;
180 req->wb_bytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 req->wb_context = get_nfs_open_context(ctx);
Trond Myklebustc03b4022007-06-17 13:26:38 -0400182 kref_init(&req->wb_kref);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 return req;
184}
185
186/**
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400187 * nfs_unlock_request - Unlock request and wake up sleepers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 * @req:
189 */
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400190void nfs_unlock_request(struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
192 if (!NFS_WBACK_BUSY(req)) {
193 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
194 BUG();
195 }
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100196 smp_mb__before_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 clear_bit(PG_BUSY, &req->wb_flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100198 smp_mb__after_atomic();
Trond Myklebust464a98b2005-06-22 17:16:21 +0000199 wake_up_bit(&req->wb_flags, PG_BUSY);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400200}
201
202/**
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400203 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
204 * @req:
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400205 */
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400206void nfs_unlock_and_release_request(struct nfs_page *req)
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400207{
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400208 nfs_unlock_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 nfs_release_request(req);
210}
211
Trond Myklebust4d65c522011-03-25 14:15:11 -0400212/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 * nfs_clear_request - Free up all resources allocated to the request
214 * @req:
215 *
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500216 * Release page and open context resources associated with a read/write
217 * request after it has completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 */
Trond Myklebust4d65c522011-03-25 14:15:11 -0400219static void nfs_clear_request(struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220{
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500221 struct page *page = req->wb_page;
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500222 struct nfs_open_context *ctx = req->wb_context;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400223 struct nfs_lock_context *l_ctx = req->wb_lock_context;
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500224
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500225 if (page != NULL) {
Trond Myklebustcd52ed32006-03-20 13:44:04 -0500226 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 req->wb_page = NULL;
228 }
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400229 if (l_ctx != NULL) {
Trond Myklebust577b4232013-04-08 21:38:12 -0400230 nfs_iocounter_dec(&l_ctx->io_count);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400231 nfs_put_lock_context(l_ctx);
232 req->wb_lock_context = NULL;
233 }
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500234 if (ctx != NULL) {
235 put_nfs_open_context(ctx);
236 req->wb_context = NULL;
237 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
240
241/**
242 * nfs_release_request - Release the count on an NFS read/write request
243 * @req: request to release
244 *
245 * Note: Should never be called with the spinlock held!
246 */
Trond Myklebustc03b4022007-06-17 13:26:38 -0400247static void nfs_free_request(struct kref *kref)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
Trond Myklebustc03b4022007-06-17 13:26:38 -0400249 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Trond Myklebustbb6fbc42010-03-11 09:19:35 -0500251 /* Release struct file and open context */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 nfs_clear_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 nfs_page_free(req);
254}
255
Trond Myklebustc03b4022007-06-17 13:26:38 -0400256void nfs_release_request(struct nfs_page *req)
257{
258 kref_put(&req->wb_kref, nfs_free_request);
259}
260
Trond Myklebust9f557cd2010-02-03 08:27:22 -0500261static int nfs_wait_bit_uninterruptible(void *word)
262{
263 io_schedule();
264 return 0;
265}
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267/**
268 * nfs_wait_on_request - Wait for a request to complete.
269 * @req: request to wait upon.
270 *
Matthew Wilcox150030b2007-12-06 16:24:39 -0500271 * Interruptible by fatal signals only.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 * The user is responsible for holding a count on the request.
273 */
274int
275nfs_wait_on_request(struct nfs_page *req)
276{
Trond Myklebust9f557cd2010-02-03 08:27:22 -0500277 return wait_on_bit(&req->wb_flags, PG_BUSY,
278 nfs_wait_bit_uninterruptible,
279 TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280}
281
Benny Halevy19345cb2011-06-19 18:33:46 -0400282bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300283{
284 /*
285 * FIXME: ideally we should be able to coalesce all requests
286 * that are not block boundary aligned, but currently this
287 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
288 * since nfs_flush_multi and nfs_pagein_multi assume you
289 * can have only one struct nfs_page.
290 */
291 if (desc->pg_bsize < PAGE_SIZE)
292 return 0;
293
294 return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
295}
Benny Halevy19345cb2011-06-19 18:33:46 -0400296EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298/**
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400299 * nfs_pageio_init - initialise a page io descriptor
300 * @desc: pointer to descriptor
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400301 * @inode: pointer to inode
302 * @doio: pointer to io function
303 * @bsize: io block size
304 * @io_flags: extra parameters for the io function
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400305 */
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400306void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
307 struct inode *inode,
Trond Myklebust1751c362011-06-10 13:30:23 -0400308 const struct nfs_pageio_ops *pg_ops,
Fred Isaman061ae2e2012-04-20 14:47:48 -0400309 const struct nfs_pgio_completion_ops *compl_ops,
Trond Myklebust84dde762007-05-04 14:44:06 -0400310 size_t bsize,
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400311 int io_flags)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400312{
313 INIT_LIST_HEAD(&desc->pg_list);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400314 desc->pg_bytes_written = 0;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400315 desc->pg_count = 0;
316 desc->pg_bsize = bsize;
317 desc->pg_base = 0;
Trond Myklebustb31268a2011-03-21 17:02:00 -0400318 desc->pg_moreio = 0;
Trond Myklebustd9156f92011-07-12 13:42:02 -0400319 desc->pg_recoalesce = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400320 desc->pg_inode = inode;
Trond Myklebust1751c362011-06-10 13:30:23 -0400321 desc->pg_ops = pg_ops;
Fred Isaman061ae2e2012-04-20 14:47:48 -0400322 desc->pg_completion_ops = compl_ops;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400323 desc->pg_ioflags = io_flags;
324 desc->pg_error = 0;
Fred Isaman94ad1c82011-03-01 01:34:14 +0000325 desc->pg_lseg = NULL;
Fred Isaman584aa812012-04-20 14:47:51 -0400326 desc->pg_dreq = NULL;
Peng Taof6166382012-08-02 15:36:09 +0300327 desc->pg_layout_private = NULL;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400328}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400329EXPORT_SYMBOL_GPL(nfs_pageio_init);
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400330
Trond Myklebust4109bb72013-09-06 11:09:38 -0400331static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
332 const struct nfs_open_context *ctx2)
333{
334 return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
335}
336
337static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
338 const struct nfs_lock_context *l2)
339{
340 return l1->lockowner.l_owner == l2->lockowner.l_owner
341 && l1->lockowner.l_pid == l2->lockowner.l_pid;
342}
343
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400344/**
345 * nfs_can_coalesce_requests - test two requests for compatibility
346 * @prev: pointer to nfs_page
347 * @req: pointer to nfs_page
348 *
349 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
350 * page data area they describe is contiguous, and that their RPC
351 * credentials, NFSv4 open state, and lockowners are the same.
352 *
353 * Return 'true' if this is the case, else return 'false'.
354 */
Benny Halevy18ad0a92011-05-25 21:03:56 +0300355static bool nfs_can_coalesce_requests(struct nfs_page *prev,
356 struct nfs_page *req,
357 struct nfs_pageio_descriptor *pgio)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400358{
Trond Myklebust4109bb72013-09-06 11:09:38 -0400359 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
Benny Halevy18ad0a92011-05-25 21:03:56 +0300360 return false;
Trond Myklebust4109bb72013-09-06 11:09:38 -0400361 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
362 !nfs_match_lock_context(req->wb_lock_context, prev->wb_lock_context))
Benny Halevy18ad0a92011-05-25 21:03:56 +0300363 return false;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400364 if (req->wb_pgbase != 0)
Benny Halevy18ad0a92011-05-25 21:03:56 +0300365 return false;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400366 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
Benny Halevy18ad0a92011-05-25 21:03:56 +0300367 return false;
Fred Isaman1825a0d2012-04-20 19:55:31 -0400368 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
369 return false;
Trond Myklebust1751c362011-06-10 13:30:23 -0400370 return pgio->pg_ops->pg_test(pgio, prev, req);
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400371}
372
373/**
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400374 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400375 * @desc: destination io descriptor
376 * @req: request
377 *
378 * Returns true if the request 'req' was successfully coalesced into the
379 * existing list of pages 'desc'.
380 */
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400381static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
382 struct nfs_page *req)
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400383{
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400384 if (desc->pg_count != 0) {
385 struct nfs_page *prev;
386
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400387 prev = nfs_list_entry(desc->pg_list.prev);
Fred Isaman94ad1c82011-03-01 01:34:14 +0000388 if (!nfs_can_coalesce_requests(prev, req, desc))
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400389 return 0;
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300390 } else {
Trond Myklebustd8007d42011-06-10 13:30:23 -0400391 if (desc->pg_ops->pg_init)
392 desc->pg_ops->pg_init(desc, req);
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400393 desc->pg_base = req->wb_pgbase;
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300394 }
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400395 nfs_list_remove_request(req);
396 nfs_list_add_request(req, &desc->pg_list);
Boaz Harrosh5b36c7d2011-05-29 11:45:39 +0300397 desc->pg_count += req->wb_bytes;
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400398 return 1;
399}
400
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400401/*
402 * Helper for nfs_pageio_add_request and nfs_pageio_complete
403 */
404static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
405{
406 if (!list_empty(&desc->pg_list)) {
Trond Myklebust1751c362011-06-10 13:30:23 -0400407 int error = desc->pg_ops->pg_doio(desc);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400408 if (error < 0)
409 desc->pg_error = error;
410 else
411 desc->pg_bytes_written += desc->pg_count;
412 }
413 if (list_empty(&desc->pg_list)) {
414 desc->pg_count = 0;
415 desc->pg_base = 0;
416 }
417}
418
419/**
420 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
421 * @desc: destination io descriptor
422 * @req: request
423 *
424 * Returns true if the request 'req' was successfully coalesced into the
425 * existing list of pages 'desc'.
426 */
Trond Myklebustd9156f92011-07-12 13:42:02 -0400427static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
Trond Myklebust8b09bee2007-04-02 18:48:28 -0400428 struct nfs_page *req)
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400429{
430 while (!nfs_pageio_do_add_request(desc, req)) {
Trond Myklebustb31268a2011-03-21 17:02:00 -0400431 desc->pg_moreio = 1;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400432 nfs_pageio_doio(desc);
433 if (desc->pg_error < 0)
434 return 0;
Trond Myklebustb31268a2011-03-21 17:02:00 -0400435 desc->pg_moreio = 0;
Trond Myklebustd9156f92011-07-12 13:42:02 -0400436 if (desc->pg_recoalesce)
437 return 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400438 }
439 return 1;
440}
441
Trond Myklebustd9156f92011-07-12 13:42:02 -0400442static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
443{
444 LIST_HEAD(head);
445
446 do {
447 list_splice_init(&desc->pg_list, &head);
448 desc->pg_bytes_written -= desc->pg_count;
449 desc->pg_count = 0;
450 desc->pg_base = 0;
451 desc->pg_recoalesce = 0;
452
453 while (!list_empty(&head)) {
454 struct nfs_page *req;
455
456 req = list_first_entry(&head, struct nfs_page, wb_list);
457 nfs_list_remove_request(req);
458 if (__nfs_pageio_add_request(desc, req))
459 continue;
460 if (desc->pg_error < 0)
461 return 0;
462 break;
463 }
464 } while (desc->pg_recoalesce);
465 return 1;
466}
467
468int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
469 struct nfs_page *req)
470{
471 int ret;
472
473 do {
474 ret = __nfs_pageio_add_request(desc, req);
475 if (ret)
476 break;
477 if (desc->pg_error < 0)
478 break;
479 ret = nfs_do_recoalesce(desc);
480 } while (ret);
481 return ret;
482}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400483EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
Trond Myklebustd9156f92011-07-12 13:42:02 -0400484
Trond Myklebustd8a5ad72007-04-02 18:48:28 -0400485/**
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400486 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
487 * @desc: pointer to io descriptor
488 */
489void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
490{
Trond Myklebustd9156f92011-07-12 13:42:02 -0400491 for (;;) {
492 nfs_pageio_doio(desc);
493 if (!desc->pg_recoalesce)
494 break;
495 if (!nfs_do_recoalesce(desc))
496 break;
497 }
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400498}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400499EXPORT_SYMBOL_GPL(nfs_pageio_complete);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400500
Trond Myklebust7fe7f842007-05-20 10:18:27 -0400501/**
502 * nfs_pageio_cond_complete - Conditional I/O completion
503 * @desc: pointer to io descriptor
504 * @index: page index
505 *
506 * It is important to ensure that processes don't try to take locks
507 * on non-contiguous ranges of pages as that might deadlock. This
508 * function should be called before attempting to wait on a locked
509 * nfs_page. It will complete the I/O if the page index 'index'
510 * is not contiguous with the existing list of pages in 'desc'.
511 */
512void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
513{
514 if (!list_empty(&desc->pg_list)) {
515 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
516 if (index != prev->wb_index + 1)
Trond Myklebustd9156f92011-07-12 13:42:02 -0400517 nfs_pageio_complete(desc);
Trond Myklebust7fe7f842007-05-20 10:18:27 -0400518 }
519}
520
David Howellsf7b422b2006-06-09 09:34:33 -0400521int __init nfs_init_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522{
523 nfs_page_cachep = kmem_cache_create("nfs_page",
524 sizeof(struct nfs_page),
525 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +0900526 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 if (nfs_page_cachep == NULL)
528 return -ENOMEM;
529
530 return 0;
531}
532
David Brownell266bee82006-06-27 12:59:15 -0700533void nfs_destroy_nfspagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -0700535 kmem_cache_destroy(nfs_page_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537