aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--fs/nfs/direct.c435
-rw-r--r--include/linux/nfs_xdr.h2
2 files changed, 234 insertions, 203 deletions
diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
index e25b7595b7ad..402005c35ab3 100644
--- a/fs/nfs/direct.c
+++ b/fs/nfs/direct.c
@@ -68,19 +68,25 @@ struct nfs_direct_req {
68 struct kref kref; /* release manager */ 68 struct kref kref; /* release manager */
69 69
70 /* I/O parameters */ 70 /* I/O parameters */
71 struct list_head list, /* nfs_read/write_data structs */
72 rewrite_list; /* saved nfs_write_data structs */
71 struct nfs_open_context *ctx; /* file open context info */ 73 struct nfs_open_context *ctx; /* file open context info */
72 struct kiocb * iocb; /* controlling i/o request */ 74 struct kiocb * iocb; /* controlling i/o request */
73 struct inode * inode; /* target file of i/o */ 75 struct inode * inode; /* target file of i/o */
76 unsigned long user_addr; /* location of user's buffer */
77 size_t user_count; /* total bytes to move */
78 loff_t pos; /* starting offset in file */
79 struct page ** pages; /* pages in our buffer */
80 unsigned int npages; /* count of pages */
74 81
75 /* completion state */ 82 /* completion state */
76 atomic_t io_count; /* i/os we're waiting for */
77 spinlock_t lock; /* protect completion state */ 83 spinlock_t lock; /* protect completion state */
84 int outstanding; /* i/os we're waiting for */
78 ssize_t count, /* bytes actually processed */ 85 ssize_t count, /* bytes actually processed */
79 error; /* any reported error */ 86 error; /* any reported error */
80 struct completion completion; /* wait for i/o completion */ 87 struct completion completion; /* wait for i/o completion */
81 88
82 /* commit state */ 89 /* commit state */
83 struct list_head rewrite_list; /* saved nfs_write_data structs */
84 struct nfs_write_data * commit_data; /* special write_data for commits */ 90 struct nfs_write_data * commit_data; /* special write_data for commits */
85 int flags; 91 int flags;
86#define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */ 92#define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */
@@ -88,37 +94,8 @@ struct nfs_direct_req {
88 struct nfs_writeverf verf; /* unstable write verifier */ 94 struct nfs_writeverf verf; /* unstable write verifier */
89}; 95};
90 96
97static void nfs_direct_write_schedule(struct nfs_direct_req *dreq, int sync);
91static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode); 98static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode);
92static const struct rpc_call_ops nfs_write_direct_ops;
93
94static inline void get_dreq(struct nfs_direct_req *dreq)
95{
96 atomic_inc(&dreq->io_count);
97}
98
99static inline int put_dreq(struct nfs_direct_req *dreq)
100{
101 return atomic_dec_and_test(&dreq->io_count);
102}
103
104/*
105 * "size" is never larger than rsize or wsize.
106 */
107static inline int nfs_direct_count_pages(unsigned long user_addr, size_t size)
108{
109 int page_count;
110
111 page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
112 page_count -= user_addr >> PAGE_SHIFT;
113 BUG_ON(page_count < 0);
114
115 return page_count;
116}
117
118static inline unsigned int nfs_max_pages(unsigned int size)
119{
120 return (size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
121}
122 99
123/** 100/**
124 * nfs_direct_IO - NFS address space operation for direct I/O 101 * nfs_direct_IO - NFS address space operation for direct I/O
@@ -142,21 +119,50 @@ ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_
142 return -EINVAL; 119 return -EINVAL;
143} 120}
144 121
145static void nfs_direct_dirty_pages(struct page **pages, int npages) 122static void nfs_free_user_pages(struct page **pages, int npages, int do_dirty)
146{ 123{
147 int i; 124 int i;
148 for (i = 0; i < npages; i++) { 125 for (i = 0; i < npages; i++) {
149 struct page *page = pages[i]; 126 struct page *page = pages[i];
150 if (!PageCompound(page)) 127 if (do_dirty && !PageCompound(page))
151 set_page_dirty_lock(page); 128 set_page_dirty_lock(page);
129 page_cache_release(page);
152 } 130 }
131 kfree(pages);
153} 132}
154 133
155static void nfs_direct_release_pages(struct page **pages, int npages) 134static inline int nfs_get_user_pages(int rw, unsigned long user_addr, size_t size, struct page ***pages)
156{ 135{
157 int i; 136 int result = -ENOMEM;
158 for (i = 0; i < npages; i++) 137 unsigned long page_count;
159 page_cache_release(pages[i]); 138 size_t array_size;
139
140 page_count = (user_addr + size + PAGE_SIZE - 1) >> PAGE_SHIFT;
141 page_count -= user_addr >> PAGE_SHIFT;
142
143 array_size = (page_count * sizeof(struct page *));
144 *pages = kmalloc(array_size, GFP_KERNEL);
145 if (*pages) {
146 down_read(&current->mm->mmap_sem);
147 result = get_user_pages(current, current->mm, user_addr,
148 page_count, (rw == READ), 0,
149 *pages, NULL);
150 up_read(&current->mm->mmap_sem);
151 if (result != page_count) {
152 /*
153 * If we got fewer pages than expected from
154 * get_user_pages(), the user buffer runs off the
155 * end of a mapping; return EFAULT.
156 */
157 if (result >= 0) {
158 nfs_free_user_pages(*pages, result, 0);
159 result = -EFAULT;
160 } else
161 kfree(*pages);
162 *pages = NULL;
163 }
164 }
165 return result;
160} 166}
161 167
162static inline struct nfs_direct_req *nfs_direct_req_alloc(void) 168static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
@@ -168,13 +174,13 @@ static inline struct nfs_direct_req *nfs_direct_req_alloc(void)
168 return NULL; 174 return NULL;
169 175
170 kref_init(&dreq->kref); 176 kref_init(&dreq->kref);
171 kref_get(&dreq->kref);
172 init_completion(&dreq->completion); 177 init_completion(&dreq->completion);
178 INIT_LIST_HEAD(&dreq->list);
173 INIT_LIST_HEAD(&dreq->rewrite_list); 179 INIT_LIST_HEAD(&dreq->rewrite_list);
174 dreq->iocb = NULL; 180 dreq->iocb = NULL;
175 dreq->ctx = NULL; 181 dreq->ctx = NULL;
176 spin_lock_init(&dreq->lock); 182 spin_lock_init(&dreq->lock);
177 atomic_set(&dreq->io_count, 0); 183 dreq->outstanding = 0;
178 dreq->count = 0; 184 dreq->count = 0;
179 dreq->error = 0; 185 dreq->error = 0;
180 dreq->flags = 0; 186 dreq->flags = 0;
@@ -215,11 +221,18 @@ out:
215} 221}
216 222
217/* 223/*
218 * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust 224 * We must hold a reference to all the pages in this direct read request
219 * the iocb is still valid here if this is a synchronous request. 225 * until the RPCs complete. This could be long *after* we are woken up in
226 * nfs_direct_wait (for instance, if someone hits ^C on a slow server).
227 *
228 * In addition, synchronous I/O uses a stack-allocated iocb. Thus we
229 * can't trust the iocb is still valid here if this is a synchronous
230 * request. If the waiter is woken prematurely, the iocb is long gone.
220 */ 231 */
221static void nfs_direct_complete(struct nfs_direct_req *dreq) 232static void nfs_direct_complete(struct nfs_direct_req *dreq)
222{ 233{
234 nfs_free_user_pages(dreq->pages, dreq->npages, 1);
235
223 if (dreq->iocb) { 236 if (dreq->iocb) {
224 long res = (long) dreq->error; 237 long res = (long) dreq->error;
225 if (!res) 238 if (!res)
@@ -232,10 +245,48 @@ static void nfs_direct_complete(struct nfs_direct_req *dreq)
232} 245}
233 246
234/* 247/*
235 * We must hold a reference to all the pages in this direct read request 248 * Note we also set the number of requests we have in the dreq when we are
236 * until the RPCs complete. This could be long *after* we are woken up in 249 * done. This prevents races with I/O completion so we will always wait
237 * nfs_direct_wait (for instance, if someone hits ^C on a slow server). 250 * until all requests have been dispatched and completed.
238 */ 251 */
252static struct nfs_direct_req *nfs_direct_read_alloc(size_t nbytes, size_t rsize)
253{
254 struct list_head *list;
255 struct nfs_direct_req *dreq;
256 unsigned int rpages = (rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
257
258 dreq = nfs_direct_req_alloc();
259 if (!dreq)
260 return NULL;
261
262 list = &dreq->list;
263 for(;;) {
264 struct nfs_read_data *data = nfs_readdata_alloc(rpages);
265
266 if (unlikely(!data)) {
267 while (!list_empty(list)) {
268 data = list_entry(list->next,
269 struct nfs_read_data, pages);
270 list_del(&data->pages);
271 nfs_readdata_free(data);
272 }
273 kref_put(&dreq->kref, nfs_direct_req_release);
274 return NULL;
275 }
276
277 INIT_LIST_HEAD(&data->pages);
278 list_add(&data->pages, list);
279
280 data->req = (struct nfs_page *) dreq;
281 dreq->outstanding++;
282 if (nbytes <= rsize)
283 break;
284 nbytes -= rsize;
285 }
286 kref_get(&dreq->kref);
287 return dreq;
288}
289
239static void nfs_direct_read_result(struct rpc_task *task, void *calldata) 290static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
240{ 291{
241 struct nfs_read_data *data = calldata; 292 struct nfs_read_data *data = calldata;
@@ -244,9 +295,6 @@ static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
244 if (nfs_readpage_result(task, data) != 0) 295 if (nfs_readpage_result(task, data) != 0)
245 return; 296 return;
246 297
247 nfs_direct_dirty_pages(data->pagevec, data->npages);
248 nfs_direct_release_pages(data->pagevec, data->npages);
249
250 spin_lock(&dreq->lock); 298 spin_lock(&dreq->lock);
251 299
252 if (likely(task->tk_status >= 0)) 300 if (likely(task->tk_status >= 0))
@@ -254,10 +302,13 @@ static void nfs_direct_read_result(struct rpc_task *task, void *calldata)
254 else 302 else
255 dreq->error = task->tk_status; 303 dreq->error = task->tk_status;
256 304
257 spin_unlock(&dreq->lock); 305 if (--dreq->outstanding) {
306 spin_unlock(&dreq->lock);
307 return;
308 }
258 309
259 if (put_dreq(dreq)) 310 spin_unlock(&dreq->lock);
260 nfs_direct_complete(dreq); 311 nfs_direct_complete(dreq);
261} 312}
262 313
263static const struct rpc_call_ops nfs_read_direct_ops = { 314static const struct rpc_call_ops nfs_read_direct_ops = {
@@ -266,60 +317,41 @@ static const struct rpc_call_ops nfs_read_direct_ops = {
266}; 317};
267 318
268/* 319/*
269 * For each rsize'd chunk of the user's buffer, dispatch an NFS READ 320 * For each nfs_read_data struct that was allocated on the list, dispatch
270 * operation. If nfs_readdata_alloc() or get_user_pages() fails, 321 * an NFS READ operation
271 * bail and stop sending more reads. Read length accounting is
272 * handled automatically by nfs_direct_read_result(). Otherwise, if
273 * no requests have been sent, just return an error.
274 */ 322 */
275static ssize_t nfs_direct_read_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos) 323static void nfs_direct_read_schedule(struct nfs_direct_req *dreq)
276{ 324{
277 struct nfs_open_context *ctx = dreq->ctx; 325 struct nfs_open_context *ctx = dreq->ctx;
278 struct inode *inode = ctx->dentry->d_inode; 326 struct inode *inode = ctx->dentry->d_inode;
327 struct list_head *list = &dreq->list;
328 struct page **pages = dreq->pages;
329 size_t count = dreq->user_count;
330 loff_t pos = dreq->pos;
279 size_t rsize = NFS_SERVER(inode)->rsize; 331 size_t rsize = NFS_SERVER(inode)->rsize;
280 unsigned int rpages = nfs_max_pages(rsize); 332 unsigned int curpage, pgbase;
281 unsigned int pgbase;
282 int result;
283 ssize_t started = 0;
284
285 get_dreq(dreq);
286 333
287 pgbase = user_addr & ~PAGE_MASK; 334 curpage = 0;
335 pgbase = dreq->user_addr & ~PAGE_MASK;
288 do { 336 do {
289 struct nfs_read_data *data; 337 struct nfs_read_data *data;
290 size_t bytes; 338 size_t bytes;
291 339
292 result = -ENOMEM;
293 data = nfs_readdata_alloc(rpages);
294 if (unlikely(!data))
295 break;
296
297 bytes = rsize; 340 bytes = rsize;
298 if (count < rsize) 341 if (count < rsize)
299 bytes = count; 342 bytes = count;
300 343
301 data->npages = nfs_direct_count_pages(user_addr, bytes); 344 BUG_ON(list_empty(list));
302 down_read(&current->mm->mmap_sem); 345 data = list_entry(list->next, struct nfs_read_data, pages);
303 result = get_user_pages(current, current->mm, user_addr, 346 list_del_init(&data->pages);
304 data->npages, 1, 0, data->pagevec, NULL);
305 up_read(&current->mm->mmap_sem);
306 if (unlikely(result < data->npages)) {
307 if (result > 0)
308 nfs_direct_release_pages(data->pagevec, result);
309 nfs_readdata_release(data);
310 break;
311 }
312
313 get_dreq(dreq);
314 347
315 data->req = (struct nfs_page *) dreq;
316 data->inode = inode; 348 data->inode = inode;
317 data->cred = ctx->cred; 349 data->cred = ctx->cred;
318 data->args.fh = NFS_FH(inode); 350 data->args.fh = NFS_FH(inode);
319 data->args.context = ctx; 351 data->args.context = ctx;
320 data->args.offset = pos; 352 data->args.offset = pos;
321 data->args.pgbase = pgbase; 353 data->args.pgbase = pgbase;
322 data->args.pages = data->pagevec; 354 data->args.pages = &pages[curpage];
323 data->args.count = bytes; 355 data->args.count = bytes;
324 data->res.fattr = &data->fattr; 356 data->res.fattr = &data->fattr;
325 data->res.eof = 0; 357 data->res.eof = 0;
@@ -342,35 +374,33 @@ static ssize_t nfs_direct_read_schedule(struct nfs_direct_req *dreq, unsigned lo
342 bytes, 374 bytes,
343 (unsigned long long)data->args.offset); 375 (unsigned long long)data->args.offset);
344 376
345 started += bytes;
346 user_addr += bytes;
347 pos += bytes; 377 pos += bytes;
348 pgbase += bytes; 378 pgbase += bytes;
379 curpage += pgbase >> PAGE_SHIFT;
349 pgbase &= ~PAGE_MASK; 380 pgbase &= ~PAGE_MASK;
350 381
351 count -= bytes; 382 count -= bytes;
352 } while (count != 0); 383 } while (count != 0);
353 384 BUG_ON(!list_empty(list));
354 if (put_dreq(dreq))
355 nfs_direct_complete(dreq);
356
357 if (started)
358 return 0;
359 return result < 0 ? (ssize_t) result : -EFAULT;
360} 385}
361 386
362static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos) 387static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos, struct page **pages, unsigned int nr_pages)
363{ 388{
364 ssize_t result = 0; 389 ssize_t result;
365 sigset_t oldset; 390 sigset_t oldset;
366 struct inode *inode = iocb->ki_filp->f_mapping->host; 391 struct inode *inode = iocb->ki_filp->f_mapping->host;
367 struct rpc_clnt *clnt = NFS_CLIENT(inode); 392 struct rpc_clnt *clnt = NFS_CLIENT(inode);
368 struct nfs_direct_req *dreq; 393 struct nfs_direct_req *dreq;
369 394
370 dreq = nfs_direct_req_alloc(); 395 dreq = nfs_direct_read_alloc(count, NFS_SERVER(inode)->rsize);
371 if (!dreq) 396 if (!dreq)
372 return -ENOMEM; 397 return -ENOMEM;
373 398
399 dreq->user_addr = user_addr;
400 dreq->user_count = count;
401 dreq->pos = pos;
402 dreq->pages = pages;
403 dreq->npages = nr_pages;
374 dreq->inode = inode; 404 dreq->inode = inode;
375 dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data); 405 dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
376 if (!is_sync_kiocb(iocb)) 406 if (!is_sync_kiocb(iocb))
@@ -378,9 +408,8 @@ static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size
378 408
379 nfs_add_stats(inode, NFSIOS_DIRECTREADBYTES, count); 409 nfs_add_stats(inode, NFSIOS_DIRECTREADBYTES, count);
380 rpc_clnt_sigmask(clnt, &oldset); 410 rpc_clnt_sigmask(clnt, &oldset);
381 result = nfs_direct_read_schedule(dreq, user_addr, count, pos); 411 nfs_direct_read_schedule(dreq);
382 if (!result) 412 result = nfs_direct_wait(dreq);
383 result = nfs_direct_wait(dreq);
384 rpc_clnt_sigunmask(clnt, &oldset); 413 rpc_clnt_sigunmask(clnt, &oldset);
385 414
386 return result; 415 return result;
@@ -388,10 +417,10 @@ static ssize_t nfs_direct_read(struct kiocb *iocb, unsigned long user_addr, size
388 417
389static void nfs_direct_free_writedata(struct nfs_direct_req *dreq) 418static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
390{ 419{
391 while (!list_empty(&dreq->rewrite_list)) { 420 list_splice_init(&dreq->rewrite_list, &dreq->list);
392 struct nfs_write_data *data = list_entry(dreq->rewrite_list.next, struct nfs_write_data, pages); 421 while (!list_empty(&dreq->list)) {
422 struct nfs_write_data *data = list_entry(dreq->list.next, struct nfs_write_data, pages);
393 list_del(&data->pages); 423 list_del(&data->pages);
394 nfs_direct_release_pages(data->pagevec, data->npages);
395 nfs_writedata_release(data); 424 nfs_writedata_release(data);
396 } 425 }
397} 426}
@@ -399,51 +428,14 @@ static void nfs_direct_free_writedata(struct nfs_direct_req *dreq)
399#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 428#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
400static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) 429static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
401{ 430{
402 struct inode *inode = dreq->inode; 431 struct list_head *pos;
403 struct list_head *p;
404 struct nfs_write_data *data;
405 432
433 list_splice_init(&dreq->rewrite_list, &dreq->list);
434 list_for_each(pos, &dreq->list)
435 dreq->outstanding++;
406 dreq->count = 0; 436 dreq->count = 0;
407 get_dreq(dreq);
408
409 list_for_each(p, &dreq->rewrite_list) {
410 data = list_entry(p, struct nfs_write_data, pages);
411
412 get_dreq(dreq);
413
414 /*
415 * Reset data->res.
416 */
417 nfs_fattr_init(&data->fattr);
418 data->res.count = data->args.count;
419 memset(&data->verf, 0, sizeof(data->verf));
420
421 /*
422 * Reuse data->task; data->args should not have changed
423 * since the original request was sent.
424 */
425 rpc_init_task(&data->task, NFS_CLIENT(inode), RPC_TASK_ASYNC,
426 &nfs_write_direct_ops, data);
427 NFS_PROTO(inode)->write_setup(data, FLUSH_STABLE);
428
429 data->task.tk_priority = RPC_PRIORITY_NORMAL;
430 data->task.tk_cookie = (unsigned long) inode;
431
432 /*
433 * We're called via an RPC callback, so BKL is already held.
434 */
435 rpc_execute(&data->task);
436
437 dprintk("NFS: %5u rescheduled direct write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
438 data->task.tk_pid,
439 inode->i_sb->s_id,
440 (long long)NFS_FILEID(inode),
441 data->args.count,
442 (unsigned long long)data->args.offset);
443 }
444 437
445 if (put_dreq(dreq)) 438 nfs_direct_write_schedule(dreq, FLUSH_STABLE);
446 nfs_direct_write_complete(dreq, inode);
447} 439}
448 440
449static void nfs_direct_commit_result(struct rpc_task *task, void *calldata) 441static void nfs_direct_commit_result(struct rpc_task *task, void *calldata)
@@ -480,8 +472,8 @@ static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq)
480 data->cred = dreq->ctx->cred; 472 data->cred = dreq->ctx->cred;
481 473
482 data->args.fh = NFS_FH(data->inode); 474 data->args.fh = NFS_FH(data->inode);
483 data->args.offset = 0; 475 data->args.offset = dreq->pos;
484 data->args.count = 0; 476 data->args.count = dreq->user_count;
485 data->res.count = 0; 477 data->res.count = 0;
486 data->res.fattr = &data->fattr; 478 data->res.fattr = &data->fattr;
487 data->res.verf = &data->verf; 479 data->res.verf = &data->verf;
@@ -543,6 +535,47 @@ static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode
543} 535}
544#endif 536#endif
545 537
538static struct nfs_direct_req *nfs_direct_write_alloc(size_t nbytes, size_t wsize)
539{
540 struct list_head *list;
541 struct nfs_direct_req *dreq;
542 unsigned int wpages = (wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
543
544 dreq = nfs_direct_req_alloc();
545 if (!dreq)
546 return NULL;
547
548 list = &dreq->list;
549 for(;;) {
550 struct nfs_write_data *data = nfs_writedata_alloc(wpages);
551
552 if (unlikely(!data)) {
553 while (!list_empty(list)) {
554 data = list_entry(list->next,
555 struct nfs_write_data, pages);
556 list_del(&data->pages);
557 nfs_writedata_free(data);
558 }
559 kref_put(&dreq->kref, nfs_direct_req_release);
560 return NULL;
561 }
562
563 INIT_LIST_HEAD(&data->pages);
564 list_add(&data->pages, list);
565
566 data->req = (struct nfs_page *) dreq;
567 dreq->outstanding++;
568 if (nbytes <= wsize)
569 break;
570 nbytes -= wsize;
571 }
572
573 nfs_alloc_commit_data(dreq);
574
575 kref_get(&dreq->kref);
576 return dreq;
577}
578
546static void nfs_direct_write_result(struct rpc_task *task, void *calldata) 579static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
547{ 580{
548 struct nfs_write_data *data = calldata; 581 struct nfs_write_data *data = calldata;
@@ -572,6 +605,8 @@ static void nfs_direct_write_result(struct rpc_task *task, void *calldata)
572 } 605 }
573 } 606 }
574 } 607 }
608 /* In case we have to resend */
609 data->args.stable = NFS_FILE_SYNC;
575 610
576 spin_unlock(&dreq->lock); 611 spin_unlock(&dreq->lock);
577} 612}
@@ -585,8 +620,14 @@ static void nfs_direct_write_release(void *calldata)
585 struct nfs_write_data *data = calldata; 620 struct nfs_write_data *data = calldata;
586 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req; 621 struct nfs_direct_req *dreq = (struct nfs_direct_req *) data->req;
587 622
588 if (put_dreq(dreq)) 623 spin_lock(&dreq->lock);
589 nfs_direct_write_complete(dreq, data->inode); 624 if (--dreq->outstanding) {
625 spin_unlock(&dreq->lock);
626 return;
627 }
628 spin_unlock(&dreq->lock);
629
630 nfs_direct_write_complete(dreq, data->inode);
590} 631}
591 632
592static const struct rpc_call_ops nfs_write_direct_ops = { 633static const struct rpc_call_ops nfs_write_direct_ops = {
@@ -595,62 +636,41 @@ static const struct rpc_call_ops nfs_write_direct_ops = {
595}; 636};
596 637
597/* 638/*
598 * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE 639 * For each nfs_write_data struct that was allocated on the list, dispatch
599 * operation. If nfs_writedata_alloc() or get_user_pages() fails, 640 * an NFS WRITE operation
600 * bail and stop sending more writes. Write length accounting is
601 * handled automatically by nfs_direct_write_result(). Otherwise, if
602 * no requests have been sent, just return an error.
603 */ 641 */
604static ssize_t nfs_direct_write_schedule(struct nfs_direct_req *dreq, unsigned long user_addr, size_t count, loff_t pos, int sync) 642static void nfs_direct_write_schedule(struct nfs_direct_req *dreq, int sync)
605{ 643{
606 struct nfs_open_context *ctx = dreq->ctx; 644 struct nfs_open_context *ctx = dreq->ctx;
607 struct inode *inode = ctx->dentry->d_inode; 645 struct inode *inode = ctx->dentry->d_inode;
646 struct list_head *list = &dreq->list;
647 struct page **pages = dreq->pages;
648 size_t count = dreq->user_count;
649 loff_t pos = dreq->pos;
608 size_t wsize = NFS_SERVER(inode)->wsize; 650 size_t wsize = NFS_SERVER(inode)->wsize;
609 unsigned int wpages = nfs_max_pages(wsize); 651 unsigned int curpage, pgbase;
610 unsigned int pgbase;
611 int result;
612 ssize_t started = 0;
613 652
614 get_dreq(dreq); 653 curpage = 0;
615 654 pgbase = dreq->user_addr & ~PAGE_MASK;
616 pgbase = user_addr & ~PAGE_MASK;
617 do { 655 do {
618 struct nfs_write_data *data; 656 struct nfs_write_data *data;
619 size_t bytes; 657 size_t bytes;
620 658
621 result = -ENOMEM;
622 data = nfs_writedata_alloc(wpages);
623 if (unlikely(!data))
624 break;
625
626 bytes = wsize; 659 bytes = wsize;
627 if (count < wsize) 660 if (count < wsize)
628 bytes = count; 661 bytes = count;
629 662
630 data->npages = nfs_direct_count_pages(user_addr, bytes); 663 BUG_ON(list_empty(list));
631 down_read(&current->mm->mmap_sem); 664 data = list_entry(list->next, struct nfs_write_data, pages);
632 result = get_user_pages(current, current->mm, user_addr,
633 data->npages, 0, 0, data->pagevec, NULL);
634 up_read(&current->mm->mmap_sem);
635 if (unlikely(result < data->npages)) {
636 if (result > 0)
637 nfs_direct_release_pages(data->pagevec, result);
638 nfs_writedata_release(data);
639 break;
640 }
641
642 get_dreq(dreq);
643
644 list_move_tail(&data->pages, &dreq->rewrite_list); 665 list_move_tail(&data->pages, &dreq->rewrite_list);
645 666
646 data->req = (struct nfs_page *) dreq;
647 data->inode = inode; 667 data->inode = inode;
648 data->cred = ctx->cred; 668 data->cred = ctx->cred;
649 data->args.fh = NFS_FH(inode); 669 data->args.fh = NFS_FH(inode);
650 data->args.context = ctx; 670 data->args.context = ctx;
651 data->args.offset = pos; 671 data->args.offset = pos;
652 data->args.pgbase = pgbase; 672 data->args.pgbase = pgbase;
653 data->args.pages = data->pagevec; 673 data->args.pages = &pages[curpage];
654 data->args.count = bytes; 674 data->args.count = bytes;
655 data->res.fattr = &data->fattr; 675 data->res.fattr = &data->fattr;
656 data->res.count = bytes; 676 data->res.count = bytes;
@@ -674,26 +694,19 @@ static ssize_t nfs_direct_write_schedule(struct nfs_direct_req *dreq, unsigned l
674 bytes, 694 bytes,
675 (unsigned long long)data->args.offset); 695 (unsigned long long)data->args.offset);
676 696
677 started += bytes;
678 user_addr += bytes;
679 pos += bytes; 697 pos += bytes;
680 pgbase += bytes; 698 pgbase += bytes;
699 curpage += pgbase >> PAGE_SHIFT;
681 pgbase &= ~PAGE_MASK; 700 pgbase &= ~PAGE_MASK;
682 701
683 count -= bytes; 702 count -= bytes;
684 } while (count != 0); 703 } while (count != 0);
685 704 BUG_ON(!list_empty(list));
686 if (put_dreq(dreq))
687 nfs_direct_write_complete(dreq, inode);
688
689 if (started)
690 return 0;
691 return result < 0 ? (ssize_t) result : -EFAULT;
692} 705}
693 706
694static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos) 707static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, size_t count, loff_t pos, struct page **pages, int nr_pages)
695{ 708{
696 ssize_t result = 0; 709 ssize_t result;
697 sigset_t oldset; 710 sigset_t oldset;
698 struct inode *inode = iocb->ki_filp->f_mapping->host; 711 struct inode *inode = iocb->ki_filp->f_mapping->host;
699 struct rpc_clnt *clnt = NFS_CLIENT(inode); 712 struct rpc_clnt *clnt = NFS_CLIENT(inode);
@@ -701,14 +714,17 @@ static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, siz
701 size_t wsize = NFS_SERVER(inode)->wsize; 714 size_t wsize = NFS_SERVER(inode)->wsize;
702 int sync = 0; 715 int sync = 0;
703 716
704 dreq = nfs_direct_req_alloc(); 717 dreq = nfs_direct_write_alloc(count, wsize);
705 if (!dreq) 718 if (!dreq)
706 return -ENOMEM; 719 return -ENOMEM;
707 nfs_alloc_commit_data(dreq);
708
709 if (dreq->commit_data == NULL || count < wsize) 720 if (dreq->commit_data == NULL || count < wsize)
710 sync = FLUSH_STABLE; 721 sync = FLUSH_STABLE;
711 722
723 dreq->user_addr = user_addr;
724 dreq->user_count = count;
725 dreq->pos = pos;
726 dreq->pages = pages;
727 dreq->npages = nr_pages;
712 dreq->inode = inode; 728 dreq->inode = inode;
713 dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data); 729 dreq->ctx = get_nfs_open_context((struct nfs_open_context *)iocb->ki_filp->private_data);
714 if (!is_sync_kiocb(iocb)) 730 if (!is_sync_kiocb(iocb))
@@ -719,9 +735,8 @@ static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, siz
719 nfs_begin_data_update(inode); 735 nfs_begin_data_update(inode);
720 736
721 rpc_clnt_sigmask(clnt, &oldset); 737 rpc_clnt_sigmask(clnt, &oldset);
722 result = nfs_direct_write_schedule(dreq, user_addr, count, pos, sync); 738 nfs_direct_write_schedule(dreq, sync);
723 if (!result) 739 result = nfs_direct_wait(dreq);
724 result = nfs_direct_wait(dreq);
725 rpc_clnt_sigunmask(clnt, &oldset); 740 rpc_clnt_sigunmask(clnt, &oldset);
726 741
727 return result; 742 return result;
@@ -751,6 +766,8 @@ static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, siz
751ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos) 766ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
752{ 767{
753 ssize_t retval = -EINVAL; 768 ssize_t retval = -EINVAL;
769 int page_count;
770 struct page **pages;
754 struct file *file = iocb->ki_filp; 771 struct file *file = iocb->ki_filp;
755 struct address_space *mapping = file->f_mapping; 772 struct address_space *mapping = file->f_mapping;
756 773
@@ -772,7 +789,14 @@ ssize_t nfs_file_direct_read(struct kiocb *iocb, char __user *buf, size_t count,
772 if (retval) 789 if (retval)
773 goto out; 790 goto out;
774 791
775 retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos); 792 retval = nfs_get_user_pages(READ, (unsigned long) buf,
793 count, &pages);
794 if (retval < 0)
795 goto out;
796 page_count = retval;
797
798 retval = nfs_direct_read(iocb, (unsigned long) buf, count, pos,
799 pages, page_count);
776 if (retval > 0) 800 if (retval > 0)
777 iocb->ki_pos = pos + retval; 801 iocb->ki_pos = pos + retval;
778 802
@@ -808,6 +832,8 @@ out:
808ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos) 832ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t count, loff_t pos)
809{ 833{
810 ssize_t retval; 834 ssize_t retval;
835 int page_count;
836 struct page **pages;
811 struct file *file = iocb->ki_filp; 837 struct file *file = iocb->ki_filp;
812 struct address_space *mapping = file->f_mapping; 838 struct address_space *mapping = file->f_mapping;
813 839
@@ -835,7 +861,14 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t
835 if (retval) 861 if (retval)
836 goto out; 862 goto out;
837 863
838 retval = nfs_direct_write(iocb, (unsigned long) buf, count, pos); 864 retval = nfs_get_user_pages(WRITE, (unsigned long) buf,
865 count, &pages);
866 if (retval < 0)
867 goto out;
868 page_count = retval;
869
870 retval = nfs_direct_write(iocb, (unsigned long) buf, count,
871 pos, pages, page_count);
839 872
840 /* 873 /*
841 * XXX: nfs_end_data_update() already ensures this file's 874 * XXX: nfs_end_data_update() already ensures this file's
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index 2d3fb6416d91..7c7320fa51aa 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -729,7 +729,6 @@ struct nfs_read_data {
729 struct list_head pages; /* Coalesced read requests */ 729 struct list_head pages; /* Coalesced read requests */
730 struct nfs_page *req; /* multi ops per nfs_page */ 730 struct nfs_page *req; /* multi ops per nfs_page */
731 struct page **pagevec; 731 struct page **pagevec;
732 unsigned int npages; /* active pages in pagevec */
733 struct nfs_readargs args; 732 struct nfs_readargs args;
734 struct nfs_readres res; 733 struct nfs_readres res;
735#ifdef CONFIG_NFS_V4 734#ifdef CONFIG_NFS_V4
@@ -748,7 +747,6 @@ struct nfs_write_data {
748 struct list_head pages; /* Coalesced requests we wish to flush */ 747 struct list_head pages; /* Coalesced requests we wish to flush */
749 struct nfs_page *req; /* multi ops per nfs_page */ 748 struct nfs_page *req; /* multi ops per nfs_page */
750 struct page **pagevec; 749 struct page **pagevec;
751 unsigned int npages; /* active pages in pagevec */
752 struct nfs_writeargs args; /* argument struct */ 750 struct nfs_writeargs args; /* argument struct */
753 struct nfs_writeres res; /* result struct */ 751 struct nfs_writeres res; /* result struct */
754#ifdef CONFIG_NFS_V4 752#ifdef CONFIG_NFS_V4
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-24 16:04:29 -0500