1 /* 2 * linux/fs/nfs/direct.c 3 * 4 * Copyright (C) 2003 by Chuck Lever <cel@netapp.com> 5 * 6 * High-performance uncached I/O for the Linux NFS client 7 * 8 * There are important applications whose performance or correctness 9 * depends on uncached access to file data. Database clusters 10 * (multiple copies of the same instance running on separate hosts) 11 * implement their own cache coherency protocol that subsumes file 12 * system cache protocols. Applications that process datasets 13 * considerably larger than the client's memory do not always benefit 14 * from a local cache. A streaming video server, for instance, has no 15 * need to cache the contents of a file. 16 * 17 * When an application requests uncached I/O, all read and write requests 18 * are made directly to the server; data stored or fetched via these 19 * requests is not cached in the Linux page cache. The client does not 20 * correct unaligned requests from applications. All requested bytes are 21 * held on permanent storage before a direct write system call returns to 22 * an application. 23 * 24 * Solaris implements an uncached I/O facility called directio() that 25 * is used for backups and sequential I/O to very large files. Solaris 26 * also supports uncaching whole NFS partitions with "-o forcedirectio," 27 * an undocumented mount option. 28 * 29 * Designed by Jeff Kimmel, Chuck Lever, and Trond Myklebust, with 30 * help from Andrew Morton. 31 * 32 * 18 Dec 2001 Initial implementation for 2.4 --cel 33 * 08 Jul 2002 Version for 2.4.19, with bug fixes --trondmy 34 * 08 Jun 2003 Port to 2.5 APIs --cel 35 * 31 Mar 2004 Handle direct I/O without VFS support --cel 36 * 15 Sep 2004 Parallel async reads --cel 37 * 04 May 2005 support O_DIRECT with aio --cel 38 * 39 */ 40 41 #include <linux/errno.h> 42 #include <linux/sched.h> 43 #include <linux/kernel.h> 44 #include <linux/file.h> 45 #include <linux/pagemap.h> 46 #include <linux/kref.h> 47 #include <linux/slab.h> 48 #include <linux/task_io_accounting_ops.h> 49 50 #include <linux/nfs_fs.h> 51 #include <linux/nfs_page.h> 52 #include <linux/sunrpc/clnt.h> 53 54 #include <asm/uaccess.h> 55 #include <linux/atomic.h> 56 57 #include "internal.h" 58 #include "iostat.h" 59 #include "pnfs.h" 60 61 #define NFSDBG_FACILITY NFSDBG_VFS 62 63 static struct kmem_cache *nfs_direct_cachep; 64 65 /* 66 * This represents a set of asynchronous requests that we're waiting on 67 */ 68 struct nfs_direct_req { 69 struct kref kref; /* release manager */ 70 71 /* I/O parameters */ 72 struct nfs_open_context *ctx; /* file open context info */ 73 struct nfs_lock_context *l_ctx; /* Lock context info */ 74 struct kiocb * iocb; /* controlling i/o request */ 75 struct inode * inode; /* target file of i/o */ 76 77 /* completion state */ 78 atomic_t io_count; /* i/os we're waiting for */ 79 spinlock_t lock; /* protect completion state */ 80 ssize_t count, /* bytes actually processed */ 81 error; /* any reported error */ 82 struct completion completion; /* wait for i/o completion */ 83 84 /* commit state */ 85 struct nfs_mds_commit_info mds_cinfo; /* Storage for cinfo */ 86 struct pnfs_ds_commit_info ds_cinfo; /* Storage for cinfo */ 87 struct work_struct work; 88 int flags; 89 #define NFS_ODIRECT_DO_COMMIT (1) /* an unstable reply was received */ 90 #define NFS_ODIRECT_RESCHED_WRITES (2) /* write verification failed */ 91 struct nfs_writeverf verf; /* unstable write verifier */ 92 }; 93 94 static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops; 95 static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops; 96 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode); 97 static void nfs_direct_write_schedule_work(struct work_struct *work); 98 99 static inline void get_dreq(struct nfs_direct_req *dreq) 100 { 101 atomic_inc(&dreq->io_count); 102 } 103 104 static inline int put_dreq(struct nfs_direct_req *dreq) 105 { 106 return atomic_dec_and_test(&dreq->io_count); 107 } 108 109 /** 110 * nfs_direct_IO - NFS address space operation for direct I/O 111 * @rw: direction (read or write) 112 * @iocb: target I/O control block 113 * @iov: array of vectors that define I/O buffer 114 * @pos: offset in file to begin the operation 115 * @nr_segs: size of iovec array 116 * 117 * The presence of this routine in the address space ops vector means 118 * the NFS client supports direct I/O. However, we shunt off direct 119 * read and write requests before the VFS gets them, so this method 120 * should never be called. 121 */ 122 ssize_t nfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, loff_t pos, unsigned long nr_segs) 123 { 124 dprintk("NFS: nfs_direct_IO (%s) off/no(%Ld/%lu) EINVAL\n", 125 iocb->ki_filp->f_path.dentry->d_name.name, 126 (long long) pos, nr_segs); 127 128 return -EINVAL; 129 } 130 131 static void nfs_direct_release_pages(struct page **pages, unsigned int npages) 132 { 133 unsigned int i; 134 for (i = 0; i < npages; i++) 135 page_cache_release(pages[i]); 136 } 137 138 void nfs_init_cinfo_from_dreq(struct nfs_commit_info *cinfo, 139 struct nfs_direct_req *dreq) 140 { 141 cinfo->lock = &dreq->lock; 142 cinfo->mds = &dreq->mds_cinfo; 143 cinfo->ds = &dreq->ds_cinfo; 144 cinfo->dreq = dreq; 145 cinfo->completion_ops = &nfs_direct_commit_completion_ops; 146 } 147 148 static inline struct nfs_direct_req *nfs_direct_req_alloc(void) 149 { 150 struct nfs_direct_req *dreq; 151 152 dreq = kmem_cache_alloc(nfs_direct_cachep, GFP_KERNEL); 153 if (!dreq) 154 return NULL; 155 156 kref_init(&dreq->kref); 157 kref_get(&dreq->kref); 158 init_completion(&dreq->completion); 159 dreq->mds_cinfo.ncommit = 0; 160 atomic_set(&dreq->mds_cinfo.rpcs_out, 0); 161 INIT_LIST_HEAD(&dreq->mds_cinfo.list); 162 INIT_WORK(&dreq->work, nfs_direct_write_schedule_work); 163 memset(&dreq->ds_cinfo, 0, sizeof(dreq->ds_cinfo)); 164 dreq->iocb = NULL; 165 dreq->ctx = NULL; 166 dreq->l_ctx = NULL; 167 spin_lock_init(&dreq->lock); 168 atomic_set(&dreq->io_count, 0); 169 dreq->count = 0; 170 dreq->error = 0; 171 dreq->flags = 0; 172 173 return dreq; 174 } 175 176 static void nfs_direct_req_free(struct kref *kref) 177 { 178 struct nfs_direct_req *dreq = container_of(kref, struct nfs_direct_req, kref); 179 180 if (dreq->l_ctx != NULL) 181 nfs_put_lock_context(dreq->l_ctx); 182 if (dreq->ctx != NULL) 183 put_nfs_open_context(dreq->ctx); 184 kmem_cache_free(nfs_direct_cachep, dreq); 185 } 186 187 static void nfs_direct_req_release(struct nfs_direct_req *dreq) 188 { 189 kref_put(&dreq->kref, nfs_direct_req_free); 190 } 191 192 /* 193 * Collects and returns the final error value/byte-count. 194 */ 195 static ssize_t nfs_direct_wait(struct nfs_direct_req *dreq) 196 { 197 ssize_t result = -EIOCBQUEUED; 198 199 /* Async requests don't wait here */ 200 if (dreq->iocb) 201 goto out; 202 203 result = wait_for_completion_killable(&dreq->completion); 204 205 if (!result) 206 result = dreq->error; 207 if (!result) 208 result = dreq->count; 209 210 out: 211 return (ssize_t) result; 212 } 213 214 /* 215 * Synchronous I/O uses a stack-allocated iocb. Thus we can't trust 216 * the iocb is still valid here if this is a synchronous request. 217 */ 218 static void nfs_direct_complete(struct nfs_direct_req *dreq) 219 { 220 if (dreq->iocb) { 221 long res = (long) dreq->error; 222 if (!res) 223 res = (long) dreq->count; 224 aio_complete(dreq->iocb, res, 0); 225 } 226 complete_all(&dreq->completion); 227 228 nfs_direct_req_release(dreq); 229 } 230 231 void nfs_direct_readpage_release(struct nfs_page *req) 232 { 233 dprintk("NFS: direct read done (%s/%lld %d@%lld)\n", 234 req->wb_context->dentry->d_inode->i_sb->s_id, 235 (long long)NFS_FILEID(req->wb_context->dentry->d_inode), 236 req->wb_bytes, 237 (long long)req_offset(req)); 238 nfs_release_request(req); 239 } 240 241 static void nfs_direct_read_completion(struct nfs_pgio_header *hdr) 242 { 243 unsigned long bytes = 0; 244 struct nfs_direct_req *dreq = hdr->dreq; 245 246 if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 247 goto out_put; 248 249 spin_lock(&dreq->lock); 250 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && (hdr->good_bytes == 0)) 251 dreq->error = hdr->error; 252 else 253 dreq->count += hdr->good_bytes; 254 spin_unlock(&dreq->lock); 255 256 if (!test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { 257 while (!list_empty(&hdr->pages)) { 258 struct nfs_page *req = nfs_list_entry(hdr->pages.next); 259 struct page *page = req->wb_page; 260 261 if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) { 262 if (bytes > hdr->good_bytes) 263 zero_user(page, 0, PAGE_SIZE); 264 else if (hdr->good_bytes - bytes < PAGE_SIZE) 265 zero_user_segment(page, 266 hdr->good_bytes & ~PAGE_MASK, 267 PAGE_SIZE); 268 } 269 bytes += req->wb_bytes; 270 nfs_list_remove_request(req); 271 nfs_direct_readpage_release(req); 272 if (!PageCompound(page)) 273 set_page_dirty(page); 274 page_cache_release(page); 275 } 276 } else { 277 while (!list_empty(&hdr->pages)) { 278 struct nfs_page *req = nfs_list_entry(hdr->pages.next); 279 280 if (bytes < hdr->good_bytes) 281 if (!PageCompound(req->wb_page)) 282 set_page_dirty(req->wb_page); 283 bytes += req->wb_bytes; 284 page_cache_release(req->wb_page); 285 nfs_list_remove_request(req); 286 nfs_direct_readpage_release(req); 287 } 288 } 289 out_put: 290 if (put_dreq(dreq)) 291 nfs_direct_complete(dreq); 292 hdr->release(hdr); 293 } 294 295 static void nfs_read_sync_pgio_error(struct list_head *head) 296 { 297 struct nfs_page *req; 298 299 while (!list_empty(head)) { 300 req = nfs_list_entry(head->next); 301 nfs_list_remove_request(req); 302 nfs_release_request(req); 303 } 304 } 305 306 static void nfs_direct_pgio_init(struct nfs_pgio_header *hdr) 307 { 308 get_dreq(hdr->dreq); 309 } 310 311 static const struct nfs_pgio_completion_ops nfs_direct_read_completion_ops = { 312 .error_cleanup = nfs_read_sync_pgio_error, 313 .init_hdr = nfs_direct_pgio_init, 314 .completion = nfs_direct_read_completion, 315 }; 316 317 /* 318 * For each rsize'd chunk of the user's buffer, dispatch an NFS READ 319 * operation. If nfs_readdata_alloc() or get_user_pages() fails, 320 * bail and stop sending more reads. Read length accounting is 321 * handled automatically by nfs_direct_read_result(). Otherwise, if 322 * no requests have been sent, just return an error. 323 */ 324 static ssize_t nfs_direct_read_schedule_segment(struct nfs_pageio_descriptor *desc, 325 const struct iovec *iov, 326 loff_t pos) 327 { 328 struct nfs_direct_req *dreq = desc->pg_dreq; 329 struct nfs_open_context *ctx = dreq->ctx; 330 struct inode *inode = ctx->dentry->d_inode; 331 unsigned long user_addr = (unsigned long)iov->iov_base; 332 size_t count = iov->iov_len; 333 size_t rsize = NFS_SERVER(inode)->rsize; 334 unsigned int pgbase; 335 int result; 336 ssize_t started = 0; 337 struct page **pagevec = NULL; 338 unsigned int npages; 339 340 do { 341 size_t bytes; 342 int i; 343 344 pgbase = user_addr & ~PAGE_MASK; 345 bytes = min(max(rsize, PAGE_SIZE), count); 346 347 result = -ENOMEM; 348 npages = nfs_page_array_len(pgbase, bytes); 349 if (!pagevec) 350 pagevec = kmalloc(npages * sizeof(struct page *), 351 GFP_KERNEL); 352 if (!pagevec) 353 break; 354 down_read(¤t->mm->mmap_sem); 355 result = get_user_pages(current, current->mm, user_addr, 356 npages, 1, 0, pagevec, NULL); 357 up_read(¤t->mm->mmap_sem); 358 if (result < 0) 359 break; 360 if ((unsigned)result < npages) { 361 bytes = result * PAGE_SIZE; 362 if (bytes <= pgbase) { 363 nfs_direct_release_pages(pagevec, result); 364 break; 365 } 366 bytes -= pgbase; 367 npages = result; 368 } 369 370 for (i = 0; i < npages; i++) { 371 struct nfs_page *req; 372 unsigned int req_len = min(bytes, PAGE_SIZE - pgbase); 373 /* XXX do we need to do the eof zeroing found in async_filler? */ 374 req = nfs_create_request(dreq->ctx, dreq->inode, 375 pagevec[i], 376 pgbase, req_len); 377 if (IS_ERR(req)) { 378 nfs_direct_release_pages(pagevec + i, 379 npages - i); 380 result = PTR_ERR(req); 381 break; 382 } 383 req->wb_index = pos >> PAGE_SHIFT; 384 req->wb_offset = pos & ~PAGE_MASK; 385 if (!nfs_pageio_add_request(desc, req)) { 386 result = desc->pg_error; 387 nfs_release_request(req); 388 nfs_direct_release_pages(pagevec + i, 389 npages - i); 390 break; 391 } 392 pgbase = 0; 393 bytes -= req_len; 394 started += req_len; 395 user_addr += req_len; 396 pos += req_len; 397 count -= req_len; 398 } 399 } while (count != 0 && result >= 0); 400 401 kfree(pagevec); 402 403 if (started) 404 return started; 405 return result < 0 ? (ssize_t) result : -EFAULT; 406 } 407 408 static ssize_t nfs_direct_read_schedule_iovec(struct nfs_direct_req *dreq, 409 const struct iovec *iov, 410 unsigned long nr_segs, 411 loff_t pos) 412 { 413 struct nfs_pageio_descriptor desc; 414 ssize_t result = -EINVAL; 415 size_t requested_bytes = 0; 416 unsigned long seg; 417 418 nfs_pageio_init_read(&desc, dreq->inode, 419 &nfs_direct_read_completion_ops); 420 get_dreq(dreq); 421 desc.pg_dreq = dreq; 422 423 for (seg = 0; seg < nr_segs; seg++) { 424 const struct iovec *vec = &iov[seg]; 425 result = nfs_direct_read_schedule_segment(&desc, vec, pos); 426 if (result < 0) 427 break; 428 requested_bytes += result; 429 if ((size_t)result < vec->iov_len) 430 break; 431 pos += vec->iov_len; 432 } 433 434 nfs_pageio_complete(&desc); 435 436 /* 437 * If no bytes were started, return the error, and let the 438 * generic layer handle the completion. 439 */ 440 if (requested_bytes == 0) { 441 nfs_direct_req_release(dreq); 442 return result < 0 ? result : -EIO; 443 } 444 445 if (put_dreq(dreq)) 446 nfs_direct_complete(dreq); 447 return 0; 448 } 449 450 static ssize_t nfs_direct_read(struct kiocb *iocb, const struct iovec *iov, 451 unsigned long nr_segs, loff_t pos) 452 { 453 ssize_t result = -ENOMEM; 454 struct inode *inode = iocb->ki_filp->f_mapping->host; 455 struct nfs_direct_req *dreq; 456 457 dreq = nfs_direct_req_alloc(); 458 if (dreq == NULL) 459 goto out; 460 461 dreq->inode = inode; 462 dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); 463 dreq->l_ctx = nfs_get_lock_context(dreq->ctx); 464 if (dreq->l_ctx == NULL) 465 goto out_release; 466 if (!is_sync_kiocb(iocb)) 467 dreq->iocb = iocb; 468 469 result = nfs_direct_read_schedule_iovec(dreq, iov, nr_segs, pos); 470 if (!result) 471 result = nfs_direct_wait(dreq); 472 out_release: 473 nfs_direct_req_release(dreq); 474 out: 475 return result; 476 } 477 478 #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4) 479 static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) 480 { 481 struct nfs_pageio_descriptor desc; 482 struct nfs_page *req, *tmp; 483 LIST_HEAD(reqs); 484 struct nfs_commit_info cinfo; 485 LIST_HEAD(failed); 486 487 nfs_init_cinfo_from_dreq(&cinfo, dreq); 488 pnfs_recover_commit_reqs(dreq->inode, &reqs, &cinfo); 489 spin_lock(cinfo.lock); 490 nfs_scan_commit_list(&cinfo.mds->list, &reqs, &cinfo, 0); 491 spin_unlock(cinfo.lock); 492 493 dreq->count = 0; 494 get_dreq(dreq); 495 496 nfs_pageio_init_write(&desc, dreq->inode, FLUSH_STABLE, 497 &nfs_direct_write_completion_ops); 498 desc.pg_dreq = dreq; 499 500 list_for_each_entry_safe(req, tmp, &reqs, wb_list) { 501 if (!nfs_pageio_add_request(&desc, req)) { 502 nfs_list_add_request(req, &failed); 503 spin_lock(cinfo.lock); 504 dreq->flags = 0; 505 dreq->error = -EIO; 506 spin_unlock(cinfo.lock); 507 } 508 } 509 nfs_pageio_complete(&desc); 510 511 while (!list_empty(&failed)) { 512 page_cache_release(req->wb_page); 513 nfs_release_request(req); 514 nfs_unlock_request(req); 515 } 516 517 if (put_dreq(dreq)) 518 nfs_direct_write_complete(dreq, dreq->inode); 519 } 520 521 static void nfs_direct_commit_complete(struct nfs_commit_data *data) 522 { 523 struct nfs_direct_req *dreq = data->dreq; 524 struct nfs_commit_info cinfo; 525 struct nfs_page *req; 526 int status = data->task.tk_status; 527 528 nfs_init_cinfo_from_dreq(&cinfo, dreq); 529 if (status < 0) { 530 dprintk("NFS: %5u commit failed with error %d.\n", 531 data->task.tk_pid, status); 532 dreq->flags = NFS_ODIRECT_RESCHED_WRITES; 533 } else if (memcmp(&dreq->verf, &data->verf, sizeof(data->verf))) { 534 dprintk("NFS: %5u commit verify failed\n", data->task.tk_pid); 535 dreq->flags = NFS_ODIRECT_RESCHED_WRITES; 536 } 537 538 dprintk("NFS: %5u commit returned %d\n", data->task.tk_pid, status); 539 while (!list_empty(&data->pages)) { 540 req = nfs_list_entry(data->pages.next); 541 nfs_list_remove_request(req); 542 if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) { 543 /* Note the rewrite will go through mds */ 544 nfs_mark_request_commit(req, NULL, &cinfo); 545 } else { 546 page_cache_release(req->wb_page); 547 nfs_release_request(req); 548 } 549 nfs_unlock_request(req); 550 } 551 552 if (atomic_dec_and_test(&cinfo.mds->rpcs_out)) 553 nfs_direct_write_complete(dreq, data->inode); 554 } 555 556 static void nfs_direct_error_cleanup(struct nfs_inode *nfsi) 557 { 558 /* There is no lock to clear */ 559 } 560 561 static const struct nfs_commit_completion_ops nfs_direct_commit_completion_ops = { 562 .completion = nfs_direct_commit_complete, 563 .error_cleanup = nfs_direct_error_cleanup, 564 }; 565 566 static void nfs_direct_commit_schedule(struct nfs_direct_req *dreq) 567 { 568 int res; 569 struct nfs_commit_info cinfo; 570 LIST_HEAD(mds_list); 571 572 nfs_init_cinfo_from_dreq(&cinfo, dreq); 573 nfs_scan_commit(dreq->inode, &mds_list, &cinfo); 574 res = nfs_generic_commit_list(dreq->inode, &mds_list, 0, &cinfo); 575 if (res < 0) /* res == -ENOMEM */ 576 nfs_direct_write_reschedule(dreq); 577 } 578 579 static void nfs_direct_write_schedule_work(struct work_struct *work) 580 { 581 struct nfs_direct_req *dreq = container_of(work, struct nfs_direct_req, work); 582 int flags = dreq->flags; 583 584 dreq->flags = 0; 585 switch (flags) { 586 case NFS_ODIRECT_DO_COMMIT: 587 nfs_direct_commit_schedule(dreq); 588 break; 589 case NFS_ODIRECT_RESCHED_WRITES: 590 nfs_direct_write_reschedule(dreq); 591 break; 592 default: 593 nfs_zap_mapping(dreq->inode, dreq->inode->i_mapping); 594 nfs_direct_complete(dreq); 595 } 596 } 597 598 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode) 599 { 600 schedule_work(&dreq->work); /* Calls nfs_direct_write_schedule_work */ 601 } 602 603 #else 604 static void nfs_direct_write_schedule_work(struct work_struct *work) 605 { 606 } 607 608 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode) 609 { 610 nfs_zap_mapping(inode, inode->i_mapping); 611 nfs_direct_complete(dreq); 612 } 613 #endif 614 615 /* 616 * NB: Return the value of the first error return code. Subsequent 617 * errors after the first one are ignored. 618 */ 619 /* 620 * For each wsize'd chunk of the user's buffer, dispatch an NFS WRITE 621 * operation. If nfs_writedata_alloc() or get_user_pages() fails, 622 * bail and stop sending more writes. Write length accounting is 623 * handled automatically by nfs_direct_write_result(). Otherwise, if 624 * no requests have been sent, just return an error. 625 */ 626 static ssize_t nfs_direct_write_schedule_segment(struct nfs_pageio_descriptor *desc, 627 const struct iovec *iov, 628 loff_t pos) 629 { 630 struct nfs_direct_req *dreq = desc->pg_dreq; 631 struct nfs_open_context *ctx = dreq->ctx; 632 struct inode *inode = ctx->dentry->d_inode; 633 unsigned long user_addr = (unsigned long)iov->iov_base; 634 size_t count = iov->iov_len; 635 size_t wsize = NFS_SERVER(inode)->wsize; 636 unsigned int pgbase; 637 int result; 638 ssize_t started = 0; 639 struct page **pagevec = NULL; 640 unsigned int npages; 641 642 do { 643 size_t bytes; 644 int i; 645 646 pgbase = user_addr & ~PAGE_MASK; 647 bytes = min(max(wsize, PAGE_SIZE), count); 648 649 result = -ENOMEM; 650 npages = nfs_page_array_len(pgbase, bytes); 651 if (!pagevec) 652 pagevec = kmalloc(npages * sizeof(struct page *), GFP_KERNEL); 653 if (!pagevec) 654 break; 655 656 down_read(¤t->mm->mmap_sem); 657 result = get_user_pages(current, current->mm, user_addr, 658 npages, 0, 0, pagevec, NULL); 659 up_read(¤t->mm->mmap_sem); 660 if (result < 0) 661 break; 662 663 if ((unsigned)result < npages) { 664 bytes = result * PAGE_SIZE; 665 if (bytes <= pgbase) { 666 nfs_direct_release_pages(pagevec, result); 667 break; 668 } 669 bytes -= pgbase; 670 npages = result; 671 } 672 673 for (i = 0; i < npages; i++) { 674 struct nfs_page *req; 675 unsigned int req_len = min(bytes, PAGE_SIZE - pgbase); 676 677 req = nfs_create_request(dreq->ctx, dreq->inode, 678 pagevec[i], 679 pgbase, req_len); 680 if (IS_ERR(req)) { 681 nfs_direct_release_pages(pagevec + i, 682 npages - i); 683 result = PTR_ERR(req); 684 break; 685 } 686 nfs_lock_request(req); 687 req->wb_index = pos >> PAGE_SHIFT; 688 req->wb_offset = pos & ~PAGE_MASK; 689 if (!nfs_pageio_add_request(desc, req)) { 690 result = desc->pg_error; 691 nfs_unlock_request(req); 692 nfs_release_request(req); 693 nfs_direct_release_pages(pagevec + i, 694 npages - i); 695 break; 696 } 697 pgbase = 0; 698 bytes -= req_len; 699 started += req_len; 700 user_addr += req_len; 701 pos += req_len; 702 count -= req_len; 703 } 704 } while (count != 0 && result >= 0); 705 706 kfree(pagevec); 707 708 if (started) 709 return started; 710 return result < 0 ? (ssize_t) result : -EFAULT; 711 } 712 713 static void nfs_direct_write_completion(struct nfs_pgio_header *hdr) 714 { 715 struct nfs_direct_req *dreq = hdr->dreq; 716 struct nfs_commit_info cinfo; 717 int bit = -1; 718 struct nfs_page *req = nfs_list_entry(hdr->pages.next); 719 720 if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 721 goto out_put; 722 723 nfs_init_cinfo_from_dreq(&cinfo, dreq); 724 725 spin_lock(&dreq->lock); 726 727 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags)) { 728 dreq->flags = 0; 729 dreq->error = hdr->error; 730 } 731 if (dreq->error != 0) 732 bit = NFS_IOHDR_ERROR; 733 else { 734 dreq->count += hdr->good_bytes; 735 if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) { 736 dreq->flags = NFS_ODIRECT_RESCHED_WRITES; 737 bit = NFS_IOHDR_NEED_RESCHED; 738 } else if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) { 739 if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) 740 bit = NFS_IOHDR_NEED_RESCHED; 741 else if (dreq->flags == 0) { 742 memcpy(&dreq->verf, &req->wb_verf, 743 sizeof(dreq->verf)); 744 bit = NFS_IOHDR_NEED_COMMIT; 745 dreq->flags = NFS_ODIRECT_DO_COMMIT; 746 } else if (dreq->flags == NFS_ODIRECT_DO_COMMIT) { 747 if (memcmp(&dreq->verf, &req->wb_verf, sizeof(dreq->verf))) { 748 dreq->flags = NFS_ODIRECT_RESCHED_WRITES; 749 bit = NFS_IOHDR_NEED_RESCHED; 750 } else 751 bit = NFS_IOHDR_NEED_COMMIT; 752 } 753 } 754 } 755 spin_unlock(&dreq->lock); 756 757 while (!list_empty(&hdr->pages)) { 758 req = nfs_list_entry(hdr->pages.next); 759 nfs_list_remove_request(req); 760 switch (bit) { 761 case NFS_IOHDR_NEED_RESCHED: 762 case NFS_IOHDR_NEED_COMMIT: 763 nfs_mark_request_commit(req, hdr->lseg, &cinfo); 764 break; 765 default: 766 page_cache_release(req->wb_page); 767 nfs_release_request(req); 768 } 769 nfs_unlock_request(req); 770 } 771 772 out_put: 773 if (put_dreq(dreq)) 774 nfs_direct_write_complete(dreq, hdr->inode); 775 hdr->release(hdr); 776 } 777 778 static void nfs_write_sync_pgio_error(struct list_head *head) 779 { 780 struct nfs_page *req; 781 782 while (!list_empty(head)) { 783 req = nfs_list_entry(head->next); 784 nfs_list_remove_request(req); 785 nfs_release_request(req); 786 nfs_unlock_request(req); 787 } 788 } 789 790 static const struct nfs_pgio_completion_ops nfs_direct_write_completion_ops = { 791 .error_cleanup = nfs_write_sync_pgio_error, 792 .init_hdr = nfs_direct_pgio_init, 793 .completion = nfs_direct_write_completion, 794 }; 795 796 static ssize_t nfs_direct_write_schedule_iovec(struct nfs_direct_req *dreq, 797 const struct iovec *iov, 798 unsigned long nr_segs, 799 loff_t pos) 800 { 801 struct nfs_pageio_descriptor desc; 802 ssize_t result = 0; 803 size_t requested_bytes = 0; 804 unsigned long seg; 805 806 nfs_pageio_init_write(&desc, dreq->inode, FLUSH_COND_STABLE, 807 &nfs_direct_write_completion_ops); 808 desc.pg_dreq = dreq; 809 get_dreq(dreq); 810 811 for (seg = 0; seg < nr_segs; seg++) { 812 const struct iovec *vec = &iov[seg]; 813 result = nfs_direct_write_schedule_segment(&desc, vec, pos); 814 if (result < 0) 815 break; 816 requested_bytes += result; 817 if ((size_t)result < vec->iov_len) 818 break; 819 pos += vec->iov_len; 820 } 821 nfs_pageio_complete(&desc); 822 823 /* 824 * If no bytes were started, return the error, and let the 825 * generic layer handle the completion. 826 */ 827 if (requested_bytes == 0) { 828 nfs_direct_req_release(dreq); 829 return result < 0 ? result : -EIO; 830 } 831 832 if (put_dreq(dreq)) 833 nfs_direct_write_complete(dreq, dreq->inode); 834 return 0; 835 } 836 837 static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov, 838 unsigned long nr_segs, loff_t pos, 839 size_t count) 840 { 841 ssize_t result = -ENOMEM; 842 struct inode *inode = iocb->ki_filp->f_mapping->host; 843 struct nfs_direct_req *dreq; 844 845 dreq = nfs_direct_req_alloc(); 846 if (!dreq) 847 goto out; 848 849 dreq->inode = inode; 850 dreq->ctx = get_nfs_open_context(nfs_file_open_context(iocb->ki_filp)); 851 dreq->l_ctx = nfs_get_lock_context(dreq->ctx); 852 if (dreq->l_ctx == NULL) 853 goto out_release; 854 if (!is_sync_kiocb(iocb)) 855 dreq->iocb = iocb; 856 857 result = nfs_direct_write_schedule_iovec(dreq, iov, nr_segs, pos); 858 if (!result) 859 result = nfs_direct_wait(dreq); 860 out_release: 861 nfs_direct_req_release(dreq); 862 out: 863 return result; 864 } 865 866 /** 867 * nfs_file_direct_read - file direct read operation for NFS files 868 * @iocb: target I/O control block 869 * @iov: vector of user buffers into which to read data 870 * @nr_segs: size of iov vector 871 * @pos: byte offset in file where reading starts 872 * 873 * We use this function for direct reads instead of calling 874 * generic_file_aio_read() in order to avoid gfar's check to see if 875 * the request starts before the end of the file. For that check 876 * to work, we must generate a GETATTR before each direct read, and 877 * even then there is a window between the GETATTR and the subsequent 878 * READ where the file size could change. Our preference is simply 879 * to do all reads the application wants, and the server will take 880 * care of managing the end of file boundary. 881 * 882 * This function also eliminates unnecessarily updating the file's 883 * atime locally, as the NFS server sets the file's atime, and this 884 * client must read the updated atime from the server back into its 885 * cache. 886 */ 887 ssize_t nfs_file_direct_read(struct kiocb *iocb, const struct iovec *iov, 888 unsigned long nr_segs, loff_t pos) 889 { 890 ssize_t retval = -EINVAL; 891 struct file *file = iocb->ki_filp; 892 struct address_space *mapping = file->f_mapping; 893 size_t count; 894 895 count = iov_length(iov, nr_segs); 896 nfs_add_stats(mapping->host, NFSIOS_DIRECTREADBYTES, count); 897 898 dfprintk(FILE, "NFS: direct read(%s/%s, %zd@%Ld)\n", 899 file->f_path.dentry->d_parent->d_name.name, 900 file->f_path.dentry->d_name.name, 901 count, (long long) pos); 902 903 retval = 0; 904 if (!count) 905 goto out; 906 907 retval = nfs_sync_mapping(mapping); 908 if (retval) 909 goto out; 910 911 task_io_account_read(count); 912 913 retval = nfs_direct_read(iocb, iov, nr_segs, pos); 914 if (retval > 0) 915 iocb->ki_pos = pos + retval; 916 917 out: 918 return retval; 919 } 920 921 /** 922 * nfs_file_direct_write - file direct write operation for NFS files 923 * @iocb: target I/O control block 924 * @iov: vector of user buffers from which to write data 925 * @nr_segs: size of iov vector 926 * @pos: byte offset in file where writing starts 927 * 928 * We use this function for direct writes instead of calling 929 * generic_file_aio_write() in order to avoid taking the inode 930 * semaphore and updating the i_size. The NFS server will set 931 * the new i_size and this client must read the updated size 932 * back into its cache. We let the server do generic write 933 * parameter checking and report problems. 934 * 935 * We eliminate local atime updates, see direct read above. 936 * 937 * We avoid unnecessary page cache invalidations for normal cached 938 * readers of this file. 939 * 940 * Note that O_APPEND is not supported for NFS direct writes, as there 941 * is no atomic O_APPEND write facility in the NFS protocol. 942 */ 943 ssize_t nfs_file_direct_write(struct kiocb *iocb, const struct iovec *iov, 944 unsigned long nr_segs, loff_t pos) 945 { 946 ssize_t retval = -EINVAL; 947 struct file *file = iocb->ki_filp; 948 struct address_space *mapping = file->f_mapping; 949 size_t count; 950 951 count = iov_length(iov, nr_segs); 952 nfs_add_stats(mapping->host, NFSIOS_DIRECTWRITTENBYTES, count); 953 954 dfprintk(FILE, "NFS: direct write(%s/%s, %zd@%Ld)\n", 955 file->f_path.dentry->d_parent->d_name.name, 956 file->f_path.dentry->d_name.name, 957 count, (long long) pos); 958 959 retval = generic_write_checks(file, &pos, &count, 0); 960 if (retval) 961 goto out; 962 963 retval = -EINVAL; 964 if ((ssize_t) count < 0) 965 goto out; 966 retval = 0; 967 if (!count) 968 goto out; 969 970 retval = nfs_sync_mapping(mapping); 971 if (retval) 972 goto out; 973 974 task_io_account_write(count); 975 976 retval = nfs_direct_write(iocb, iov, nr_segs, pos, count); 977 if (retval > 0) { 978 struct inode *inode = mapping->host; 979 980 iocb->ki_pos = pos + retval; 981 spin_lock(&inode->i_lock); 982 if (i_size_read(inode) < iocb->ki_pos) 983 i_size_write(inode, iocb->ki_pos); 984 spin_unlock(&inode->i_lock); 985 } 986 out: 987 return retval; 988 } 989 990 /** 991 * nfs_init_directcache - create a slab cache for nfs_direct_req structures 992 * 993 */ 994 int __init nfs_init_directcache(void) 995 { 996 nfs_direct_cachep = kmem_cache_create("nfs_direct_cache", 997 sizeof(struct nfs_direct_req), 998 0, (SLAB_RECLAIM_ACCOUNT| 999 SLAB_MEM_SPREAD), 1000 NULL); 1001 if (nfs_direct_cachep == NULL) 1002 return -ENOMEM; 1003 1004 return 0; 1005 } 1006 1007 /** 1008 * nfs_destroy_directcache - destroy the slab cache for nfs_direct_req structures 1009 * 1010 */ 1011 void nfs_destroy_directcache(void) 1012 { 1013 kmem_cache_destroy(nfs_direct_cachep); 1014 } 1015