4ec67f8d70aa300c528829f4b2dfd55a9203c35c
[firefly-linux-kernel-4.4.55.git] / fs / nfs / pagelist.c
1 /*
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
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/sched.h>
15 #include <linux/sunrpc/clnt.h>
16 #include <linux/nfs.h>
17 #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>
22 #include <linux/export.h>
23
24 #include "internal.h"
25 #include "pnfs.h"
26
27 #define NFSDBG_FACILITY         NFSDBG_PAGECACHE
28
29 static struct kmem_cache *nfs_page_cachep;
30 static const struct rpc_call_ops nfs_pgio_common_ops;
31
32 static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
33 {
34         p->npages = pagecount;
35         if (pagecount <= ARRAY_SIZE(p->page_array))
36                 p->pagevec = p->page_array;
37         else {
38                 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
39                 if (!p->pagevec)
40                         p->npages = 0;
41         }
42         return p->pagevec != NULL;
43 }
44
45 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
46                        struct nfs_pgio_header *hdr,
47                        void (*release)(struct nfs_pgio_header *hdr))
48 {
49         hdr->req = nfs_list_entry(desc->pg_list.next);
50         hdr->inode = desc->pg_inode;
51         hdr->cred = hdr->req->wb_context->cred;
52         hdr->io_start = req_offset(hdr->req);
53         hdr->good_bytes = desc->pg_count;
54         hdr->dreq = desc->pg_dreq;
55         hdr->layout_private = desc->pg_layout_private;
56         hdr->release = release;
57         hdr->completion_ops = desc->pg_completion_ops;
58         if (hdr->completion_ops->init_hdr)
59                 hdr->completion_ops->init_hdr(hdr);
60 }
61 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
62
63 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
64 {
65         spin_lock(&hdr->lock);
66         if (pos < hdr->io_start + hdr->good_bytes) {
67                 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
68                 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
69                 hdr->good_bytes = pos - hdr->io_start;
70                 hdr->error = error;
71         }
72         spin_unlock(&hdr->lock);
73 }
74
75 static inline struct nfs_page *
76 nfs_page_alloc(void)
77 {
78         struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
79         if (p)
80                 INIT_LIST_HEAD(&p->wb_list);
81         return p;
82 }
83
84 static inline void
85 nfs_page_free(struct nfs_page *p)
86 {
87         kmem_cache_free(nfs_page_cachep, p);
88 }
89
90 static void
91 nfs_iocounter_inc(struct nfs_io_counter *c)
92 {
93         atomic_inc(&c->io_count);
94 }
95
96 static void
97 nfs_iocounter_dec(struct nfs_io_counter *c)
98 {
99         if (atomic_dec_and_test(&c->io_count)) {
100                 clear_bit(NFS_IO_INPROGRESS, &c->flags);
101                 smp_mb__after_atomic();
102                 wake_up_bit(&c->flags, NFS_IO_INPROGRESS);
103         }
104 }
105
106 static int
107 __nfs_iocounter_wait(struct nfs_io_counter *c)
108 {
109         wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS);
110         DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS);
111         int ret = 0;
112
113         do {
114                 prepare_to_wait(wq, &q.wait, TASK_KILLABLE);
115                 set_bit(NFS_IO_INPROGRESS, &c->flags);
116                 if (atomic_read(&c->io_count) == 0)
117                         break;
118                 ret = nfs_wait_bit_killable(&q.key);
119         } while (atomic_read(&c->io_count) != 0);
120         finish_wait(wq, &q.wait);
121         return ret;
122 }
123
124 /**
125  * nfs_iocounter_wait - wait for i/o to complete
126  * @c: nfs_io_counter to use
127  *
128  * returns -ERESTARTSYS if interrupted by a fatal signal.
129  * Otherwise returns 0 once the io_count hits 0.
130  */
131 int
132 nfs_iocounter_wait(struct nfs_io_counter *c)
133 {
134         if (atomic_read(&c->io_count) == 0)
135                 return 0;
136         return __nfs_iocounter_wait(c);
137 }
138
139 /*
140  * nfs_page_group_lock - lock the head of the page group
141  * @req - request in group that is to be locked
142  * @nonblock - if true don't block waiting for lock
143  *
144  * this lock must be held if modifying the page group list
145  *
146  * return 0 on success, < 0 on error: -EDELAY if nonblocking or the
147  * result from wait_on_bit_lock
148  *
149  * NOTE: calling with nonblock=false should always have set the
150  *       lock bit (see fs/buffer.c and other uses of wait_on_bit_lock
151  *       with TASK_UNINTERRUPTIBLE), so there is no need to check the result.
152  */
153 int
154 nfs_page_group_lock(struct nfs_page *req, bool nonblock)
155 {
156         struct nfs_page *head = req->wb_head;
157
158         WARN_ON_ONCE(head != head->wb_head);
159
160         if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags))
161                 return 0;
162
163         if (!nonblock)
164                 return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
165                                 TASK_UNINTERRUPTIBLE);
166
167         return -EAGAIN;
168 }
169
170 /*
171  * nfs_page_group_lock_wait - wait for the lock to clear, but don't grab it
172  * @req - a request in the group
173  *
174  * This is a blocking call to wait for the group lock to be cleared.
175  */
176 void
177 nfs_page_group_lock_wait(struct nfs_page *req)
178 {
179         struct nfs_page *head = req->wb_head;
180
181         WARN_ON_ONCE(head != head->wb_head);
182
183         wait_on_bit(&head->wb_flags, PG_HEADLOCK,
184                 TASK_UNINTERRUPTIBLE);
185 }
186
187 /*
188  * nfs_page_group_unlock - unlock the head of the page group
189  * @req - request in group that is to be unlocked
190  */
191 void
192 nfs_page_group_unlock(struct nfs_page *req)
193 {
194         struct nfs_page *head = req->wb_head;
195
196         WARN_ON_ONCE(head != head->wb_head);
197
198         smp_mb__before_atomic();
199         clear_bit(PG_HEADLOCK, &head->wb_flags);
200         smp_mb__after_atomic();
201         wake_up_bit(&head->wb_flags, PG_HEADLOCK);
202 }
203
204 /*
205  * nfs_page_group_sync_on_bit_locked
206  *
207  * must be called with page group lock held
208  */
209 static bool
210 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
211 {
212         struct nfs_page *head = req->wb_head;
213         struct nfs_page *tmp;
214
215         WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
216         WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
217
218         tmp = req->wb_this_page;
219         while (tmp != req) {
220                 if (!test_bit(bit, &tmp->wb_flags))
221                         return false;
222                 tmp = tmp->wb_this_page;
223         }
224
225         /* true! reset all bits */
226         tmp = req;
227         do {
228                 clear_bit(bit, &tmp->wb_flags);
229                 tmp = tmp->wb_this_page;
230         } while (tmp != req);
231
232         return true;
233 }
234
235 /*
236  * nfs_page_group_sync_on_bit - set bit on current request, but only
237  *   return true if the bit is set for all requests in page group
238  * @req - request in page group
239  * @bit - PG_* bit that is used to sync page group
240  */
241 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
242 {
243         bool ret;
244
245         nfs_page_group_lock(req, false);
246         ret = nfs_page_group_sync_on_bit_locked(req, bit);
247         nfs_page_group_unlock(req);
248
249         return ret;
250 }
251
252 /*
253  * nfs_page_group_init - Initialize the page group linkage for @req
254  * @req - a new nfs request
255  * @prev - the previous request in page group, or NULL if @req is the first
256  *         or only request in the group (the head).
257  */
258 static inline void
259 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
260 {
261         WARN_ON_ONCE(prev == req);
262
263         if (!prev) {
264                 /* a head request */
265                 req->wb_head = req;
266                 req->wb_this_page = req;
267         } else {
268                 /* a subrequest */
269                 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
270                 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
271                 req->wb_head = prev->wb_head;
272                 req->wb_this_page = prev->wb_this_page;
273                 prev->wb_this_page = req;
274
275                 /* All subrequests take a ref on the head request until
276                  * nfs_page_group_destroy is called */
277                 kref_get(&req->wb_head->wb_kref);
278
279                 /* grab extra ref if head request has extra ref from
280                  * the write/commit path to handle handoff between write
281                  * and commit lists */
282                 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
283                         set_bit(PG_INODE_REF, &req->wb_flags);
284                         kref_get(&req->wb_kref);
285                 }
286         }
287 }
288
289 /*
290  * nfs_page_group_destroy - sync the destruction of page groups
291  * @req - request that no longer needs the page group
292  *
293  * releases the page group reference from each member once all
294  * members have called this function.
295  */
296 static void
297 nfs_page_group_destroy(struct kref *kref)
298 {
299         struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
300         struct nfs_page *tmp, *next;
301
302         /* subrequests must release the ref on the head request */
303         if (req->wb_head != req)
304                 nfs_release_request(req->wb_head);
305
306         if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
307                 return;
308
309         tmp = req;
310         do {
311                 next = tmp->wb_this_page;
312                 /* unlink and free */
313                 tmp->wb_this_page = tmp;
314                 tmp->wb_head = tmp;
315                 nfs_free_request(tmp);
316                 tmp = next;
317         } while (tmp != req);
318 }
319
320 /**
321  * nfs_create_request - Create an NFS read/write request.
322  * @ctx: open context to use
323  * @page: page to write
324  * @last: last nfs request created for this page group or NULL if head
325  * @offset: starting offset within the page for the write
326  * @count: number of bytes to read/write
327  *
328  * The page must be locked by the caller. This makes sure we never
329  * create two different requests for the same page.
330  * User should ensure it is safe to sleep in this function.
331  */
332 struct nfs_page *
333 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
334                    struct nfs_page *last, unsigned int offset,
335                    unsigned int count)
336 {
337         struct nfs_page         *req;
338         struct nfs_lock_context *l_ctx;
339
340         if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
341                 return ERR_PTR(-EBADF);
342         /* try to allocate the request struct */
343         req = nfs_page_alloc();
344         if (req == NULL)
345                 return ERR_PTR(-ENOMEM);
346
347         /* get lock context early so we can deal with alloc failures */
348         l_ctx = nfs_get_lock_context(ctx);
349         if (IS_ERR(l_ctx)) {
350                 nfs_page_free(req);
351                 return ERR_CAST(l_ctx);
352         }
353         req->wb_lock_context = l_ctx;
354         nfs_iocounter_inc(&l_ctx->io_count);
355
356         /* Initialize the request struct. Initially, we assume a
357          * long write-back delay. This will be adjusted in
358          * update_nfs_request below if the region is not locked. */
359         req->wb_page    = page;
360         req->wb_index   = page_file_index(page);
361         page_cache_get(page);
362         req->wb_offset  = offset;
363         req->wb_pgbase  = offset;
364         req->wb_bytes   = count;
365         req->wb_context = get_nfs_open_context(ctx);
366         kref_init(&req->wb_kref);
367         nfs_page_group_init(req, last);
368         return req;
369 }
370
371 /**
372  * nfs_unlock_request - Unlock request and wake up sleepers.
373  * @req:
374  */
375 void nfs_unlock_request(struct nfs_page *req)
376 {
377         if (!NFS_WBACK_BUSY(req)) {
378                 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
379                 BUG();
380         }
381         smp_mb__before_atomic();
382         clear_bit(PG_BUSY, &req->wb_flags);
383         smp_mb__after_atomic();
384         wake_up_bit(&req->wb_flags, PG_BUSY);
385 }
386
387 /**
388  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
389  * @req:
390  */
391 void nfs_unlock_and_release_request(struct nfs_page *req)
392 {
393         nfs_unlock_request(req);
394         nfs_release_request(req);
395 }
396
397 /*
398  * nfs_clear_request - Free up all resources allocated to the request
399  * @req:
400  *
401  * Release page and open context resources associated with a read/write
402  * request after it has completed.
403  */
404 static void nfs_clear_request(struct nfs_page *req)
405 {
406         struct page *page = req->wb_page;
407         struct nfs_open_context *ctx = req->wb_context;
408         struct nfs_lock_context *l_ctx = req->wb_lock_context;
409
410         if (page != NULL) {
411                 page_cache_release(page);
412                 req->wb_page = NULL;
413         }
414         if (l_ctx != NULL) {
415                 nfs_iocounter_dec(&l_ctx->io_count);
416                 nfs_put_lock_context(l_ctx);
417                 req->wb_lock_context = NULL;
418         }
419         if (ctx != NULL) {
420                 put_nfs_open_context(ctx);
421                 req->wb_context = NULL;
422         }
423 }
424
425 /**
426  * nfs_release_request - Release the count on an NFS read/write request
427  * @req: request to release
428  *
429  * Note: Should never be called with the spinlock held!
430  */
431 void nfs_free_request(struct nfs_page *req)
432 {
433         WARN_ON_ONCE(req->wb_this_page != req);
434
435         /* extra debug: make sure no sync bits are still set */
436         WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
437         WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
438         WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
439         WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
440         WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
441
442         /* Release struct file and open context */
443         nfs_clear_request(req);
444         nfs_page_free(req);
445 }
446
447 void nfs_release_request(struct nfs_page *req)
448 {
449         kref_put(&req->wb_kref, nfs_page_group_destroy);
450 }
451
452 /**
453  * nfs_wait_on_request - Wait for a request to complete.
454  * @req: request to wait upon.
455  *
456  * Interruptible by fatal signals only.
457  * The user is responsible for holding a count on the request.
458  */
459 int
460 nfs_wait_on_request(struct nfs_page *req)
461 {
462         return wait_on_bit_io(&req->wb_flags, PG_BUSY,
463                               TASK_UNINTERRUPTIBLE);
464 }
465
466 /*
467  * nfs_generic_pg_test - determine if requests can be coalesced
468  * @desc: pointer to descriptor
469  * @prev: previous request in desc, or NULL
470  * @req: this request
471  *
472  * Returns zero if @req can be coalesced into @desc, otherwise it returns
473  * the size of the request.
474  */
475 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
476                            struct nfs_page *prev, struct nfs_page *req)
477 {
478         if (desc->pg_count > desc->pg_bsize) {
479                 /* should never happen */
480                 WARN_ON_ONCE(1);
481                 return 0;
482         }
483
484         return min(desc->pg_bsize - desc->pg_count, (size_t)req->wb_bytes);
485 }
486 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
487
488 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
489 {
490         struct nfs_pgio_header *hdr = ops->rw_alloc_header();
491
492         if (hdr) {
493                 INIT_LIST_HEAD(&hdr->pages);
494                 spin_lock_init(&hdr->lock);
495                 hdr->rw_ops = ops;
496         }
497         return hdr;
498 }
499 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
500
501 /*
502  * nfs_pgio_header_free - Free a read or write header
503  * @hdr: The header to free
504  */
505 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
506 {
507         hdr->rw_ops->rw_free_header(hdr);
508 }
509 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
510
511 /**
512  * nfs_pgio_data_destroy - make @hdr suitable for reuse
513  *
514  * Frees memory and releases refs from nfs_generic_pgio, so that it may
515  * be called again.
516  *
517  * @hdr: A header that has had nfs_generic_pgio called
518  */
519 void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
520 {
521         put_nfs_open_context(hdr->args.context);
522         if (hdr->page_array.pagevec != hdr->page_array.page_array)
523                 kfree(hdr->page_array.pagevec);
524 }
525 EXPORT_SYMBOL_GPL(nfs_pgio_data_destroy);
526
527 /**
528  * nfs_pgio_rpcsetup - Set up arguments for a pageio call
529  * @hdr: The pageio hdr
530  * @count: Number of bytes to read
531  * @offset: Initial offset
532  * @how: How to commit data (writes only)
533  * @cinfo: Commit information for the call (writes only)
534  */
535 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
536                               unsigned int count, unsigned int offset,
537                               int how, struct nfs_commit_info *cinfo)
538 {
539         struct nfs_page *req = hdr->req;
540
541         /* Set up the RPC argument and reply structs
542          * NB: take care not to mess about with hdr->commit et al. */
543
544         hdr->args.fh     = NFS_FH(hdr->inode);
545         hdr->args.offset = req_offset(req) + offset;
546         /* pnfs_set_layoutcommit needs this */
547         hdr->mds_offset = hdr->args.offset;
548         hdr->args.pgbase = req->wb_pgbase + offset;
549         hdr->args.pages  = hdr->page_array.pagevec;
550         hdr->args.count  = count;
551         hdr->args.context = get_nfs_open_context(req->wb_context);
552         hdr->args.lock_context = req->wb_lock_context;
553         hdr->args.stable  = NFS_UNSTABLE;
554         switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
555         case 0:
556                 break;
557         case FLUSH_COND_STABLE:
558                 if (nfs_reqs_to_commit(cinfo))
559                         break;
560         default:
561                 hdr->args.stable = NFS_FILE_SYNC;
562         }
563
564         hdr->res.fattr   = &hdr->fattr;
565         hdr->res.count   = count;
566         hdr->res.eof     = 0;
567         hdr->res.verf    = &hdr->verf;
568         nfs_fattr_init(&hdr->fattr);
569 }
570
571 /**
572  * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
573  * @task: The current task
574  * @calldata: pageio header to prepare
575  */
576 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
577 {
578         struct nfs_pgio_header *hdr = calldata;
579         int err;
580         err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
581         if (err)
582                 rpc_exit(task, err);
583 }
584
585 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
586                       const struct rpc_call_ops *call_ops, int how, int flags)
587 {
588         struct rpc_task *task;
589         struct rpc_message msg = {
590                 .rpc_argp = &hdr->args,
591                 .rpc_resp = &hdr->res,
592                 .rpc_cred = hdr->cred,
593         };
594         struct rpc_task_setup task_setup_data = {
595                 .rpc_client = clnt,
596                 .task = &hdr->task,
597                 .rpc_message = &msg,
598                 .callback_ops = call_ops,
599                 .callback_data = hdr,
600                 .workqueue = nfsiod_workqueue,
601                 .flags = RPC_TASK_ASYNC | flags,
602         };
603         int ret = 0;
604
605         hdr->rw_ops->rw_initiate(hdr, &msg, &task_setup_data, how);
606
607         dprintk("NFS: %5u initiated pgio call "
608                 "(req %s/%llu, %u bytes @ offset %llu)\n",
609                 hdr->task.tk_pid,
610                 hdr->inode->i_sb->s_id,
611                 (unsigned long long)NFS_FILEID(hdr->inode),
612                 hdr->args.count,
613                 (unsigned long long)hdr->args.offset);
614
615         task = rpc_run_task(&task_setup_data);
616         if (IS_ERR(task)) {
617                 ret = PTR_ERR(task);
618                 goto out;
619         }
620         if (how & FLUSH_SYNC) {
621                 ret = rpc_wait_for_completion_task(task);
622                 if (ret == 0)
623                         ret = task->tk_status;
624         }
625         rpc_put_task(task);
626 out:
627         return ret;
628 }
629 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
630
631 /**
632  * nfs_pgio_error - Clean up from a pageio error
633  * @desc: IO descriptor
634  * @hdr: pageio header
635  */
636 static int nfs_pgio_error(struct nfs_pageio_descriptor *desc,
637                           struct nfs_pgio_header *hdr)
638 {
639         set_bit(NFS_IOHDR_REDO, &hdr->flags);
640         nfs_pgio_data_destroy(hdr);
641         hdr->completion_ops->completion(hdr);
642         desc->pg_completion_ops->error_cleanup(&desc->pg_list);
643         return -ENOMEM;
644 }
645
646 /**
647  * nfs_pgio_release - Release pageio data
648  * @calldata: The pageio header to release
649  */
650 static void nfs_pgio_release(void *calldata)
651 {
652         struct nfs_pgio_header *hdr = calldata;
653         if (hdr->rw_ops->rw_release)
654                 hdr->rw_ops->rw_release(hdr);
655         nfs_pgio_data_destroy(hdr);
656         hdr->completion_ops->completion(hdr);
657 }
658
659 /**
660  * nfs_pageio_init - initialise a page io descriptor
661  * @desc: pointer to descriptor
662  * @inode: pointer to inode
663  * @doio: pointer to io function
664  * @bsize: io block size
665  * @io_flags: extra parameters for the io function
666  */
667 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
668                      struct inode *inode,
669                      const struct nfs_pageio_ops *pg_ops,
670                      const struct nfs_pgio_completion_ops *compl_ops,
671                      const struct nfs_rw_ops *rw_ops,
672                      size_t bsize,
673                      int io_flags)
674 {
675         INIT_LIST_HEAD(&desc->pg_list);
676         desc->pg_bytes_written = 0;
677         desc->pg_count = 0;
678         desc->pg_bsize = bsize;
679         desc->pg_base = 0;
680         desc->pg_moreio = 0;
681         desc->pg_recoalesce = 0;
682         desc->pg_inode = inode;
683         desc->pg_ops = pg_ops;
684         desc->pg_completion_ops = compl_ops;
685         desc->pg_rw_ops = rw_ops;
686         desc->pg_ioflags = io_flags;
687         desc->pg_error = 0;
688         desc->pg_lseg = NULL;
689         desc->pg_dreq = NULL;
690         desc->pg_layout_private = NULL;
691 }
692 EXPORT_SYMBOL_GPL(nfs_pageio_init);
693
694 /**
695  * nfs_pgio_result - Basic pageio error handling
696  * @task: The task that ran
697  * @calldata: Pageio header to check
698  */
699 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
700 {
701         struct nfs_pgio_header *hdr = calldata;
702         struct inode *inode = hdr->inode;
703
704         dprintk("NFS: %s: %5u, (status %d)\n", __func__,
705                 task->tk_pid, task->tk_status);
706
707         if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
708                 return;
709         if (task->tk_status < 0)
710                 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
711         else
712                 hdr->rw_ops->rw_result(task, hdr);
713 }
714
715 /*
716  * Create an RPC task for the given read or write request and kick it.
717  * The page must have been locked by the caller.
718  *
719  * It may happen that the page we're passed is not marked dirty.
720  * This is the case if nfs_updatepage detects a conflicting request
721  * that has been written but not committed.
722  */
723 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
724                      struct nfs_pgio_header *hdr)
725 {
726         struct nfs_page         *req;
727         struct page             **pages;
728         struct list_head *head = &desc->pg_list;
729         struct nfs_commit_info cinfo;
730         unsigned int pagecount;
731
732         pagecount = nfs_page_array_len(desc->pg_base, desc->pg_count);
733         if (!nfs_pgarray_set(&hdr->page_array, pagecount))
734                 return nfs_pgio_error(desc, hdr);
735
736         nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
737         pages = hdr->page_array.pagevec;
738         while (!list_empty(head)) {
739                 req = nfs_list_entry(head->next);
740                 nfs_list_remove_request(req);
741                 nfs_list_add_request(req, &hdr->pages);
742                 *pages++ = req->wb_page;
743         }
744
745         if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
746             (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
747                 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
748
749         /* Set up the argument struct */
750         nfs_pgio_rpcsetup(hdr, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
751         desc->pg_rpc_callops = &nfs_pgio_common_ops;
752         return 0;
753 }
754 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
755
756 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
757 {
758         struct nfs_pgio_header *hdr;
759         int ret;
760
761         hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
762         if (!hdr) {
763                 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
764                 return -ENOMEM;
765         }
766         nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
767         ret = nfs_generic_pgio(desc, hdr);
768         if (ret == 0)
769                 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
770                                         hdr, desc->pg_rpc_callops,
771                                         desc->pg_ioflags, 0);
772         return ret;
773 }
774
775 static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
776                 const struct nfs_open_context *ctx2)
777 {
778         return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
779 }
780
781 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
782                 const struct nfs_lock_context *l2)
783 {
784         return l1->lockowner.l_owner == l2->lockowner.l_owner
785                 && l1->lockowner.l_pid == l2->lockowner.l_pid;
786 }
787
788 /**
789  * nfs_can_coalesce_requests - test two requests for compatibility
790  * @prev: pointer to nfs_page
791  * @req: pointer to nfs_page
792  *
793  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
794  * page data area they describe is contiguous, and that their RPC
795  * credentials, NFSv4 open state, and lockowners are the same.
796  *
797  * Return 'true' if this is the case, else return 'false'.
798  */
799 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
800                                       struct nfs_page *req,
801                                       struct nfs_pageio_descriptor *pgio)
802 {
803         size_t size;
804
805         if (prev) {
806                 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
807                         return false;
808                 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
809                     !nfs_match_lock_context(req->wb_lock_context,
810                                             prev->wb_lock_context))
811                         return false;
812                 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
813                         return false;
814         }
815         size = pgio->pg_ops->pg_test(pgio, prev, req);
816         WARN_ON_ONCE(size > req->wb_bytes);
817         if (size && size < req->wb_bytes)
818                 req->wb_bytes = size;
819         return size > 0;
820 }
821
822 /**
823  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
824  * @desc: destination io descriptor
825  * @req: request
826  *
827  * Returns true if the request 'req' was successfully coalesced into the
828  * existing list of pages 'desc'.
829  */
830 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
831                                      struct nfs_page *req)
832 {
833         struct nfs_page *prev = NULL;
834         if (desc->pg_count != 0) {
835                 prev = nfs_list_entry(desc->pg_list.prev);
836         } else {
837                 if (desc->pg_ops->pg_init)
838                         desc->pg_ops->pg_init(desc, req);
839                 desc->pg_base = req->wb_pgbase;
840         }
841         if (!nfs_can_coalesce_requests(prev, req, desc))
842                 return 0;
843         nfs_list_remove_request(req);
844         nfs_list_add_request(req, &desc->pg_list);
845         desc->pg_count += req->wb_bytes;
846         return 1;
847 }
848
849 /*
850  * Helper for nfs_pageio_add_request and nfs_pageio_complete
851  */
852 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
853 {
854         if (!list_empty(&desc->pg_list)) {
855                 int error = desc->pg_ops->pg_doio(desc);
856                 if (error < 0)
857                         desc->pg_error = error;
858                 else
859                         desc->pg_bytes_written += desc->pg_count;
860         }
861         if (list_empty(&desc->pg_list)) {
862                 desc->pg_count = 0;
863                 desc->pg_base = 0;
864         }
865 }
866
867 /**
868  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
869  * @desc: destination io descriptor
870  * @req: request
871  *
872  * This may split a request into subrequests which are all part of the
873  * same page group.
874  *
875  * Returns true if the request 'req' was successfully coalesced into the
876  * existing list of pages 'desc'.
877  */
878 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
879                            struct nfs_page *req)
880 {
881         struct nfs_page *subreq;
882         unsigned int bytes_left = 0;
883         unsigned int offset, pgbase;
884
885         nfs_page_group_lock(req, false);
886
887         subreq = req;
888         bytes_left = subreq->wb_bytes;
889         offset = subreq->wb_offset;
890         pgbase = subreq->wb_pgbase;
891
892         do {
893                 if (!nfs_pageio_do_add_request(desc, subreq)) {
894                         /* make sure pg_test call(s) did nothing */
895                         WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
896                         WARN_ON_ONCE(subreq->wb_offset != offset);
897                         WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
898
899                         nfs_page_group_unlock(req);
900                         desc->pg_moreio = 1;
901                         nfs_pageio_doio(desc);
902                         if (desc->pg_error < 0)
903                                 return 0;
904                         if (desc->pg_recoalesce)
905                                 return 0;
906                         /* retry add_request for this subreq */
907                         nfs_page_group_lock(req, false);
908                         continue;
909                 }
910
911                 /* check for buggy pg_test call(s) */
912                 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
913                 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
914                 WARN_ON_ONCE(subreq->wb_bytes == 0);
915
916                 bytes_left -= subreq->wb_bytes;
917                 offset += subreq->wb_bytes;
918                 pgbase += subreq->wb_bytes;
919
920                 if (bytes_left) {
921                         subreq = nfs_create_request(req->wb_context,
922                                         req->wb_page,
923                                         subreq, pgbase, bytes_left);
924                         if (IS_ERR(subreq))
925                                 goto err_ptr;
926                         nfs_lock_request(subreq);
927                         subreq->wb_offset  = offset;
928                         subreq->wb_index = req->wb_index;
929                 }
930         } while (bytes_left > 0);
931
932         nfs_page_group_unlock(req);
933         return 1;
934 err_ptr:
935         desc->pg_error = PTR_ERR(subreq);
936         nfs_page_group_unlock(req);
937         return 0;
938 }
939
940 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
941 {
942         LIST_HEAD(head);
943
944         do {
945                 list_splice_init(&desc->pg_list, &head);
946                 desc->pg_bytes_written -= desc->pg_count;
947                 desc->pg_count = 0;
948                 desc->pg_base = 0;
949                 desc->pg_recoalesce = 0;
950                 desc->pg_moreio = 0;
951
952                 while (!list_empty(&head)) {
953                         struct nfs_page *req;
954
955                         req = list_first_entry(&head, struct nfs_page, wb_list);
956                         nfs_list_remove_request(req);
957                         if (__nfs_pageio_add_request(desc, req))
958                                 continue;
959                         if (desc->pg_error < 0)
960                                 return 0;
961                         break;
962                 }
963         } while (desc->pg_recoalesce);
964         return 1;
965 }
966
967 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
968                 struct nfs_page *req)
969 {
970         int ret;
971
972         do {
973                 ret = __nfs_pageio_add_request(desc, req);
974                 if (ret)
975                         break;
976                 if (desc->pg_error < 0)
977                         break;
978                 ret = nfs_do_recoalesce(desc);
979         } while (ret);
980         return ret;
981 }
982
983 /*
984  * nfs_pageio_resend - Transfer requests to new descriptor and resend
985  * @hdr - the pgio header to move request from
986  * @desc - the pageio descriptor to add requests to
987  *
988  * Try to move each request (nfs_page) from @hdr to @desc then attempt
989  * to send them.
990  *
991  * Returns 0 on success and < 0 on error.
992  */
993 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
994                       struct nfs_pgio_header *hdr)
995 {
996         LIST_HEAD(failed);
997
998         desc->pg_dreq = hdr->dreq;
999         while (!list_empty(&hdr->pages)) {
1000                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
1001
1002                 nfs_list_remove_request(req);
1003                 if (!nfs_pageio_add_request(desc, req))
1004                         nfs_list_add_request(req, &failed);
1005         }
1006         nfs_pageio_complete(desc);
1007         if (!list_empty(&failed)) {
1008                 list_move(&failed, &hdr->pages);
1009                 return -EIO;
1010         }
1011         return 0;
1012 }
1013 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1014
1015 /**
1016  * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
1017  * @desc: pointer to io descriptor
1018  */
1019 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1020 {
1021         for (;;) {
1022                 nfs_pageio_doio(desc);
1023                 if (!desc->pg_recoalesce)
1024                         break;
1025                 if (!nfs_do_recoalesce(desc))
1026                         break;
1027         }
1028 }
1029
1030 /**
1031  * nfs_pageio_cond_complete - Conditional I/O completion
1032  * @desc: pointer to io descriptor
1033  * @index: page index
1034  *
1035  * It is important to ensure that processes don't try to take locks
1036  * on non-contiguous ranges of pages as that might deadlock. This
1037  * function should be called before attempting to wait on a locked
1038  * nfs_page. It will complete the I/O if the page index 'index'
1039  * is not contiguous with the existing list of pages in 'desc'.
1040  */
1041 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1042 {
1043         if (!list_empty(&desc->pg_list)) {
1044                 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
1045                 if (index != prev->wb_index + 1)
1046                         nfs_pageio_complete(desc);
1047         }
1048 }
1049
1050 int __init nfs_init_nfspagecache(void)
1051 {
1052         nfs_page_cachep = kmem_cache_create("nfs_page",
1053                                             sizeof(struct nfs_page),
1054                                             0, SLAB_HWCACHE_ALIGN,
1055                                             NULL);
1056         if (nfs_page_cachep == NULL)
1057                 return -ENOMEM;
1058
1059         return 0;
1060 }
1061
1062 void nfs_destroy_nfspagecache(void)
1063 {
1064         kmem_cache_destroy(nfs_page_cachep);
1065 }
1066
1067 static const struct rpc_call_ops nfs_pgio_common_ops = {
1068         .rpc_call_prepare = nfs_pgio_prepare,
1069         .rpc_call_done = nfs_pgio_result,
1070         .rpc_release = nfs_pgio_release,
1071 };
1072
1073 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1074         .pg_test = nfs_generic_pg_test,
1075         .pg_doio = nfs_generic_pg_pgios,
1076 };