Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
1da177e4 LT |
12 | #include <linux/slab.h> |
13 | #include <linux/file.h> | |
e8edc6e0 | 14 | #include <linux/sched.h> |
1da177e4 | 15 | #include <linux/sunrpc/clnt.h> |
1313e603 | 16 | #include <linux/nfs.h> |
1da177e4 LT |
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> | |
afeacc8c | 22 | #include <linux/export.h> |
1da177e4 | 23 | |
8d5658c9 | 24 | #include "internal.h" |
bae724ef | 25 | #include "pnfs.h" |
8d5658c9 | 26 | |
0eecb214 AS |
27 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
28 | ||
e18b890b | 29 | static struct kmem_cache *nfs_page_cachep; |
ef2c488c | 30 | static const struct rpc_call_ops nfs_pgio_common_ops; |
1da177e4 | 31 | |
00bfa30a | 32 | static bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount) |
30dd374f FI |
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 | ||
4db6e0b7 FI |
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; | |
584aa810 | 54 | hdr->dreq = desc->pg_dreq; |
f6166384 | 55 | hdr->layout_private = desc->pg_layout_private; |
4db6e0b7 | 56 | hdr->release = release; |
061ae2ed | 57 | hdr->completion_ops = desc->pg_completion_ops; |
584aa810 FI |
58 | if (hdr->completion_ops->init_hdr) |
59 | hdr->completion_ops->init_hdr(hdr); | |
4db6e0b7 | 60 | } |
89d77c8f | 61 | EXPORT_SYMBOL_GPL(nfs_pgheader_init); |
4db6e0b7 FI |
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 | ||
1da177e4 LT |
75 | static inline struct nfs_page * |
76 | nfs_page_alloc(void) | |
77 | { | |
192e501b | 78 | struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO); |
72895b1a | 79 | if (p) |
1da177e4 | 80 | INIT_LIST_HEAD(&p->wb_list); |
1da177e4 LT |
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 | ||
577b4232 TM |
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_clear_bit(); | |
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(&c->flags); | |
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 | ||
1da177e4 LT |
139 | /** |
140 | * nfs_create_request - Create an NFS read/write request. | |
c02f557d | 141 | * @ctx: open context to use |
1da177e4 LT |
142 | * @inode: inode to which the request is attached |
143 | * @page: page to write | |
144 | * @offset: starting offset within the page for the write | |
145 | * @count: number of bytes to read/write | |
146 | * | |
147 | * The page must be locked by the caller. This makes sure we never | |
a19b89ca | 148 | * create two different requests for the same page. |
1da177e4 LT |
149 | * User should ensure it is safe to sleep in this function. |
150 | */ | |
151 | struct nfs_page * | |
152 | nfs_create_request(struct nfs_open_context *ctx, struct inode *inode, | |
153 | struct page *page, | |
154 | unsigned int offset, unsigned int count) | |
155 | { | |
1da177e4 | 156 | struct nfs_page *req; |
b3c54de6 | 157 | struct nfs_lock_context *l_ctx; |
1da177e4 | 158 | |
c58c8441 TM |
159 | if (test_bit(NFS_CONTEXT_BAD, &ctx->flags)) |
160 | return ERR_PTR(-EBADF); | |
18eb8842 TM |
161 | /* try to allocate the request struct */ |
162 | req = nfs_page_alloc(); | |
163 | if (req == NULL) | |
164 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 165 | |
015f0212 | 166 | /* get lock context early so we can deal with alloc failures */ |
b3c54de6 TM |
167 | l_ctx = nfs_get_lock_context(ctx); |
168 | if (IS_ERR(l_ctx)) { | |
015f0212 | 169 | nfs_page_free(req); |
b3c54de6 | 170 | return ERR_CAST(l_ctx); |
015f0212 | 171 | } |
b3c54de6 | 172 | req->wb_lock_context = l_ctx; |
577b4232 | 173 | nfs_iocounter_inc(&l_ctx->io_count); |
015f0212 | 174 | |
1da177e4 LT |
175 | /* Initialize the request struct. Initially, we assume a |
176 | * long write-back delay. This will be adjusted in | |
177 | * update_nfs_request below if the region is not locked. */ | |
178 | req->wb_page = page; | |
d56b4ddf | 179 | req->wb_index = page_file_index(page); |
1da177e4 LT |
180 | page_cache_get(page); |
181 | req->wb_offset = offset; | |
182 | req->wb_pgbase = offset; | |
183 | req->wb_bytes = count; | |
1da177e4 | 184 | req->wb_context = get_nfs_open_context(ctx); |
c03b4024 | 185 | kref_init(&req->wb_kref); |
1da177e4 LT |
186 | return req; |
187 | } | |
188 | ||
189 | /** | |
1d1afcbc | 190 | * nfs_unlock_request - Unlock request and wake up sleepers. |
1da177e4 LT |
191 | * @req: |
192 | */ | |
1d1afcbc | 193 | void nfs_unlock_request(struct nfs_page *req) |
1da177e4 LT |
194 | { |
195 | if (!NFS_WBACK_BUSY(req)) { | |
196 | printk(KERN_ERR "NFS: Invalid unlock attempted\n"); | |
197 | BUG(); | |
198 | } | |
199 | smp_mb__before_clear_bit(); | |
200 | clear_bit(PG_BUSY, &req->wb_flags); | |
201 | smp_mb__after_clear_bit(); | |
464a98bd | 202 | wake_up_bit(&req->wb_flags, PG_BUSY); |
3aff4ebb TM |
203 | } |
204 | ||
205 | /** | |
1d1afcbc TM |
206 | * nfs_unlock_and_release_request - Unlock request and release the nfs_page |
207 | * @req: | |
3aff4ebb | 208 | */ |
1d1afcbc | 209 | void nfs_unlock_and_release_request(struct nfs_page *req) |
3aff4ebb | 210 | { |
1d1afcbc | 211 | nfs_unlock_request(req); |
1da177e4 LT |
212 | nfs_release_request(req); |
213 | } | |
214 | ||
4d65c520 | 215 | /* |
1da177e4 LT |
216 | * nfs_clear_request - Free up all resources allocated to the request |
217 | * @req: | |
218 | * | |
bb6fbc45 TM |
219 | * Release page and open context resources associated with a read/write |
220 | * request after it has completed. | |
1da177e4 | 221 | */ |
4d65c520 | 222 | static void nfs_clear_request(struct nfs_page *req) |
1da177e4 | 223 | { |
cd52ed35 | 224 | struct page *page = req->wb_page; |
bb6fbc45 | 225 | struct nfs_open_context *ctx = req->wb_context; |
f11ac8db | 226 | struct nfs_lock_context *l_ctx = req->wb_lock_context; |
bb6fbc45 | 227 | |
cd52ed35 | 228 | if (page != NULL) { |
cd52ed35 | 229 | page_cache_release(page); |
1da177e4 LT |
230 | req->wb_page = NULL; |
231 | } | |
f11ac8db | 232 | if (l_ctx != NULL) { |
577b4232 | 233 | nfs_iocounter_dec(&l_ctx->io_count); |
f11ac8db TM |
234 | nfs_put_lock_context(l_ctx); |
235 | req->wb_lock_context = NULL; | |
236 | } | |
bb6fbc45 TM |
237 | if (ctx != NULL) { |
238 | put_nfs_open_context(ctx); | |
239 | req->wb_context = NULL; | |
240 | } | |
1da177e4 LT |
241 | } |
242 | ||
243 | ||
244 | /** | |
245 | * nfs_release_request - Release the count on an NFS read/write request | |
246 | * @req: request to release | |
247 | * | |
248 | * Note: Should never be called with the spinlock held! | |
249 | */ | |
c03b4024 | 250 | static void nfs_free_request(struct kref *kref) |
1da177e4 | 251 | { |
c03b4024 | 252 | struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref); |
1da177e4 | 253 | |
bb6fbc45 | 254 | /* Release struct file and open context */ |
1da177e4 | 255 | nfs_clear_request(req); |
1da177e4 LT |
256 | nfs_page_free(req); |
257 | } | |
258 | ||
c03b4024 TM |
259 | void nfs_release_request(struct nfs_page *req) |
260 | { | |
261 | kref_put(&req->wb_kref, nfs_free_request); | |
262 | } | |
263 | ||
9f557cd8 TM |
264 | static int nfs_wait_bit_uninterruptible(void *word) |
265 | { | |
266 | io_schedule(); | |
267 | return 0; | |
268 | } | |
269 | ||
1da177e4 LT |
270 | /** |
271 | * nfs_wait_on_request - Wait for a request to complete. | |
272 | * @req: request to wait upon. | |
273 | * | |
150030b7 | 274 | * Interruptible by fatal signals only. |
1da177e4 LT |
275 | * The user is responsible for holding a count on the request. |
276 | */ | |
277 | int | |
278 | nfs_wait_on_request(struct nfs_page *req) | |
279 | { | |
9f557cd8 TM |
280 | return wait_on_bit(&req->wb_flags, PG_BUSY, |
281 | nfs_wait_bit_uninterruptible, | |
282 | TASK_UNINTERRUPTIBLE); | |
1da177e4 LT |
283 | } |
284 | ||
19345cb2 | 285 | bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req) |
5b36c7dc BH |
286 | { |
287 | /* | |
288 | * FIXME: ideally we should be able to coalesce all requests | |
289 | * that are not block boundary aligned, but currently this | |
290 | * is problematic for the case of bsize < PAGE_CACHE_SIZE, | |
291 | * since nfs_flush_multi and nfs_pagein_multi assume you | |
292 | * can have only one struct nfs_page. | |
293 | */ | |
294 | if (desc->pg_bsize < PAGE_SIZE) | |
295 | return 0; | |
296 | ||
297 | return desc->pg_count + req->wb_bytes <= desc->pg_bsize; | |
298 | } | |
19345cb2 | 299 | EXPORT_SYMBOL_GPL(nfs_generic_pg_test); |
5b36c7dc | 300 | |
00bfa30a AS |
301 | static inline struct nfs_rw_header *NFS_RW_HEADER(struct nfs_pgio_header *hdr) |
302 | { | |
303 | return container_of(hdr, struct nfs_rw_header, header); | |
304 | } | |
305 | ||
4a0de55c AS |
306 | /** |
307 | * nfs_rw_header_alloc - Allocate a header for a read or write | |
308 | * @ops: Read or write function vector | |
309 | */ | |
310 | struct nfs_rw_header *nfs_rw_header_alloc(const struct nfs_rw_ops *ops) | |
311 | { | |
312 | struct nfs_rw_header *header = ops->rw_alloc_header(); | |
313 | ||
314 | if (header) { | |
315 | struct nfs_pgio_header *hdr = &header->header; | |
316 | ||
317 | INIT_LIST_HEAD(&hdr->pages); | |
318 | INIT_LIST_HEAD(&hdr->rpc_list); | |
319 | spin_lock_init(&hdr->lock); | |
320 | atomic_set(&hdr->refcnt, 0); | |
321 | hdr->rw_ops = ops; | |
322 | } | |
323 | return header; | |
324 | } | |
325 | EXPORT_SYMBOL_GPL(nfs_rw_header_alloc); | |
326 | ||
327 | /* | |
328 | * nfs_rw_header_free - Free a read or write header | |
329 | * @hdr: The header to free | |
330 | */ | |
331 | void nfs_rw_header_free(struct nfs_pgio_header *hdr) | |
332 | { | |
333 | hdr->rw_ops->rw_free_header(NFS_RW_HEADER(hdr)); | |
334 | } | |
335 | EXPORT_SYMBOL_GPL(nfs_rw_header_free); | |
336 | ||
00bfa30a AS |
337 | /** |
338 | * nfs_pgio_data_alloc - Allocate pageio data | |
339 | * @hdr: The header making a request | |
340 | * @pagecount: Number of pages to create | |
341 | */ | |
ef2c488c AS |
342 | static struct nfs_pgio_data *nfs_pgio_data_alloc(struct nfs_pgio_header *hdr, |
343 | unsigned int pagecount) | |
00bfa30a AS |
344 | { |
345 | struct nfs_pgio_data *data, *prealloc; | |
346 | ||
347 | prealloc = &NFS_RW_HEADER(hdr)->rpc_data; | |
348 | if (prealloc->header == NULL) | |
349 | data = prealloc; | |
350 | else | |
351 | data = kzalloc(sizeof(*data), GFP_KERNEL); | |
352 | if (!data) | |
353 | goto out; | |
354 | ||
355 | if (nfs_pgarray_set(&data->pages, pagecount)) { | |
356 | data->header = hdr; | |
357 | atomic_inc(&hdr->refcnt); | |
358 | } else { | |
359 | if (data != prealloc) | |
360 | kfree(data); | |
361 | data = NULL; | |
362 | } | |
363 | out: | |
364 | return data; | |
365 | } | |
366 | ||
367 | /** | |
368 | * nfs_pgio_data_release - Properly free pageio data | |
369 | * @data: The data to release | |
370 | */ | |
371 | void nfs_pgio_data_release(struct nfs_pgio_data *data) | |
372 | { | |
373 | struct nfs_pgio_header *hdr = data->header; | |
374 | struct nfs_rw_header *pageio_header = NFS_RW_HEADER(hdr); | |
375 | ||
376 | put_nfs_open_context(data->args.context); | |
377 | if (data->pages.pagevec != data->pages.page_array) | |
378 | kfree(data->pages.pagevec); | |
379 | if (data == &pageio_header->rpc_data) { | |
380 | data->header = NULL; | |
381 | data = NULL; | |
382 | } | |
383 | if (atomic_dec_and_test(&hdr->refcnt)) | |
384 | hdr->completion_ops->completion(hdr); | |
385 | /* Note: we only free the rpc_task after callbacks are done. | |
386 | * See the comment in rpc_free_task() for why | |
387 | */ | |
388 | kfree(data); | |
389 | } | |
390 | EXPORT_SYMBOL_GPL(nfs_pgio_data_release); | |
391 | ||
ce59515c AS |
392 | /** |
393 | * nfs_pgio_rpcsetup - Set up arguments for a pageio call | |
394 | * @data: The pageio data | |
395 | * @count: Number of bytes to read | |
396 | * @offset: Initial offset | |
397 | * @how: How to commit data (writes only) | |
398 | * @cinfo: Commit information for the call (writes only) | |
399 | */ | |
ef2c488c | 400 | static void nfs_pgio_rpcsetup(struct nfs_pgio_data *data, |
ce59515c AS |
401 | unsigned int count, unsigned int offset, |
402 | int how, struct nfs_commit_info *cinfo) | |
403 | { | |
404 | struct nfs_page *req = data->header->req; | |
405 | ||
406 | /* Set up the RPC argument and reply structs | |
407 | * NB: take care not to mess about with data->commit et al. */ | |
408 | ||
409 | data->args.fh = NFS_FH(data->header->inode); | |
410 | data->args.offset = req_offset(req) + offset; | |
411 | /* pnfs_set_layoutcommit needs this */ | |
412 | data->mds_offset = data->args.offset; | |
413 | data->args.pgbase = req->wb_pgbase + offset; | |
414 | data->args.pages = data->pages.pagevec; | |
415 | data->args.count = count; | |
416 | data->args.context = get_nfs_open_context(req->wb_context); | |
417 | data->args.lock_context = req->wb_lock_context; | |
418 | data->args.stable = NFS_UNSTABLE; | |
419 | switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) { | |
420 | case 0: | |
421 | break; | |
422 | case FLUSH_COND_STABLE: | |
423 | if (nfs_reqs_to_commit(cinfo)) | |
424 | break; | |
425 | default: | |
426 | data->args.stable = NFS_FILE_SYNC; | |
427 | } | |
428 | ||
429 | data->res.fattr = &data->fattr; | |
430 | data->res.count = count; | |
431 | data->res.eof = 0; | |
432 | data->res.verf = &data->verf; | |
433 | nfs_fattr_init(&data->fattr); | |
434 | } | |
435 | ||
a4cdda59 AS |
436 | /** |
437 | * nfs_pgio_prepare - Prepare pageio data to go over the wire | |
438 | * @task: The current task | |
439 | * @calldata: pageio data to prepare | |
440 | */ | |
6f92fa45 | 441 | static void nfs_pgio_prepare(struct rpc_task *task, void *calldata) |
a4cdda59 AS |
442 | { |
443 | struct nfs_pgio_data *data = calldata; | |
444 | int err; | |
445 | err = NFS_PROTO(data->header->inode)->pgio_rpc_prepare(task, data); | |
446 | if (err) | |
447 | rpc_exit(task, err); | |
448 | } | |
449 | ||
1ed26f33 AS |
450 | int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_data *data, |
451 | const struct rpc_call_ops *call_ops, int how, int flags) | |
452 | { | |
453 | struct rpc_task *task; | |
454 | struct rpc_message msg = { | |
455 | .rpc_argp = &data->args, | |
456 | .rpc_resp = &data->res, | |
457 | .rpc_cred = data->header->cred, | |
458 | }; | |
459 | struct rpc_task_setup task_setup_data = { | |
460 | .rpc_client = clnt, | |
461 | .task = &data->task, | |
462 | .rpc_message = &msg, | |
463 | .callback_ops = call_ops, | |
464 | .callback_data = data, | |
465 | .workqueue = nfsiod_workqueue, | |
466 | .flags = RPC_TASK_ASYNC | flags, | |
467 | }; | |
468 | int ret = 0; | |
469 | ||
470 | data->header->rw_ops->rw_initiate(data, &msg, &task_setup_data, how); | |
471 | ||
472 | dprintk("NFS: %5u initiated pgio call " | |
473 | "(req %s/%llu, %u bytes @ offset %llu)\n", | |
474 | data->task.tk_pid, | |
475 | data->header->inode->i_sb->s_id, | |
476 | (unsigned long long)NFS_FILEID(data->header->inode), | |
477 | data->args.count, | |
478 | (unsigned long long)data->args.offset); | |
479 | ||
480 | task = rpc_run_task(&task_setup_data); | |
481 | if (IS_ERR(task)) { | |
482 | ret = PTR_ERR(task); | |
483 | goto out; | |
484 | } | |
485 | if (how & FLUSH_SYNC) { | |
486 | ret = rpc_wait_for_completion_task(task); | |
487 | if (ret == 0) | |
488 | ret = task->tk_status; | |
489 | } | |
490 | rpc_put_task(task); | |
491 | out: | |
492 | return ret; | |
493 | } | |
494 | EXPORT_SYMBOL_GPL(nfs_initiate_pgio); | |
495 | ||
cf485fcd | 496 | static int nfs_do_multiple_pgios(struct list_head *head, |
c3766276 AS |
497 | const struct rpc_call_ops *call_ops, |
498 | int how) | |
499 | { | |
500 | struct nfs_pgio_data *data; | |
501 | int ret = 0; | |
502 | ||
503 | while (!list_empty(head)) { | |
504 | int ret2; | |
505 | ||
506 | data = list_first_entry(head, struct nfs_pgio_data, list); | |
507 | list_del_init(&data->list); | |
508 | ||
509 | ret2 = nfs_initiate_pgio(NFS_CLIENT(data->header->inode), | |
510 | data, call_ops, how, 0); | |
511 | if (ret == 0) | |
512 | ret = ret2; | |
513 | } | |
514 | return ret; | |
515 | } | |
516 | ||
844c9e69 AS |
517 | /** |
518 | * nfs_pgio_error - Clean up from a pageio error | |
519 | * @desc: IO descriptor | |
520 | * @hdr: pageio header | |
521 | */ | |
ef2c488c | 522 | static int nfs_pgio_error(struct nfs_pageio_descriptor *desc, |
844c9e69 AS |
523 | struct nfs_pgio_header *hdr) |
524 | { | |
525 | struct nfs_pgio_data *data; | |
526 | ||
527 | set_bit(NFS_IOHDR_REDO, &hdr->flags); | |
528 | while (!list_empty(&hdr->rpc_list)) { | |
529 | data = list_first_entry(&hdr->rpc_list, struct nfs_pgio_data, list); | |
530 | list_del(&data->list); | |
531 | nfs_pgio_data_release(data); | |
532 | } | |
533 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); | |
534 | return -ENOMEM; | |
535 | } | |
536 | ||
a4cdda59 AS |
537 | /** |
538 | * nfs_pgio_release - Release pageio data | |
539 | * @calldata: The pageio data to release | |
540 | */ | |
6f92fa45 | 541 | static void nfs_pgio_release(void *calldata) |
a4cdda59 AS |
542 | { |
543 | struct nfs_pgio_data *data = calldata; | |
544 | if (data->header->rw_ops->rw_release) | |
545 | data->header->rw_ops->rw_release(data); | |
546 | nfs_pgio_data_release(data); | |
547 | } | |
548 | ||
1da177e4 | 549 | /** |
d8a5ad75 TM |
550 | * nfs_pageio_init - initialise a page io descriptor |
551 | * @desc: pointer to descriptor | |
bcb71bba TM |
552 | * @inode: pointer to inode |
553 | * @doio: pointer to io function | |
554 | * @bsize: io block size | |
555 | * @io_flags: extra parameters for the io function | |
d8a5ad75 | 556 | */ |
bcb71bba TM |
557 | void nfs_pageio_init(struct nfs_pageio_descriptor *desc, |
558 | struct inode *inode, | |
1751c363 | 559 | const struct nfs_pageio_ops *pg_ops, |
061ae2ed | 560 | const struct nfs_pgio_completion_ops *compl_ops, |
4a0de55c | 561 | const struct nfs_rw_ops *rw_ops, |
84dde76c | 562 | size_t bsize, |
bcb71bba | 563 | int io_flags) |
d8a5ad75 TM |
564 | { |
565 | INIT_LIST_HEAD(&desc->pg_list); | |
bcb71bba | 566 | desc->pg_bytes_written = 0; |
d8a5ad75 TM |
567 | desc->pg_count = 0; |
568 | desc->pg_bsize = bsize; | |
569 | desc->pg_base = 0; | |
b31268ac | 570 | desc->pg_moreio = 0; |
d9156f9f | 571 | desc->pg_recoalesce = 0; |
bcb71bba | 572 | desc->pg_inode = inode; |
1751c363 | 573 | desc->pg_ops = pg_ops; |
061ae2ed | 574 | desc->pg_completion_ops = compl_ops; |
4a0de55c | 575 | desc->pg_rw_ops = rw_ops; |
bcb71bba TM |
576 | desc->pg_ioflags = io_flags; |
577 | desc->pg_error = 0; | |
94ad1c80 | 578 | desc->pg_lseg = NULL; |
584aa810 | 579 | desc->pg_dreq = NULL; |
f6166384 | 580 | desc->pg_layout_private = NULL; |
d8a5ad75 | 581 | } |
89d77c8f | 582 | EXPORT_SYMBOL_GPL(nfs_pageio_init); |
d8a5ad75 | 583 | |
0eecb214 AS |
584 | /** |
585 | * nfs_pgio_result - Basic pageio error handling | |
586 | * @task: The task that ran | |
587 | * @calldata: Pageio data to check | |
588 | */ | |
6f92fa45 | 589 | static void nfs_pgio_result(struct rpc_task *task, void *calldata) |
0eecb214 AS |
590 | { |
591 | struct nfs_pgio_data *data = calldata; | |
592 | struct inode *inode = data->header->inode; | |
593 | ||
594 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, | |
595 | task->tk_pid, task->tk_status); | |
596 | ||
597 | if (data->header->rw_ops->rw_done(task, data, inode) != 0) | |
598 | return; | |
599 | if (task->tk_status < 0) | |
600 | nfs_set_pgio_error(data->header, task->tk_status, data->args.offset); | |
601 | else | |
602 | data->header->rw_ops->rw_result(task, data); | |
603 | } | |
604 | ||
ef2c488c AS |
605 | /* |
606 | * Generate multiple small requests to read or write a single | |
607 | * contiguous dirty on one page. | |
608 | */ | |
609 | static int nfs_pgio_multi(struct nfs_pageio_descriptor *desc, | |
610 | struct nfs_pgio_header *hdr) | |
611 | { | |
612 | struct nfs_page *req = hdr->req; | |
613 | struct page *page = req->wb_page; | |
614 | struct nfs_pgio_data *data; | |
615 | size_t wsize = desc->pg_bsize, nbytes; | |
616 | unsigned int offset; | |
617 | int requests = 0; | |
618 | struct nfs_commit_info cinfo; | |
619 | ||
620 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); | |
621 | ||
622 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && | |
623 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) || | |
624 | desc->pg_count > wsize)) | |
625 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
626 | ||
627 | offset = 0; | |
628 | nbytes = desc->pg_count; | |
629 | do { | |
630 | size_t len = min(nbytes, wsize); | |
631 | ||
632 | data = nfs_pgio_data_alloc(hdr, 1); | |
633 | if (!data) | |
634 | return nfs_pgio_error(desc, hdr); | |
635 | data->pages.pagevec[0] = page; | |
636 | nfs_pgio_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo); | |
637 | list_add(&data->list, &hdr->rpc_list); | |
638 | requests++; | |
639 | nbytes -= len; | |
640 | offset += len; | |
641 | } while (nbytes != 0); | |
642 | ||
643 | nfs_list_remove_request(req); | |
644 | nfs_list_add_request(req, &hdr->pages); | |
645 | desc->pg_rpc_callops = &nfs_pgio_common_ops; | |
646 | return 0; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Create an RPC task for the given read or write request and kick it. | |
651 | * The page must have been locked by the caller. | |
652 | * | |
653 | * It may happen that the page we're passed is not marked dirty. | |
654 | * This is the case if nfs_updatepage detects a conflicting request | |
655 | * that has been written but not committed. | |
656 | */ | |
657 | static int nfs_pgio_one(struct nfs_pageio_descriptor *desc, | |
658 | struct nfs_pgio_header *hdr) | |
659 | { | |
660 | struct nfs_page *req; | |
661 | struct page **pages; | |
662 | struct nfs_pgio_data *data; | |
663 | struct list_head *head = &desc->pg_list; | |
664 | struct nfs_commit_info cinfo; | |
665 | ||
666 | data = nfs_pgio_data_alloc(hdr, nfs_page_array_len(desc->pg_base, | |
667 | desc->pg_count)); | |
668 | if (!data) | |
669 | return nfs_pgio_error(desc, hdr); | |
670 | ||
671 | nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq); | |
672 | pages = data->pages.pagevec; | |
673 | while (!list_empty(head)) { | |
674 | req = nfs_list_entry(head->next); | |
675 | nfs_list_remove_request(req); | |
676 | nfs_list_add_request(req, &hdr->pages); | |
677 | *pages++ = req->wb_page; | |
678 | } | |
679 | ||
680 | if ((desc->pg_ioflags & FLUSH_COND_STABLE) && | |
681 | (desc->pg_moreio || nfs_reqs_to_commit(&cinfo))) | |
682 | desc->pg_ioflags &= ~FLUSH_COND_STABLE; | |
683 | ||
684 | /* Set up the argument struct */ | |
685 | nfs_pgio_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo); | |
686 | list_add(&data->list, &hdr->rpc_list); | |
687 | desc->pg_rpc_callops = &nfs_pgio_common_ops; | |
688 | return 0; | |
689 | } | |
690 | ||
cf485fcd AS |
691 | int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc) |
692 | { | |
693 | struct nfs_rw_header *rw_hdr; | |
694 | struct nfs_pgio_header *hdr; | |
695 | int ret; | |
696 | ||
697 | rw_hdr = nfs_rw_header_alloc(desc->pg_rw_ops); | |
698 | if (!rw_hdr) { | |
699 | desc->pg_completion_ops->error_cleanup(&desc->pg_list); | |
700 | return -ENOMEM; | |
701 | } | |
702 | hdr = &rw_hdr->header; | |
703 | nfs_pgheader_init(desc, hdr, nfs_rw_header_free); | |
704 | atomic_inc(&hdr->refcnt); | |
705 | ret = nfs_generic_pgio(desc, hdr); | |
706 | if (ret == 0) | |
707 | ret = nfs_do_multiple_pgios(&hdr->rpc_list, | |
708 | desc->pg_rpc_callops, | |
709 | desc->pg_ioflags); | |
710 | if (atomic_dec_and_test(&hdr->refcnt)) | |
711 | hdr->completion_ops->completion(hdr); | |
712 | return ret; | |
713 | } | |
714 | ||
ef2c488c AS |
715 | int nfs_generic_pgio(struct nfs_pageio_descriptor *desc, |
716 | struct nfs_pgio_header *hdr) | |
717 | { | |
718 | if (desc->pg_bsize < PAGE_CACHE_SIZE) | |
719 | return nfs_pgio_multi(desc, hdr); | |
720 | return nfs_pgio_one(desc, hdr); | |
721 | } | |
722 | EXPORT_SYMBOL_GPL(nfs_generic_pgio); | |
723 | ||
4109bb74 TM |
724 | static bool nfs_match_open_context(const struct nfs_open_context *ctx1, |
725 | const struct nfs_open_context *ctx2) | |
726 | { | |
727 | return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state; | |
728 | } | |
729 | ||
730 | static bool nfs_match_lock_context(const struct nfs_lock_context *l1, | |
731 | const struct nfs_lock_context *l2) | |
732 | { | |
733 | return l1->lockowner.l_owner == l2->lockowner.l_owner | |
734 | && l1->lockowner.l_pid == l2->lockowner.l_pid; | |
735 | } | |
736 | ||
d8a5ad75 TM |
737 | /** |
738 | * nfs_can_coalesce_requests - test two requests for compatibility | |
739 | * @prev: pointer to nfs_page | |
740 | * @req: pointer to nfs_page | |
741 | * | |
742 | * The nfs_page structures 'prev' and 'req' are compared to ensure that the | |
743 | * page data area they describe is contiguous, and that their RPC | |
744 | * credentials, NFSv4 open state, and lockowners are the same. | |
745 | * | |
746 | * Return 'true' if this is the case, else return 'false'. | |
747 | */ | |
18ad0a9f BH |
748 | static bool nfs_can_coalesce_requests(struct nfs_page *prev, |
749 | struct nfs_page *req, | |
750 | struct nfs_pageio_descriptor *pgio) | |
d8a5ad75 | 751 | { |
4109bb74 | 752 | if (!nfs_match_open_context(req->wb_context, prev->wb_context)) |
18ad0a9f | 753 | return false; |
4109bb74 TM |
754 | if (req->wb_context->dentry->d_inode->i_flock != NULL && |
755 | !nfs_match_lock_context(req->wb_lock_context, prev->wb_lock_context)) | |
18ad0a9f | 756 | return false; |
d8a5ad75 | 757 | if (req->wb_pgbase != 0) |
18ad0a9f | 758 | return false; |
d8a5ad75 | 759 | if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE) |
18ad0a9f | 760 | return false; |
1825a0d0 FI |
761 | if (req_offset(req) != req_offset(prev) + prev->wb_bytes) |
762 | return false; | |
1751c363 | 763 | return pgio->pg_ops->pg_test(pgio, prev, req); |
d8a5ad75 TM |
764 | } |
765 | ||
766 | /** | |
bcb71bba | 767 | * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list. |
d8a5ad75 TM |
768 | * @desc: destination io descriptor |
769 | * @req: request | |
770 | * | |
771 | * Returns true if the request 'req' was successfully coalesced into the | |
772 | * existing list of pages 'desc'. | |
773 | */ | |
bcb71bba TM |
774 | static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc, |
775 | struct nfs_page *req) | |
d8a5ad75 | 776 | { |
d8a5ad75 TM |
777 | if (desc->pg_count != 0) { |
778 | struct nfs_page *prev; | |
779 | ||
d8a5ad75 | 780 | prev = nfs_list_entry(desc->pg_list.prev); |
94ad1c80 | 781 | if (!nfs_can_coalesce_requests(prev, req, desc)) |
d8a5ad75 | 782 | return 0; |
5b36c7dc | 783 | } else { |
d8007d4d TM |
784 | if (desc->pg_ops->pg_init) |
785 | desc->pg_ops->pg_init(desc, req); | |
d8a5ad75 | 786 | desc->pg_base = req->wb_pgbase; |
5b36c7dc | 787 | } |
d8a5ad75 TM |
788 | nfs_list_remove_request(req); |
789 | nfs_list_add_request(req, &desc->pg_list); | |
5b36c7dc | 790 | desc->pg_count += req->wb_bytes; |
d8a5ad75 TM |
791 | return 1; |
792 | } | |
793 | ||
bcb71bba TM |
794 | /* |
795 | * Helper for nfs_pageio_add_request and nfs_pageio_complete | |
796 | */ | |
797 | static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc) | |
798 | { | |
799 | if (!list_empty(&desc->pg_list)) { | |
1751c363 | 800 | int error = desc->pg_ops->pg_doio(desc); |
bcb71bba TM |
801 | if (error < 0) |
802 | desc->pg_error = error; | |
803 | else | |
804 | desc->pg_bytes_written += desc->pg_count; | |
805 | } | |
806 | if (list_empty(&desc->pg_list)) { | |
807 | desc->pg_count = 0; | |
808 | desc->pg_base = 0; | |
809 | } | |
810 | } | |
811 | ||
812 | /** | |
813 | * nfs_pageio_add_request - Attempt to coalesce a request into a page list. | |
814 | * @desc: destination io descriptor | |
815 | * @req: request | |
816 | * | |
817 | * Returns true if the request 'req' was successfully coalesced into the | |
818 | * existing list of pages 'desc'. | |
819 | */ | |
d9156f9f | 820 | static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, |
8b09bee3 | 821 | struct nfs_page *req) |
bcb71bba TM |
822 | { |
823 | while (!nfs_pageio_do_add_request(desc, req)) { | |
b31268ac | 824 | desc->pg_moreio = 1; |
bcb71bba TM |
825 | nfs_pageio_doio(desc); |
826 | if (desc->pg_error < 0) | |
827 | return 0; | |
b31268ac | 828 | desc->pg_moreio = 0; |
d9156f9f TM |
829 | if (desc->pg_recoalesce) |
830 | return 0; | |
bcb71bba TM |
831 | } |
832 | return 1; | |
833 | } | |
834 | ||
d9156f9f TM |
835 | static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc) |
836 | { | |
837 | LIST_HEAD(head); | |
838 | ||
839 | do { | |
840 | list_splice_init(&desc->pg_list, &head); | |
841 | desc->pg_bytes_written -= desc->pg_count; | |
842 | desc->pg_count = 0; | |
843 | desc->pg_base = 0; | |
844 | desc->pg_recoalesce = 0; | |
845 | ||
846 | while (!list_empty(&head)) { | |
847 | struct nfs_page *req; | |
848 | ||
849 | req = list_first_entry(&head, struct nfs_page, wb_list); | |
850 | nfs_list_remove_request(req); | |
851 | if (__nfs_pageio_add_request(desc, req)) | |
852 | continue; | |
853 | if (desc->pg_error < 0) | |
854 | return 0; | |
855 | break; | |
856 | } | |
857 | } while (desc->pg_recoalesce); | |
858 | return 1; | |
859 | } | |
860 | ||
861 | int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc, | |
862 | struct nfs_page *req) | |
863 | { | |
864 | int ret; | |
865 | ||
866 | do { | |
867 | ret = __nfs_pageio_add_request(desc, req); | |
868 | if (ret) | |
869 | break; | |
870 | if (desc->pg_error < 0) | |
871 | break; | |
872 | ret = nfs_do_recoalesce(desc); | |
873 | } while (ret); | |
874 | return ret; | |
875 | } | |
89d77c8f | 876 | EXPORT_SYMBOL_GPL(nfs_pageio_add_request); |
d9156f9f | 877 | |
bcb71bba TM |
878 | /** |
879 | * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor | |
880 | * @desc: pointer to io descriptor | |
881 | */ | |
882 | void nfs_pageio_complete(struct nfs_pageio_descriptor *desc) | |
883 | { | |
d9156f9f TM |
884 | for (;;) { |
885 | nfs_pageio_doio(desc); | |
886 | if (!desc->pg_recoalesce) | |
887 | break; | |
888 | if (!nfs_do_recoalesce(desc)) | |
889 | break; | |
890 | } | |
bcb71bba | 891 | } |
89d77c8f | 892 | EXPORT_SYMBOL_GPL(nfs_pageio_complete); |
bcb71bba | 893 | |
7fe7f848 TM |
894 | /** |
895 | * nfs_pageio_cond_complete - Conditional I/O completion | |
896 | * @desc: pointer to io descriptor | |
897 | * @index: page index | |
898 | * | |
899 | * It is important to ensure that processes don't try to take locks | |
900 | * on non-contiguous ranges of pages as that might deadlock. This | |
901 | * function should be called before attempting to wait on a locked | |
902 | * nfs_page. It will complete the I/O if the page index 'index' | |
903 | * is not contiguous with the existing list of pages in 'desc'. | |
904 | */ | |
905 | void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index) | |
906 | { | |
907 | if (!list_empty(&desc->pg_list)) { | |
908 | struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev); | |
909 | if (index != prev->wb_index + 1) | |
d9156f9f | 910 | nfs_pageio_complete(desc); |
7fe7f848 TM |
911 | } |
912 | } | |
913 | ||
f7b422b1 | 914 | int __init nfs_init_nfspagecache(void) |
1da177e4 LT |
915 | { |
916 | nfs_page_cachep = kmem_cache_create("nfs_page", | |
917 | sizeof(struct nfs_page), | |
918 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 919 | NULL); |
1da177e4 LT |
920 | if (nfs_page_cachep == NULL) |
921 | return -ENOMEM; | |
922 | ||
923 | return 0; | |
924 | } | |
925 | ||
266bee88 | 926 | void nfs_destroy_nfspagecache(void) |
1da177e4 | 927 | { |
1a1d92c1 | 928 | kmem_cache_destroy(nfs_page_cachep); |
1da177e4 LT |
929 | } |
930 | ||
ef2c488c | 931 | static const struct rpc_call_ops nfs_pgio_common_ops = { |
6f92fa45 AS |
932 | .rpc_call_prepare = nfs_pgio_prepare, |
933 | .rpc_call_done = nfs_pgio_result, | |
934 | .rpc_release = nfs_pgio_release, | |
935 | }; |