Commit | Line | Data |
---|---|---|
155e7524 FI |
1 | /* |
2 | * linux/fs/nfs/blocklayout/blocklayout.c | |
3 | * | |
4 | * Module for the NFSv4.1 pNFS block layout driver. | |
5 | * | |
6 | * Copyright (c) 2006 The Regents of the University of Michigan. | |
7 | * All rights reserved. | |
8 | * | |
9 | * Andy Adamson <andros@citi.umich.edu> | |
10 | * Fred Isaman <iisaman@umich.edu> | |
11 | * | |
12 | * permission is granted to use, copy, create derivative works and | |
13 | * redistribute this software and such derivative works for any purpose, | |
14 | * so long as the name of the university of michigan is not used in | |
15 | * any advertising or publicity pertaining to the use or distribution | |
16 | * of this software without specific, written prior authorization. if | |
17 | * the above copyright notice or any other identification of the | |
18 | * university of michigan is included in any copy of any portion of | |
19 | * this software, then the disclaimer below must also be included. | |
20 | * | |
21 | * this software is provided as is, without representation from the | |
22 | * university of michigan as to its fitness for any purpose, and without | |
23 | * warranty by the university of michigan of any kind, either express | |
24 | * or implied, including without limitation the implied warranties of | |
25 | * merchantability and fitness for a particular purpose. the regents | |
26 | * of the university of michigan shall not be liable for any damages, | |
27 | * including special, indirect, incidental, or consequential damages, | |
28 | * with respect to any claim arising out or in connection with the use | |
29 | * of the software, even if it has been or is hereafter advised of the | |
30 | * possibility of such damages. | |
31 | */ | |
9549ec01 | 32 | |
155e7524 FI |
33 | #include <linux/module.h> |
34 | #include <linux/init.h> | |
fe0a9b74 JR |
35 | #include <linux/mount.h> |
36 | #include <linux/namei.h> | |
9549ec01 | 37 | #include <linux/bio.h> /* struct bio */ |
88c9e421 | 38 | #include <linux/prefetch.h> |
6296556f | 39 | #include <linux/pagevec.h> |
155e7524 | 40 | |
10bd295a | 41 | #include "../pnfs.h" |
76e697ba | 42 | #include "../nfs4session.h" |
10bd295a | 43 | #include "../internal.h" |
155e7524 FI |
44 | #include "blocklayout.h" |
45 | ||
46 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD | |
47 | ||
48 | MODULE_LICENSE("GPL"); | |
49 | MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>"); | |
50 | MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver"); | |
51 | ||
8067253c | 52 | static bool is_hole(struct pnfs_block_extent *be) |
9549ec01 | 53 | { |
8067253c CH |
54 | switch (be->be_state) { |
55 | case PNFS_BLOCK_NONE_DATA: | |
56 | return true; | |
57 | case PNFS_BLOCK_INVALID_DATA: | |
58 | return be->be_tag ? false : true; | |
59 | default: | |
60 | return false; | |
61 | } | |
650e2d39 FI |
62 | } |
63 | ||
9549ec01 FI |
64 | /* The data we are handed might be spread across several bios. We need |
65 | * to track when the last one is finished. | |
66 | */ | |
67 | struct parallel_io { | |
68 | struct kref refcnt; | |
8067253c | 69 | void (*pnfs_callback) (void *data); |
9549ec01 FI |
70 | void *data; |
71 | }; | |
72 | ||
73 | static inline struct parallel_io *alloc_parallel(void *data) | |
74 | { | |
75 | struct parallel_io *rv; | |
76 | ||
77 | rv = kmalloc(sizeof(*rv), GFP_NOFS); | |
78 | if (rv) { | |
79 | rv->data = data; | |
80 | kref_init(&rv->refcnt); | |
81 | } | |
82 | return rv; | |
83 | } | |
84 | ||
85 | static inline void get_parallel(struct parallel_io *p) | |
86 | { | |
87 | kref_get(&p->refcnt); | |
88 | } | |
89 | ||
90 | static void destroy_parallel(struct kref *kref) | |
91 | { | |
92 | struct parallel_io *p = container_of(kref, struct parallel_io, refcnt); | |
93 | ||
94 | dprintk("%s enter\n", __func__); | |
8067253c | 95 | p->pnfs_callback(p->data); |
9549ec01 FI |
96 | kfree(p); |
97 | } | |
98 | ||
99 | static inline void put_parallel(struct parallel_io *p) | |
100 | { | |
101 | kref_put(&p->refcnt, destroy_parallel); | |
102 | } | |
103 | ||
104 | static struct bio * | |
105 | bl_submit_bio(int rw, struct bio *bio) | |
106 | { | |
107 | if (bio) { | |
108 | get_parallel(bio->bi_private); | |
109 | dprintk("%s submitting %s bio %u@%llu\n", __func__, | |
4f024f37 KO |
110 | rw == READ ? "read" : "write", bio->bi_iter.bi_size, |
111 | (unsigned long long)bio->bi_iter.bi_sector); | |
9549ec01 FI |
112 | submit_bio(rw, bio); |
113 | } | |
114 | return NULL; | |
115 | } | |
116 | ||
117 | static struct bio *bl_alloc_init_bio(int npg, sector_t isect, | |
118 | struct pnfs_block_extent *be, | |
119 | void (*end_io)(struct bio *, int err), | |
120 | struct parallel_io *par) | |
121 | { | |
20d655d6 CH |
122 | struct pnfs_block_dev *dev = |
123 | container_of(be->be_device, struct pnfs_block_dev, d_node); | |
9549ec01 FI |
124 | struct bio *bio; |
125 | ||
74a6eeb4 | 126 | npg = min(npg, BIO_MAX_PAGES); |
9549ec01 | 127 | bio = bio_alloc(GFP_NOIO, npg); |
74a6eeb4 PT |
128 | if (!bio && (current->flags & PF_MEMALLOC)) { |
129 | while (!bio && (npg /= 2)) | |
130 | bio = bio_alloc(GFP_NOIO, npg); | |
131 | } | |
9549ec01 | 132 | |
74a6eeb4 | 133 | if (bio) { |
4f024f37 KO |
134 | bio->bi_iter.bi_sector = isect - be->be_f_offset + |
135 | be->be_v_offset; | |
20d655d6 | 136 | bio->bi_bdev = dev->d_bdev; |
74a6eeb4 PT |
137 | bio->bi_end_io = end_io; |
138 | bio->bi_private = par; | |
139 | } | |
9549ec01 FI |
140 | return bio; |
141 | } | |
142 | ||
fe6e1e8d | 143 | static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw, |
9549ec01 FI |
144 | sector_t isect, struct page *page, |
145 | struct pnfs_block_extent *be, | |
146 | void (*end_io)(struct bio *, int err), | |
fe6e1e8d PT |
147 | struct parallel_io *par, |
148 | unsigned int offset, int len) | |
9549ec01 | 149 | { |
fe6e1e8d PT |
150 | isect = isect + (offset >> SECTOR_SHIFT); |
151 | dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__, | |
152 | npg, rw, (unsigned long long)isect, offset, len); | |
9549ec01 FI |
153 | retry: |
154 | if (!bio) { | |
155 | bio = bl_alloc_init_bio(npg, isect, be, end_io, par); | |
156 | if (!bio) | |
157 | return ERR_PTR(-ENOMEM); | |
158 | } | |
fe6e1e8d | 159 | if (bio_add_page(bio, page, len, offset) < len) { |
9549ec01 FI |
160 | bio = bl_submit_bio(rw, bio); |
161 | goto retry; | |
162 | } | |
163 | return bio; | |
164 | } | |
165 | ||
9549ec01 FI |
166 | static void bl_end_io_read(struct bio *bio, int err) |
167 | { | |
168 | struct parallel_io *par = bio->bi_private; | |
9549ec01 | 169 | |
2c30c71b | 170 | if (err) { |
d45f60c6 | 171 | struct nfs_pgio_header *header = par->data; |
cd841605 FI |
172 | |
173 | if (!header->pnfs_error) | |
174 | header->pnfs_error = -EIO; | |
175 | pnfs_set_lo_fail(header->lseg); | |
9549ec01 | 176 | } |
8c792ea9 | 177 | |
9549ec01 FI |
178 | bio_put(bio); |
179 | put_parallel(par); | |
180 | } | |
181 | ||
182 | static void bl_read_cleanup(struct work_struct *work) | |
183 | { | |
184 | struct rpc_task *task; | |
d45f60c6 | 185 | struct nfs_pgio_header *hdr; |
9549ec01 FI |
186 | dprintk("%s enter\n", __func__); |
187 | task = container_of(work, struct rpc_task, u.tk_work); | |
d45f60c6 WAA |
188 | hdr = container_of(task, struct nfs_pgio_header, task); |
189 | pnfs_ld_read_done(hdr); | |
9549ec01 FI |
190 | } |
191 | ||
192 | static void | |
8067253c | 193 | bl_end_par_io_read(void *data) |
9549ec01 | 194 | { |
d45f60c6 | 195 | struct nfs_pgio_header *hdr = data; |
9549ec01 | 196 | |
d45f60c6 WAA |
197 | hdr->task.tk_status = hdr->pnfs_error; |
198 | INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup); | |
199 | schedule_work(&hdr->task.u.tk_work); | |
9549ec01 FI |
200 | } |
201 | ||
155e7524 | 202 | static enum pnfs_try_status |
8067253c | 203 | bl_read_pagelist(struct nfs_pgio_header *header) |
155e7524 | 204 | { |
8067253c | 205 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); |
9549ec01 | 206 | struct bio *bio = NULL; |
8067253c | 207 | struct pnfs_block_extent be; |
9549ec01 FI |
208 | sector_t isect, extent_length = 0; |
209 | struct parallel_io *par; | |
8067253c CH |
210 | loff_t f_offset = header->args.offset; |
211 | size_t bytes_left = header->args.count; | |
f742dc4a | 212 | unsigned int pg_offset, pg_len; |
8067253c CH |
213 | struct page **pages = header->args.pages; |
214 | int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT; | |
f742dc4a | 215 | const bool is_dio = (header->dreq != NULL); |
be98fd0a | 216 | struct blk_plug plug; |
8067253c | 217 | int i; |
9549ec01 | 218 | |
6f00866d | 219 | dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__, |
8067253c CH |
220 | header->page_array.npages, f_offset, |
221 | (unsigned int)header->args.count); | |
9549ec01 | 222 | |
8067253c | 223 | par = alloc_parallel(header); |
9549ec01 | 224 | if (!par) |
8067253c | 225 | return PNFS_NOT_ATTEMPTED; |
9549ec01 | 226 | par->pnfs_callback = bl_end_par_io_read; |
9549ec01 | 227 | |
be98fd0a CH |
228 | blk_start_plug(&plug); |
229 | ||
9549ec01 FI |
230 | isect = (sector_t) (f_offset >> SECTOR_SHIFT); |
231 | /* Code assumes extents are page-aligned */ | |
8067253c | 232 | for (i = pg_index; i < header->page_array.npages; i++) { |
921b81a8 | 233 | if (extent_length <= 0) { |
9549ec01 | 234 | /* We've used up the previous extent */ |
9549ec01 | 235 | bio = bl_submit_bio(READ, bio); |
8067253c | 236 | |
9549ec01 | 237 | /* Get the next one */ |
8067253c | 238 | if (!ext_tree_lookup(bl, isect, &be, false)) { |
cd841605 | 239 | header->pnfs_error = -EIO; |
9549ec01 FI |
240 | goto out; |
241 | } | |
8067253c | 242 | extent_length = be.be_length - (isect - be.be_f_offset); |
9549ec01 | 243 | } |
f742dc4a | 244 | |
3a6fd1f0 | 245 | pg_offset = f_offset & ~PAGE_CACHE_MASK; |
f742dc4a | 246 | if (is_dio) { |
f742dc4a PT |
247 | if (pg_offset + bytes_left > PAGE_CACHE_SIZE) |
248 | pg_len = PAGE_CACHE_SIZE - pg_offset; | |
249 | else | |
250 | pg_len = bytes_left; | |
251 | ||
252 | f_offset += pg_len; | |
253 | bytes_left -= pg_len; | |
254 | isect += (pg_offset >> SECTOR_SHIFT); | |
921b81a8 | 255 | extent_length -= (pg_offset >> SECTOR_SHIFT); |
f742dc4a | 256 | } else { |
3a6fd1f0 | 257 | BUG_ON(pg_offset != 0); |
f742dc4a PT |
258 | pg_len = PAGE_CACHE_SIZE; |
259 | } | |
260 | ||
8067253c | 261 | if (is_hole(&be)) { |
9549ec01 FI |
262 | bio = bl_submit_bio(READ, bio); |
263 | /* Fill hole w/ zeroes w/o accessing device */ | |
264 | dprintk("%s Zeroing page for hole\n", __func__); | |
f742dc4a | 265 | zero_user_segment(pages[i], pg_offset, pg_len); |
9549ec01 | 266 | } else { |
823b0c9d | 267 | bio = do_add_page_to_bio(bio, |
8067253c | 268 | header->page_array.npages - i, |
30dd374f | 269 | READ, |
8067253c | 270 | isect, pages[i], &be, |
f742dc4a PT |
271 | bl_end_io_read, par, |
272 | pg_offset, pg_len); | |
9549ec01 | 273 | if (IS_ERR(bio)) { |
cd841605 | 274 | header->pnfs_error = PTR_ERR(bio); |
e6d05a75 | 275 | bio = NULL; |
9549ec01 FI |
276 | goto out; |
277 | } | |
278 | } | |
f742dc4a | 279 | isect += (pg_len >> SECTOR_SHIFT); |
921b81a8 | 280 | extent_length -= (pg_len >> SECTOR_SHIFT); |
9549ec01 | 281 | } |
cd841605 | 282 | if ((isect << SECTOR_SHIFT) >= header->inode->i_size) { |
8067253c CH |
283 | header->res.eof = 1; |
284 | header->res.count = header->inode->i_size - header->args.offset; | |
9549ec01 | 285 | } else { |
8067253c | 286 | header->res.count = (isect << SECTOR_SHIFT) - header->args.offset; |
9549ec01 FI |
287 | } |
288 | out: | |
9549ec01 | 289 | bl_submit_bio(READ, bio); |
be98fd0a | 290 | blk_finish_plug(&plug); |
9549ec01 FI |
291 | put_parallel(par); |
292 | return PNFS_ATTEMPTED; | |
31e6306a FI |
293 | } |
294 | ||
650e2d39 FI |
295 | static void bl_end_io_write(struct bio *bio, int err) |
296 | { | |
297 | struct parallel_io *par = bio->bi_private; | |
298 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d45f60c6 | 299 | struct nfs_pgio_header *header = par->data; |
650e2d39 FI |
300 | |
301 | if (!uptodate) { | |
cd841605 FI |
302 | if (!header->pnfs_error) |
303 | header->pnfs_error = -EIO; | |
304 | pnfs_set_lo_fail(header->lseg); | |
650e2d39 FI |
305 | } |
306 | bio_put(bio); | |
307 | put_parallel(par); | |
308 | } | |
309 | ||
310 | /* Function scheduled for call during bl_end_par_io_write, | |
311 | * it marks sectors as written and extends the commitlist. | |
312 | */ | |
313 | static void bl_write_cleanup(struct work_struct *work) | |
314 | { | |
8067253c CH |
315 | struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work); |
316 | struct nfs_pgio_header *hdr = | |
317 | container_of(task, struct nfs_pgio_header, task); | |
318 | ||
650e2d39 | 319 | dprintk("%s enter\n", __func__); |
8067253c | 320 | |
d45f60c6 | 321 | if (likely(!hdr->pnfs_error)) { |
8067253c CH |
322 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg); |
323 | u64 start = hdr->args.offset & (loff_t)PAGE_CACHE_MASK; | |
324 | u64 end = (hdr->args.offset + hdr->args.count + | |
325 | PAGE_CACHE_SIZE - 1) & (loff_t)PAGE_CACHE_MASK; | |
326 | ||
327 | ext_tree_mark_written(bl, start >> SECTOR_SHIFT, | |
328 | (end - start) >> SECTOR_SHIFT); | |
31e6306a | 329 | } |
8067253c | 330 | |
d45f60c6 | 331 | pnfs_ld_write_done(hdr); |
650e2d39 FI |
332 | } |
333 | ||
334 | /* Called when last of bios associated with a bl_write_pagelist call finishes */ | |
8067253c | 335 | static void bl_end_par_io_write(void *data) |
650e2d39 | 336 | { |
d45f60c6 | 337 | struct nfs_pgio_header *hdr = data; |
650e2d39 | 338 | |
d45f60c6 | 339 | hdr->task.tk_status = hdr->pnfs_error; |
c65e6254 | 340 | hdr->verf.committed = NFS_FILE_SYNC; |
d45f60c6 WAA |
341 | INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup); |
342 | schedule_work(&hdr->task.u.tk_work); | |
650e2d39 FI |
343 | } |
344 | ||
155e7524 | 345 | static enum pnfs_try_status |
d45f60c6 | 346 | bl_write_pagelist(struct nfs_pgio_header *header, int sync) |
155e7524 | 347 | { |
8067253c | 348 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); |
650e2d39 | 349 | struct bio *bio = NULL; |
8067253c | 350 | struct pnfs_block_extent be; |
3a6fd1f0 | 351 | sector_t isect, extent_length = 0; |
96c9eae6 | 352 | struct parallel_io *par = NULL; |
d45f60c6 WAA |
353 | loff_t offset = header->args.offset; |
354 | size_t count = header->args.count; | |
d45f60c6 | 355 | struct page **pages = header->args.pages; |
3a6fd1f0 | 356 | int pg_index = pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT; |
be98fd0a | 357 | struct blk_plug plug; |
8067253c | 358 | int i; |
650e2d39 FI |
359 | |
360 | dprintk("%s enter, %Zu@%lld\n", __func__, count, offset); | |
96c9eae6 | 361 | |
d45f60c6 | 362 | /* At this point, header->page_aray is a (sequential) list of nfs_pages. |
71cdd40f PT |
363 | * We want to write each, and if there is an error set pnfs_error |
364 | * to have it redone using nfs. | |
650e2d39 | 365 | */ |
d45f60c6 | 366 | par = alloc_parallel(header); |
650e2d39 | 367 | if (!par) |
8067253c | 368 | return PNFS_NOT_ATTEMPTED; |
650e2d39 | 369 | par->pnfs_callback = bl_end_par_io_write; |
650e2d39 | 370 | |
3a6fd1f0 | 371 | blk_start_plug(&plug); |
71cdd40f | 372 | |
3a6fd1f0 CH |
373 | /* we always write out the whole page */ |
374 | offset = offset & (loff_t)PAGE_CACHE_MASK; | |
375 | isect = offset >> SECTOR_SHIFT; | |
71cdd40f | 376 | |
d45f60c6 | 377 | for (i = pg_index; i < header->page_array.npages; i++) { |
921b81a8 | 378 | if (extent_length <= 0) { |
650e2d39 | 379 | /* We've used up the previous extent */ |
650e2d39 FI |
380 | bio = bl_submit_bio(WRITE, bio); |
381 | /* Get the next one */ | |
8067253c | 382 | if (!ext_tree_lookup(bl, isect, &be, true)) { |
cd841605 | 383 | header->pnfs_error = -EINVAL; |
650e2d39 FI |
384 | goto out; |
385 | } | |
fe6e1e8d | 386 | |
8067253c | 387 | extent_length = be.be_length - (isect - be.be_f_offset); |
71cdd40f | 388 | } |
fe6e1e8d | 389 | |
d45f60c6 | 390 | bio = do_add_page_to_bio(bio, header->page_array.npages - i, |
8067253c | 391 | WRITE, isect, pages[i], &be, |
fe6e1e8d | 392 | bl_end_io_write, par, |
3a6fd1f0 | 393 | 0, PAGE_CACHE_SIZE); |
71cdd40f | 394 | if (IS_ERR(bio)) { |
cd841605 | 395 | header->pnfs_error = PTR_ERR(bio); |
e6d05a75 | 396 | bio = NULL; |
71cdd40f | 397 | goto out; |
650e2d39 | 398 | } |
3a6fd1f0 CH |
399 | offset += PAGE_CACHE_SIZE; |
400 | count -= PAGE_CACHE_SIZE; | |
650e2d39 FI |
401 | isect += PAGE_CACHE_SECTORS; |
402 | extent_length -= PAGE_CACHE_SECTORS; | |
403 | } | |
71cdd40f | 404 | |
d45f60c6 | 405 | header->res.count = header->args.count; |
650e2d39 | 406 | out: |
650e2d39 | 407 | bl_submit_bio(WRITE, bio); |
be98fd0a | 408 | blk_finish_plug(&plug); |
650e2d39 FI |
409 | put_parallel(par); |
410 | return PNFS_ATTEMPTED; | |
155e7524 FI |
411 | } |
412 | ||
413 | static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
414 | { | |
415 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); | |
8067253c | 416 | int err; |
155e7524 FI |
417 | |
418 | dprintk("%s enter\n", __func__); | |
8067253c CH |
419 | |
420 | err = ext_tree_remove(bl, true, 0, LLONG_MAX); | |
421 | WARN_ON(err); | |
422 | ||
155e7524 FI |
423 | kfree(bl); |
424 | } | |
425 | ||
426 | static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode, | |
427 | gfp_t gfp_flags) | |
428 | { | |
429 | struct pnfs_block_layout *bl; | |
430 | ||
431 | dprintk("%s enter\n", __func__); | |
432 | bl = kzalloc(sizeof(*bl), gfp_flags); | |
433 | if (!bl) | |
434 | return NULL; | |
8067253c CH |
435 | |
436 | bl->bl_ext_rw = RB_ROOT; | |
437 | bl->bl_ext_ro = RB_ROOT; | |
155e7524 | 438 | spin_lock_init(&bl->bl_ext_lock); |
8067253c | 439 | |
155e7524 FI |
440 | return &bl->bl_layout; |
441 | } | |
442 | ||
a60d2ebd | 443 | static void bl_free_lseg(struct pnfs_layout_segment *lseg) |
155e7524 | 444 | { |
a60d2ebd FI |
445 | dprintk("%s enter\n", __func__); |
446 | kfree(lseg); | |
155e7524 FI |
447 | } |
448 | ||
a60d2ebd FI |
449 | /* We pretty much ignore lseg, and store all data layout wide, so we |
450 | * can correctly merge. | |
451 | */ | |
452 | static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo, | |
453 | struct nfs4_layoutget_res *lgr, | |
454 | gfp_t gfp_flags) | |
155e7524 | 455 | { |
a60d2ebd FI |
456 | struct pnfs_layout_segment *lseg; |
457 | int status; | |
458 | ||
459 | dprintk("%s enter\n", __func__); | |
460 | lseg = kzalloc(sizeof(*lseg), gfp_flags); | |
461 | if (!lseg) | |
462 | return ERR_PTR(-ENOMEM); | |
463 | status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags); | |
464 | if (status) { | |
465 | /* We don't want to call the full-blown bl_free_lseg, | |
466 | * since on error extents were not touched. | |
467 | */ | |
468 | kfree(lseg); | |
469 | return ERR_PTR(status); | |
470 | } | |
471 | return lseg; | |
155e7524 FI |
472 | } |
473 | ||
71d5b763 CH |
474 | static void |
475 | bl_return_range(struct pnfs_layout_hdr *lo, | |
476 | struct pnfs_layout_range *range) | |
477 | { | |
478 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); | |
479 | sector_t offset = range->offset >> SECTOR_SHIFT, end; | |
480 | int err; | |
481 | ||
482 | if (range->offset % 8) { | |
483 | dprintk("%s: offset %lld not block size aligned\n", | |
484 | __func__, range->offset); | |
485 | return; | |
486 | } | |
487 | ||
488 | if (range->length != NFS4_MAX_UINT64) { | |
489 | if (range->length % 8) { | |
490 | dprintk("%s: length %lld not block size aligned\n", | |
491 | __func__, range->length); | |
492 | return; | |
493 | } | |
494 | ||
495 | end = offset + (range->length >> SECTOR_SHIFT); | |
496 | } else { | |
497 | end = round_down(NFS4_MAX_UINT64, PAGE_SIZE); | |
498 | } | |
499 | ||
500 | err = ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end); | |
501 | } | |
502 | ||
34dc93c2 CH |
503 | static int |
504 | bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg) | |
155e7524 | 505 | { |
34dc93c2 | 506 | return ext_tree_prepare_commit(arg); |
155e7524 FI |
507 | } |
508 | ||
509 | static void | |
510 | bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata) | |
511 | { | |
34dc93c2 | 512 | ext_tree_mark_committed(&lcdata->args, lcdata->res.status); |
155e7524 FI |
513 | } |
514 | ||
515 | static int | |
516 | bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh) | |
517 | { | |
518 | dprintk("%s enter\n", __func__); | |
2f9fd182 FI |
519 | |
520 | if (server->pnfs_blksize == 0) { | |
521 | dprintk("%s Server did not return blksize\n", __func__); | |
522 | return -EINVAL; | |
523 | } | |
e3aaf7f2 CH |
524 | if (server->pnfs_blksize > PAGE_SIZE) { |
525 | printk(KERN_ERR "%s: pNFS blksize %d not supported.\n", | |
526 | __func__, server->pnfs_blksize); | |
527 | return -EINVAL; | |
528 | } | |
529 | ||
d4b18c3e | 530 | return 0; |
155e7524 FI |
531 | } |
532 | ||
f742dc4a | 533 | static bool |
3a6fd1f0 CH |
534 | is_aligned_req(struct nfs_pageio_descriptor *pgio, |
535 | struct nfs_page *req, unsigned int alignment) | |
f742dc4a | 536 | { |
3a6fd1f0 CH |
537 | /* |
538 | * Always accept buffered writes, higher layers take care of the | |
539 | * right alignment. | |
540 | */ | |
541 | if (pgio->pg_dreq == NULL) | |
542 | return true; | |
543 | ||
544 | if (!IS_ALIGNED(req->wb_offset, alignment)) | |
545 | return false; | |
546 | ||
547 | if (IS_ALIGNED(req->wb_bytes, alignment)) | |
548 | return true; | |
549 | ||
550 | if (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode)) { | |
551 | /* | |
552 | * If the write goes up to the inode size, just write | |
553 | * the full page. Data past the inode size is | |
554 | * guaranteed to be zeroed by the higher level client | |
555 | * code, and this behaviour is mandated by RFC 5663 | |
556 | * section 2.3.2. | |
557 | */ | |
558 | return true; | |
559 | } | |
560 | ||
561 | return false; | |
f742dc4a PT |
562 | } |
563 | ||
564 | static void | |
565 | bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
566 | { | |
3a6fd1f0 | 567 | if (!is_aligned_req(pgio, req, SECTOR_SIZE)) { |
f742dc4a | 568 | nfs_pageio_reset_read_mds(pgio); |
3a6fd1f0 CH |
569 | return; |
570 | } | |
571 | ||
572 | pnfs_generic_pg_init_read(pgio, req); | |
f742dc4a PT |
573 | } |
574 | ||
b4fdac1a WAA |
575 | /* |
576 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
577 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
578 | */ | |
579 | static size_t | |
f742dc4a PT |
580 | bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
581 | struct nfs_page *req) | |
582 | { | |
3a6fd1f0 | 583 | if (!is_aligned_req(pgio, req, SECTOR_SIZE)) |
b4fdac1a | 584 | return 0; |
f742dc4a PT |
585 | return pnfs_generic_pg_test(pgio, prev, req); |
586 | } | |
587 | ||
6296556f PT |
588 | /* |
589 | * Return the number of contiguous bytes for a given inode | |
590 | * starting at page frame idx. | |
591 | */ | |
592 | static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx) | |
593 | { | |
594 | struct address_space *mapping = inode->i_mapping; | |
595 | pgoff_t end; | |
596 | ||
597 | /* Optimize common case that writes from 0 to end of file */ | |
598 | end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE); | |
599 | if (end != NFS_I(inode)->npages) { | |
600 | rcu_read_lock(); | |
e7b563bb | 601 | end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX); |
6296556f PT |
602 | rcu_read_unlock(); |
603 | } | |
604 | ||
605 | if (!end) | |
606 | return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT); | |
607 | else | |
608 | return (end - idx) << PAGE_CACHE_SHIFT; | |
609 | } | |
610 | ||
6f018efa | 611 | static void |
96c9eae6 PT |
612 | bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) |
613 | { | |
3a6fd1f0 CH |
614 | u64 wb_size; |
615 | ||
616 | if (!is_aligned_req(pgio, req, PAGE_SIZE)) { | |
96c9eae6 | 617 | nfs_pageio_reset_write_mds(pgio); |
3a6fd1f0 | 618 | return; |
6296556f | 619 | } |
3a6fd1f0 CH |
620 | |
621 | if (pgio->pg_dreq == NULL) | |
622 | wb_size = pnfs_num_cont_bytes(pgio->pg_inode, | |
623 | req->wb_index); | |
624 | else | |
625 | wb_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
626 | ||
627 | pnfs_generic_pg_init_write(pgio, req, wb_size); | |
96c9eae6 PT |
628 | } |
629 | ||
b4fdac1a WAA |
630 | /* |
631 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
632 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
633 | */ | |
634 | static size_t | |
96c9eae6 PT |
635 | bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
636 | struct nfs_page *req) | |
637 | { | |
3a6fd1f0 | 638 | if (!is_aligned_req(pgio, req, PAGE_SIZE)) |
b4fdac1a | 639 | return 0; |
96c9eae6 PT |
640 | return pnfs_generic_pg_test(pgio, prev, req); |
641 | } | |
642 | ||
e9643fe8 | 643 | static const struct nfs_pageio_ops bl_pg_read_ops = { |
f742dc4a PT |
644 | .pg_init = bl_pg_init_read, |
645 | .pg_test = bl_pg_test_read, | |
e9643fe8 BH |
646 | .pg_doio = pnfs_generic_pg_readpages, |
647 | }; | |
648 | ||
649 | static const struct nfs_pageio_ops bl_pg_write_ops = { | |
96c9eae6 PT |
650 | .pg_init = bl_pg_init_write, |
651 | .pg_test = bl_pg_test_write, | |
e9643fe8 BH |
652 | .pg_doio = pnfs_generic_pg_writepages, |
653 | }; | |
654 | ||
155e7524 FI |
655 | static struct pnfs_layoutdriver_type blocklayout_type = { |
656 | .id = LAYOUT_BLOCK_VOLUME, | |
657 | .name = "LAYOUT_BLOCK_VOLUME", | |
5a12cca6 | 658 | .owner = THIS_MODULE, |
848746bd CH |
659 | .flags = PNFS_LAYOUTRET_ON_SETATTR | |
660 | PNFS_READ_WHOLE_PAGE, | |
155e7524 FI |
661 | .read_pagelist = bl_read_pagelist, |
662 | .write_pagelist = bl_write_pagelist, | |
663 | .alloc_layout_hdr = bl_alloc_layout_hdr, | |
664 | .free_layout_hdr = bl_free_layout_hdr, | |
665 | .alloc_lseg = bl_alloc_lseg, | |
666 | .free_lseg = bl_free_lseg, | |
71d5b763 | 667 | .return_range = bl_return_range, |
34dc93c2 | 668 | .prepare_layoutcommit = bl_prepare_layoutcommit, |
155e7524 FI |
669 | .cleanup_layoutcommit = bl_cleanup_layoutcommit, |
670 | .set_layoutdriver = bl_set_layoutdriver, | |
20d655d6 CH |
671 | .alloc_deviceid_node = bl_alloc_deviceid_node, |
672 | .free_deviceid_node = bl_free_deviceid_node, | |
e9643fe8 BH |
673 | .pg_read_ops = &bl_pg_read_ops, |
674 | .pg_write_ops = &bl_pg_write_ops, | |
155e7524 FI |
675 | }; |
676 | ||
fe0a9b74 | 677 | static const struct rpc_pipe_ops bl_upcall_ops = { |
c1225158 | 678 | .upcall = rpc_pipe_generic_upcall, |
fe0a9b74 JR |
679 | .downcall = bl_pipe_downcall, |
680 | .destroy_msg = bl_pipe_destroy_msg, | |
681 | }; | |
682 | ||
332dfab6 SK |
683 | static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb, |
684 | struct rpc_pipe *pipe) | |
685 | { | |
686 | struct dentry *dir, *dentry; | |
687 | ||
688 | dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME); | |
689 | if (dir == NULL) | |
690 | return ERR_PTR(-ENOENT); | |
691 | dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe); | |
692 | dput(dir); | |
693 | return dentry; | |
694 | } | |
695 | ||
696 | static void nfs4blocklayout_unregister_sb(struct super_block *sb, | |
697 | struct rpc_pipe *pipe) | |
698 | { | |
699 | if (pipe->dentry) | |
700 | rpc_unlink(pipe->dentry); | |
701 | } | |
702 | ||
627f3066 SK |
703 | static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event, |
704 | void *ptr) | |
705 | { | |
706 | struct super_block *sb = ptr; | |
707 | struct net *net = sb->s_fs_info; | |
708 | struct nfs_net *nn = net_generic(net, nfs_net_id); | |
709 | struct dentry *dentry; | |
710 | int ret = 0; | |
711 | ||
712 | if (!try_module_get(THIS_MODULE)) | |
713 | return 0; | |
714 | ||
715 | if (nn->bl_device_pipe == NULL) { | |
716 | module_put(THIS_MODULE); | |
717 | return 0; | |
718 | } | |
719 | ||
720 | switch (event) { | |
721 | case RPC_PIPEFS_MOUNT: | |
722 | dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe); | |
723 | if (IS_ERR(dentry)) { | |
724 | ret = PTR_ERR(dentry); | |
725 | break; | |
726 | } | |
727 | nn->bl_device_pipe->dentry = dentry; | |
728 | break; | |
729 | case RPC_PIPEFS_UMOUNT: | |
730 | if (nn->bl_device_pipe->dentry) | |
731 | nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe); | |
732 | break; | |
733 | default: | |
734 | ret = -ENOTSUPP; | |
735 | break; | |
736 | } | |
737 | module_put(THIS_MODULE); | |
738 | return ret; | |
739 | } | |
740 | ||
741 | static struct notifier_block nfs4blocklayout_block = { | |
742 | .notifier_call = rpc_pipefs_event, | |
743 | }; | |
744 | ||
332dfab6 SK |
745 | static struct dentry *nfs4blocklayout_register_net(struct net *net, |
746 | struct rpc_pipe *pipe) | |
747 | { | |
748 | struct super_block *pipefs_sb; | |
749 | struct dentry *dentry; | |
750 | ||
751 | pipefs_sb = rpc_get_sb_net(net); | |
752 | if (!pipefs_sb) | |
2561d618 | 753 | return NULL; |
332dfab6 SK |
754 | dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe); |
755 | rpc_put_sb_net(net); | |
756 | return dentry; | |
757 | } | |
758 | ||
759 | static void nfs4blocklayout_unregister_net(struct net *net, | |
760 | struct rpc_pipe *pipe) | |
761 | { | |
762 | struct super_block *pipefs_sb; | |
763 | ||
764 | pipefs_sb = rpc_get_sb_net(net); | |
765 | if (pipefs_sb) { | |
766 | nfs4blocklayout_unregister_sb(pipefs_sb, pipe); | |
767 | rpc_put_sb_net(net); | |
768 | } | |
769 | } | |
770 | ||
9e2e74db SK |
771 | static int nfs4blocklayout_net_init(struct net *net) |
772 | { | |
773 | struct nfs_net *nn = net_generic(net, nfs_net_id); | |
774 | struct dentry *dentry; | |
775 | ||
5ffaf855 | 776 | init_waitqueue_head(&nn->bl_wq); |
9e2e74db SK |
777 | nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0); |
778 | if (IS_ERR(nn->bl_device_pipe)) | |
779 | return PTR_ERR(nn->bl_device_pipe); | |
780 | dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe); | |
781 | if (IS_ERR(dentry)) { | |
782 | rpc_destroy_pipe_data(nn->bl_device_pipe); | |
783 | return PTR_ERR(dentry); | |
784 | } | |
785 | nn->bl_device_pipe->dentry = dentry; | |
786 | return 0; | |
787 | } | |
788 | ||
789 | static void nfs4blocklayout_net_exit(struct net *net) | |
790 | { | |
791 | struct nfs_net *nn = net_generic(net, nfs_net_id); | |
792 | ||
793 | nfs4blocklayout_unregister_net(net, nn->bl_device_pipe); | |
794 | rpc_destroy_pipe_data(nn->bl_device_pipe); | |
795 | nn->bl_device_pipe = NULL; | |
796 | } | |
797 | ||
798 | static struct pernet_operations nfs4blocklayout_net_ops = { | |
799 | .init = nfs4blocklayout_net_init, | |
800 | .exit = nfs4blocklayout_net_exit, | |
801 | }; | |
802 | ||
155e7524 FI |
803 | static int __init nfs4blocklayout_init(void) |
804 | { | |
805 | int ret; | |
806 | ||
807 | dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__); | |
808 | ||
809 | ret = pnfs_register_layoutdriver(&blocklayout_type); | |
fe0a9b74 JR |
810 | if (ret) |
811 | goto out; | |
812 | ||
627f3066 | 813 | ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block); |
9e2e74db SK |
814 | if (ret) |
815 | goto out_remove; | |
627f3066 SK |
816 | ret = register_pernet_subsys(&nfs4blocklayout_net_ops); |
817 | if (ret) | |
818 | goto out_notifier; | |
fe0a9b74 JR |
819 | out: |
820 | return ret; | |
821 | ||
627f3066 SK |
822 | out_notifier: |
823 | rpc_pipefs_notifier_unregister(&nfs4blocklayout_block); | |
fe0a9b74 JR |
824 | out_remove: |
825 | pnfs_unregister_layoutdriver(&blocklayout_type); | |
155e7524 FI |
826 | return ret; |
827 | } | |
828 | ||
829 | static void __exit nfs4blocklayout_exit(void) | |
830 | { | |
831 | dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n", | |
832 | __func__); | |
833 | ||
627f3066 | 834 | rpc_pipefs_notifier_unregister(&nfs4blocklayout_block); |
9e2e74db | 835 | unregister_pernet_subsys(&nfs4blocklayout_net_ops); |
155e7524 FI |
836 | pnfs_unregister_layoutdriver(&blocklayout_type); |
837 | } | |
838 | ||
839 | MODULE_ALIAS("nfs-layouttype4-3"); | |
840 | ||
841 | module_init(nfs4blocklayout_init); | |
842 | module_exit(nfs4blocklayout_exit); |