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 | ||
5c83746a CH |
117 | static struct bio * |
118 | bl_alloc_init_bio(int npg, struct block_device *bdev, sector_t disk_sector, | |
119 | void (*end_io)(struct bio *, int err), struct parallel_io *par) | |
9549ec01 FI |
120 | { |
121 | struct bio *bio; | |
122 | ||
74a6eeb4 | 123 | npg = min(npg, BIO_MAX_PAGES); |
9549ec01 | 124 | bio = bio_alloc(GFP_NOIO, npg); |
74a6eeb4 PT |
125 | if (!bio && (current->flags & PF_MEMALLOC)) { |
126 | while (!bio && (npg /= 2)) | |
127 | bio = bio_alloc(GFP_NOIO, npg); | |
128 | } | |
9549ec01 | 129 | |
74a6eeb4 | 130 | if (bio) { |
5c83746a CH |
131 | bio->bi_iter.bi_sector = disk_sector; |
132 | bio->bi_bdev = bdev; | |
74a6eeb4 PT |
133 | bio->bi_end_io = end_io; |
134 | bio->bi_private = par; | |
135 | } | |
9549ec01 FI |
136 | return bio; |
137 | } | |
138 | ||
5c83746a CH |
139 | static struct bio * |
140 | do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect, | |
141 | struct page *page, struct pnfs_block_dev_map *map, | |
142 | struct pnfs_block_extent *be, | |
143 | void (*end_io)(struct bio *, int err), | |
144 | struct parallel_io *par, unsigned int offset, int *len) | |
9549ec01 | 145 | { |
5c83746a CH |
146 | struct pnfs_block_dev *dev = |
147 | container_of(be->be_device, struct pnfs_block_dev, node); | |
148 | u64 disk_addr, end; | |
149 | ||
fe6e1e8d | 150 | dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__, |
5c83746a CH |
151 | npg, rw, (unsigned long long)isect, offset, *len); |
152 | ||
153 | /* translate to device offset */ | |
154 | isect += be->be_v_offset; | |
155 | isect -= be->be_f_offset; | |
156 | ||
157 | /* translate to physical disk offset */ | |
158 | disk_addr = (u64)isect << SECTOR_SHIFT; | |
159 | if (disk_addr < map->start || disk_addr >= map->start + map->len) { | |
160 | if (!dev->map(dev, disk_addr, map)) | |
161 | return ERR_PTR(-EIO); | |
162 | bio = bl_submit_bio(rw, bio); | |
163 | } | |
164 | disk_addr += map->disk_offset; | |
165 | disk_addr -= map->start; | |
166 | ||
167 | /* limit length to what the device mapping allows */ | |
168 | end = disk_addr + *len; | |
169 | if (end >= map->start + map->len) | |
170 | *len = map->start + map->len - disk_addr; | |
171 | ||
9549ec01 FI |
172 | retry: |
173 | if (!bio) { | |
5c83746a CH |
174 | bio = bl_alloc_init_bio(npg, map->bdev, |
175 | disk_addr >> SECTOR_SHIFT, end_io, par); | |
9549ec01 FI |
176 | if (!bio) |
177 | return ERR_PTR(-ENOMEM); | |
178 | } | |
5c83746a | 179 | if (bio_add_page(bio, page, *len, offset) < *len) { |
9549ec01 FI |
180 | bio = bl_submit_bio(rw, bio); |
181 | goto retry; | |
182 | } | |
183 | return bio; | |
184 | } | |
185 | ||
9549ec01 FI |
186 | static void bl_end_io_read(struct bio *bio, int err) |
187 | { | |
188 | struct parallel_io *par = bio->bi_private; | |
9549ec01 | 189 | |
2c30c71b | 190 | if (err) { |
d45f60c6 | 191 | struct nfs_pgio_header *header = par->data; |
cd841605 FI |
192 | |
193 | if (!header->pnfs_error) | |
194 | header->pnfs_error = -EIO; | |
195 | pnfs_set_lo_fail(header->lseg); | |
9549ec01 | 196 | } |
8c792ea9 | 197 | |
9549ec01 FI |
198 | bio_put(bio); |
199 | put_parallel(par); | |
200 | } | |
201 | ||
202 | static void bl_read_cleanup(struct work_struct *work) | |
203 | { | |
204 | struct rpc_task *task; | |
d45f60c6 | 205 | struct nfs_pgio_header *hdr; |
9549ec01 FI |
206 | dprintk("%s enter\n", __func__); |
207 | task = container_of(work, struct rpc_task, u.tk_work); | |
d45f60c6 WAA |
208 | hdr = container_of(task, struct nfs_pgio_header, task); |
209 | pnfs_ld_read_done(hdr); | |
9549ec01 FI |
210 | } |
211 | ||
212 | static void | |
8067253c | 213 | bl_end_par_io_read(void *data) |
9549ec01 | 214 | { |
d45f60c6 | 215 | struct nfs_pgio_header *hdr = data; |
9549ec01 | 216 | |
d45f60c6 WAA |
217 | hdr->task.tk_status = hdr->pnfs_error; |
218 | INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup); | |
219 | schedule_work(&hdr->task.u.tk_work); | |
9549ec01 FI |
220 | } |
221 | ||
155e7524 | 222 | static enum pnfs_try_status |
8067253c | 223 | bl_read_pagelist(struct nfs_pgio_header *header) |
155e7524 | 224 | { |
8067253c | 225 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); |
5c83746a | 226 | struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 }; |
9549ec01 | 227 | struct bio *bio = NULL; |
8067253c | 228 | struct pnfs_block_extent be; |
9549ec01 FI |
229 | sector_t isect, extent_length = 0; |
230 | struct parallel_io *par; | |
8067253c CH |
231 | loff_t f_offset = header->args.offset; |
232 | size_t bytes_left = header->args.count; | |
f742dc4a | 233 | unsigned int pg_offset, pg_len; |
8067253c CH |
234 | struct page **pages = header->args.pages; |
235 | int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT; | |
f742dc4a | 236 | const bool is_dio = (header->dreq != NULL); |
be98fd0a | 237 | struct blk_plug plug; |
8067253c | 238 | int i; |
9549ec01 | 239 | |
6f00866d | 240 | dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__, |
8067253c CH |
241 | header->page_array.npages, f_offset, |
242 | (unsigned int)header->args.count); | |
9549ec01 | 243 | |
8067253c | 244 | par = alloc_parallel(header); |
9549ec01 | 245 | if (!par) |
8067253c | 246 | return PNFS_NOT_ATTEMPTED; |
9549ec01 | 247 | par->pnfs_callback = bl_end_par_io_read; |
9549ec01 | 248 | |
be98fd0a CH |
249 | blk_start_plug(&plug); |
250 | ||
9549ec01 FI |
251 | isect = (sector_t) (f_offset >> SECTOR_SHIFT); |
252 | /* Code assumes extents are page-aligned */ | |
8067253c | 253 | for (i = pg_index; i < header->page_array.npages; i++) { |
921b81a8 | 254 | if (extent_length <= 0) { |
9549ec01 | 255 | /* We've used up the previous extent */ |
9549ec01 | 256 | bio = bl_submit_bio(READ, bio); |
8067253c | 257 | |
9549ec01 | 258 | /* Get the next one */ |
8067253c | 259 | if (!ext_tree_lookup(bl, isect, &be, false)) { |
cd841605 | 260 | header->pnfs_error = -EIO; |
9549ec01 FI |
261 | goto out; |
262 | } | |
8067253c | 263 | extent_length = be.be_length - (isect - be.be_f_offset); |
9549ec01 | 264 | } |
f742dc4a | 265 | |
3a6fd1f0 | 266 | pg_offset = f_offset & ~PAGE_CACHE_MASK; |
f742dc4a | 267 | if (is_dio) { |
f742dc4a PT |
268 | if (pg_offset + bytes_left > PAGE_CACHE_SIZE) |
269 | pg_len = PAGE_CACHE_SIZE - pg_offset; | |
270 | else | |
271 | pg_len = bytes_left; | |
f742dc4a | 272 | } else { |
3a6fd1f0 | 273 | BUG_ON(pg_offset != 0); |
f742dc4a PT |
274 | pg_len = PAGE_CACHE_SIZE; |
275 | } | |
276 | ||
5c83746a CH |
277 | isect += (pg_offset >> SECTOR_SHIFT); |
278 | extent_length -= (pg_offset >> SECTOR_SHIFT); | |
279 | ||
8067253c | 280 | if (is_hole(&be)) { |
9549ec01 FI |
281 | bio = bl_submit_bio(READ, bio); |
282 | /* Fill hole w/ zeroes w/o accessing device */ | |
283 | dprintk("%s Zeroing page for hole\n", __func__); | |
f742dc4a | 284 | zero_user_segment(pages[i], pg_offset, pg_len); |
5c83746a CH |
285 | |
286 | /* invalidate map */ | |
287 | map.start = NFS4_MAX_UINT64; | |
9549ec01 | 288 | } else { |
823b0c9d | 289 | bio = do_add_page_to_bio(bio, |
8067253c | 290 | header->page_array.npages - i, |
30dd374f | 291 | READ, |
5c83746a | 292 | isect, pages[i], &map, &be, |
f742dc4a | 293 | bl_end_io_read, par, |
5c83746a | 294 | pg_offset, &pg_len); |
9549ec01 | 295 | if (IS_ERR(bio)) { |
cd841605 | 296 | header->pnfs_error = PTR_ERR(bio); |
e6d05a75 | 297 | bio = NULL; |
9549ec01 FI |
298 | goto out; |
299 | } | |
300 | } | |
f742dc4a | 301 | isect += (pg_len >> SECTOR_SHIFT); |
921b81a8 | 302 | extent_length -= (pg_len >> SECTOR_SHIFT); |
5c83746a CH |
303 | f_offset += pg_len; |
304 | bytes_left -= pg_len; | |
9549ec01 | 305 | } |
cd841605 | 306 | if ((isect << SECTOR_SHIFT) >= header->inode->i_size) { |
8067253c CH |
307 | header->res.eof = 1; |
308 | header->res.count = header->inode->i_size - header->args.offset; | |
9549ec01 | 309 | } else { |
8067253c | 310 | header->res.count = (isect << SECTOR_SHIFT) - header->args.offset; |
9549ec01 FI |
311 | } |
312 | out: | |
9549ec01 | 313 | bl_submit_bio(READ, bio); |
be98fd0a | 314 | blk_finish_plug(&plug); |
9549ec01 FI |
315 | put_parallel(par); |
316 | return PNFS_ATTEMPTED; | |
31e6306a FI |
317 | } |
318 | ||
650e2d39 FI |
319 | static void bl_end_io_write(struct bio *bio, int err) |
320 | { | |
321 | struct parallel_io *par = bio->bi_private; | |
322 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d45f60c6 | 323 | struct nfs_pgio_header *header = par->data; |
650e2d39 FI |
324 | |
325 | if (!uptodate) { | |
cd841605 FI |
326 | if (!header->pnfs_error) |
327 | header->pnfs_error = -EIO; | |
328 | pnfs_set_lo_fail(header->lseg); | |
650e2d39 FI |
329 | } |
330 | bio_put(bio); | |
331 | put_parallel(par); | |
332 | } | |
333 | ||
334 | /* Function scheduled for call during bl_end_par_io_write, | |
335 | * it marks sectors as written and extends the commitlist. | |
336 | */ | |
337 | static void bl_write_cleanup(struct work_struct *work) | |
338 | { | |
8067253c CH |
339 | struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work); |
340 | struct nfs_pgio_header *hdr = | |
341 | container_of(task, struct nfs_pgio_header, task); | |
342 | ||
650e2d39 | 343 | dprintk("%s enter\n", __func__); |
8067253c | 344 | |
d45f60c6 | 345 | if (likely(!hdr->pnfs_error)) { |
8067253c CH |
346 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg); |
347 | u64 start = hdr->args.offset & (loff_t)PAGE_CACHE_MASK; | |
348 | u64 end = (hdr->args.offset + hdr->args.count + | |
349 | PAGE_CACHE_SIZE - 1) & (loff_t)PAGE_CACHE_MASK; | |
350 | ||
351 | ext_tree_mark_written(bl, start >> SECTOR_SHIFT, | |
352 | (end - start) >> SECTOR_SHIFT); | |
31e6306a | 353 | } |
8067253c | 354 | |
d45f60c6 | 355 | pnfs_ld_write_done(hdr); |
650e2d39 FI |
356 | } |
357 | ||
358 | /* Called when last of bios associated with a bl_write_pagelist call finishes */ | |
8067253c | 359 | static void bl_end_par_io_write(void *data) |
650e2d39 | 360 | { |
d45f60c6 | 361 | struct nfs_pgio_header *hdr = data; |
650e2d39 | 362 | |
d45f60c6 | 363 | hdr->task.tk_status = hdr->pnfs_error; |
c65e6254 | 364 | hdr->verf.committed = NFS_FILE_SYNC; |
d45f60c6 WAA |
365 | INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup); |
366 | schedule_work(&hdr->task.u.tk_work); | |
650e2d39 FI |
367 | } |
368 | ||
155e7524 | 369 | static enum pnfs_try_status |
d45f60c6 | 370 | bl_write_pagelist(struct nfs_pgio_header *header, int sync) |
155e7524 | 371 | { |
8067253c | 372 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); |
5c83746a | 373 | struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 }; |
650e2d39 | 374 | struct bio *bio = NULL; |
8067253c | 375 | struct pnfs_block_extent be; |
3a6fd1f0 | 376 | sector_t isect, extent_length = 0; |
96c9eae6 | 377 | struct parallel_io *par = NULL; |
d45f60c6 WAA |
378 | loff_t offset = header->args.offset; |
379 | size_t count = header->args.count; | |
d45f60c6 | 380 | struct page **pages = header->args.pages; |
b283f944 | 381 | int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT; |
5c83746a | 382 | unsigned int pg_len; |
be98fd0a | 383 | struct blk_plug plug; |
8067253c | 384 | int i; |
650e2d39 FI |
385 | |
386 | dprintk("%s enter, %Zu@%lld\n", __func__, count, offset); | |
96c9eae6 | 387 | |
d45f60c6 | 388 | /* At this point, header->page_aray is a (sequential) list of nfs_pages. |
71cdd40f PT |
389 | * We want to write each, and if there is an error set pnfs_error |
390 | * to have it redone using nfs. | |
650e2d39 | 391 | */ |
d45f60c6 | 392 | par = alloc_parallel(header); |
650e2d39 | 393 | if (!par) |
8067253c | 394 | return PNFS_NOT_ATTEMPTED; |
650e2d39 | 395 | par->pnfs_callback = bl_end_par_io_write; |
650e2d39 | 396 | |
3a6fd1f0 | 397 | blk_start_plug(&plug); |
71cdd40f | 398 | |
3a6fd1f0 CH |
399 | /* we always write out the whole page */ |
400 | offset = offset & (loff_t)PAGE_CACHE_MASK; | |
401 | isect = offset >> SECTOR_SHIFT; | |
71cdd40f | 402 | |
d45f60c6 | 403 | for (i = pg_index; i < header->page_array.npages; i++) { |
921b81a8 | 404 | if (extent_length <= 0) { |
650e2d39 | 405 | /* We've used up the previous extent */ |
650e2d39 FI |
406 | bio = bl_submit_bio(WRITE, bio); |
407 | /* Get the next one */ | |
8067253c | 408 | if (!ext_tree_lookup(bl, isect, &be, true)) { |
cd841605 | 409 | header->pnfs_error = -EINVAL; |
650e2d39 FI |
410 | goto out; |
411 | } | |
fe6e1e8d | 412 | |
8067253c | 413 | extent_length = be.be_length - (isect - be.be_f_offset); |
71cdd40f | 414 | } |
fe6e1e8d | 415 | |
5c83746a | 416 | pg_len = PAGE_CACHE_SIZE; |
d45f60c6 | 417 | bio = do_add_page_to_bio(bio, header->page_array.npages - i, |
5c83746a | 418 | WRITE, isect, pages[i], &map, &be, |
fe6e1e8d | 419 | bl_end_io_write, par, |
5c83746a | 420 | 0, &pg_len); |
71cdd40f | 421 | if (IS_ERR(bio)) { |
cd841605 | 422 | header->pnfs_error = PTR_ERR(bio); |
e6d05a75 | 423 | bio = NULL; |
71cdd40f | 424 | goto out; |
650e2d39 | 425 | } |
5c83746a CH |
426 | |
427 | offset += pg_len; | |
428 | count -= pg_len; | |
429 | isect += (pg_len >> SECTOR_SHIFT); | |
430 | extent_length -= (pg_len >> SECTOR_SHIFT); | |
650e2d39 | 431 | } |
71cdd40f | 432 | |
d45f60c6 | 433 | header->res.count = header->args.count; |
650e2d39 | 434 | out: |
650e2d39 | 435 | bl_submit_bio(WRITE, bio); |
be98fd0a | 436 | blk_finish_plug(&plug); |
650e2d39 FI |
437 | put_parallel(par); |
438 | return PNFS_ATTEMPTED; | |
155e7524 FI |
439 | } |
440 | ||
441 | static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
442 | { | |
443 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); | |
8067253c | 444 | int err; |
155e7524 FI |
445 | |
446 | dprintk("%s enter\n", __func__); | |
8067253c CH |
447 | |
448 | err = ext_tree_remove(bl, true, 0, LLONG_MAX); | |
449 | WARN_ON(err); | |
450 | ||
155e7524 FI |
451 | kfree(bl); |
452 | } | |
453 | ||
454 | static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode, | |
455 | gfp_t gfp_flags) | |
456 | { | |
457 | struct pnfs_block_layout *bl; | |
458 | ||
459 | dprintk("%s enter\n", __func__); | |
460 | bl = kzalloc(sizeof(*bl), gfp_flags); | |
461 | if (!bl) | |
462 | return NULL; | |
8067253c CH |
463 | |
464 | bl->bl_ext_rw = RB_ROOT; | |
465 | bl->bl_ext_ro = RB_ROOT; | |
155e7524 | 466 | spin_lock_init(&bl->bl_ext_lock); |
8067253c | 467 | |
155e7524 FI |
468 | return &bl->bl_layout; |
469 | } | |
470 | ||
a60d2ebd | 471 | static void bl_free_lseg(struct pnfs_layout_segment *lseg) |
155e7524 | 472 | { |
a60d2ebd FI |
473 | dprintk("%s enter\n", __func__); |
474 | kfree(lseg); | |
155e7524 FI |
475 | } |
476 | ||
9cc47541 CH |
477 | /* Tracks info needed to ensure extents in layout obey constraints of spec */ |
478 | struct layout_verification { | |
479 | u32 mode; /* R or RW */ | |
480 | u64 start; /* Expected start of next non-COW extent */ | |
481 | u64 inval; /* Start of INVAL coverage */ | |
482 | u64 cowread; /* End of COW read coverage */ | |
483 | }; | |
484 | ||
485 | /* Verify the extent meets the layout requirements of the pnfs-block draft, | |
486 | * section 2.3.1. | |
487 | */ | |
488 | static int verify_extent(struct pnfs_block_extent *be, | |
489 | struct layout_verification *lv) | |
490 | { | |
491 | if (lv->mode == IOMODE_READ) { | |
492 | if (be->be_state == PNFS_BLOCK_READWRITE_DATA || | |
493 | be->be_state == PNFS_BLOCK_INVALID_DATA) | |
494 | return -EIO; | |
495 | if (be->be_f_offset != lv->start) | |
496 | return -EIO; | |
497 | lv->start += be->be_length; | |
498 | return 0; | |
499 | } | |
500 | /* lv->mode == IOMODE_RW */ | |
501 | if (be->be_state == PNFS_BLOCK_READWRITE_DATA) { | |
502 | if (be->be_f_offset != lv->start) | |
503 | return -EIO; | |
504 | if (lv->cowread > lv->start) | |
505 | return -EIO; | |
506 | lv->start += be->be_length; | |
507 | lv->inval = lv->start; | |
508 | return 0; | |
509 | } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) { | |
510 | if (be->be_f_offset != lv->start) | |
511 | return -EIO; | |
512 | lv->start += be->be_length; | |
513 | return 0; | |
514 | } else if (be->be_state == PNFS_BLOCK_READ_DATA) { | |
515 | if (be->be_f_offset > lv->start) | |
516 | return -EIO; | |
517 | if (be->be_f_offset < lv->inval) | |
518 | return -EIO; | |
519 | if (be->be_f_offset < lv->cowread) | |
520 | return -EIO; | |
521 | /* It looks like you might want to min this with lv->start, | |
522 | * but you really don't. | |
523 | */ | |
524 | lv->inval = lv->inval + be->be_length; | |
525 | lv->cowread = be->be_f_offset + be->be_length; | |
526 | return 0; | |
527 | } else | |
528 | return -EIO; | |
529 | } | |
530 | ||
531 | static int decode_sector_number(__be32 **rp, sector_t *sp) | |
532 | { | |
533 | uint64_t s; | |
534 | ||
535 | *rp = xdr_decode_hyper(*rp, &s); | |
536 | if (s & 0x1ff) { | |
537 | printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__); | |
538 | return -1; | |
539 | } | |
540 | *sp = s >> SECTOR_SHIFT; | |
541 | return 0; | |
542 | } | |
543 | ||
9cc47541 | 544 | static int |
ca0fe1df CH |
545 | bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo, |
546 | struct layout_verification *lv, struct list_head *extents, | |
547 | gfp_t gfp_mask) | |
9cc47541 | 548 | { |
ca0fe1df CH |
549 | struct pnfs_block_extent *be; |
550 | struct nfs4_deviceid id; | |
551 | int error; | |
9cc47541 | 552 | __be32 *p; |
ca0fe1df CH |
553 | |
554 | p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE); | |
555 | if (!p) | |
556 | return -EIO; | |
557 | ||
558 | be = kzalloc(sizeof(*be), GFP_NOFS); | |
559 | if (!be) | |
560 | return -ENOMEM; | |
561 | ||
562 | memcpy(&id, p, NFS4_DEVICEID4_SIZE); | |
563 | p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE); | |
564 | ||
565 | error = -EIO; | |
566 | be->be_device = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id, | |
567 | lo->plh_lc_cred, gfp_mask); | |
568 | if (!be->be_device) | |
569 | goto out_free_be; | |
570 | ||
571 | /* | |
572 | * The next three values are read in as bytes, but stored in the | |
573 | * extent structure in 512-byte granularity. | |
574 | */ | |
575 | if (decode_sector_number(&p, &be->be_f_offset) < 0) | |
576 | goto out_put_deviceid; | |
577 | if (decode_sector_number(&p, &be->be_length) < 0) | |
578 | goto out_put_deviceid; | |
579 | if (decode_sector_number(&p, &be->be_v_offset) < 0) | |
580 | goto out_put_deviceid; | |
581 | be->be_state = be32_to_cpup(p++); | |
582 | ||
583 | error = verify_extent(be, lv); | |
584 | if (error) { | |
585 | dprintk("%s: extent verification failed\n", __func__); | |
586 | goto out_put_deviceid; | |
587 | } | |
588 | ||
589 | list_add_tail(&be->be_list, extents); | |
590 | return 0; | |
591 | ||
592 | out_put_deviceid: | |
593 | nfs4_put_deviceid_node(be->be_device); | |
594 | out_free_be: | |
595 | kfree(be); | |
596 | return error; | |
597 | } | |
598 | ||
599 | static struct pnfs_layout_segment * | |
600 | bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr, | |
601 | gfp_t gfp_mask) | |
602 | { | |
9cc47541 CH |
603 | struct layout_verification lv = { |
604 | .mode = lgr->range.iomode, | |
605 | .start = lgr->range.offset >> SECTOR_SHIFT, | |
606 | .inval = lgr->range.offset >> SECTOR_SHIFT, | |
607 | .cowread = lgr->range.offset >> SECTOR_SHIFT, | |
608 | }; | |
ca0fe1df CH |
609 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); |
610 | struct pnfs_layout_segment *lseg; | |
611 | struct xdr_buf buf; | |
612 | struct xdr_stream xdr; | |
613 | struct page *scratch; | |
614 | int status, i; | |
615 | uint32_t count; | |
616 | __be32 *p; | |
9cc47541 CH |
617 | LIST_HEAD(extents); |
618 | ||
619 | dprintk("---> %s\n", __func__); | |
620 | ||
ca0fe1df CH |
621 | lseg = kzalloc(sizeof(*lseg), gfp_mask); |
622 | if (!lseg) | |
623 | return ERR_PTR(-ENOMEM); | |
624 | ||
625 | status = -ENOMEM; | |
626 | scratch = alloc_page(gfp_mask); | |
9cc47541 | 627 | if (!scratch) |
ca0fe1df | 628 | goto out; |
9cc47541 | 629 | |
ca0fe1df CH |
630 | xdr_init_decode_pages(&xdr, &buf, |
631 | lgr->layoutp->pages, lgr->layoutp->len); | |
632 | xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE); | |
9cc47541 | 633 | |
ca0fe1df CH |
634 | status = -EIO; |
635 | p = xdr_inline_decode(&xdr, 4); | |
9cc47541 | 636 | if (unlikely(!p)) |
ca0fe1df | 637 | goto out_free_scratch; |
9cc47541 CH |
638 | |
639 | count = be32_to_cpup(p++); | |
ca0fe1df | 640 | dprintk("%s: number of extents %d\n", __func__, count); |
9cc47541 | 641 | |
ca0fe1df CH |
642 | /* |
643 | * Decode individual extents, putting them in temporary staging area | |
644 | * until whole layout is decoded to make error recovery easier. | |
9cc47541 CH |
645 | */ |
646 | for (i = 0; i < count; i++) { | |
ca0fe1df CH |
647 | status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask); |
648 | if (status) | |
649 | goto process_extents; | |
9cc47541 | 650 | } |
ca0fe1df | 651 | |
9cc47541 CH |
652 | if (lgr->range.offset + lgr->range.length != |
653 | lv.start << SECTOR_SHIFT) { | |
654 | dprintk("%s Final length mismatch\n", __func__); | |
ca0fe1df CH |
655 | status = -EIO; |
656 | goto process_extents; | |
9cc47541 | 657 | } |
ca0fe1df | 658 | |
9cc47541 CH |
659 | if (lv.start < lv.cowread) { |
660 | dprintk("%s Final uncovered COW extent\n", __func__); | |
ca0fe1df | 661 | status = -EIO; |
9cc47541 | 662 | } |
9cc47541 | 663 | |
ca0fe1df | 664 | process_extents: |
9cc47541 | 665 | while (!list_empty(&extents)) { |
ca0fe1df CH |
666 | struct pnfs_block_extent *be = |
667 | list_first_entry(&extents, struct pnfs_block_extent, | |
668 | be_list); | |
9cc47541 | 669 | list_del(&be->be_list); |
9cc47541 | 670 | |
ca0fe1df CH |
671 | if (!status) |
672 | status = ext_tree_insert(bl, be); | |
a60d2ebd | 673 | |
ca0fe1df CH |
674 | if (status) { |
675 | nfs4_put_deviceid_node(be->be_device); | |
676 | kfree(be); | |
677 | } | |
678 | } | |
679 | ||
680 | out_free_scratch: | |
681 | __free_page(scratch); | |
682 | out: | |
683 | dprintk("%s returns %d\n", __func__, status); | |
a60d2ebd | 684 | if (status) { |
a60d2ebd FI |
685 | kfree(lseg); |
686 | return ERR_PTR(status); | |
687 | } | |
688 | return lseg; | |
155e7524 FI |
689 | } |
690 | ||
71d5b763 CH |
691 | static void |
692 | bl_return_range(struct pnfs_layout_hdr *lo, | |
693 | struct pnfs_layout_range *range) | |
694 | { | |
695 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); | |
696 | sector_t offset = range->offset >> SECTOR_SHIFT, end; | |
71d5b763 CH |
697 | |
698 | if (range->offset % 8) { | |
699 | dprintk("%s: offset %lld not block size aligned\n", | |
700 | __func__, range->offset); | |
701 | return; | |
702 | } | |
703 | ||
704 | if (range->length != NFS4_MAX_UINT64) { | |
705 | if (range->length % 8) { | |
706 | dprintk("%s: length %lld not block size aligned\n", | |
707 | __func__, range->length); | |
708 | return; | |
709 | } | |
710 | ||
711 | end = offset + (range->length >> SECTOR_SHIFT); | |
712 | } else { | |
713 | end = round_down(NFS4_MAX_UINT64, PAGE_SIZE); | |
714 | } | |
715 | ||
164ae58c | 716 | ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end); |
71d5b763 CH |
717 | } |
718 | ||
34dc93c2 CH |
719 | static int |
720 | bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg) | |
155e7524 | 721 | { |
34dc93c2 | 722 | return ext_tree_prepare_commit(arg); |
155e7524 FI |
723 | } |
724 | ||
725 | static void | |
726 | bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata) | |
727 | { | |
34dc93c2 | 728 | ext_tree_mark_committed(&lcdata->args, lcdata->res.status); |
155e7524 FI |
729 | } |
730 | ||
731 | static int | |
732 | bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh) | |
733 | { | |
734 | dprintk("%s enter\n", __func__); | |
2f9fd182 FI |
735 | |
736 | if (server->pnfs_blksize == 0) { | |
737 | dprintk("%s Server did not return blksize\n", __func__); | |
738 | return -EINVAL; | |
739 | } | |
e3aaf7f2 CH |
740 | if (server->pnfs_blksize > PAGE_SIZE) { |
741 | printk(KERN_ERR "%s: pNFS blksize %d not supported.\n", | |
742 | __func__, server->pnfs_blksize); | |
743 | return -EINVAL; | |
744 | } | |
745 | ||
d4b18c3e | 746 | return 0; |
155e7524 FI |
747 | } |
748 | ||
f742dc4a | 749 | static bool |
3a6fd1f0 CH |
750 | is_aligned_req(struct nfs_pageio_descriptor *pgio, |
751 | struct nfs_page *req, unsigned int alignment) | |
f742dc4a | 752 | { |
3a6fd1f0 CH |
753 | /* |
754 | * Always accept buffered writes, higher layers take care of the | |
755 | * right alignment. | |
756 | */ | |
757 | if (pgio->pg_dreq == NULL) | |
758 | return true; | |
759 | ||
760 | if (!IS_ALIGNED(req->wb_offset, alignment)) | |
761 | return false; | |
762 | ||
763 | if (IS_ALIGNED(req->wb_bytes, alignment)) | |
764 | return true; | |
765 | ||
766 | if (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode)) { | |
767 | /* | |
768 | * If the write goes up to the inode size, just write | |
769 | * the full page. Data past the inode size is | |
770 | * guaranteed to be zeroed by the higher level client | |
771 | * code, and this behaviour is mandated by RFC 5663 | |
772 | * section 2.3.2. | |
773 | */ | |
774 | return true; | |
775 | } | |
776 | ||
777 | return false; | |
f742dc4a PT |
778 | } |
779 | ||
780 | static void | |
781 | bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
782 | { | |
3a6fd1f0 | 783 | if (!is_aligned_req(pgio, req, SECTOR_SIZE)) { |
f742dc4a | 784 | nfs_pageio_reset_read_mds(pgio); |
3a6fd1f0 CH |
785 | return; |
786 | } | |
787 | ||
788 | pnfs_generic_pg_init_read(pgio, req); | |
f742dc4a PT |
789 | } |
790 | ||
b4fdac1a WAA |
791 | /* |
792 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
793 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
794 | */ | |
795 | static size_t | |
f742dc4a PT |
796 | bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
797 | struct nfs_page *req) | |
798 | { | |
3a6fd1f0 | 799 | if (!is_aligned_req(pgio, req, SECTOR_SIZE)) |
b4fdac1a | 800 | return 0; |
f742dc4a PT |
801 | return pnfs_generic_pg_test(pgio, prev, req); |
802 | } | |
803 | ||
6296556f PT |
804 | /* |
805 | * Return the number of contiguous bytes for a given inode | |
806 | * starting at page frame idx. | |
807 | */ | |
808 | static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx) | |
809 | { | |
810 | struct address_space *mapping = inode->i_mapping; | |
811 | pgoff_t end; | |
812 | ||
813 | /* Optimize common case that writes from 0 to end of file */ | |
814 | end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE); | |
6a74c0c9 | 815 | if (end != inode->i_mapping->nrpages) { |
6296556f | 816 | rcu_read_lock(); |
e7b563bb | 817 | end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX); |
6296556f PT |
818 | rcu_read_unlock(); |
819 | } | |
820 | ||
821 | if (!end) | |
822 | return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT); | |
823 | else | |
824 | return (end - idx) << PAGE_CACHE_SHIFT; | |
825 | } | |
826 | ||
6f018efa | 827 | static void |
96c9eae6 PT |
828 | bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) |
829 | { | |
3a6fd1f0 CH |
830 | u64 wb_size; |
831 | ||
832 | if (!is_aligned_req(pgio, req, PAGE_SIZE)) { | |
96c9eae6 | 833 | nfs_pageio_reset_write_mds(pgio); |
3a6fd1f0 | 834 | return; |
6296556f | 835 | } |
3a6fd1f0 CH |
836 | |
837 | if (pgio->pg_dreq == NULL) | |
838 | wb_size = pnfs_num_cont_bytes(pgio->pg_inode, | |
839 | req->wb_index); | |
840 | else | |
841 | wb_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
842 | ||
843 | pnfs_generic_pg_init_write(pgio, req, wb_size); | |
96c9eae6 PT |
844 | } |
845 | ||
b4fdac1a WAA |
846 | /* |
847 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
848 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
849 | */ | |
850 | static size_t | |
96c9eae6 PT |
851 | bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
852 | struct nfs_page *req) | |
853 | { | |
3a6fd1f0 | 854 | if (!is_aligned_req(pgio, req, PAGE_SIZE)) |
b4fdac1a | 855 | return 0; |
96c9eae6 PT |
856 | return pnfs_generic_pg_test(pgio, prev, req); |
857 | } | |
858 | ||
e9643fe8 | 859 | static const struct nfs_pageio_ops bl_pg_read_ops = { |
f742dc4a PT |
860 | .pg_init = bl_pg_init_read, |
861 | .pg_test = bl_pg_test_read, | |
e9643fe8 BH |
862 | .pg_doio = pnfs_generic_pg_readpages, |
863 | }; | |
864 | ||
865 | static const struct nfs_pageio_ops bl_pg_write_ops = { | |
96c9eae6 PT |
866 | .pg_init = bl_pg_init_write, |
867 | .pg_test = bl_pg_test_write, | |
e9643fe8 BH |
868 | .pg_doio = pnfs_generic_pg_writepages, |
869 | }; | |
870 | ||
155e7524 FI |
871 | static struct pnfs_layoutdriver_type blocklayout_type = { |
872 | .id = LAYOUT_BLOCK_VOLUME, | |
873 | .name = "LAYOUT_BLOCK_VOLUME", | |
5a12cca6 | 874 | .owner = THIS_MODULE, |
848746bd CH |
875 | .flags = PNFS_LAYOUTRET_ON_SETATTR | |
876 | PNFS_READ_WHOLE_PAGE, | |
155e7524 FI |
877 | .read_pagelist = bl_read_pagelist, |
878 | .write_pagelist = bl_write_pagelist, | |
879 | .alloc_layout_hdr = bl_alloc_layout_hdr, | |
880 | .free_layout_hdr = bl_free_layout_hdr, | |
881 | .alloc_lseg = bl_alloc_lseg, | |
882 | .free_lseg = bl_free_lseg, | |
71d5b763 | 883 | .return_range = bl_return_range, |
34dc93c2 | 884 | .prepare_layoutcommit = bl_prepare_layoutcommit, |
155e7524 FI |
885 | .cleanup_layoutcommit = bl_cleanup_layoutcommit, |
886 | .set_layoutdriver = bl_set_layoutdriver, | |
20d655d6 CH |
887 | .alloc_deviceid_node = bl_alloc_deviceid_node, |
888 | .free_deviceid_node = bl_free_deviceid_node, | |
e9643fe8 BH |
889 | .pg_read_ops = &bl_pg_read_ops, |
890 | .pg_write_ops = &bl_pg_write_ops, | |
155e7524 FI |
891 | }; |
892 | ||
893 | static int __init nfs4blocklayout_init(void) | |
894 | { | |
895 | int ret; | |
896 | ||
897 | dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__); | |
898 | ||
899 | ret = pnfs_register_layoutdriver(&blocklayout_type); | |
fe0a9b74 JR |
900 | if (ret) |
901 | goto out; | |
871760ce | 902 | ret = bl_init_pipefs(); |
627f3066 | 903 | if (ret) |
871760ce CH |
904 | goto out_unregister; |
905 | return 0; | |
fe0a9b74 | 906 | |
871760ce | 907 | out_unregister: |
fe0a9b74 | 908 | pnfs_unregister_layoutdriver(&blocklayout_type); |
871760ce | 909 | out: |
155e7524 FI |
910 | return ret; |
911 | } | |
912 | ||
913 | static void __exit nfs4blocklayout_exit(void) | |
914 | { | |
915 | dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n", | |
916 | __func__); | |
917 | ||
871760ce | 918 | bl_cleanup_pipefs(); |
155e7524 FI |
919 | pnfs_unregister_layoutdriver(&blocklayout_type); |
920 | } | |
921 | ||
922 | MODULE_ALIAS("nfs-layouttype4-3"); | |
923 | ||
924 | module_init(nfs4blocklayout_init); | |
925 | module_exit(nfs4blocklayout_exit); |