Commit | Line | Data |
---|---|---|
e8062719 BH |
1 | /* |
2 | * Copyright (C) 2005, 2006 | |
27d2e149 | 3 | * Avishay Traeger (avishay@gmail.com) |
e8062719 BH |
4 | * Copyright (C) 2008, 2009 |
5 | * Boaz Harrosh <bharrosh@panasas.com> | |
6 | * | |
7 | * Copyrights for code taken from ext2: | |
8 | * Copyright (C) 1992, 1993, 1994, 1995 | |
9 | * Remy Card (card@masi.ibp.fr) | |
10 | * Laboratoire MASI - Institut Blaise Pascal | |
11 | * Universite Pierre et Marie Curie (Paris VI) | |
12 | * from | |
13 | * linux/fs/minix/inode.c | |
14 | * Copyright (C) 1991, 1992 Linus Torvalds | |
15 | * | |
16 | * This file is part of exofs. | |
17 | * | |
18 | * exofs is free software; you can redistribute it and/or modify | |
19 | * it under the terms of the GNU General Public License as published by | |
20 | * the Free Software Foundation. Since it is based on ext2, and the only | |
21 | * valid version of GPL for the Linux kernel is version 2, the only valid | |
22 | * version of GPL for exofs is version 2. | |
23 | * | |
24 | * exofs is distributed in the hope that it will be useful, | |
25 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
26 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
27 | * GNU General Public License for more details. | |
28 | * | |
29 | * You should have received a copy of the GNU General Public License | |
30 | * along with exofs; if not, write to the Free Software | |
31 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
32 | */ | |
33 | ||
5a0e3ad6 | 34 | #include <linux/slab.h> |
e8062719 BH |
35 | |
36 | #include "exofs.h" | |
37 | ||
fe33cc1e BH |
38 | #define EXOFS_DBGMSG2(M...) do {} while (0) |
39 | ||
5a51c0c7 | 40 | enum {MAX_PAGES_KMALLOC = PAGE_SIZE / sizeof(struct page *), }; |
06886a5a | 41 | |
8ff660ab | 42 | unsigned exofs_max_io_pages(struct ore_layout *layout, |
66cd6cad | 43 | unsigned expected_pages) |
44 | { | |
45 | unsigned pages = min_t(unsigned, expected_pages, MAX_PAGES_KMALLOC); | |
46 | ||
47 | /* TODO: easily support bio chaining */ | |
5a51c0c7 | 48 | pages = min_t(unsigned, pages, layout->max_io_length / PAGE_SIZE); |
66cd6cad | 49 | return pages; |
50 | } | |
51 | ||
beaec07b BH |
52 | struct page_collect { |
53 | struct exofs_sb_info *sbi; | |
beaec07b BH |
54 | struct inode *inode; |
55 | unsigned expected_pages; | |
8ff660ab | 56 | struct ore_io_state *ios; |
beaec07b | 57 | |
86093aaf BH |
58 | struct page **pages; |
59 | unsigned alloc_pages; | |
beaec07b BH |
60 | unsigned nr_pages; |
61 | unsigned long length; | |
62 | loff_t pg_first; /* keep 64bit also in 32-arches */ | |
f17b1f9f BH |
63 | bool read_4_write; /* This means two things: that the read is sync |
64 | * And the pages should not be unlocked. | |
65 | */ | |
beaec07b BH |
66 | }; |
67 | ||
68 | static void _pcol_init(struct page_collect *pcol, unsigned expected_pages, | |
06886a5a | 69 | struct inode *inode) |
beaec07b BH |
70 | { |
71 | struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; | |
beaec07b BH |
72 | |
73 | pcol->sbi = sbi; | |
beaec07b BH |
74 | pcol->inode = inode; |
75 | pcol->expected_pages = expected_pages; | |
76 | ||
06886a5a | 77 | pcol->ios = NULL; |
86093aaf BH |
78 | pcol->pages = NULL; |
79 | pcol->alloc_pages = 0; | |
beaec07b BH |
80 | pcol->nr_pages = 0; |
81 | pcol->length = 0; | |
82 | pcol->pg_first = -1; | |
f17b1f9f | 83 | pcol->read_4_write = false; |
beaec07b BH |
84 | } |
85 | ||
86 | static void _pcol_reset(struct page_collect *pcol) | |
87 | { | |
88 | pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages); | |
89 | ||
86093aaf BH |
90 | pcol->pages = NULL; |
91 | pcol->alloc_pages = 0; | |
beaec07b BH |
92 | pcol->nr_pages = 0; |
93 | pcol->length = 0; | |
94 | pcol->pg_first = -1; | |
06886a5a | 95 | pcol->ios = NULL; |
beaec07b BH |
96 | |
97 | /* this is probably the end of the loop but in writes | |
98 | * it might not end here. don't be left with nothing | |
99 | */ | |
100 | if (!pcol->expected_pages) | |
86093aaf | 101 | pcol->expected_pages = MAX_PAGES_KMALLOC; |
beaec07b BH |
102 | } |
103 | ||
104 | static int pcol_try_alloc(struct page_collect *pcol) | |
105 | { | |
66cd6cad | 106 | unsigned pages; |
06886a5a | 107 | |
86093aaf | 108 | /* TODO: easily support bio chaining */ |
66cd6cad | 109 | pages = exofs_max_io_pages(&pcol->sbi->layout, pcol->expected_pages); |
86093aaf | 110 | |
beaec07b | 111 | for (; pages; pages >>= 1) { |
86093aaf BH |
112 | pcol->pages = kmalloc(pages * sizeof(struct page *), |
113 | GFP_KERNEL); | |
114 | if (likely(pcol->pages)) { | |
115 | pcol->alloc_pages = pages; | |
beaec07b | 116 | return 0; |
86093aaf | 117 | } |
beaec07b BH |
118 | } |
119 | ||
86093aaf | 120 | EXOFS_ERR("Failed to kmalloc expected_pages=%u\n", |
beaec07b BH |
121 | pcol->expected_pages); |
122 | return -ENOMEM; | |
123 | } | |
124 | ||
125 | static void pcol_free(struct page_collect *pcol) | |
126 | { | |
86093aaf BH |
127 | kfree(pcol->pages); |
128 | pcol->pages = NULL; | |
06886a5a BH |
129 | |
130 | if (pcol->ios) { | |
8ff660ab | 131 | ore_put_io_state(pcol->ios); |
06886a5a BH |
132 | pcol->ios = NULL; |
133 | } | |
beaec07b BH |
134 | } |
135 | ||
136 | static int pcol_add_page(struct page_collect *pcol, struct page *page, | |
137 | unsigned len) | |
138 | { | |
86093aaf | 139 | if (unlikely(pcol->nr_pages >= pcol->alloc_pages)) |
beaec07b BH |
140 | return -ENOMEM; |
141 | ||
86093aaf | 142 | pcol->pages[pcol->nr_pages++] = page; |
beaec07b BH |
143 | pcol->length += len; |
144 | return 0; | |
145 | } | |
146 | ||
154a9300 | 147 | enum {PAGE_WAS_NOT_IN_IO = 17}; |
beaec07b BH |
148 | static int update_read_page(struct page *page, int ret) |
149 | { | |
154a9300 BH |
150 | switch (ret) { |
151 | case 0: | |
beaec07b BH |
152 | /* Everything is OK */ |
153 | SetPageUptodate(page); | |
154 | if (PageError(page)) | |
155 | ClearPageError(page); | |
154a9300 BH |
156 | break; |
157 | case -EFAULT: | |
beaec07b BH |
158 | /* In this case we were trying to read something that wasn't on |
159 | * disk yet - return a page full of zeroes. This should be OK, | |
160 | * because the object should be empty (if there was a write | |
161 | * before this read, the read would be waiting with the page | |
162 | * locked */ | |
163 | clear_highpage(page); | |
164 | ||
165 | SetPageUptodate(page); | |
166 | if (PageError(page)) | |
167 | ClearPageError(page); | |
beaec07b | 168 | EXOFS_DBGMSG("recovered read error\n"); |
154a9300 BH |
169 | /* fall through */ |
170 | case PAGE_WAS_NOT_IN_IO: | |
171 | ret = 0; /* recovered error */ | |
172 | break; | |
173 | default: | |
beaec07b | 174 | SetPageError(page); |
154a9300 | 175 | } |
beaec07b BH |
176 | return ret; |
177 | } | |
178 | ||
179 | static void update_write_page(struct page *page, int ret) | |
180 | { | |
154a9300 BH |
181 | if (unlikely(ret == PAGE_WAS_NOT_IN_IO)) |
182 | return; /* don't pass start don't collect $200 */ | |
183 | ||
beaec07b BH |
184 | if (ret) { |
185 | mapping_set_error(page->mapping, ret); | |
186 | SetPageError(page); | |
187 | } | |
188 | end_page_writeback(page); | |
189 | } | |
190 | ||
191 | /* Called at the end of reads, to optionally unlock pages and update their | |
192 | * status. | |
193 | */ | |
7aebf410 | 194 | static int __readpages_done(struct page_collect *pcol) |
beaec07b | 195 | { |
beaec07b | 196 | int i; |
beaec07b BH |
197 | u64 good_bytes; |
198 | u64 length = 0; | |
4b46c9f5 | 199 | int ret = ore_check_io(pcol->ios, NULL); |
beaec07b | 200 | |
154a9300 | 201 | if (likely(!ret)) { |
beaec07b | 202 | good_bytes = pcol->length; |
154a9300 BH |
203 | ret = PAGE_WAS_NOT_IN_IO; |
204 | } else { | |
4b46c9f5 | 205 | good_bytes = 0; |
154a9300 | 206 | } |
beaec07b | 207 | |
34ce4e7c | 208 | EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx" |
beaec07b BH |
209 | " length=0x%lx nr_pages=%u\n", |
210 | pcol->inode->i_ino, _LLU(good_bytes), pcol->length, | |
211 | pcol->nr_pages); | |
212 | ||
86093aaf BH |
213 | for (i = 0; i < pcol->nr_pages; i++) { |
214 | struct page *page = pcol->pages[i]; | |
beaec07b BH |
215 | struct inode *inode = page->mapping->host; |
216 | int page_stat; | |
217 | ||
218 | if (inode != pcol->inode) | |
219 | continue; /* osd might add more pages at end */ | |
220 | ||
221 | if (likely(length < good_bytes)) | |
222 | page_stat = 0; | |
223 | else | |
224 | page_stat = ret; | |
225 | ||
fe33cc1e | 226 | EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n", |
beaec07b BH |
227 | inode->i_ino, page->index, |
228 | page_stat ? "bad_bytes" : "good_bytes"); | |
229 | ||
230 | ret = update_read_page(page, page_stat); | |
7aebf410 | 231 | if (!pcol->read_4_write) |
beaec07b | 232 | unlock_page(page); |
86093aaf | 233 | length += PAGE_SIZE; |
beaec07b BH |
234 | } |
235 | ||
236 | pcol_free(pcol); | |
34ce4e7c | 237 | EXOFS_DBGMSG2("readpages_done END\n"); |
beaec07b BH |
238 | return ret; |
239 | } | |
240 | ||
241 | /* callback of async reads */ | |
8ff660ab | 242 | static void readpages_done(struct ore_io_state *ios, void *p) |
beaec07b BH |
243 | { |
244 | struct page_collect *pcol = p; | |
245 | ||
7aebf410 | 246 | __readpages_done(pcol); |
beaec07b | 247 | atomic_dec(&pcol->sbi->s_curr_pending); |
06886a5a | 248 | kfree(pcol); |
beaec07b BH |
249 | } |
250 | ||
251 | static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw) | |
252 | { | |
beaec07b BH |
253 | int i; |
254 | ||
86093aaf BH |
255 | for (i = 0; i < pcol->nr_pages; i++) { |
256 | struct page *page = pcol->pages[i]; | |
beaec07b BH |
257 | |
258 | if (rw == READ) | |
259 | update_read_page(page, ret); | |
260 | else | |
261 | update_write_page(page, ret); | |
262 | ||
263 | unlock_page(page); | |
264 | } | |
beaec07b BH |
265 | } |
266 | ||
b916c5cd BH |
267 | static int _maybe_not_all_in_one_io(struct ore_io_state *ios, |
268 | struct page_collect *pcol_src, struct page_collect *pcol) | |
269 | { | |
270 | /* length was wrong or offset was not page aligned */ | |
271 | BUG_ON(pcol_src->nr_pages < ios->nr_pages); | |
272 | ||
273 | if (pcol_src->nr_pages > ios->nr_pages) { | |
274 | struct page **src_page; | |
275 | unsigned pages_less = pcol_src->nr_pages - ios->nr_pages; | |
276 | unsigned long len_less = pcol_src->length - ios->length; | |
277 | unsigned i; | |
278 | int ret; | |
279 | ||
280 | /* This IO was trimmed */ | |
281 | pcol_src->nr_pages = ios->nr_pages; | |
282 | pcol_src->length = ios->length; | |
283 | ||
284 | /* Left over pages are passed to the next io */ | |
285 | pcol->expected_pages += pages_less; | |
286 | pcol->nr_pages = pages_less; | |
287 | pcol->length = len_less; | |
288 | src_page = pcol_src->pages + pcol_src->nr_pages; | |
289 | pcol->pg_first = (*src_page)->index; | |
290 | ||
291 | ret = pcol_try_alloc(pcol); | |
292 | if (unlikely(ret)) | |
293 | return ret; | |
294 | ||
295 | for (i = 0; i < pages_less; ++i) | |
296 | pcol->pages[i] = *src_page++; | |
297 | ||
298 | EXOFS_DBGMSG("Length was adjusted nr_pages=0x%x " | |
299 | "pages_less=0x%x expected_pages=0x%x " | |
300 | "next_offset=0x%llx next_len=0x%lx\n", | |
301 | pcol_src->nr_pages, pages_less, pcol->expected_pages, | |
302 | pcol->pg_first * PAGE_SIZE, pcol->length); | |
303 | } | |
304 | return 0; | |
305 | } | |
306 | ||
7aebf410 | 307 | static int read_exec(struct page_collect *pcol) |
beaec07b BH |
308 | { |
309 | struct exofs_i_info *oi = exofs_i(pcol->inode); | |
8ff660ab | 310 | struct ore_io_state *ios; |
beaec07b | 311 | struct page_collect *pcol_copy = NULL; |
beaec07b BH |
312 | int ret; |
313 | ||
86093aaf | 314 | if (!pcol->pages) |
beaec07b BH |
315 | return 0; |
316 | ||
e1042ba0 | 317 | if (!pcol->ios) { |
5bf696da | 318 | int ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, true, |
e1042ba0 BH |
319 | pcol->pg_first << PAGE_CACHE_SHIFT, |
320 | pcol->length, &pcol->ios); | |
321 | ||
322 | if (ret) | |
323 | return ret; | |
324 | } | |
325 | ||
326 | ios = pcol->ios; | |
86093aaf | 327 | ios->pages = pcol->pages; |
beaec07b | 328 | |
7aebf410 | 329 | if (pcol->read_4_write) { |
8ff660ab | 330 | ore_read(pcol->ios); |
7aebf410 | 331 | return __readpages_done(pcol); |
beaec07b BH |
332 | } |
333 | ||
334 | pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); | |
335 | if (!pcol_copy) { | |
336 | ret = -ENOMEM; | |
337 | goto err; | |
338 | } | |
339 | ||
340 | *pcol_copy = *pcol; | |
06886a5a BH |
341 | ios->done = readpages_done; |
342 | ios->private = pcol_copy; | |
b916c5cd BH |
343 | |
344 | /* pages ownership was passed to pcol_copy */ | |
345 | _pcol_reset(pcol); | |
346 | ||
347 | ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol); | |
348 | if (unlikely(ret)) | |
349 | goto err; | |
350 | ||
351 | EXOFS_DBGMSG2("read_exec(0x%lx) offset=0x%llx length=0x%llx\n", | |
352 | pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length)); | |
353 | ||
8ff660ab | 354 | ret = ore_read(ios); |
beaec07b BH |
355 | if (unlikely(ret)) |
356 | goto err; | |
357 | ||
358 | atomic_inc(&pcol->sbi->s_curr_pending); | |
359 | ||
beaec07b BH |
360 | return 0; |
361 | ||
362 | err: | |
7aebf410 | 363 | if (!pcol->read_4_write) |
beaec07b | 364 | _unlock_pcol_pages(pcol, ret, READ); |
06886a5a BH |
365 | |
366 | pcol_free(pcol); | |
b76a3f93 | 367 | |
beaec07b | 368 | kfree(pcol_copy); |
beaec07b BH |
369 | return ret; |
370 | } | |
371 | ||
372 | /* readpage_strip is called either directly from readpage() or by the VFS from | |
373 | * within read_cache_pages(), to add one more page to be read. It will try to | |
374 | * collect as many contiguous pages as posible. If a discontinuity is | |
375 | * encountered, or it runs out of resources, it will submit the previous segment | |
376 | * and will start a new collection. Eventually caller must submit the last | |
377 | * segment if present. | |
378 | */ | |
379 | static int readpage_strip(void *data, struct page *page) | |
380 | { | |
381 | struct page_collect *pcol = data; | |
382 | struct inode *inode = pcol->inode; | |
383 | struct exofs_i_info *oi = exofs_i(inode); | |
384 | loff_t i_size = i_size_read(inode); | |
385 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
386 | size_t len; | |
387 | int ret; | |
388 | ||
389 | /* FIXME: Just for debugging, will be removed */ | |
390 | if (PageUptodate(page)) | |
391 | EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino, | |
392 | page->index); | |
393 | ||
394 | if (page->index < end_index) | |
395 | len = PAGE_CACHE_SIZE; | |
396 | else if (page->index == end_index) | |
397 | len = i_size & ~PAGE_CACHE_MASK; | |
398 | else | |
399 | len = 0; | |
400 | ||
401 | if (!len || !obj_created(oi)) { | |
402 | /* this will be out of bounds, or doesn't exist yet. | |
403 | * Current page is cleared and the request is split | |
404 | */ | |
405 | clear_highpage(page); | |
406 | ||
407 | SetPageUptodate(page); | |
408 | if (PageError(page)) | |
409 | ClearPageError(page); | |
410 | ||
f17b1f9f BH |
411 | if (!pcol->read_4_write) |
412 | unlock_page(page); | |
a8f1418f BH |
413 | EXOFS_DBGMSG("readpage_strip(0x%lx) empty page len=%zx " |
414 | "read_4_write=%d index=0x%lx end_index=0x%lx " | |
415 | "splitting\n", inode->i_ino, len, | |
416 | pcol->read_4_write, page->index, end_index); | |
beaec07b | 417 | |
7aebf410 | 418 | return read_exec(pcol); |
beaec07b BH |
419 | } |
420 | ||
421 | try_again: | |
422 | ||
423 | if (unlikely(pcol->pg_first == -1)) { | |
424 | pcol->pg_first = page->index; | |
425 | } else if (unlikely((pcol->pg_first + pcol->nr_pages) != | |
426 | page->index)) { | |
427 | /* Discontinuity detected, split the request */ | |
7aebf410 | 428 | ret = read_exec(pcol); |
beaec07b BH |
429 | if (unlikely(ret)) |
430 | goto fail; | |
431 | goto try_again; | |
432 | } | |
433 | ||
86093aaf | 434 | if (!pcol->pages) { |
beaec07b BH |
435 | ret = pcol_try_alloc(pcol); |
436 | if (unlikely(ret)) | |
437 | goto fail; | |
438 | } | |
439 | ||
440 | if (len != PAGE_CACHE_SIZE) | |
441 | zero_user(page, len, PAGE_CACHE_SIZE - len); | |
442 | ||
fe33cc1e | 443 | EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n", |
beaec07b BH |
444 | inode->i_ino, page->index, len); |
445 | ||
446 | ret = pcol_add_page(pcol, page, len); | |
447 | if (ret) { | |
fe33cc1e | 448 | EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p " |
beaec07b BH |
449 | "this_len=0x%zx nr_pages=%u length=0x%lx\n", |
450 | page, len, pcol->nr_pages, pcol->length); | |
451 | ||
452 | /* split the request, and start again with current page */ | |
7aebf410 | 453 | ret = read_exec(pcol); |
beaec07b BH |
454 | if (unlikely(ret)) |
455 | goto fail; | |
456 | ||
457 | goto try_again; | |
458 | } | |
459 | ||
460 | return 0; | |
461 | ||
462 | fail: | |
463 | /* SetPageError(page); ??? */ | |
464 | unlock_page(page); | |
465 | return ret; | |
466 | } | |
467 | ||
468 | static int exofs_readpages(struct file *file, struct address_space *mapping, | |
469 | struct list_head *pages, unsigned nr_pages) | |
470 | { | |
471 | struct page_collect pcol; | |
472 | int ret; | |
473 | ||
474 | _pcol_init(&pcol, nr_pages, mapping->host); | |
475 | ||
476 | ret = read_cache_pages(mapping, pages, readpage_strip, &pcol); | |
477 | if (ret) { | |
478 | EXOFS_ERR("read_cache_pages => %d\n", ret); | |
479 | return ret; | |
480 | } | |
481 | ||
b916c5cd BH |
482 | ret = read_exec(&pcol); |
483 | if (unlikely(ret)) | |
484 | return ret; | |
485 | ||
7aebf410 | 486 | return read_exec(&pcol); |
beaec07b BH |
487 | } |
488 | ||
7aebf410 | 489 | static int _readpage(struct page *page, bool read_4_write) |
beaec07b BH |
490 | { |
491 | struct page_collect pcol; | |
492 | int ret; | |
493 | ||
494 | _pcol_init(&pcol, 1, page->mapping->host); | |
495 | ||
7aebf410 | 496 | pcol.read_4_write = read_4_write; |
beaec07b BH |
497 | ret = readpage_strip(&pcol, page); |
498 | if (ret) { | |
499 | EXOFS_ERR("_readpage => %d\n", ret); | |
500 | return ret; | |
501 | } | |
502 | ||
7aebf410 | 503 | return read_exec(&pcol); |
beaec07b BH |
504 | } |
505 | ||
506 | /* | |
507 | * We don't need the file | |
508 | */ | |
509 | static int exofs_readpage(struct file *file, struct page *page) | |
510 | { | |
511 | return _readpage(page, false); | |
512 | } | |
513 | ||
06886a5a | 514 | /* Callback for osd_write. All writes are asynchronous */ |
8ff660ab | 515 | static void writepages_done(struct ore_io_state *ios, void *p) |
beaec07b BH |
516 | { |
517 | struct page_collect *pcol = p; | |
beaec07b | 518 | int i; |
beaec07b BH |
519 | u64 good_bytes; |
520 | u64 length = 0; | |
4b46c9f5 | 521 | int ret = ore_check_io(ios, NULL); |
beaec07b | 522 | |
beaec07b BH |
523 | atomic_dec(&pcol->sbi->s_curr_pending); |
524 | ||
154a9300 | 525 | if (likely(!ret)) { |
beaec07b | 526 | good_bytes = pcol->length; |
154a9300 BH |
527 | ret = PAGE_WAS_NOT_IN_IO; |
528 | } else { | |
4b46c9f5 | 529 | good_bytes = 0; |
154a9300 | 530 | } |
beaec07b | 531 | |
34ce4e7c | 532 | EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx" |
beaec07b BH |
533 | " length=0x%lx nr_pages=%u\n", |
534 | pcol->inode->i_ino, _LLU(good_bytes), pcol->length, | |
535 | pcol->nr_pages); | |
536 | ||
86093aaf BH |
537 | for (i = 0; i < pcol->nr_pages; i++) { |
538 | struct page *page = pcol->pages[i]; | |
beaec07b BH |
539 | struct inode *inode = page->mapping->host; |
540 | int page_stat; | |
541 | ||
542 | if (inode != pcol->inode) | |
543 | continue; /* osd might add more pages to a bio */ | |
544 | ||
545 | if (likely(length < good_bytes)) | |
546 | page_stat = 0; | |
547 | else | |
548 | page_stat = ret; | |
549 | ||
550 | update_write_page(page, page_stat); | |
551 | unlock_page(page); | |
fe33cc1e | 552 | EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n", |
beaec07b BH |
553 | inode->i_ino, page->index, page_stat); |
554 | ||
86093aaf | 555 | length += PAGE_SIZE; |
beaec07b BH |
556 | } |
557 | ||
558 | pcol_free(pcol); | |
559 | kfree(pcol); | |
34ce4e7c | 560 | EXOFS_DBGMSG2("writepages_done END\n"); |
beaec07b BH |
561 | } |
562 | ||
563 | static int write_exec(struct page_collect *pcol) | |
564 | { | |
565 | struct exofs_i_info *oi = exofs_i(pcol->inode); | |
8ff660ab | 566 | struct ore_io_state *ios; |
beaec07b | 567 | struct page_collect *pcol_copy = NULL; |
beaec07b BH |
568 | int ret; |
569 | ||
86093aaf | 570 | if (!pcol->pages) |
beaec07b BH |
571 | return 0; |
572 | ||
e1042ba0 | 573 | BUG_ON(pcol->ios); |
5bf696da | 574 | ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, false, |
e1042ba0 BH |
575 | pcol->pg_first << PAGE_CACHE_SHIFT, |
576 | pcol->length, &pcol->ios); | |
e1042ba0 BH |
577 | if (unlikely(ret)) |
578 | goto err; | |
579 | ||
beaec07b BH |
580 | pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); |
581 | if (!pcol_copy) { | |
571f7f46 | 582 | EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n"); |
beaec07b BH |
583 | ret = -ENOMEM; |
584 | goto err; | |
585 | } | |
586 | ||
587 | *pcol_copy = *pcol; | |
588 | ||
e1042ba0 | 589 | ios = pcol->ios; |
86093aaf | 590 | ios->pages = pcol_copy->pages; |
06886a5a BH |
591 | ios->done = writepages_done; |
592 | ios->private = pcol_copy; | |
593 | ||
b916c5cd BH |
594 | /* pages ownership was passed to pcol_copy */ |
595 | _pcol_reset(pcol); | |
596 | ||
597 | ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol); | |
598 | if (unlikely(ret)) | |
599 | goto err; | |
600 | ||
601 | EXOFS_DBGMSG2("write_exec(0x%lx) offset=0x%llx length=0x%llx\n", | |
602 | pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length)); | |
603 | ||
8ff660ab | 604 | ret = ore_write(ios); |
beaec07b | 605 | if (unlikely(ret)) { |
8ff660ab | 606 | EXOFS_ERR("write_exec: ore_write() Failed\n"); |
beaec07b BH |
607 | goto err; |
608 | } | |
609 | ||
610 | atomic_inc(&pcol->sbi->s_curr_pending); | |
beaec07b BH |
611 | return 0; |
612 | ||
613 | err: | |
614 | _unlock_pcol_pages(pcol, ret, WRITE); | |
06886a5a | 615 | pcol_free(pcol); |
beaec07b | 616 | kfree(pcol_copy); |
06886a5a | 617 | |
beaec07b BH |
618 | return ret; |
619 | } | |
620 | ||
621 | /* writepage_strip is called either directly from writepage() or by the VFS from | |
622 | * within write_cache_pages(), to add one more page to be written to storage. | |
623 | * It will try to collect as many contiguous pages as possible. If a | |
624 | * discontinuity is encountered or it runs out of resources it will submit the | |
625 | * previous segment and will start a new collection. | |
626 | * Eventually caller must submit the last segment if present. | |
627 | */ | |
628 | static int writepage_strip(struct page *page, | |
629 | struct writeback_control *wbc_unused, void *data) | |
630 | { | |
631 | struct page_collect *pcol = data; | |
632 | struct inode *inode = pcol->inode; | |
633 | struct exofs_i_info *oi = exofs_i(inode); | |
634 | loff_t i_size = i_size_read(inode); | |
635 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
636 | size_t len; | |
637 | int ret; | |
638 | ||
639 | BUG_ON(!PageLocked(page)); | |
640 | ||
641 | ret = wait_obj_created(oi); | |
642 | if (unlikely(ret)) | |
643 | goto fail; | |
644 | ||
645 | if (page->index < end_index) | |
646 | /* in this case, the page is within the limits of the file */ | |
647 | len = PAGE_CACHE_SIZE; | |
648 | else { | |
649 | len = i_size & ~PAGE_CACHE_MASK; | |
650 | ||
651 | if (page->index > end_index || !len) { | |
652 | /* in this case, the page is outside the limits | |
653 | * (truncate in progress) | |
654 | */ | |
655 | ret = write_exec(pcol); | |
656 | if (unlikely(ret)) | |
657 | goto fail; | |
658 | if (PageError(page)) | |
659 | ClearPageError(page); | |
660 | unlock_page(page); | |
06886a5a BH |
661 | EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) " |
662 | "outside the limits\n", | |
663 | inode->i_ino, page->index); | |
beaec07b BH |
664 | return 0; |
665 | } | |
666 | } | |
667 | ||
668 | try_again: | |
669 | ||
670 | if (unlikely(pcol->pg_first == -1)) { | |
671 | pcol->pg_first = page->index; | |
672 | } else if (unlikely((pcol->pg_first + pcol->nr_pages) != | |
673 | page->index)) { | |
674 | /* Discontinuity detected, split the request */ | |
675 | ret = write_exec(pcol); | |
676 | if (unlikely(ret)) | |
677 | goto fail; | |
06886a5a BH |
678 | |
679 | EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n", | |
680 | inode->i_ino, page->index); | |
beaec07b BH |
681 | goto try_again; |
682 | } | |
683 | ||
86093aaf | 684 | if (!pcol->pages) { |
beaec07b BH |
685 | ret = pcol_try_alloc(pcol); |
686 | if (unlikely(ret)) | |
687 | goto fail; | |
688 | } | |
689 | ||
fe33cc1e | 690 | EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n", |
beaec07b BH |
691 | inode->i_ino, page->index, len); |
692 | ||
693 | ret = pcol_add_page(pcol, page, len); | |
694 | if (unlikely(ret)) { | |
34ce4e7c | 695 | EXOFS_DBGMSG2("Failed pcol_add_page " |
beaec07b BH |
696 | "nr_pages=%u total_length=0x%lx\n", |
697 | pcol->nr_pages, pcol->length); | |
698 | ||
699 | /* split the request, next loop will start again */ | |
700 | ret = write_exec(pcol); | |
701 | if (unlikely(ret)) { | |
571f7f46 | 702 | EXOFS_DBGMSG("write_exec failed => %d", ret); |
beaec07b BH |
703 | goto fail; |
704 | } | |
705 | ||
706 | goto try_again; | |
707 | } | |
708 | ||
709 | BUG_ON(PageWriteback(page)); | |
710 | set_page_writeback(page); | |
711 | ||
712 | return 0; | |
713 | ||
714 | fail: | |
06886a5a BH |
715 | EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n", |
716 | inode->i_ino, page->index, ret); | |
beaec07b BH |
717 | set_bit(AS_EIO, &page->mapping->flags); |
718 | unlock_page(page); | |
719 | return ret; | |
720 | } | |
721 | ||
722 | static int exofs_writepages(struct address_space *mapping, | |
723 | struct writeback_control *wbc) | |
724 | { | |
725 | struct page_collect pcol; | |
726 | long start, end, expected_pages; | |
727 | int ret; | |
728 | ||
729 | start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
730 | end = (wbc->range_end == LLONG_MAX) ? | |
731 | start + mapping->nrpages : | |
732 | wbc->range_end >> PAGE_CACHE_SHIFT; | |
733 | ||
734 | if (start || end) | |
06886a5a | 735 | expected_pages = end - start + 1; |
beaec07b BH |
736 | else |
737 | expected_pages = mapping->nrpages; | |
738 | ||
06886a5a BH |
739 | if (expected_pages < 32L) |
740 | expected_pages = 32L; | |
741 | ||
34ce4e7c | 742 | EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx " |
06886a5a | 743 | "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n", |
beaec07b | 744 | mapping->host->i_ino, wbc->range_start, wbc->range_end, |
06886a5a | 745 | mapping->nrpages, start, end, expected_pages); |
beaec07b BH |
746 | |
747 | _pcol_init(&pcol, expected_pages, mapping->host); | |
748 | ||
749 | ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol); | |
b916c5cd | 750 | if (unlikely(ret)) { |
beaec07b BH |
751 | EXOFS_ERR("write_cache_pages => %d\n", ret); |
752 | return ret; | |
753 | } | |
754 | ||
b916c5cd BH |
755 | ret = write_exec(&pcol); |
756 | if (unlikely(ret)) | |
757 | return ret; | |
758 | ||
759 | if (wbc->sync_mode == WB_SYNC_ALL) { | |
760 | return write_exec(&pcol); /* pump the last reminder */ | |
761 | } else if (pcol.nr_pages) { | |
762 | /* not SYNC let the reminder join the next writeout */ | |
763 | unsigned i; | |
764 | ||
765 | for (i = 0; i < pcol.nr_pages; i++) { | |
766 | struct page *page = pcol.pages[i]; | |
767 | ||
768 | end_page_writeback(page); | |
769 | set_page_dirty(page); | |
770 | unlock_page(page); | |
771 | } | |
772 | } | |
773 | return 0; | |
beaec07b BH |
774 | } |
775 | ||
776 | static int exofs_writepage(struct page *page, struct writeback_control *wbc) | |
777 | { | |
778 | struct page_collect pcol; | |
779 | int ret; | |
780 | ||
781 | _pcol_init(&pcol, 1, page->mapping->host); | |
782 | ||
783 | ret = writepage_strip(page, NULL, &pcol); | |
784 | if (ret) { | |
785 | EXOFS_ERR("exofs_writepage => %d\n", ret); | |
786 | return ret; | |
787 | } | |
788 | ||
789 | return write_exec(&pcol); | |
790 | } | |
791 | ||
2f246fd0 BH |
792 | /* i_mutex held using inode->i_size directly */ |
793 | static void _write_failed(struct inode *inode, loff_t to) | |
794 | { | |
795 | if (to > inode->i_size) | |
796 | truncate_pagecache(inode, to, inode->i_size); | |
797 | } | |
798 | ||
beaec07b BH |
799 | int exofs_write_begin(struct file *file, struct address_space *mapping, |
800 | loff_t pos, unsigned len, unsigned flags, | |
801 | struct page **pagep, void **fsdata) | |
802 | { | |
803 | int ret = 0; | |
804 | struct page *page; | |
805 | ||
806 | page = *pagep; | |
807 | if (page == NULL) { | |
808 | ret = simple_write_begin(file, mapping, pos, len, flags, pagep, | |
809 | fsdata); | |
810 | if (ret) { | |
571f7f46 | 811 | EXOFS_DBGMSG("simple_write_begin failed\n"); |
2f246fd0 | 812 | goto out; |
beaec07b BH |
813 | } |
814 | ||
815 | page = *pagep; | |
816 | } | |
817 | ||
818 | /* read modify write */ | |
819 | if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) { | |
a8f1418f BH |
820 | loff_t i_size = i_size_read(mapping->host); |
821 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
822 | size_t rlen; | |
823 | ||
824 | if (page->index < end_index) | |
825 | rlen = PAGE_CACHE_SIZE; | |
826 | else if (page->index == end_index) | |
827 | rlen = i_size & ~PAGE_CACHE_MASK; | |
828 | else | |
829 | rlen = 0; | |
830 | ||
831 | if (!rlen) { | |
832 | clear_highpage(page); | |
833 | SetPageUptodate(page); | |
834 | goto out; | |
835 | } | |
836 | ||
beaec07b BH |
837 | ret = _readpage(page, true); |
838 | if (ret) { | |
839 | /*SetPageError was done by _readpage. Is it ok?*/ | |
840 | unlock_page(page); | |
a8f1418f | 841 | EXOFS_DBGMSG("__readpage failed\n"); |
beaec07b BH |
842 | } |
843 | } | |
2f246fd0 BH |
844 | out: |
845 | if (unlikely(ret)) | |
846 | _write_failed(mapping->host, pos + len); | |
beaec07b BH |
847 | |
848 | return ret; | |
849 | } | |
850 | ||
851 | static int exofs_write_begin_export(struct file *file, | |
852 | struct address_space *mapping, | |
853 | loff_t pos, unsigned len, unsigned flags, | |
854 | struct page **pagep, void **fsdata) | |
855 | { | |
856 | *pagep = NULL; | |
857 | ||
858 | return exofs_write_begin(file, mapping, pos, len, flags, pagep, | |
859 | fsdata); | |
860 | } | |
861 | ||
efd124b9 BH |
862 | static int exofs_write_end(struct file *file, struct address_space *mapping, |
863 | loff_t pos, unsigned len, unsigned copied, | |
864 | struct page *page, void *fsdata) | |
865 | { | |
866 | struct inode *inode = mapping->host; | |
867 | /* According to comment in simple_write_end i_mutex is held */ | |
868 | loff_t i_size = inode->i_size; | |
869 | int ret; | |
870 | ||
871 | ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata); | |
2f246fd0 BH |
872 | if (unlikely(ret)) |
873 | _write_failed(inode, pos + len); | |
874 | ||
875 | /* TODO: once simple_write_end marks inode dirty remove */ | |
efd124b9 BH |
876 | if (i_size != inode->i_size) |
877 | mark_inode_dirty(inode); | |
878 | return ret; | |
879 | } | |
880 | ||
200b0700 BH |
881 | static int exofs_releasepage(struct page *page, gfp_t gfp) |
882 | { | |
883 | EXOFS_DBGMSG("page 0x%lx\n", page->index); | |
884 | WARN_ON(1); | |
85dc7878 | 885 | return 0; |
200b0700 BH |
886 | } |
887 | ||
888 | static void exofs_invalidatepage(struct page *page, unsigned long offset) | |
889 | { | |
85dc7878 | 890 | EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page->index, offset); |
200b0700 | 891 | WARN_ON(1); |
200b0700 BH |
892 | } |
893 | ||
beaec07b BH |
894 | const struct address_space_operations exofs_aops = { |
895 | .readpage = exofs_readpage, | |
896 | .readpages = exofs_readpages, | |
897 | .writepage = exofs_writepage, | |
898 | .writepages = exofs_writepages, | |
899 | .write_begin = exofs_write_begin_export, | |
efd124b9 | 900 | .write_end = exofs_write_end, |
200b0700 BH |
901 | .releasepage = exofs_releasepage, |
902 | .set_page_dirty = __set_page_dirty_nobuffers, | |
903 | .invalidatepage = exofs_invalidatepage, | |
904 | ||
905 | /* Not implemented Yet */ | |
906 | .bmap = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */ | |
907 | .direct_IO = NULL, /* TODO: Should be trivial to do */ | |
908 | ||
909 | /* With these NULL has special meaning or default is not exported */ | |
200b0700 BH |
910 | .get_xip_mem = NULL, |
911 | .migratepage = NULL, | |
912 | .launder_page = NULL, | |
913 | .is_partially_uptodate = NULL, | |
914 | .error_remove_page = NULL, | |
beaec07b BH |
915 | }; |
916 | ||
e8062719 BH |
917 | /****************************************************************************** |
918 | * INODE OPERATIONS | |
919 | *****************************************************************************/ | |
920 | ||
921 | /* | |
922 | * Test whether an inode is a fast symlink. | |
923 | */ | |
924 | static inline int exofs_inode_is_fast_symlink(struct inode *inode) | |
925 | { | |
926 | struct exofs_i_info *oi = exofs_i(inode); | |
927 | ||
928 | return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0); | |
929 | } | |
930 | ||
2f246fd0 | 931 | static int _do_truncate(struct inode *inode, loff_t newsize) |
06886a5a BH |
932 | { |
933 | struct exofs_i_info *oi = exofs_i(inode); | |
9e9db456 | 934 | struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; |
06886a5a BH |
935 | int ret; |
936 | ||
937 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
938 | ||
5bf696da | 939 | ret = ore_truncate(&sbi->layout, &oi->oc, (u64)newsize); |
2f246fd0 BH |
940 | if (likely(!ret)) |
941 | truncate_setsize(inode, newsize); | |
06886a5a | 942 | |
2f246fd0 BH |
943 | EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n", |
944 | inode->i_ino, newsize, ret); | |
06886a5a BH |
945 | return ret; |
946 | } | |
947 | ||
e8062719 | 948 | /* |
2f246fd0 BH |
949 | * Set inode attributes - update size attribute on OSD if needed, |
950 | * otherwise just call generic functions. | |
e8062719 BH |
951 | */ |
952 | int exofs_setattr(struct dentry *dentry, struct iattr *iattr) | |
953 | { | |
954 | struct inode *inode = dentry->d_inode; | |
955 | int error; | |
956 | ||
2f246fd0 BH |
957 | /* if we are about to modify an object, and it hasn't been |
958 | * created yet, wait | |
959 | */ | |
960 | error = wait_obj_created(exofs_i(inode)); | |
961 | if (unlikely(error)) | |
962 | return error; | |
963 | ||
e8062719 | 964 | error = inode_change_ok(inode, iattr); |
2f246fd0 | 965 | if (unlikely(error)) |
e8062719 BH |
966 | return error; |
967 | ||
1025774c CH |
968 | if ((iattr->ia_valid & ATTR_SIZE) && |
969 | iattr->ia_size != i_size_read(inode)) { | |
2f246fd0 BH |
970 | error = _do_truncate(inode, iattr->ia_size); |
971 | if (unlikely(error)) | |
1025774c CH |
972 | return error; |
973 | } | |
974 | ||
975 | setattr_copy(inode, iattr); | |
976 | mark_inode_dirty(inode); | |
977 | return 0; | |
e8062719 | 978 | } |
e6af00f1 | 979 | |
d9c740d2 BH |
980 | static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF( |
981 | EXOFS_APAGE_FS_DATA, | |
982 | EXOFS_ATTR_INODE_FILE_LAYOUT, | |
983 | 0); | |
984 | static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF( | |
985 | EXOFS_APAGE_FS_DATA, | |
986 | EXOFS_ATTR_INODE_DIR_LAYOUT, | |
987 | 0); | |
988 | ||
e6af00f1 | 989 | /* |
5d952b83 BH |
990 | * Read the Linux inode info from the OSD, and return it as is. In exofs the |
991 | * inode info is in an application specific page/attribute of the osd-object. | |
e6af00f1 BH |
992 | */ |
993 | static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi, | |
5d952b83 | 994 | struct exofs_fcb *inode) |
e6af00f1 BH |
995 | { |
996 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
d9c740d2 BH |
997 | struct osd_attr attrs[] = { |
998 | [0] = g_attr_inode_data, | |
999 | [1] = g_attr_inode_file_layout, | |
1000 | [2] = g_attr_inode_dir_layout, | |
d9c740d2 | 1001 | }; |
8ff660ab | 1002 | struct ore_io_state *ios; |
d9c740d2 | 1003 | struct exofs_on_disk_inode_layout *layout; |
e6af00f1 BH |
1004 | int ret; |
1005 | ||
5bf696da | 1006 | ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios); |
06886a5a | 1007 | if (unlikely(ret)) { |
8ff660ab | 1008 | EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__); |
06886a5a | 1009 | return ret; |
e6af00f1 | 1010 | } |
e6af00f1 | 1011 | |
5bf696da BH |
1012 | attrs[1].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs); |
1013 | attrs[2].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs); | |
d9c740d2 | 1014 | |
06886a5a BH |
1015 | ios->in_attr = attrs; |
1016 | ios->in_attr_len = ARRAY_SIZE(attrs); | |
e6af00f1 | 1017 | |
8ff660ab | 1018 | ret = ore_read(ios); |
96391e2b BH |
1019 | if (unlikely(ret)) { |
1020 | EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n", | |
9e9db456 | 1021 | _LLU(oi->one_comp.obj.id), ret); |
96391e2b BH |
1022 | memset(inode, 0, sizeof(*inode)); |
1023 | inode->i_mode = 0040000 | (0777 & ~022); | |
1024 | /* If object is lost on target we might as well enable it's | |
1025 | * delete. | |
1026 | */ | |
1027 | if ((ret == -ENOENT) || (ret == -EINVAL)) | |
1028 | ret = 0; | |
e6af00f1 | 1029 | goto out; |
96391e2b | 1030 | } |
e6af00f1 | 1031 | |
06886a5a | 1032 | ret = extract_attr_from_ios(ios, &attrs[0]); |
e6af00f1 | 1033 | if (ret) { |
06886a5a | 1034 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); |
e6af00f1 BH |
1035 | goto out; |
1036 | } | |
06886a5a BH |
1037 | WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE); |
1038 | memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE); | |
e6af00f1 | 1039 | |
06886a5a | 1040 | ret = extract_attr_from_ios(ios, &attrs[1]); |
d9c740d2 BH |
1041 | if (ret) { |
1042 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); | |
1043 | goto out; | |
1044 | } | |
1045 | if (attrs[1].len) { | |
1046 | layout = attrs[1].val_ptr; | |
1047 | if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) { | |
1048 | EXOFS_ERR("%s: unsupported files layout %d\n", | |
1049 | __func__, layout->gen_func); | |
1050 | ret = -ENOTSUPP; | |
1051 | goto out; | |
1052 | } | |
1053 | } | |
1054 | ||
1055 | ret = extract_attr_from_ios(ios, &attrs[2]); | |
1056 | if (ret) { | |
1057 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); | |
1058 | goto out; | |
1059 | } | |
1060 | if (attrs[2].len) { | |
1061 | layout = attrs[2].val_ptr; | |
1062 | if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) { | |
1063 | EXOFS_ERR("%s: unsupported meta-data layout %d\n", | |
1064 | __func__, layout->gen_func); | |
1065 | ret = -ENOTSUPP; | |
1066 | goto out; | |
1067 | } | |
1068 | } | |
1069 | ||
e6af00f1 | 1070 | out: |
8ff660ab | 1071 | ore_put_io_state(ios); |
e6af00f1 BH |
1072 | return ret; |
1073 | } | |
1074 | ||
9cfdc7aa BH |
1075 | static void __oi_init(struct exofs_i_info *oi) |
1076 | { | |
1077 | init_waitqueue_head(&oi->i_wq); | |
1078 | oi->i_flags = 0; | |
1079 | } | |
e6af00f1 BH |
1080 | /* |
1081 | * Fill in an inode read from the OSD and set it up for use | |
1082 | */ | |
1083 | struct inode *exofs_iget(struct super_block *sb, unsigned long ino) | |
1084 | { | |
1085 | struct exofs_i_info *oi; | |
1086 | struct exofs_fcb fcb; | |
1087 | struct inode *inode; | |
e6af00f1 BH |
1088 | int ret; |
1089 | ||
1090 | inode = iget_locked(sb, ino); | |
1091 | if (!inode) | |
1092 | return ERR_PTR(-ENOMEM); | |
1093 | if (!(inode->i_state & I_NEW)) | |
1094 | return inode; | |
1095 | oi = exofs_i(inode); | |
9cfdc7aa | 1096 | __oi_init(oi); |
5bf696da | 1097 | exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info, |
9e9db456 | 1098 | exofs_oi_objno(oi)); |
e6af00f1 BH |
1099 | |
1100 | /* read the inode from the osd */ | |
5d952b83 | 1101 | ret = exofs_get_inode(sb, oi, &fcb); |
e6af00f1 BH |
1102 | if (ret) |
1103 | goto bad_inode; | |
1104 | ||
e6af00f1 BH |
1105 | set_obj_created(oi); |
1106 | ||
1107 | /* copy stuff from on-disk struct to in-memory struct */ | |
1108 | inode->i_mode = le16_to_cpu(fcb.i_mode); | |
1109 | inode->i_uid = le32_to_cpu(fcb.i_uid); | |
1110 | inode->i_gid = le32_to_cpu(fcb.i_gid); | |
1111 | inode->i_nlink = le16_to_cpu(fcb.i_links_count); | |
1112 | inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime); | |
1113 | inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime); | |
1114 | inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime); | |
1115 | inode->i_ctime.tv_nsec = | |
1116 | inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0; | |
1117 | oi->i_commit_size = le64_to_cpu(fcb.i_size); | |
1118 | i_size_write(inode, oi->i_commit_size); | |
1119 | inode->i_blkbits = EXOFS_BLKSHIFT; | |
1120 | inode->i_generation = le32_to_cpu(fcb.i_generation); | |
1121 | ||
e6af00f1 BH |
1122 | oi->i_dir_start_lookup = 0; |
1123 | ||
1124 | if ((inode->i_nlink == 0) && (inode->i_mode == 0)) { | |
1125 | ret = -ESTALE; | |
1126 | goto bad_inode; | |
1127 | } | |
1128 | ||
1129 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
1130 | if (fcb.i_data[0]) | |
1131 | inode->i_rdev = | |
1132 | old_decode_dev(le32_to_cpu(fcb.i_data[0])); | |
1133 | else | |
1134 | inode->i_rdev = | |
1135 | new_decode_dev(le32_to_cpu(fcb.i_data[1])); | |
1136 | } else { | |
1137 | memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data)); | |
1138 | } | |
1139 | ||
66cd6cad | 1140 | inode->i_mapping->backing_dev_info = sb->s_bdi; |
e6af00f1 BH |
1141 | if (S_ISREG(inode->i_mode)) { |
1142 | inode->i_op = &exofs_file_inode_operations; | |
1143 | inode->i_fop = &exofs_file_operations; | |
1144 | inode->i_mapping->a_ops = &exofs_aops; | |
1145 | } else if (S_ISDIR(inode->i_mode)) { | |
1146 | inode->i_op = &exofs_dir_inode_operations; | |
1147 | inode->i_fop = &exofs_dir_operations; | |
1148 | inode->i_mapping->a_ops = &exofs_aops; | |
1149 | } else if (S_ISLNK(inode->i_mode)) { | |
1150 | if (exofs_inode_is_fast_symlink(inode)) | |
1151 | inode->i_op = &exofs_fast_symlink_inode_operations; | |
1152 | else { | |
1153 | inode->i_op = &exofs_symlink_inode_operations; | |
1154 | inode->i_mapping->a_ops = &exofs_aops; | |
1155 | } | |
1156 | } else { | |
1157 | inode->i_op = &exofs_special_inode_operations; | |
1158 | if (fcb.i_data[0]) | |
1159 | init_special_inode(inode, inode->i_mode, | |
1160 | old_decode_dev(le32_to_cpu(fcb.i_data[0]))); | |
1161 | else | |
1162 | init_special_inode(inode, inode->i_mode, | |
1163 | new_decode_dev(le32_to_cpu(fcb.i_data[1]))); | |
1164 | } | |
1165 | ||
1166 | unlock_new_inode(inode); | |
1167 | return inode; | |
1168 | ||
1169 | bad_inode: | |
1170 | iget_failed(inode); | |
1171 | return ERR_PTR(ret); | |
1172 | } | |
1173 | ||
1174 | int __exofs_wait_obj_created(struct exofs_i_info *oi) | |
1175 | { | |
1176 | if (!obj_created(oi)) { | |
fe2fd9ed | 1177 | EXOFS_DBGMSG("!obj_created\n"); |
e6af00f1 BH |
1178 | BUG_ON(!obj_2bcreated(oi)); |
1179 | wait_event(oi->i_wq, obj_created(oi)); | |
fe2fd9ed | 1180 | EXOFS_DBGMSG("wait_event done\n"); |
e6af00f1 BH |
1181 | } |
1182 | return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0; | |
1183 | } | |
1cea312a | 1184 | |
e6af00f1 BH |
1185 | /* |
1186 | * Callback function from exofs_new_inode(). The important thing is that we | |
1187 | * set the obj_created flag so that other methods know that the object exists on | |
1188 | * the OSD. | |
1189 | */ | |
8ff660ab | 1190 | static void create_done(struct ore_io_state *ios, void *p) |
e6af00f1 BH |
1191 | { |
1192 | struct inode *inode = p; | |
1193 | struct exofs_i_info *oi = exofs_i(inode); | |
1194 | struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; | |
1195 | int ret; | |
1196 | ||
8ff660ab BH |
1197 | ret = ore_check_io(ios, NULL); |
1198 | ore_put_io_state(ios); | |
06886a5a | 1199 | |
e6af00f1 BH |
1200 | atomic_dec(&sbi->s_curr_pending); |
1201 | ||
1202 | if (unlikely(ret)) { | |
571f7f46 | 1203 | EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx", |
9e9db456 BH |
1204 | _LLU(exofs_oi_objno(oi)), |
1205 | _LLU(oi->one_comp.obj.partition)); | |
06886a5a BH |
1206 | /*TODO: When FS is corrupted creation can fail, object already |
1207 | * exist. Get rid of this asynchronous creation, if exist | |
1208 | * increment the obj counter and try the next object. Until we | |
1209 | * succeed. All these dangling objects will be made into lost | |
1210 | * files by chkfs.exofs | |
1211 | */ | |
1212 | } | |
1213 | ||
1214 | set_obj_created(oi); | |
e6af00f1 | 1215 | |
e6af00f1 BH |
1216 | wake_up(&oi->i_wq); |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * Set up a new inode and create an object for it on the OSD | |
1221 | */ | |
1222 | struct inode *exofs_new_inode(struct inode *dir, int mode) | |
1223 | { | |
9e9db456 BH |
1224 | struct super_block *sb = dir->i_sb; |
1225 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
e6af00f1 BH |
1226 | struct inode *inode; |
1227 | struct exofs_i_info *oi; | |
8ff660ab | 1228 | struct ore_io_state *ios; |
e6af00f1 BH |
1229 | int ret; |
1230 | ||
e6af00f1 BH |
1231 | inode = new_inode(sb); |
1232 | if (!inode) | |
1233 | return ERR_PTR(-ENOMEM); | |
1234 | ||
1235 | oi = exofs_i(inode); | |
9cfdc7aa | 1236 | __oi_init(oi); |
e6af00f1 | 1237 | |
e6af00f1 BH |
1238 | set_obj_2bcreated(oi); |
1239 | ||
66cd6cad | 1240 | inode->i_mapping->backing_dev_info = sb->s_bdi; |
e00117f1 | 1241 | inode_init_owner(inode, dir, mode); |
e6af00f1 BH |
1242 | inode->i_ino = sbi->s_nextid++; |
1243 | inode->i_blkbits = EXOFS_BLKSHIFT; | |
1244 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
1245 | oi->i_commit_size = inode->i_size = 0; | |
1246 | spin_lock(&sbi->s_next_gen_lock); | |
1247 | inode->i_generation = sbi->s_next_generation++; | |
1248 | spin_unlock(&sbi->s_next_gen_lock); | |
1249 | insert_inode_hash(inode); | |
1250 | ||
5bf696da | 1251 | exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info, |
9e9db456 | 1252 | exofs_oi_objno(oi)); |
1cea312a BH |
1253 | exofs_sbi_write_stats(sbi); /* Make sure new sbi->s_nextid is on disk */ |
1254 | ||
e6af00f1 BH |
1255 | mark_inode_dirty(inode); |
1256 | ||
5bf696da | 1257 | ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios); |
06886a5a | 1258 | if (unlikely(ret)) { |
8ff660ab | 1259 | EXOFS_ERR("exofs_new_inode: ore_get_io_state failed\n"); |
06886a5a | 1260 | return ERR_PTR(ret); |
e6af00f1 BH |
1261 | } |
1262 | ||
06886a5a BH |
1263 | ios->done = create_done; |
1264 | ios->private = inode; | |
9e9db456 | 1265 | |
8ff660ab | 1266 | ret = ore_create(ios); |
e6af00f1 | 1267 | if (ret) { |
8ff660ab | 1268 | ore_put_io_state(ios); |
06886a5a | 1269 | return ERR_PTR(ret); |
e6af00f1 BH |
1270 | } |
1271 | atomic_inc(&sbi->s_curr_pending); | |
1272 | ||
1273 | return inode; | |
1274 | } | |
ba9e5e98 BH |
1275 | |
1276 | /* | |
1277 | * struct to pass two arguments to update_inode's callback | |
1278 | */ | |
1279 | struct updatei_args { | |
1280 | struct exofs_sb_info *sbi; | |
1281 | struct exofs_fcb fcb; | |
1282 | }; | |
1283 | ||
1284 | /* | |
1285 | * Callback function from exofs_update_inode(). | |
1286 | */ | |
8ff660ab | 1287 | static void updatei_done(struct ore_io_state *ios, void *p) |
ba9e5e98 BH |
1288 | { |
1289 | struct updatei_args *args = p; | |
1290 | ||
8ff660ab | 1291 | ore_put_io_state(ios); |
ba9e5e98 BH |
1292 | |
1293 | atomic_dec(&args->sbi->s_curr_pending); | |
1294 | ||
1295 | kfree(args); | |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * Write the inode to the OSD. Just fill up the struct, and set the attribute | |
1300 | * synchronously or asynchronously depending on the do_sync flag. | |
1301 | */ | |
1302 | static int exofs_update_inode(struct inode *inode, int do_sync) | |
1303 | { | |
1304 | struct exofs_i_info *oi = exofs_i(inode); | |
1305 | struct super_block *sb = inode->i_sb; | |
1306 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
8ff660ab | 1307 | struct ore_io_state *ios; |
ba9e5e98 BH |
1308 | struct osd_attr attr; |
1309 | struct exofs_fcb *fcb; | |
1310 | struct updatei_args *args; | |
1311 | int ret; | |
1312 | ||
1313 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
34ce4e7c | 1314 | if (!args) { |
571f7f46 | 1315 | EXOFS_DBGMSG("Failed kzalloc of args\n"); |
ba9e5e98 | 1316 | return -ENOMEM; |
34ce4e7c | 1317 | } |
ba9e5e98 BH |
1318 | |
1319 | fcb = &args->fcb; | |
1320 | ||
1321 | fcb->i_mode = cpu_to_le16(inode->i_mode); | |
1322 | fcb->i_uid = cpu_to_le32(inode->i_uid); | |
1323 | fcb->i_gid = cpu_to_le32(inode->i_gid); | |
1324 | fcb->i_links_count = cpu_to_le16(inode->i_nlink); | |
1325 | fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); | |
1326 | fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec); | |
1327 | fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); | |
1328 | oi->i_commit_size = i_size_read(inode); | |
1329 | fcb->i_size = cpu_to_le64(oi->i_commit_size); | |
1330 | fcb->i_generation = cpu_to_le32(inode->i_generation); | |
1331 | ||
1332 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
1333 | if (old_valid_dev(inode->i_rdev)) { | |
1334 | fcb->i_data[0] = | |
1335 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
1336 | fcb->i_data[1] = 0; | |
1337 | } else { | |
1338 | fcb->i_data[0] = 0; | |
1339 | fcb->i_data[1] = | |
1340 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
1341 | fcb->i_data[2] = 0; | |
1342 | } | |
1343 | } else | |
1344 | memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data)); | |
1345 | ||
5bf696da | 1346 | ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios); |
06886a5a | 1347 | if (unlikely(ret)) { |
8ff660ab | 1348 | EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__); |
ba9e5e98 BH |
1349 | goto free_args; |
1350 | } | |
1351 | ||
ba9e5e98 BH |
1352 | attr = g_attr_inode_data; |
1353 | attr.val_ptr = fcb; | |
06886a5a BH |
1354 | ios->out_attr_len = 1; |
1355 | ios->out_attr = &attr; | |
ba9e5e98 | 1356 | |
fe2fd9ed | 1357 | wait_obj_created(oi); |
ba9e5e98 | 1358 | |
06886a5a | 1359 | if (!do_sync) { |
ba9e5e98 | 1360 | args->sbi = sbi; |
06886a5a BH |
1361 | ios->done = updatei_done; |
1362 | ios->private = args; | |
1363 | } | |
ba9e5e98 | 1364 | |
8ff660ab | 1365 | ret = ore_write(ios); |
06886a5a | 1366 | if (!do_sync && !ret) { |
ba9e5e98 BH |
1367 | atomic_inc(&sbi->s_curr_pending); |
1368 | goto out; /* deallocation in updatei_done */ | |
1369 | } | |
1370 | ||
8ff660ab | 1371 | ore_put_io_state(ios); |
ba9e5e98 BH |
1372 | free_args: |
1373 | kfree(args); | |
1374 | out: | |
34ce4e7c BH |
1375 | EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n", |
1376 | inode->i_ino, do_sync, ret); | |
ba9e5e98 BH |
1377 | return ret; |
1378 | } | |
1379 | ||
a9185b41 | 1380 | int exofs_write_inode(struct inode *inode, struct writeback_control *wbc) |
ba9e5e98 | 1381 | { |
97178b7b NP |
1382 | /* FIXME: fix fsync and use wbc->sync_mode == WB_SYNC_ALL */ |
1383 | return exofs_update_inode(inode, 1); | |
ba9e5e98 BH |
1384 | } |
1385 | ||
1386 | /* | |
1387 | * Callback function from exofs_delete_inode() - don't have much cleaning up to | |
1388 | * do. | |
1389 | */ | |
8ff660ab | 1390 | static void delete_done(struct ore_io_state *ios, void *p) |
ba9e5e98 | 1391 | { |
06886a5a BH |
1392 | struct exofs_sb_info *sbi = p; |
1393 | ||
8ff660ab | 1394 | ore_put_io_state(ios); |
06886a5a | 1395 | |
ba9e5e98 BH |
1396 | atomic_dec(&sbi->s_curr_pending); |
1397 | } | |
1398 | ||
1399 | /* | |
1400 | * Called when the refcount of an inode reaches zero. We remove the object | |
1401 | * from the OSD here. We make sure the object was created before we try and | |
1402 | * delete it. | |
1403 | */ | |
4ec70c9b | 1404 | void exofs_evict_inode(struct inode *inode) |
ba9e5e98 BH |
1405 | { |
1406 | struct exofs_i_info *oi = exofs_i(inode); | |
1407 | struct super_block *sb = inode->i_sb; | |
1408 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
8ff660ab | 1409 | struct ore_io_state *ios; |
ba9e5e98 BH |
1410 | int ret; |
1411 | ||
1412 | truncate_inode_pages(&inode->i_data, 0); | |
1413 | ||
2f246fd0 | 1414 | /* TODO: should do better here */ |
4ec70c9b | 1415 | if (inode->i_nlink || is_bad_inode(inode)) |
ba9e5e98 BH |
1416 | goto no_delete; |
1417 | ||
ba9e5e98 | 1418 | inode->i_size = 0; |
4ec70c9b | 1419 | end_writeback(inode); |
ba9e5e98 | 1420 | |
fe2fd9ed BH |
1421 | /* if we are deleting an obj that hasn't been created yet, wait. |
1422 | * This also makes sure that create_done cannot be called with an | |
1423 | * already evicted inode. | |
1424 | */ | |
1425 | wait_obj_created(oi); | |
1426 | /* ignore the error, attempt a remove anyway */ | |
2f246fd0 BH |
1427 | |
1428 | /* Now Remove the OSD objects */ | |
5bf696da | 1429 | ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios); |
2f246fd0 | 1430 | if (unlikely(ret)) { |
8ff660ab | 1431 | EXOFS_ERR("%s: ore_get_io_state failed\n", __func__); |
2f246fd0 | 1432 | return; |
ba9e5e98 BH |
1433 | } |
1434 | ||
06886a5a BH |
1435 | ios->done = delete_done; |
1436 | ios->private = sbi; | |
9e9db456 | 1437 | |
8ff660ab | 1438 | ret = ore_remove(ios); |
ba9e5e98 | 1439 | if (ret) { |
8ff660ab BH |
1440 | EXOFS_ERR("%s: ore_remove failed\n", __func__); |
1441 | ore_put_io_state(ios); | |
ba9e5e98 BH |
1442 | return; |
1443 | } | |
1444 | atomic_inc(&sbi->s_curr_pending); | |
1445 | ||
1446 | return; | |
1447 | ||
1448 | no_delete: | |
4ec70c9b | 1449 | end_writeback(inode); |
ba9e5e98 | 1450 | } |