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