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 | ||
34 | #include <linux/writeback.h> | |
35 | #include <linux/buffer_head.h> | |
beaec07b | 36 | #include <scsi/scsi_device.h> |
e8062719 BH |
37 | |
38 | #include "exofs.h" | |
39 | ||
fe33cc1e BH |
40 | #define EXOFS_DBGMSG2(M...) do {} while (0) |
41 | ||
06886a5a BH |
42 | enum { BIO_MAX_PAGES_KMALLOC = |
43 | (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec), | |
86093aaf BH |
44 | MAX_PAGES_KMALLOC = |
45 | PAGE_SIZE / sizeof(struct page *), | |
06886a5a BH |
46 | }; |
47 | ||
beaec07b BH |
48 | struct page_collect { |
49 | struct exofs_sb_info *sbi; | |
beaec07b BH |
50 | struct inode *inode; |
51 | unsigned expected_pages; | |
06886a5a | 52 | struct exofs_io_state *ios; |
beaec07b | 53 | |
86093aaf BH |
54 | struct page **pages; |
55 | unsigned alloc_pages; | |
beaec07b BH |
56 | unsigned nr_pages; |
57 | unsigned long length; | |
58 | loff_t pg_first; /* keep 64bit also in 32-arches */ | |
59 | }; | |
60 | ||
61 | static void _pcol_init(struct page_collect *pcol, unsigned expected_pages, | |
06886a5a | 62 | struct inode *inode) |
beaec07b BH |
63 | { |
64 | struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; | |
beaec07b BH |
65 | |
66 | pcol->sbi = sbi; | |
beaec07b BH |
67 | pcol->inode = inode; |
68 | pcol->expected_pages = expected_pages; | |
69 | ||
06886a5a | 70 | pcol->ios = NULL; |
86093aaf BH |
71 | pcol->pages = NULL; |
72 | pcol->alloc_pages = 0; | |
beaec07b BH |
73 | pcol->nr_pages = 0; |
74 | pcol->length = 0; | |
75 | pcol->pg_first = -1; | |
beaec07b BH |
76 | } |
77 | ||
78 | static void _pcol_reset(struct page_collect *pcol) | |
79 | { | |
80 | pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages); | |
81 | ||
86093aaf BH |
82 | pcol->pages = NULL; |
83 | pcol->alloc_pages = 0; | |
beaec07b BH |
84 | pcol->nr_pages = 0; |
85 | pcol->length = 0; | |
86 | pcol->pg_first = -1; | |
06886a5a | 87 | pcol->ios = NULL; |
beaec07b BH |
88 | |
89 | /* this is probably the end of the loop but in writes | |
90 | * it might not end here. don't be left with nothing | |
91 | */ | |
92 | if (!pcol->expected_pages) | |
86093aaf | 93 | pcol->expected_pages = MAX_PAGES_KMALLOC; |
beaec07b BH |
94 | } |
95 | ||
96 | static int pcol_try_alloc(struct page_collect *pcol) | |
97 | { | |
86093aaf BH |
98 | unsigned pages = min_t(unsigned, pcol->expected_pages, |
99 | MAX_PAGES_KMALLOC); | |
06886a5a BH |
100 | |
101 | if (!pcol->ios) { /* First time allocate io_state */ | |
45d3abcb | 102 | int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios); |
06886a5a BH |
103 | |
104 | if (ret) | |
105 | return ret; | |
106 | } | |
beaec07b | 107 | |
86093aaf BH |
108 | /* TODO: easily support bio chaining */ |
109 | pages = min_t(unsigned, pages, | |
110 | pcol->sbi->layout.group_width * BIO_MAX_PAGES_KMALLOC); | |
111 | ||
beaec07b | 112 | for (; pages; pages >>= 1) { |
86093aaf BH |
113 | pcol->pages = kmalloc(pages * sizeof(struct page *), |
114 | GFP_KERNEL); | |
115 | if (likely(pcol->pages)) { | |
116 | pcol->alloc_pages = pages; | |
beaec07b | 117 | return 0; |
86093aaf | 118 | } |
beaec07b BH |
119 | } |
120 | ||
86093aaf | 121 | EXOFS_ERR("Failed to kmalloc expected_pages=%u\n", |
beaec07b BH |
122 | pcol->expected_pages); |
123 | return -ENOMEM; | |
124 | } | |
125 | ||
126 | static void pcol_free(struct page_collect *pcol) | |
127 | { | |
86093aaf BH |
128 | kfree(pcol->pages); |
129 | pcol->pages = NULL; | |
06886a5a BH |
130 | |
131 | if (pcol->ios) { | |
132 | exofs_put_io_state(pcol->ios); | |
133 | pcol->ios = NULL; | |
134 | } | |
beaec07b BH |
135 | } |
136 | ||
137 | static int pcol_add_page(struct page_collect *pcol, struct page *page, | |
138 | unsigned len) | |
139 | { | |
86093aaf | 140 | if (unlikely(pcol->nr_pages >= pcol->alloc_pages)) |
beaec07b BH |
141 | return -ENOMEM; |
142 | ||
86093aaf | 143 | pcol->pages[pcol->nr_pages++] = page; |
beaec07b BH |
144 | pcol->length += len; |
145 | return 0; | |
146 | } | |
147 | ||
148 | static int update_read_page(struct page *page, int ret) | |
149 | { | |
150 | if (ret == 0) { | |
151 | /* Everything is OK */ | |
152 | SetPageUptodate(page); | |
153 | if (PageError(page)) | |
154 | ClearPageError(page); | |
155 | } else if (ret == -EFAULT) { | |
156 | /* In this case we were trying to read something that wasn't on | |
157 | * disk yet - return a page full of zeroes. This should be OK, | |
158 | * because the object should be empty (if there was a write | |
159 | * before this read, the read would be waiting with the page | |
160 | * locked */ | |
161 | clear_highpage(page); | |
162 | ||
163 | SetPageUptodate(page); | |
164 | if (PageError(page)) | |
165 | ClearPageError(page); | |
166 | ret = 0; /* recovered error */ | |
167 | EXOFS_DBGMSG("recovered read error\n"); | |
168 | } else /* Error */ | |
169 | SetPageError(page); | |
170 | ||
171 | return ret; | |
172 | } | |
173 | ||
174 | static void update_write_page(struct page *page, int ret) | |
175 | { | |
176 | if (ret) { | |
177 | mapping_set_error(page->mapping, ret); | |
178 | SetPageError(page); | |
179 | } | |
180 | end_page_writeback(page); | |
181 | } | |
182 | ||
183 | /* Called at the end of reads, to optionally unlock pages and update their | |
184 | * status. | |
185 | */ | |
06886a5a | 186 | static int __readpages_done(struct page_collect *pcol, bool do_unlock) |
beaec07b | 187 | { |
beaec07b BH |
188 | int i; |
189 | u64 resid; | |
190 | u64 good_bytes; | |
191 | u64 length = 0; | |
06886a5a | 192 | int ret = exofs_check_io(pcol->ios, &resid); |
beaec07b BH |
193 | |
194 | if (likely(!ret)) | |
195 | good_bytes = pcol->length; | |
beaec07b BH |
196 | else |
197 | good_bytes = pcol->length - resid; | |
198 | ||
34ce4e7c | 199 | EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx" |
beaec07b BH |
200 | " length=0x%lx nr_pages=%u\n", |
201 | pcol->inode->i_ino, _LLU(good_bytes), pcol->length, | |
202 | pcol->nr_pages); | |
203 | ||
86093aaf BH |
204 | for (i = 0; i < pcol->nr_pages; i++) { |
205 | struct page *page = pcol->pages[i]; | |
beaec07b BH |
206 | struct inode *inode = page->mapping->host; |
207 | int page_stat; | |
208 | ||
209 | if (inode != pcol->inode) | |
210 | continue; /* osd might add more pages at end */ | |
211 | ||
212 | if (likely(length < good_bytes)) | |
213 | page_stat = 0; | |
214 | else | |
215 | page_stat = ret; | |
216 | ||
fe33cc1e | 217 | EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n", |
beaec07b BH |
218 | inode->i_ino, page->index, |
219 | page_stat ? "bad_bytes" : "good_bytes"); | |
220 | ||
221 | ret = update_read_page(page, page_stat); | |
222 | if (do_unlock) | |
223 | unlock_page(page); | |
86093aaf | 224 | length += PAGE_SIZE; |
beaec07b BH |
225 | } |
226 | ||
227 | pcol_free(pcol); | |
34ce4e7c | 228 | EXOFS_DBGMSG2("readpages_done END\n"); |
beaec07b BH |
229 | return ret; |
230 | } | |
231 | ||
232 | /* callback of async reads */ | |
06886a5a | 233 | static void readpages_done(struct exofs_io_state *ios, void *p) |
beaec07b BH |
234 | { |
235 | struct page_collect *pcol = p; | |
236 | ||
06886a5a | 237 | __readpages_done(pcol, true); |
beaec07b | 238 | atomic_dec(&pcol->sbi->s_curr_pending); |
06886a5a | 239 | kfree(pcol); |
beaec07b BH |
240 | } |
241 | ||
242 | static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw) | |
243 | { | |
beaec07b BH |
244 | int i; |
245 | ||
86093aaf BH |
246 | for (i = 0; i < pcol->nr_pages; i++) { |
247 | struct page *page = pcol->pages[i]; | |
beaec07b BH |
248 | |
249 | if (rw == READ) | |
250 | update_read_page(page, ret); | |
251 | else | |
252 | update_write_page(page, ret); | |
253 | ||
254 | unlock_page(page); | |
255 | } | |
beaec07b BH |
256 | } |
257 | ||
258 | static int read_exec(struct page_collect *pcol, bool is_sync) | |
259 | { | |
260 | struct exofs_i_info *oi = exofs_i(pcol->inode); | |
06886a5a | 261 | struct exofs_io_state *ios = pcol->ios; |
beaec07b | 262 | struct page_collect *pcol_copy = NULL; |
beaec07b BH |
263 | int ret; |
264 | ||
86093aaf | 265 | if (!pcol->pages) |
beaec07b BH |
266 | return 0; |
267 | ||
268 | /* see comment in _readpage() about sync reads */ | |
269 | WARN_ON(is_sync && (pcol->nr_pages != 1)); | |
270 | ||
86093aaf BH |
271 | ios->pages = pcol->pages; |
272 | ios->nr_pages = pcol->nr_pages; | |
06886a5a BH |
273 | ios->length = pcol->length; |
274 | ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT; | |
beaec07b BH |
275 | |
276 | if (is_sync) { | |
06886a5a BH |
277 | exofs_oi_read(oi, pcol->ios); |
278 | return __readpages_done(pcol, false); | |
beaec07b BH |
279 | } |
280 | ||
281 | pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); | |
282 | if (!pcol_copy) { | |
283 | ret = -ENOMEM; | |
284 | goto err; | |
285 | } | |
286 | ||
287 | *pcol_copy = *pcol; | |
06886a5a BH |
288 | ios->done = readpages_done; |
289 | ios->private = pcol_copy; | |
290 | ret = exofs_oi_read(oi, ios); | |
beaec07b BH |
291 | if (unlikely(ret)) |
292 | goto err; | |
293 | ||
294 | atomic_inc(&pcol->sbi->s_curr_pending); | |
295 | ||
34ce4e7c | 296 | EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n", |
06886a5a | 297 | ios->obj.id, _LLU(ios->offset), pcol->length); |
beaec07b BH |
298 | |
299 | /* pages ownership was passed to pcol_copy */ | |
300 | _pcol_reset(pcol); | |
301 | return 0; | |
302 | ||
303 | err: | |
304 | if (!is_sync) | |
305 | _unlock_pcol_pages(pcol, ret, READ); | |
06886a5a BH |
306 | |
307 | pcol_free(pcol); | |
b76a3f93 | 308 | |
beaec07b | 309 | kfree(pcol_copy); |
beaec07b BH |
310 | return ret; |
311 | } | |
312 | ||
313 | /* readpage_strip is called either directly from readpage() or by the VFS from | |
314 | * within read_cache_pages(), to add one more page to be read. It will try to | |
315 | * collect as many contiguous pages as posible. If a discontinuity is | |
316 | * encountered, or it runs out of resources, it will submit the previous segment | |
317 | * and will start a new collection. Eventually caller must submit the last | |
318 | * segment if present. | |
319 | */ | |
320 | static int readpage_strip(void *data, struct page *page) | |
321 | { | |
322 | struct page_collect *pcol = data; | |
323 | struct inode *inode = pcol->inode; | |
324 | struct exofs_i_info *oi = exofs_i(inode); | |
325 | loff_t i_size = i_size_read(inode); | |
326 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
327 | size_t len; | |
328 | int ret; | |
329 | ||
330 | /* FIXME: Just for debugging, will be removed */ | |
331 | if (PageUptodate(page)) | |
332 | EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino, | |
333 | page->index); | |
334 | ||
335 | if (page->index < end_index) | |
336 | len = PAGE_CACHE_SIZE; | |
337 | else if (page->index == end_index) | |
338 | len = i_size & ~PAGE_CACHE_MASK; | |
339 | else | |
340 | len = 0; | |
341 | ||
342 | if (!len || !obj_created(oi)) { | |
343 | /* this will be out of bounds, or doesn't exist yet. | |
344 | * Current page is cleared and the request is split | |
345 | */ | |
346 | clear_highpage(page); | |
347 | ||
348 | SetPageUptodate(page); | |
349 | if (PageError(page)) | |
350 | ClearPageError(page); | |
351 | ||
352 | unlock_page(page); | |
353 | EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page," | |
354 | " splitting\n", inode->i_ino, page->index); | |
355 | ||
356 | return read_exec(pcol, false); | |
357 | } | |
358 | ||
359 | try_again: | |
360 | ||
361 | if (unlikely(pcol->pg_first == -1)) { | |
362 | pcol->pg_first = page->index; | |
363 | } else if (unlikely((pcol->pg_first + pcol->nr_pages) != | |
364 | page->index)) { | |
365 | /* Discontinuity detected, split the request */ | |
366 | ret = read_exec(pcol, false); | |
367 | if (unlikely(ret)) | |
368 | goto fail; | |
369 | goto try_again; | |
370 | } | |
371 | ||
86093aaf | 372 | if (!pcol->pages) { |
beaec07b BH |
373 | ret = pcol_try_alloc(pcol); |
374 | if (unlikely(ret)) | |
375 | goto fail; | |
376 | } | |
377 | ||
378 | if (len != PAGE_CACHE_SIZE) | |
379 | zero_user(page, len, PAGE_CACHE_SIZE - len); | |
380 | ||
fe33cc1e | 381 | EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n", |
beaec07b BH |
382 | inode->i_ino, page->index, len); |
383 | ||
384 | ret = pcol_add_page(pcol, page, len); | |
385 | if (ret) { | |
fe33cc1e | 386 | EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p " |
beaec07b BH |
387 | "this_len=0x%zx nr_pages=%u length=0x%lx\n", |
388 | page, len, pcol->nr_pages, pcol->length); | |
389 | ||
390 | /* split the request, and start again with current page */ | |
391 | ret = read_exec(pcol, false); | |
392 | if (unlikely(ret)) | |
393 | goto fail; | |
394 | ||
395 | goto try_again; | |
396 | } | |
397 | ||
398 | return 0; | |
399 | ||
400 | fail: | |
401 | /* SetPageError(page); ??? */ | |
402 | unlock_page(page); | |
403 | return ret; | |
404 | } | |
405 | ||
406 | static int exofs_readpages(struct file *file, struct address_space *mapping, | |
407 | struct list_head *pages, unsigned nr_pages) | |
408 | { | |
409 | struct page_collect pcol; | |
410 | int ret; | |
411 | ||
412 | _pcol_init(&pcol, nr_pages, mapping->host); | |
413 | ||
414 | ret = read_cache_pages(mapping, pages, readpage_strip, &pcol); | |
415 | if (ret) { | |
416 | EXOFS_ERR("read_cache_pages => %d\n", ret); | |
417 | return ret; | |
418 | } | |
419 | ||
420 | return read_exec(&pcol, false); | |
421 | } | |
422 | ||
423 | static int _readpage(struct page *page, bool is_sync) | |
424 | { | |
425 | struct page_collect pcol; | |
426 | int ret; | |
427 | ||
428 | _pcol_init(&pcol, 1, page->mapping->host); | |
429 | ||
06886a5a BH |
430 | /* readpage_strip might call read_exec(,is_sync==false) at several |
431 | * places but not if we have a single page. | |
beaec07b BH |
432 | */ |
433 | ret = readpage_strip(&pcol, page); | |
434 | if (ret) { | |
435 | EXOFS_ERR("_readpage => %d\n", ret); | |
436 | return ret; | |
437 | } | |
438 | ||
439 | return read_exec(&pcol, is_sync); | |
440 | } | |
441 | ||
442 | /* | |
443 | * We don't need the file | |
444 | */ | |
445 | static int exofs_readpage(struct file *file, struct page *page) | |
446 | { | |
447 | return _readpage(page, false); | |
448 | } | |
449 | ||
06886a5a BH |
450 | /* Callback for osd_write. All writes are asynchronous */ |
451 | static void writepages_done(struct exofs_io_state *ios, void *p) | |
beaec07b BH |
452 | { |
453 | struct page_collect *pcol = p; | |
beaec07b BH |
454 | int i; |
455 | u64 resid; | |
456 | u64 good_bytes; | |
457 | u64 length = 0; | |
06886a5a | 458 | int ret = exofs_check_io(ios, &resid); |
beaec07b | 459 | |
beaec07b BH |
460 | atomic_dec(&pcol->sbi->s_curr_pending); |
461 | ||
462 | if (likely(!ret)) | |
463 | good_bytes = pcol->length; | |
beaec07b BH |
464 | else |
465 | good_bytes = pcol->length - resid; | |
466 | ||
34ce4e7c | 467 | EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx" |
beaec07b BH |
468 | " length=0x%lx nr_pages=%u\n", |
469 | pcol->inode->i_ino, _LLU(good_bytes), pcol->length, | |
470 | pcol->nr_pages); | |
471 | ||
86093aaf BH |
472 | for (i = 0; i < pcol->nr_pages; i++) { |
473 | struct page *page = pcol->pages[i]; | |
beaec07b BH |
474 | struct inode *inode = page->mapping->host; |
475 | int page_stat; | |
476 | ||
477 | if (inode != pcol->inode) | |
478 | continue; /* osd might add more pages to a bio */ | |
479 | ||
480 | if (likely(length < good_bytes)) | |
481 | page_stat = 0; | |
482 | else | |
483 | page_stat = ret; | |
484 | ||
485 | update_write_page(page, page_stat); | |
486 | unlock_page(page); | |
fe33cc1e | 487 | EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n", |
beaec07b BH |
488 | inode->i_ino, page->index, page_stat); |
489 | ||
86093aaf | 490 | length += PAGE_SIZE; |
beaec07b BH |
491 | } |
492 | ||
493 | pcol_free(pcol); | |
494 | kfree(pcol); | |
34ce4e7c | 495 | EXOFS_DBGMSG2("writepages_done END\n"); |
beaec07b BH |
496 | } |
497 | ||
498 | static int write_exec(struct page_collect *pcol) | |
499 | { | |
500 | struct exofs_i_info *oi = exofs_i(pcol->inode); | |
06886a5a | 501 | struct exofs_io_state *ios = pcol->ios; |
beaec07b | 502 | struct page_collect *pcol_copy = NULL; |
beaec07b BH |
503 | int ret; |
504 | ||
86093aaf | 505 | if (!pcol->pages) |
beaec07b BH |
506 | return 0; |
507 | ||
beaec07b BH |
508 | pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); |
509 | if (!pcol_copy) { | |
510 | EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n"); | |
511 | ret = -ENOMEM; | |
512 | goto err; | |
513 | } | |
514 | ||
515 | *pcol_copy = *pcol; | |
516 | ||
86093aaf BH |
517 | ios->pages = pcol_copy->pages; |
518 | ios->nr_pages = pcol_copy->nr_pages; | |
06886a5a BH |
519 | ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT; |
520 | ios->length = pcol_copy->length; | |
521 | ios->done = writepages_done; | |
522 | ios->private = pcol_copy; | |
523 | ||
524 | ret = exofs_oi_write(oi, ios); | |
beaec07b | 525 | if (unlikely(ret)) { |
06886a5a | 526 | EXOFS_ERR("write_exec: exofs_oi_write() Faild\n"); |
beaec07b BH |
527 | goto err; |
528 | } | |
529 | ||
530 | atomic_inc(&pcol->sbi->s_curr_pending); | |
34ce4e7c | 531 | EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n", |
06886a5a | 532 | pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset), |
beaec07b BH |
533 | pcol->length); |
534 | /* pages ownership was passed to pcol_copy */ | |
535 | _pcol_reset(pcol); | |
536 | return 0; | |
537 | ||
538 | err: | |
539 | _unlock_pcol_pages(pcol, ret, WRITE); | |
06886a5a | 540 | pcol_free(pcol); |
beaec07b | 541 | kfree(pcol_copy); |
06886a5a | 542 | |
beaec07b BH |
543 | return ret; |
544 | } | |
545 | ||
546 | /* writepage_strip is called either directly from writepage() or by the VFS from | |
547 | * within write_cache_pages(), to add one more page to be written to storage. | |
548 | * It will try to collect as many contiguous pages as possible. If a | |
549 | * discontinuity is encountered or it runs out of resources it will submit the | |
550 | * previous segment and will start a new collection. | |
551 | * Eventually caller must submit the last segment if present. | |
552 | */ | |
553 | static int writepage_strip(struct page *page, | |
554 | struct writeback_control *wbc_unused, void *data) | |
555 | { | |
556 | struct page_collect *pcol = data; | |
557 | struct inode *inode = pcol->inode; | |
558 | struct exofs_i_info *oi = exofs_i(inode); | |
559 | loff_t i_size = i_size_read(inode); | |
560 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
561 | size_t len; | |
562 | int ret; | |
563 | ||
564 | BUG_ON(!PageLocked(page)); | |
565 | ||
566 | ret = wait_obj_created(oi); | |
567 | if (unlikely(ret)) | |
568 | goto fail; | |
569 | ||
570 | if (page->index < end_index) | |
571 | /* in this case, the page is within the limits of the file */ | |
572 | len = PAGE_CACHE_SIZE; | |
573 | else { | |
574 | len = i_size & ~PAGE_CACHE_MASK; | |
575 | ||
576 | if (page->index > end_index || !len) { | |
577 | /* in this case, the page is outside the limits | |
578 | * (truncate in progress) | |
579 | */ | |
580 | ret = write_exec(pcol); | |
581 | if (unlikely(ret)) | |
582 | goto fail; | |
583 | if (PageError(page)) | |
584 | ClearPageError(page); | |
585 | unlock_page(page); | |
06886a5a BH |
586 | EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) " |
587 | "outside the limits\n", | |
588 | inode->i_ino, page->index); | |
beaec07b BH |
589 | return 0; |
590 | } | |
591 | } | |
592 | ||
593 | try_again: | |
594 | ||
595 | if (unlikely(pcol->pg_first == -1)) { | |
596 | pcol->pg_first = page->index; | |
597 | } else if (unlikely((pcol->pg_first + pcol->nr_pages) != | |
598 | page->index)) { | |
599 | /* Discontinuity detected, split the request */ | |
600 | ret = write_exec(pcol); | |
601 | if (unlikely(ret)) | |
602 | goto fail; | |
06886a5a BH |
603 | |
604 | EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n", | |
605 | inode->i_ino, page->index); | |
beaec07b BH |
606 | goto try_again; |
607 | } | |
608 | ||
86093aaf | 609 | if (!pcol->pages) { |
beaec07b BH |
610 | ret = pcol_try_alloc(pcol); |
611 | if (unlikely(ret)) | |
612 | goto fail; | |
613 | } | |
614 | ||
fe33cc1e | 615 | EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n", |
beaec07b BH |
616 | inode->i_ino, page->index, len); |
617 | ||
618 | ret = pcol_add_page(pcol, page, len); | |
619 | if (unlikely(ret)) { | |
34ce4e7c | 620 | EXOFS_DBGMSG2("Failed pcol_add_page " |
beaec07b BH |
621 | "nr_pages=%u total_length=0x%lx\n", |
622 | pcol->nr_pages, pcol->length); | |
623 | ||
624 | /* split the request, next loop will start again */ | |
625 | ret = write_exec(pcol); | |
626 | if (unlikely(ret)) { | |
627 | EXOFS_DBGMSG("write_exec faild => %d", ret); | |
628 | goto fail; | |
629 | } | |
630 | ||
631 | goto try_again; | |
632 | } | |
633 | ||
634 | BUG_ON(PageWriteback(page)); | |
635 | set_page_writeback(page); | |
636 | ||
637 | return 0; | |
638 | ||
639 | fail: | |
06886a5a BH |
640 | EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n", |
641 | inode->i_ino, page->index, ret); | |
beaec07b BH |
642 | set_bit(AS_EIO, &page->mapping->flags); |
643 | unlock_page(page); | |
644 | return ret; | |
645 | } | |
646 | ||
647 | static int exofs_writepages(struct address_space *mapping, | |
648 | struct writeback_control *wbc) | |
649 | { | |
650 | struct page_collect pcol; | |
651 | long start, end, expected_pages; | |
652 | int ret; | |
653 | ||
654 | start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
655 | end = (wbc->range_end == LLONG_MAX) ? | |
656 | start + mapping->nrpages : | |
657 | wbc->range_end >> PAGE_CACHE_SHIFT; | |
658 | ||
659 | if (start || end) | |
06886a5a | 660 | expected_pages = end - start + 1; |
beaec07b BH |
661 | else |
662 | expected_pages = mapping->nrpages; | |
663 | ||
06886a5a BH |
664 | if (expected_pages < 32L) |
665 | expected_pages = 32L; | |
666 | ||
34ce4e7c | 667 | EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx " |
06886a5a | 668 | "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n", |
beaec07b | 669 | mapping->host->i_ino, wbc->range_start, wbc->range_end, |
06886a5a | 670 | mapping->nrpages, start, end, expected_pages); |
beaec07b BH |
671 | |
672 | _pcol_init(&pcol, expected_pages, mapping->host); | |
673 | ||
674 | ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol); | |
675 | if (ret) { | |
676 | EXOFS_ERR("write_cache_pages => %d\n", ret); | |
677 | return ret; | |
678 | } | |
679 | ||
680 | return write_exec(&pcol); | |
681 | } | |
682 | ||
683 | static int exofs_writepage(struct page *page, struct writeback_control *wbc) | |
684 | { | |
685 | struct page_collect pcol; | |
686 | int ret; | |
687 | ||
688 | _pcol_init(&pcol, 1, page->mapping->host); | |
689 | ||
690 | ret = writepage_strip(page, NULL, &pcol); | |
691 | if (ret) { | |
692 | EXOFS_ERR("exofs_writepage => %d\n", ret); | |
693 | return ret; | |
694 | } | |
695 | ||
696 | return write_exec(&pcol); | |
697 | } | |
698 | ||
699 | int exofs_write_begin(struct file *file, struct address_space *mapping, | |
700 | loff_t pos, unsigned len, unsigned flags, | |
701 | struct page **pagep, void **fsdata) | |
702 | { | |
703 | int ret = 0; | |
704 | struct page *page; | |
705 | ||
706 | page = *pagep; | |
707 | if (page == NULL) { | |
708 | ret = simple_write_begin(file, mapping, pos, len, flags, pagep, | |
709 | fsdata); | |
710 | if (ret) { | |
711 | EXOFS_DBGMSG("simple_write_begin faild\n"); | |
712 | return ret; | |
713 | } | |
714 | ||
715 | page = *pagep; | |
716 | } | |
717 | ||
718 | /* read modify write */ | |
719 | if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) { | |
720 | ret = _readpage(page, true); | |
721 | if (ret) { | |
722 | /*SetPageError was done by _readpage. Is it ok?*/ | |
723 | unlock_page(page); | |
724 | EXOFS_DBGMSG("__readpage_filler faild\n"); | |
725 | } | |
726 | } | |
727 | ||
728 | return ret; | |
729 | } | |
730 | ||
731 | static int exofs_write_begin_export(struct file *file, | |
732 | struct address_space *mapping, | |
733 | loff_t pos, unsigned len, unsigned flags, | |
734 | struct page **pagep, void **fsdata) | |
735 | { | |
736 | *pagep = NULL; | |
737 | ||
738 | return exofs_write_begin(file, mapping, pos, len, flags, pagep, | |
739 | fsdata); | |
740 | } | |
741 | ||
efd124b9 BH |
742 | static int exofs_write_end(struct file *file, struct address_space *mapping, |
743 | loff_t pos, unsigned len, unsigned copied, | |
744 | struct page *page, void *fsdata) | |
745 | { | |
746 | struct inode *inode = mapping->host; | |
747 | /* According to comment in simple_write_end i_mutex is held */ | |
748 | loff_t i_size = inode->i_size; | |
749 | int ret; | |
750 | ||
751 | ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata); | |
752 | if (i_size != inode->i_size) | |
753 | mark_inode_dirty(inode); | |
754 | return ret; | |
755 | } | |
756 | ||
beaec07b BH |
757 | const struct address_space_operations exofs_aops = { |
758 | .readpage = exofs_readpage, | |
759 | .readpages = exofs_readpages, | |
760 | .writepage = exofs_writepage, | |
761 | .writepages = exofs_writepages, | |
762 | .write_begin = exofs_write_begin_export, | |
efd124b9 | 763 | .write_end = exofs_write_end, |
beaec07b BH |
764 | }; |
765 | ||
e8062719 BH |
766 | /****************************************************************************** |
767 | * INODE OPERATIONS | |
768 | *****************************************************************************/ | |
769 | ||
770 | /* | |
771 | * Test whether an inode is a fast symlink. | |
772 | */ | |
773 | static inline int exofs_inode_is_fast_symlink(struct inode *inode) | |
774 | { | |
775 | struct exofs_i_info *oi = exofs_i(inode); | |
776 | ||
777 | return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0); | |
778 | } | |
779 | ||
780 | /* | |
781 | * get_block_t - Fill in a buffer_head | |
782 | * An OSD takes care of block allocation so we just fake an allocation by | |
783 | * putting in the inode's sector_t in the buffer_head. | |
784 | * TODO: What about the case of create==0 and @iblock does not exist in the | |
785 | * object? | |
786 | */ | |
787 | static int exofs_get_block(struct inode *inode, sector_t iblock, | |
788 | struct buffer_head *bh_result, int create) | |
789 | { | |
790 | map_bh(bh_result, inode->i_sb, iblock); | |
791 | return 0; | |
792 | } | |
793 | ||
794 | const struct osd_attr g_attr_logical_length = ATTR_DEF( | |
795 | OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8); | |
796 | ||
06886a5a BH |
797 | static int _do_truncate(struct inode *inode) |
798 | { | |
799 | struct exofs_i_info *oi = exofs_i(inode); | |
800 | loff_t isize = i_size_read(inode); | |
801 | int ret; | |
802 | ||
803 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
804 | ||
805 | nobh_truncate_page(inode->i_mapping, isize, exofs_get_block); | |
806 | ||
807 | ret = exofs_oi_truncate(oi, (u64)isize); | |
808 | EXOFS_DBGMSG("(0x%lx) size=0x%llx\n", inode->i_ino, isize); | |
809 | return ret; | |
810 | } | |
811 | ||
e8062719 BH |
812 | /* |
813 | * Truncate a file to the specified size - all we have to do is set the size | |
814 | * attribute. We make sure the object exists first. | |
815 | */ | |
816 | void exofs_truncate(struct inode *inode) | |
817 | { | |
e8062719 | 818 | struct exofs_i_info *oi = exofs_i(inode); |
e8062719 BH |
819 | int ret; |
820 | ||
821 | if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) | |
822 | || S_ISLNK(inode->i_mode))) | |
823 | return; | |
824 | if (exofs_inode_is_fast_symlink(inode)) | |
825 | return; | |
826 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
827 | return; | |
e8062719 BH |
828 | |
829 | /* if we are about to truncate an object, and it hasn't been | |
830 | * created yet, wait | |
831 | */ | |
832 | if (unlikely(wait_obj_created(oi))) | |
833 | goto fail; | |
834 | ||
06886a5a | 835 | ret = _do_truncate(inode); |
e8062719 BH |
836 | if (ret) |
837 | goto fail; | |
838 | ||
839 | out: | |
840 | mark_inode_dirty(inode); | |
841 | return; | |
842 | fail: | |
843 | make_bad_inode(inode); | |
844 | goto out; | |
845 | } | |
846 | ||
847 | /* | |
848 | * Set inode attributes - just call generic functions. | |
849 | */ | |
850 | int exofs_setattr(struct dentry *dentry, struct iattr *iattr) | |
851 | { | |
852 | struct inode *inode = dentry->d_inode; | |
853 | int error; | |
854 | ||
855 | error = inode_change_ok(inode, iattr); | |
856 | if (error) | |
857 | return error; | |
858 | ||
859 | error = inode_setattr(inode, iattr); | |
860 | return error; | |
861 | } | |
e6af00f1 | 862 | |
d9c740d2 BH |
863 | static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF( |
864 | EXOFS_APAGE_FS_DATA, | |
865 | EXOFS_ATTR_INODE_FILE_LAYOUT, | |
866 | 0); | |
867 | static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF( | |
868 | EXOFS_APAGE_FS_DATA, | |
869 | EXOFS_ATTR_INODE_DIR_LAYOUT, | |
870 | 0); | |
871 | ||
e6af00f1 | 872 | /* |
5d952b83 BH |
873 | * Read the Linux inode info from the OSD, and return it as is. In exofs the |
874 | * inode info is in an application specific page/attribute of the osd-object. | |
e6af00f1 BH |
875 | */ |
876 | static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi, | |
5d952b83 | 877 | struct exofs_fcb *inode) |
e6af00f1 BH |
878 | { |
879 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
d9c740d2 BH |
880 | struct osd_attr attrs[] = { |
881 | [0] = g_attr_inode_data, | |
882 | [1] = g_attr_inode_file_layout, | |
883 | [2] = g_attr_inode_dir_layout, | |
d9c740d2 | 884 | }; |
06886a5a | 885 | struct exofs_io_state *ios; |
d9c740d2 | 886 | struct exofs_on_disk_inode_layout *layout; |
e6af00f1 BH |
887 | int ret; |
888 | ||
45d3abcb | 889 | ret = exofs_get_io_state(&sbi->layout, &ios); |
06886a5a BH |
890 | if (unlikely(ret)) { |
891 | EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__); | |
892 | return ret; | |
e6af00f1 | 893 | } |
e6af00f1 | 894 | |
06886a5a BH |
895 | ios->obj.id = exofs_oi_objno(oi); |
896 | exofs_make_credential(oi->i_cred, &ios->obj); | |
897 | ios->cred = oi->i_cred; | |
e6af00f1 | 898 | |
d9c740d2 BH |
899 | attrs[1].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs); |
900 | attrs[2].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs); | |
901 | ||
06886a5a BH |
902 | ios->in_attr = attrs; |
903 | ios->in_attr_len = ARRAY_SIZE(attrs); | |
e6af00f1 | 904 | |
06886a5a | 905 | ret = exofs_sbi_read(ios); |
96391e2b BH |
906 | if (unlikely(ret)) { |
907 | EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n", | |
908 | _LLU(ios->obj.id), ret); | |
909 | memset(inode, 0, sizeof(*inode)); | |
910 | inode->i_mode = 0040000 | (0777 & ~022); | |
911 | /* If object is lost on target we might as well enable it's | |
912 | * delete. | |
913 | */ | |
914 | if ((ret == -ENOENT) || (ret == -EINVAL)) | |
915 | ret = 0; | |
e6af00f1 | 916 | goto out; |
96391e2b | 917 | } |
e6af00f1 | 918 | |
06886a5a | 919 | ret = extract_attr_from_ios(ios, &attrs[0]); |
e6af00f1 | 920 | if (ret) { |
06886a5a | 921 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); |
e6af00f1 BH |
922 | goto out; |
923 | } | |
06886a5a BH |
924 | WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE); |
925 | memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE); | |
e6af00f1 | 926 | |
06886a5a | 927 | ret = extract_attr_from_ios(ios, &attrs[1]); |
d9c740d2 BH |
928 | if (ret) { |
929 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); | |
930 | goto out; | |
931 | } | |
932 | if (attrs[1].len) { | |
933 | layout = attrs[1].val_ptr; | |
934 | if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) { | |
935 | EXOFS_ERR("%s: unsupported files layout %d\n", | |
936 | __func__, layout->gen_func); | |
937 | ret = -ENOTSUPP; | |
938 | goto out; | |
939 | } | |
940 | } | |
941 | ||
942 | ret = extract_attr_from_ios(ios, &attrs[2]); | |
943 | if (ret) { | |
944 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); | |
945 | goto out; | |
946 | } | |
947 | if (attrs[2].len) { | |
948 | layout = attrs[2].val_ptr; | |
949 | if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) { | |
950 | EXOFS_ERR("%s: unsupported meta-data layout %d\n", | |
951 | __func__, layout->gen_func); | |
952 | ret = -ENOTSUPP; | |
953 | goto out; | |
954 | } | |
955 | } | |
956 | ||
e6af00f1 | 957 | out: |
06886a5a | 958 | exofs_put_io_state(ios); |
e6af00f1 BH |
959 | return ret; |
960 | } | |
961 | ||
9cfdc7aa BH |
962 | static void __oi_init(struct exofs_i_info *oi) |
963 | { | |
964 | init_waitqueue_head(&oi->i_wq); | |
965 | oi->i_flags = 0; | |
966 | } | |
e6af00f1 BH |
967 | /* |
968 | * Fill in an inode read from the OSD and set it up for use | |
969 | */ | |
970 | struct inode *exofs_iget(struct super_block *sb, unsigned long ino) | |
971 | { | |
972 | struct exofs_i_info *oi; | |
973 | struct exofs_fcb fcb; | |
974 | struct inode *inode; | |
e6af00f1 BH |
975 | int ret; |
976 | ||
977 | inode = iget_locked(sb, ino); | |
978 | if (!inode) | |
979 | return ERR_PTR(-ENOMEM); | |
980 | if (!(inode->i_state & I_NEW)) | |
981 | return inode; | |
982 | oi = exofs_i(inode); | |
9cfdc7aa | 983 | __oi_init(oi); |
e6af00f1 BH |
984 | |
985 | /* read the inode from the osd */ | |
5d952b83 | 986 | ret = exofs_get_inode(sb, oi, &fcb); |
e6af00f1 BH |
987 | if (ret) |
988 | goto bad_inode; | |
989 | ||
e6af00f1 BH |
990 | set_obj_created(oi); |
991 | ||
992 | /* copy stuff from on-disk struct to in-memory struct */ | |
993 | inode->i_mode = le16_to_cpu(fcb.i_mode); | |
994 | inode->i_uid = le32_to_cpu(fcb.i_uid); | |
995 | inode->i_gid = le32_to_cpu(fcb.i_gid); | |
996 | inode->i_nlink = le16_to_cpu(fcb.i_links_count); | |
997 | inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime); | |
998 | inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime); | |
999 | inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime); | |
1000 | inode->i_ctime.tv_nsec = | |
1001 | inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0; | |
1002 | oi->i_commit_size = le64_to_cpu(fcb.i_size); | |
1003 | i_size_write(inode, oi->i_commit_size); | |
1004 | inode->i_blkbits = EXOFS_BLKSHIFT; | |
1005 | inode->i_generation = le32_to_cpu(fcb.i_generation); | |
1006 | ||
e6af00f1 BH |
1007 | oi->i_dir_start_lookup = 0; |
1008 | ||
1009 | if ((inode->i_nlink == 0) && (inode->i_mode == 0)) { | |
1010 | ret = -ESTALE; | |
1011 | goto bad_inode; | |
1012 | } | |
1013 | ||
1014 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
1015 | if (fcb.i_data[0]) | |
1016 | inode->i_rdev = | |
1017 | old_decode_dev(le32_to_cpu(fcb.i_data[0])); | |
1018 | else | |
1019 | inode->i_rdev = | |
1020 | new_decode_dev(le32_to_cpu(fcb.i_data[1])); | |
1021 | } else { | |
1022 | memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data)); | |
1023 | } | |
1024 | ||
1025 | if (S_ISREG(inode->i_mode)) { | |
1026 | inode->i_op = &exofs_file_inode_operations; | |
1027 | inode->i_fop = &exofs_file_operations; | |
1028 | inode->i_mapping->a_ops = &exofs_aops; | |
1029 | } else if (S_ISDIR(inode->i_mode)) { | |
1030 | inode->i_op = &exofs_dir_inode_operations; | |
1031 | inode->i_fop = &exofs_dir_operations; | |
1032 | inode->i_mapping->a_ops = &exofs_aops; | |
1033 | } else if (S_ISLNK(inode->i_mode)) { | |
1034 | if (exofs_inode_is_fast_symlink(inode)) | |
1035 | inode->i_op = &exofs_fast_symlink_inode_operations; | |
1036 | else { | |
1037 | inode->i_op = &exofs_symlink_inode_operations; | |
1038 | inode->i_mapping->a_ops = &exofs_aops; | |
1039 | } | |
1040 | } else { | |
1041 | inode->i_op = &exofs_special_inode_operations; | |
1042 | if (fcb.i_data[0]) | |
1043 | init_special_inode(inode, inode->i_mode, | |
1044 | old_decode_dev(le32_to_cpu(fcb.i_data[0]))); | |
1045 | else | |
1046 | init_special_inode(inode, inode->i_mode, | |
1047 | new_decode_dev(le32_to_cpu(fcb.i_data[1]))); | |
1048 | } | |
1049 | ||
1050 | unlock_new_inode(inode); | |
1051 | return inode; | |
1052 | ||
1053 | bad_inode: | |
1054 | iget_failed(inode); | |
1055 | return ERR_PTR(ret); | |
1056 | } | |
1057 | ||
1058 | int __exofs_wait_obj_created(struct exofs_i_info *oi) | |
1059 | { | |
1060 | if (!obj_created(oi)) { | |
1061 | BUG_ON(!obj_2bcreated(oi)); | |
1062 | wait_event(oi->i_wq, obj_created(oi)); | |
1063 | } | |
1064 | return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0; | |
1065 | } | |
1066 | /* | |
1067 | * Callback function from exofs_new_inode(). The important thing is that we | |
1068 | * set the obj_created flag so that other methods know that the object exists on | |
1069 | * the OSD. | |
1070 | */ | |
06886a5a | 1071 | static void create_done(struct exofs_io_state *ios, void *p) |
e6af00f1 BH |
1072 | { |
1073 | struct inode *inode = p; | |
1074 | struct exofs_i_info *oi = exofs_i(inode); | |
1075 | struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; | |
1076 | int ret; | |
1077 | ||
06886a5a BH |
1078 | ret = exofs_check_io(ios, NULL); |
1079 | exofs_put_io_state(ios); | |
1080 | ||
e6af00f1 BH |
1081 | atomic_dec(&sbi->s_curr_pending); |
1082 | ||
1083 | if (unlikely(ret)) { | |
1084 | EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx", | |
45d3abcb | 1085 | _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid)); |
06886a5a BH |
1086 | /*TODO: When FS is corrupted creation can fail, object already |
1087 | * exist. Get rid of this asynchronous creation, if exist | |
1088 | * increment the obj counter and try the next object. Until we | |
1089 | * succeed. All these dangling objects will be made into lost | |
1090 | * files by chkfs.exofs | |
1091 | */ | |
1092 | } | |
1093 | ||
1094 | set_obj_created(oi); | |
e6af00f1 BH |
1095 | |
1096 | atomic_dec(&inode->i_count); | |
1097 | wake_up(&oi->i_wq); | |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * Set up a new inode and create an object for it on the OSD | |
1102 | */ | |
1103 | struct inode *exofs_new_inode(struct inode *dir, int mode) | |
1104 | { | |
1105 | struct super_block *sb; | |
1106 | struct inode *inode; | |
1107 | struct exofs_i_info *oi; | |
1108 | struct exofs_sb_info *sbi; | |
06886a5a | 1109 | struct exofs_io_state *ios; |
e6af00f1 BH |
1110 | int ret; |
1111 | ||
1112 | sb = dir->i_sb; | |
1113 | inode = new_inode(sb); | |
1114 | if (!inode) | |
1115 | return ERR_PTR(-ENOMEM); | |
1116 | ||
1117 | oi = exofs_i(inode); | |
9cfdc7aa | 1118 | __oi_init(oi); |
e6af00f1 | 1119 | |
e6af00f1 BH |
1120 | set_obj_2bcreated(oi); |
1121 | ||
1122 | sbi = sb->s_fs_info; | |
1123 | ||
1124 | sb->s_dirt = 1; | |
1125 | inode->i_uid = current->cred->fsuid; | |
1126 | if (dir->i_mode & S_ISGID) { | |
1127 | inode->i_gid = dir->i_gid; | |
1128 | if (S_ISDIR(mode)) | |
1129 | mode |= S_ISGID; | |
1130 | } else { | |
1131 | inode->i_gid = current->cred->fsgid; | |
1132 | } | |
1133 | inode->i_mode = mode; | |
1134 | ||
1135 | inode->i_ino = sbi->s_nextid++; | |
1136 | inode->i_blkbits = EXOFS_BLKSHIFT; | |
1137 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
1138 | oi->i_commit_size = inode->i_size = 0; | |
1139 | spin_lock(&sbi->s_next_gen_lock); | |
1140 | inode->i_generation = sbi->s_next_generation++; | |
1141 | spin_unlock(&sbi->s_next_gen_lock); | |
1142 | insert_inode_hash(inode); | |
1143 | ||
1144 | mark_inode_dirty(inode); | |
1145 | ||
45d3abcb | 1146 | ret = exofs_get_io_state(&sbi->layout, &ios); |
06886a5a BH |
1147 | if (unlikely(ret)) { |
1148 | EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n"); | |
1149 | return ERR_PTR(ret); | |
e6af00f1 BH |
1150 | } |
1151 | ||
06886a5a BH |
1152 | ios->obj.id = exofs_oi_objno(oi); |
1153 | exofs_make_credential(oi->i_cred, &ios->obj); | |
e6af00f1 BH |
1154 | |
1155 | /* increment the refcount so that the inode will still be around when we | |
1156 | * reach the callback | |
1157 | */ | |
1158 | atomic_inc(&inode->i_count); | |
1159 | ||
06886a5a BH |
1160 | ios->done = create_done; |
1161 | ios->private = inode; | |
1162 | ios->cred = oi->i_cred; | |
1163 | ret = exofs_sbi_create(ios); | |
e6af00f1 BH |
1164 | if (ret) { |
1165 | atomic_dec(&inode->i_count); | |
06886a5a BH |
1166 | exofs_put_io_state(ios); |
1167 | return ERR_PTR(ret); | |
e6af00f1 BH |
1168 | } |
1169 | atomic_inc(&sbi->s_curr_pending); | |
1170 | ||
1171 | return inode; | |
1172 | } | |
ba9e5e98 BH |
1173 | |
1174 | /* | |
1175 | * struct to pass two arguments to update_inode's callback | |
1176 | */ | |
1177 | struct updatei_args { | |
1178 | struct exofs_sb_info *sbi; | |
1179 | struct exofs_fcb fcb; | |
1180 | }; | |
1181 | ||
1182 | /* | |
1183 | * Callback function from exofs_update_inode(). | |
1184 | */ | |
06886a5a | 1185 | static void updatei_done(struct exofs_io_state *ios, void *p) |
ba9e5e98 BH |
1186 | { |
1187 | struct updatei_args *args = p; | |
1188 | ||
06886a5a | 1189 | exofs_put_io_state(ios); |
ba9e5e98 BH |
1190 | |
1191 | atomic_dec(&args->sbi->s_curr_pending); | |
1192 | ||
1193 | kfree(args); | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * Write the inode to the OSD. Just fill up the struct, and set the attribute | |
1198 | * synchronously or asynchronously depending on the do_sync flag. | |
1199 | */ | |
1200 | static int exofs_update_inode(struct inode *inode, int do_sync) | |
1201 | { | |
1202 | struct exofs_i_info *oi = exofs_i(inode); | |
1203 | struct super_block *sb = inode->i_sb; | |
1204 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
06886a5a | 1205 | struct exofs_io_state *ios; |
ba9e5e98 BH |
1206 | struct osd_attr attr; |
1207 | struct exofs_fcb *fcb; | |
1208 | struct updatei_args *args; | |
1209 | int ret; | |
1210 | ||
1211 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
34ce4e7c BH |
1212 | if (!args) { |
1213 | EXOFS_DBGMSG("Faild kzalloc of args\n"); | |
ba9e5e98 | 1214 | return -ENOMEM; |
34ce4e7c | 1215 | } |
ba9e5e98 BH |
1216 | |
1217 | fcb = &args->fcb; | |
1218 | ||
1219 | fcb->i_mode = cpu_to_le16(inode->i_mode); | |
1220 | fcb->i_uid = cpu_to_le32(inode->i_uid); | |
1221 | fcb->i_gid = cpu_to_le32(inode->i_gid); | |
1222 | fcb->i_links_count = cpu_to_le16(inode->i_nlink); | |
1223 | fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); | |
1224 | fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec); | |
1225 | fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); | |
1226 | oi->i_commit_size = i_size_read(inode); | |
1227 | fcb->i_size = cpu_to_le64(oi->i_commit_size); | |
1228 | fcb->i_generation = cpu_to_le32(inode->i_generation); | |
1229 | ||
1230 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
1231 | if (old_valid_dev(inode->i_rdev)) { | |
1232 | fcb->i_data[0] = | |
1233 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
1234 | fcb->i_data[1] = 0; | |
1235 | } else { | |
1236 | fcb->i_data[0] = 0; | |
1237 | fcb->i_data[1] = | |
1238 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
1239 | fcb->i_data[2] = 0; | |
1240 | } | |
1241 | } else | |
1242 | memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data)); | |
1243 | ||
45d3abcb | 1244 | ret = exofs_get_io_state(&sbi->layout, &ios); |
06886a5a BH |
1245 | if (unlikely(ret)) { |
1246 | EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__); | |
ba9e5e98 BH |
1247 | goto free_args; |
1248 | } | |
1249 | ||
ba9e5e98 BH |
1250 | attr = g_attr_inode_data; |
1251 | attr.val_ptr = fcb; | |
06886a5a BH |
1252 | ios->out_attr_len = 1; |
1253 | ios->out_attr = &attr; | |
ba9e5e98 BH |
1254 | |
1255 | if (!obj_created(oi)) { | |
1256 | EXOFS_DBGMSG("!obj_created\n"); | |
1257 | BUG_ON(!obj_2bcreated(oi)); | |
1258 | wait_event(oi->i_wq, obj_created(oi)); | |
1259 | EXOFS_DBGMSG("wait_event done\n"); | |
1260 | } | |
1261 | ||
06886a5a | 1262 | if (!do_sync) { |
ba9e5e98 | 1263 | args->sbi = sbi; |
06886a5a BH |
1264 | ios->done = updatei_done; |
1265 | ios->private = args; | |
1266 | } | |
ba9e5e98 | 1267 | |
06886a5a BH |
1268 | ret = exofs_oi_write(oi, ios); |
1269 | if (!do_sync && !ret) { | |
ba9e5e98 BH |
1270 | atomic_inc(&sbi->s_curr_pending); |
1271 | goto out; /* deallocation in updatei_done */ | |
1272 | } | |
1273 | ||
06886a5a | 1274 | exofs_put_io_state(ios); |
ba9e5e98 BH |
1275 | free_args: |
1276 | kfree(args); | |
1277 | out: | |
34ce4e7c BH |
1278 | EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n", |
1279 | inode->i_ino, do_sync, ret); | |
ba9e5e98 BH |
1280 | return ret; |
1281 | } | |
1282 | ||
a9185b41 | 1283 | int exofs_write_inode(struct inode *inode, struct writeback_control *wbc) |
ba9e5e98 | 1284 | { |
a9185b41 | 1285 | return exofs_update_inode(inode, wbc->sync_mode == WB_SYNC_ALL); |
ba9e5e98 BH |
1286 | } |
1287 | ||
1288 | /* | |
1289 | * Callback function from exofs_delete_inode() - don't have much cleaning up to | |
1290 | * do. | |
1291 | */ | |
06886a5a | 1292 | static void delete_done(struct exofs_io_state *ios, void *p) |
ba9e5e98 | 1293 | { |
06886a5a BH |
1294 | struct exofs_sb_info *sbi = p; |
1295 | ||
1296 | exofs_put_io_state(ios); | |
1297 | ||
ba9e5e98 BH |
1298 | atomic_dec(&sbi->s_curr_pending); |
1299 | } | |
1300 | ||
1301 | /* | |
1302 | * Called when the refcount of an inode reaches zero. We remove the object | |
1303 | * from the OSD here. We make sure the object was created before we try and | |
1304 | * delete it. | |
1305 | */ | |
1306 | void exofs_delete_inode(struct inode *inode) | |
1307 | { | |
1308 | struct exofs_i_info *oi = exofs_i(inode); | |
1309 | struct super_block *sb = inode->i_sb; | |
1310 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
06886a5a | 1311 | struct exofs_io_state *ios; |
ba9e5e98 BH |
1312 | int ret; |
1313 | ||
1314 | truncate_inode_pages(&inode->i_data, 0); | |
1315 | ||
1316 | if (is_bad_inode(inode)) | |
1317 | goto no_delete; | |
1318 | ||
1319 | mark_inode_dirty(inode); | |
1320 | exofs_update_inode(inode, inode_needs_sync(inode)); | |
1321 | ||
1322 | inode->i_size = 0; | |
1323 | if (inode->i_blocks) | |
1324 | exofs_truncate(inode); | |
1325 | ||
1326 | clear_inode(inode); | |
1327 | ||
45d3abcb | 1328 | ret = exofs_get_io_state(&sbi->layout, &ios); |
06886a5a BH |
1329 | if (unlikely(ret)) { |
1330 | EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__); | |
ba9e5e98 BH |
1331 | return; |
1332 | } | |
1333 | ||
ba9e5e98 BH |
1334 | /* if we are deleting an obj that hasn't been created yet, wait */ |
1335 | if (!obj_created(oi)) { | |
1336 | BUG_ON(!obj_2bcreated(oi)); | |
1337 | wait_event(oi->i_wq, obj_created(oi)); | |
1338 | } | |
1339 | ||
06886a5a BH |
1340 | ios->obj.id = exofs_oi_objno(oi); |
1341 | ios->done = delete_done; | |
1342 | ios->private = sbi; | |
1343 | ios->cred = oi->i_cred; | |
1344 | ret = exofs_sbi_remove(ios); | |
ba9e5e98 | 1345 | if (ret) { |
06886a5a BH |
1346 | EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__); |
1347 | exofs_put_io_state(ios); | |
ba9e5e98 BH |
1348 | return; |
1349 | } | |
1350 | atomic_inc(&sbi->s_curr_pending); | |
1351 | ||
1352 | return; | |
1353 | ||
1354 | no_delete: | |
1355 | clear_inode(inode); | |
1356 | } |