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), | |
44 | }; | |
45 | ||
beaec07b BH |
46 | struct page_collect { |
47 | struct exofs_sb_info *sbi; | |
48 | struct request_queue *req_q; | |
49 | struct inode *inode; | |
50 | unsigned expected_pages; | |
06886a5a | 51 | struct exofs_io_state *ios; |
beaec07b BH |
52 | |
53 | struct bio *bio; | |
54 | unsigned nr_pages; | |
55 | unsigned long length; | |
56 | loff_t pg_first; /* keep 64bit also in 32-arches */ | |
57 | }; | |
58 | ||
59 | static void _pcol_init(struct page_collect *pcol, unsigned expected_pages, | |
06886a5a | 60 | struct inode *inode) |
beaec07b BH |
61 | { |
62 | struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; | |
beaec07b BH |
63 | |
64 | pcol->sbi = sbi; | |
04dc1e88 BH |
65 | /* Create master bios on first Q, later on cloning, each clone will be |
66 | * allocated on it's destination Q | |
67 | */ | |
45d3abcb | 68 | pcol->req_q = osd_request_queue(sbi->layout.s_ods[0]); |
beaec07b BH |
69 | pcol->inode = inode; |
70 | pcol->expected_pages = expected_pages; | |
71 | ||
06886a5a | 72 | pcol->ios = NULL; |
beaec07b BH |
73 | pcol->bio = NULL; |
74 | pcol->nr_pages = 0; | |
75 | pcol->length = 0; | |
76 | pcol->pg_first = -1; | |
beaec07b BH |
77 | } |
78 | ||
79 | static void _pcol_reset(struct page_collect *pcol) | |
80 | { | |
81 | pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages); | |
82 | ||
83 | pcol->bio = NULL; | |
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) | |
06886a5a | 93 | pcol->expected_pages = BIO_MAX_PAGES_KMALLOC; |
beaec07b BH |
94 | } |
95 | ||
96 | static int pcol_try_alloc(struct page_collect *pcol) | |
97 | { | |
06886a5a BH |
98 | int pages = min_t(unsigned, pcol->expected_pages, |
99 | BIO_MAX_PAGES_KMALLOC); | |
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 BH |
107 | |
108 | for (; pages; pages >>= 1) { | |
06886a5a | 109 | pcol->bio = bio_kmalloc(GFP_KERNEL, pages); |
beaec07b BH |
110 | if (likely(pcol->bio)) |
111 | return 0; | |
112 | } | |
113 | ||
06886a5a | 114 | EXOFS_ERR("Failed to bio_kmalloc expected_pages=%u\n", |
beaec07b BH |
115 | pcol->expected_pages); |
116 | return -ENOMEM; | |
117 | } | |
118 | ||
119 | static void pcol_free(struct page_collect *pcol) | |
120 | { | |
06886a5a BH |
121 | if (pcol->bio) { |
122 | bio_put(pcol->bio); | |
123 | pcol->bio = NULL; | |
124 | } | |
125 | ||
126 | if (pcol->ios) { | |
127 | exofs_put_io_state(pcol->ios); | |
128 | pcol->ios = NULL; | |
129 | } | |
beaec07b BH |
130 | } |
131 | ||
132 | static int pcol_add_page(struct page_collect *pcol, struct page *page, | |
133 | unsigned len) | |
134 | { | |
135 | int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0); | |
136 | if (unlikely(len != added_len)) | |
137 | return -ENOMEM; | |
138 | ||
139 | ++pcol->nr_pages; | |
140 | pcol->length += len; | |
141 | return 0; | |
142 | } | |
143 | ||
144 | static int update_read_page(struct page *page, int ret) | |
145 | { | |
146 | if (ret == 0) { | |
147 | /* Everything is OK */ | |
148 | SetPageUptodate(page); | |
149 | if (PageError(page)) | |
150 | ClearPageError(page); | |
151 | } else if (ret == -EFAULT) { | |
152 | /* In this case we were trying to read something that wasn't on | |
153 | * disk yet - return a page full of zeroes. This should be OK, | |
154 | * because the object should be empty (if there was a write | |
155 | * before this read, the read would be waiting with the page | |
156 | * locked */ | |
157 | clear_highpage(page); | |
158 | ||
159 | SetPageUptodate(page); | |
160 | if (PageError(page)) | |
161 | ClearPageError(page); | |
162 | ret = 0; /* recovered error */ | |
163 | EXOFS_DBGMSG("recovered read error\n"); | |
164 | } else /* Error */ | |
165 | SetPageError(page); | |
166 | ||
167 | return ret; | |
168 | } | |
169 | ||
170 | static void update_write_page(struct page *page, int ret) | |
171 | { | |
172 | if (ret) { | |
173 | mapping_set_error(page->mapping, ret); | |
174 | SetPageError(page); | |
175 | } | |
176 | end_page_writeback(page); | |
177 | } | |
178 | ||
179 | /* Called at the end of reads, to optionally unlock pages and update their | |
180 | * status. | |
181 | */ | |
06886a5a | 182 | static int __readpages_done(struct page_collect *pcol, bool do_unlock) |
beaec07b BH |
183 | { |
184 | struct bio_vec *bvec; | |
185 | int i; | |
186 | u64 resid; | |
187 | u64 good_bytes; | |
188 | u64 length = 0; | |
06886a5a | 189 | int ret = exofs_check_io(pcol->ios, &resid); |
beaec07b BH |
190 | |
191 | if (likely(!ret)) | |
192 | good_bytes = pcol->length; | |
beaec07b BH |
193 | else |
194 | good_bytes = pcol->length - resid; | |
195 | ||
34ce4e7c | 196 | EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx" |
beaec07b BH |
197 | " length=0x%lx nr_pages=%u\n", |
198 | pcol->inode->i_ino, _LLU(good_bytes), pcol->length, | |
199 | pcol->nr_pages); | |
200 | ||
201 | __bio_for_each_segment(bvec, pcol->bio, i, 0) { | |
202 | struct page *page = bvec->bv_page; | |
203 | struct inode *inode = page->mapping->host; | |
204 | int page_stat; | |
205 | ||
206 | if (inode != pcol->inode) | |
207 | continue; /* osd might add more pages at end */ | |
208 | ||
209 | if (likely(length < good_bytes)) | |
210 | page_stat = 0; | |
211 | else | |
212 | page_stat = ret; | |
213 | ||
fe33cc1e | 214 | EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n", |
beaec07b BH |
215 | inode->i_ino, page->index, |
216 | page_stat ? "bad_bytes" : "good_bytes"); | |
217 | ||
218 | ret = update_read_page(page, page_stat); | |
219 | if (do_unlock) | |
220 | unlock_page(page); | |
221 | length += bvec->bv_len; | |
222 | } | |
223 | ||
224 | pcol_free(pcol); | |
34ce4e7c | 225 | EXOFS_DBGMSG2("readpages_done END\n"); |
beaec07b BH |
226 | return ret; |
227 | } | |
228 | ||
229 | /* callback of async reads */ | |
06886a5a | 230 | static void readpages_done(struct exofs_io_state *ios, void *p) |
beaec07b BH |
231 | { |
232 | struct page_collect *pcol = p; | |
233 | ||
06886a5a | 234 | __readpages_done(pcol, true); |
beaec07b | 235 | atomic_dec(&pcol->sbi->s_curr_pending); |
06886a5a | 236 | kfree(pcol); |
beaec07b BH |
237 | } |
238 | ||
239 | static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw) | |
240 | { | |
241 | struct bio_vec *bvec; | |
242 | int i; | |
243 | ||
244 | __bio_for_each_segment(bvec, pcol->bio, i, 0) { | |
245 | struct page *page = bvec->bv_page; | |
246 | ||
247 | if (rw == READ) | |
248 | update_read_page(page, ret); | |
249 | else | |
250 | update_write_page(page, ret); | |
251 | ||
252 | unlock_page(page); | |
253 | } | |
beaec07b BH |
254 | } |
255 | ||
256 | static int read_exec(struct page_collect *pcol, bool is_sync) | |
257 | { | |
258 | struct exofs_i_info *oi = exofs_i(pcol->inode); | |
06886a5a | 259 | struct exofs_io_state *ios = pcol->ios; |
beaec07b | 260 | struct page_collect *pcol_copy = NULL; |
beaec07b BH |
261 | int ret; |
262 | ||
263 | if (!pcol->bio) | |
264 | return 0; | |
265 | ||
266 | /* see comment in _readpage() about sync reads */ | |
267 | WARN_ON(is_sync && (pcol->nr_pages != 1)); | |
268 | ||
06886a5a BH |
269 | ios->bio = pcol->bio; |
270 | ios->length = pcol->length; | |
271 | ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT; | |
beaec07b BH |
272 | |
273 | if (is_sync) { | |
06886a5a BH |
274 | exofs_oi_read(oi, pcol->ios); |
275 | return __readpages_done(pcol, false); | |
beaec07b BH |
276 | } |
277 | ||
278 | pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); | |
279 | if (!pcol_copy) { | |
280 | ret = -ENOMEM; | |
281 | goto err; | |
282 | } | |
283 | ||
284 | *pcol_copy = *pcol; | |
06886a5a BH |
285 | ios->done = readpages_done; |
286 | ios->private = pcol_copy; | |
287 | ret = exofs_oi_read(oi, ios); | |
beaec07b BH |
288 | if (unlikely(ret)) |
289 | goto err; | |
290 | ||
291 | atomic_inc(&pcol->sbi->s_curr_pending); | |
292 | ||
34ce4e7c | 293 | EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n", |
06886a5a | 294 | ios->obj.id, _LLU(ios->offset), pcol->length); |
beaec07b BH |
295 | |
296 | /* pages ownership was passed to pcol_copy */ | |
297 | _pcol_reset(pcol); | |
298 | return 0; | |
299 | ||
300 | err: | |
301 | if (!is_sync) | |
302 | _unlock_pcol_pages(pcol, ret, READ); | |
06886a5a BH |
303 | |
304 | pcol_free(pcol); | |
b76a3f93 | 305 | |
beaec07b | 306 | kfree(pcol_copy); |
beaec07b BH |
307 | return ret; |
308 | } | |
309 | ||
310 | /* readpage_strip is called either directly from readpage() or by the VFS from | |
311 | * within read_cache_pages(), to add one more page to be read. It will try to | |
312 | * collect as many contiguous pages as posible. If a discontinuity is | |
313 | * encountered, or it runs out of resources, it will submit the previous segment | |
314 | * and will start a new collection. Eventually caller must submit the last | |
315 | * segment if present. | |
316 | */ | |
317 | static int readpage_strip(void *data, struct page *page) | |
318 | { | |
319 | struct page_collect *pcol = data; | |
320 | struct inode *inode = pcol->inode; | |
321 | struct exofs_i_info *oi = exofs_i(inode); | |
322 | loff_t i_size = i_size_read(inode); | |
323 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
324 | size_t len; | |
325 | int ret; | |
326 | ||
327 | /* FIXME: Just for debugging, will be removed */ | |
328 | if (PageUptodate(page)) | |
329 | EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino, | |
330 | page->index); | |
331 | ||
332 | if (page->index < end_index) | |
333 | len = PAGE_CACHE_SIZE; | |
334 | else if (page->index == end_index) | |
335 | len = i_size & ~PAGE_CACHE_MASK; | |
336 | else | |
337 | len = 0; | |
338 | ||
339 | if (!len || !obj_created(oi)) { | |
340 | /* this will be out of bounds, or doesn't exist yet. | |
341 | * Current page is cleared and the request is split | |
342 | */ | |
343 | clear_highpage(page); | |
344 | ||
345 | SetPageUptodate(page); | |
346 | if (PageError(page)) | |
347 | ClearPageError(page); | |
348 | ||
349 | unlock_page(page); | |
350 | EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page," | |
351 | " splitting\n", inode->i_ino, page->index); | |
352 | ||
353 | return read_exec(pcol, false); | |
354 | } | |
355 | ||
356 | try_again: | |
357 | ||
358 | if (unlikely(pcol->pg_first == -1)) { | |
359 | pcol->pg_first = page->index; | |
360 | } else if (unlikely((pcol->pg_first + pcol->nr_pages) != | |
361 | page->index)) { | |
362 | /* Discontinuity detected, split the request */ | |
363 | ret = read_exec(pcol, false); | |
364 | if (unlikely(ret)) | |
365 | goto fail; | |
366 | goto try_again; | |
367 | } | |
368 | ||
369 | if (!pcol->bio) { | |
370 | ret = pcol_try_alloc(pcol); | |
371 | if (unlikely(ret)) | |
372 | goto fail; | |
373 | } | |
374 | ||
375 | if (len != PAGE_CACHE_SIZE) | |
376 | zero_user(page, len, PAGE_CACHE_SIZE - len); | |
377 | ||
fe33cc1e | 378 | EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n", |
beaec07b BH |
379 | inode->i_ino, page->index, len); |
380 | ||
381 | ret = pcol_add_page(pcol, page, len); | |
382 | if (ret) { | |
fe33cc1e | 383 | EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p " |
beaec07b BH |
384 | "this_len=0x%zx nr_pages=%u length=0x%lx\n", |
385 | page, len, pcol->nr_pages, pcol->length); | |
386 | ||
387 | /* split the request, and start again with current page */ | |
388 | ret = read_exec(pcol, false); | |
389 | if (unlikely(ret)) | |
390 | goto fail; | |
391 | ||
392 | goto try_again; | |
393 | } | |
394 | ||
395 | return 0; | |
396 | ||
397 | fail: | |
398 | /* SetPageError(page); ??? */ | |
399 | unlock_page(page); | |
400 | return ret; | |
401 | } | |
402 | ||
403 | static int exofs_readpages(struct file *file, struct address_space *mapping, | |
404 | struct list_head *pages, unsigned nr_pages) | |
405 | { | |
406 | struct page_collect pcol; | |
407 | int ret; | |
408 | ||
409 | _pcol_init(&pcol, nr_pages, mapping->host); | |
410 | ||
411 | ret = read_cache_pages(mapping, pages, readpage_strip, &pcol); | |
412 | if (ret) { | |
413 | EXOFS_ERR("read_cache_pages => %d\n", ret); | |
414 | return ret; | |
415 | } | |
416 | ||
417 | return read_exec(&pcol, false); | |
418 | } | |
419 | ||
420 | static int _readpage(struct page *page, bool is_sync) | |
421 | { | |
422 | struct page_collect pcol; | |
423 | int ret; | |
424 | ||
425 | _pcol_init(&pcol, 1, page->mapping->host); | |
426 | ||
06886a5a BH |
427 | /* readpage_strip might call read_exec(,is_sync==false) at several |
428 | * places but not if we have a single page. | |
beaec07b BH |
429 | */ |
430 | ret = readpage_strip(&pcol, page); | |
431 | if (ret) { | |
432 | EXOFS_ERR("_readpage => %d\n", ret); | |
433 | return ret; | |
434 | } | |
435 | ||
436 | return read_exec(&pcol, is_sync); | |
437 | } | |
438 | ||
439 | /* | |
440 | * We don't need the file | |
441 | */ | |
442 | static int exofs_readpage(struct file *file, struct page *page) | |
443 | { | |
444 | return _readpage(page, false); | |
445 | } | |
446 | ||
06886a5a BH |
447 | /* Callback for osd_write. All writes are asynchronous */ |
448 | static void writepages_done(struct exofs_io_state *ios, void *p) | |
beaec07b BH |
449 | { |
450 | struct page_collect *pcol = p; | |
451 | struct bio_vec *bvec; | |
452 | int i; | |
453 | u64 resid; | |
454 | u64 good_bytes; | |
455 | u64 length = 0; | |
06886a5a | 456 | int ret = exofs_check_io(ios, &resid); |
beaec07b | 457 | |
beaec07b BH |
458 | atomic_dec(&pcol->sbi->s_curr_pending); |
459 | ||
460 | if (likely(!ret)) | |
461 | good_bytes = pcol->length; | |
beaec07b BH |
462 | else |
463 | good_bytes = pcol->length - resid; | |
464 | ||
34ce4e7c | 465 | EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx" |
beaec07b BH |
466 | " length=0x%lx nr_pages=%u\n", |
467 | pcol->inode->i_ino, _LLU(good_bytes), pcol->length, | |
468 | pcol->nr_pages); | |
469 | ||
470 | __bio_for_each_segment(bvec, pcol->bio, i, 0) { | |
471 | struct page *page = bvec->bv_page; | |
472 | struct inode *inode = page->mapping->host; | |
473 | int page_stat; | |
474 | ||
475 | if (inode != pcol->inode) | |
476 | continue; /* osd might add more pages to a bio */ | |
477 | ||
478 | if (likely(length < good_bytes)) | |
479 | page_stat = 0; | |
480 | else | |
481 | page_stat = ret; | |
482 | ||
483 | update_write_page(page, page_stat); | |
484 | unlock_page(page); | |
fe33cc1e | 485 | EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n", |
beaec07b BH |
486 | inode->i_ino, page->index, page_stat); |
487 | ||
488 | length += bvec->bv_len; | |
489 | } | |
490 | ||
491 | pcol_free(pcol); | |
492 | kfree(pcol); | |
34ce4e7c | 493 | EXOFS_DBGMSG2("writepages_done END\n"); |
beaec07b BH |
494 | } |
495 | ||
496 | static int write_exec(struct page_collect *pcol) | |
497 | { | |
498 | struct exofs_i_info *oi = exofs_i(pcol->inode); | |
06886a5a | 499 | struct exofs_io_state *ios = pcol->ios; |
beaec07b | 500 | struct page_collect *pcol_copy = NULL; |
beaec07b BH |
501 | int ret; |
502 | ||
503 | if (!pcol->bio) | |
504 | return 0; | |
505 | ||
beaec07b BH |
506 | pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); |
507 | if (!pcol_copy) { | |
508 | EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n"); | |
509 | ret = -ENOMEM; | |
510 | goto err; | |
511 | } | |
512 | ||
513 | *pcol_copy = *pcol; | |
514 | ||
62f469b5 | 515 | pcol_copy->bio->bi_rw |= (1 << BIO_RW); /* FIXME: bio_set_dir() */ |
06886a5a BH |
516 | |
517 | ios->bio = pcol_copy->bio; | |
518 | ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT; | |
519 | ios->length = pcol_copy->length; | |
520 | ios->done = writepages_done; | |
521 | ios->private = pcol_copy; | |
522 | ||
523 | ret = exofs_oi_write(oi, ios); | |
beaec07b | 524 | if (unlikely(ret)) { |
06886a5a | 525 | EXOFS_ERR("write_exec: exofs_oi_write() Faild\n"); |
beaec07b BH |
526 | goto err; |
527 | } | |
528 | ||
529 | atomic_inc(&pcol->sbi->s_curr_pending); | |
34ce4e7c | 530 | EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n", |
06886a5a | 531 | pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset), |
beaec07b BH |
532 | pcol->length); |
533 | /* pages ownership was passed to pcol_copy */ | |
534 | _pcol_reset(pcol); | |
535 | return 0; | |
536 | ||
537 | err: | |
538 | _unlock_pcol_pages(pcol, ret, WRITE); | |
06886a5a | 539 | pcol_free(pcol); |
beaec07b | 540 | kfree(pcol_copy); |
06886a5a | 541 | |
beaec07b BH |
542 | return ret; |
543 | } | |
544 | ||
545 | /* writepage_strip is called either directly from writepage() or by the VFS from | |
546 | * within write_cache_pages(), to add one more page to be written to storage. | |
547 | * It will try to collect as many contiguous pages as possible. If a | |
548 | * discontinuity is encountered or it runs out of resources it will submit the | |
549 | * previous segment and will start a new collection. | |
550 | * Eventually caller must submit the last segment if present. | |
551 | */ | |
552 | static int writepage_strip(struct page *page, | |
553 | struct writeback_control *wbc_unused, void *data) | |
554 | { | |
555 | struct page_collect *pcol = data; | |
556 | struct inode *inode = pcol->inode; | |
557 | struct exofs_i_info *oi = exofs_i(inode); | |
558 | loff_t i_size = i_size_read(inode); | |
559 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
560 | size_t len; | |
561 | int ret; | |
562 | ||
563 | BUG_ON(!PageLocked(page)); | |
564 | ||
565 | ret = wait_obj_created(oi); | |
566 | if (unlikely(ret)) | |
567 | goto fail; | |
568 | ||
569 | if (page->index < end_index) | |
570 | /* in this case, the page is within the limits of the file */ | |
571 | len = PAGE_CACHE_SIZE; | |
572 | else { | |
573 | len = i_size & ~PAGE_CACHE_MASK; | |
574 | ||
575 | if (page->index > end_index || !len) { | |
576 | /* in this case, the page is outside the limits | |
577 | * (truncate in progress) | |
578 | */ | |
579 | ret = write_exec(pcol); | |
580 | if (unlikely(ret)) | |
581 | goto fail; | |
582 | if (PageError(page)) | |
583 | ClearPageError(page); | |
584 | unlock_page(page); | |
06886a5a BH |
585 | EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) " |
586 | "outside the limits\n", | |
587 | inode->i_ino, page->index); | |
beaec07b BH |
588 | return 0; |
589 | } | |
590 | } | |
591 | ||
592 | try_again: | |
593 | ||
594 | if (unlikely(pcol->pg_first == -1)) { | |
595 | pcol->pg_first = page->index; | |
596 | } else if (unlikely((pcol->pg_first + pcol->nr_pages) != | |
597 | page->index)) { | |
598 | /* Discontinuity detected, split the request */ | |
599 | ret = write_exec(pcol); | |
600 | if (unlikely(ret)) | |
601 | goto fail; | |
06886a5a BH |
602 | |
603 | EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n", | |
604 | inode->i_ino, page->index); | |
beaec07b BH |
605 | goto try_again; |
606 | } | |
607 | ||
608 | if (!pcol->bio) { | |
609 | ret = pcol_try_alloc(pcol); | |
610 | if (unlikely(ret)) | |
611 | goto fail; | |
612 | } | |
613 | ||
fe33cc1e | 614 | EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n", |
beaec07b BH |
615 | inode->i_ino, page->index, len); |
616 | ||
617 | ret = pcol_add_page(pcol, page, len); | |
618 | if (unlikely(ret)) { | |
34ce4e7c | 619 | EXOFS_DBGMSG2("Failed pcol_add_page " |
beaec07b BH |
620 | "nr_pages=%u total_length=0x%lx\n", |
621 | pcol->nr_pages, pcol->length); | |
622 | ||
623 | /* split the request, next loop will start again */ | |
624 | ret = write_exec(pcol); | |
625 | if (unlikely(ret)) { | |
626 | EXOFS_DBGMSG("write_exec faild => %d", ret); | |
627 | goto fail; | |
628 | } | |
629 | ||
630 | goto try_again; | |
631 | } | |
632 | ||
633 | BUG_ON(PageWriteback(page)); | |
634 | set_page_writeback(page); | |
635 | ||
636 | return 0; | |
637 | ||
638 | fail: | |
06886a5a BH |
639 | EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n", |
640 | inode->i_ino, page->index, ret); | |
beaec07b BH |
641 | set_bit(AS_EIO, &page->mapping->flags); |
642 | unlock_page(page); | |
643 | return ret; | |
644 | } | |
645 | ||
646 | static int exofs_writepages(struct address_space *mapping, | |
647 | struct writeback_control *wbc) | |
648 | { | |
649 | struct page_collect pcol; | |
650 | long start, end, expected_pages; | |
651 | int ret; | |
652 | ||
653 | start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
654 | end = (wbc->range_end == LLONG_MAX) ? | |
655 | start + mapping->nrpages : | |
656 | wbc->range_end >> PAGE_CACHE_SHIFT; | |
657 | ||
658 | if (start || end) | |
06886a5a | 659 | expected_pages = end - start + 1; |
beaec07b BH |
660 | else |
661 | expected_pages = mapping->nrpages; | |
662 | ||
06886a5a BH |
663 | if (expected_pages < 32L) |
664 | expected_pages = 32L; | |
665 | ||
34ce4e7c | 666 | EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx " |
06886a5a | 667 | "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n", |
beaec07b | 668 | mapping->host->i_ino, wbc->range_start, wbc->range_end, |
06886a5a | 669 | mapping->nrpages, start, end, expected_pages); |
beaec07b BH |
670 | |
671 | _pcol_init(&pcol, expected_pages, mapping->host); | |
672 | ||
673 | ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol); | |
674 | if (ret) { | |
675 | EXOFS_ERR("write_cache_pages => %d\n", ret); | |
676 | return ret; | |
677 | } | |
678 | ||
679 | return write_exec(&pcol); | |
680 | } | |
681 | ||
682 | static int exofs_writepage(struct page *page, struct writeback_control *wbc) | |
683 | { | |
684 | struct page_collect pcol; | |
685 | int ret; | |
686 | ||
687 | _pcol_init(&pcol, 1, page->mapping->host); | |
688 | ||
689 | ret = writepage_strip(page, NULL, &pcol); | |
690 | if (ret) { | |
691 | EXOFS_ERR("exofs_writepage => %d\n", ret); | |
692 | return ret; | |
693 | } | |
694 | ||
695 | return write_exec(&pcol); | |
696 | } | |
697 | ||
698 | int exofs_write_begin(struct file *file, struct address_space *mapping, | |
699 | loff_t pos, unsigned len, unsigned flags, | |
700 | struct page **pagep, void **fsdata) | |
701 | { | |
702 | int ret = 0; | |
703 | struct page *page; | |
704 | ||
705 | page = *pagep; | |
706 | if (page == NULL) { | |
707 | ret = simple_write_begin(file, mapping, pos, len, flags, pagep, | |
708 | fsdata); | |
709 | if (ret) { | |
710 | EXOFS_DBGMSG("simple_write_begin faild\n"); | |
711 | return ret; | |
712 | } | |
713 | ||
714 | page = *pagep; | |
715 | } | |
716 | ||
717 | /* read modify write */ | |
718 | if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) { | |
719 | ret = _readpage(page, true); | |
720 | if (ret) { | |
721 | /*SetPageError was done by _readpage. Is it ok?*/ | |
722 | unlock_page(page); | |
723 | EXOFS_DBGMSG("__readpage_filler faild\n"); | |
724 | } | |
725 | } | |
726 | ||
727 | return ret; | |
728 | } | |
729 | ||
730 | static int exofs_write_begin_export(struct file *file, | |
731 | struct address_space *mapping, | |
732 | loff_t pos, unsigned len, unsigned flags, | |
733 | struct page **pagep, void **fsdata) | |
734 | { | |
735 | *pagep = NULL; | |
736 | ||
737 | return exofs_write_begin(file, mapping, pos, len, flags, pagep, | |
738 | fsdata); | |
739 | } | |
740 | ||
efd124b9 BH |
741 | static int exofs_write_end(struct file *file, struct address_space *mapping, |
742 | loff_t pos, unsigned len, unsigned copied, | |
743 | struct page *page, void *fsdata) | |
744 | { | |
745 | struct inode *inode = mapping->host; | |
746 | /* According to comment in simple_write_end i_mutex is held */ | |
747 | loff_t i_size = inode->i_size; | |
748 | int ret; | |
749 | ||
750 | ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata); | |
751 | if (i_size != inode->i_size) | |
752 | mark_inode_dirty(inode); | |
753 | return ret; | |
754 | } | |
755 | ||
beaec07b BH |
756 | const struct address_space_operations exofs_aops = { |
757 | .readpage = exofs_readpage, | |
758 | .readpages = exofs_readpages, | |
759 | .writepage = exofs_writepage, | |
760 | .writepages = exofs_writepages, | |
761 | .write_begin = exofs_write_begin_export, | |
efd124b9 | 762 | .write_end = exofs_write_end, |
beaec07b BH |
763 | }; |
764 | ||
e8062719 BH |
765 | /****************************************************************************** |
766 | * INODE OPERATIONS | |
767 | *****************************************************************************/ | |
768 | ||
769 | /* | |
770 | * Test whether an inode is a fast symlink. | |
771 | */ | |
772 | static inline int exofs_inode_is_fast_symlink(struct inode *inode) | |
773 | { | |
774 | struct exofs_i_info *oi = exofs_i(inode); | |
775 | ||
776 | return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0); | |
777 | } | |
778 | ||
779 | /* | |
780 | * get_block_t - Fill in a buffer_head | |
781 | * An OSD takes care of block allocation so we just fake an allocation by | |
782 | * putting in the inode's sector_t in the buffer_head. | |
783 | * TODO: What about the case of create==0 and @iblock does not exist in the | |
784 | * object? | |
785 | */ | |
786 | static int exofs_get_block(struct inode *inode, sector_t iblock, | |
787 | struct buffer_head *bh_result, int create) | |
788 | { | |
789 | map_bh(bh_result, inode->i_sb, iblock); | |
790 | return 0; | |
791 | } | |
792 | ||
793 | const struct osd_attr g_attr_logical_length = ATTR_DEF( | |
794 | OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8); | |
795 | ||
06886a5a BH |
796 | static int _do_truncate(struct inode *inode) |
797 | { | |
798 | struct exofs_i_info *oi = exofs_i(inode); | |
799 | loff_t isize = i_size_read(inode); | |
800 | int ret; | |
801 | ||
802 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
803 | ||
804 | nobh_truncate_page(inode->i_mapping, isize, exofs_get_block); | |
805 | ||
806 | ret = exofs_oi_truncate(oi, (u64)isize); | |
807 | EXOFS_DBGMSG("(0x%lx) size=0x%llx\n", inode->i_ino, isize); | |
808 | return ret; | |
809 | } | |
810 | ||
e8062719 BH |
811 | /* |
812 | * Truncate a file to the specified size - all we have to do is set the size | |
813 | * attribute. We make sure the object exists first. | |
814 | */ | |
815 | void exofs_truncate(struct inode *inode) | |
816 | { | |
e8062719 | 817 | struct exofs_i_info *oi = exofs_i(inode); |
e8062719 BH |
818 | int ret; |
819 | ||
820 | if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) | |
821 | || S_ISLNK(inode->i_mode))) | |
822 | return; | |
823 | if (exofs_inode_is_fast_symlink(inode)) | |
824 | return; | |
825 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) | |
826 | return; | |
e8062719 BH |
827 | |
828 | /* if we are about to truncate an object, and it hasn't been | |
829 | * created yet, wait | |
830 | */ | |
831 | if (unlikely(wait_obj_created(oi))) | |
832 | goto fail; | |
833 | ||
06886a5a | 834 | ret = _do_truncate(inode); |
e8062719 BH |
835 | if (ret) |
836 | goto fail; | |
837 | ||
838 | out: | |
839 | mark_inode_dirty(inode); | |
840 | return; | |
841 | fail: | |
842 | make_bad_inode(inode); | |
843 | goto out; | |
844 | } | |
845 | ||
846 | /* | |
847 | * Set inode attributes - just call generic functions. | |
848 | */ | |
849 | int exofs_setattr(struct dentry *dentry, struct iattr *iattr) | |
850 | { | |
851 | struct inode *inode = dentry->d_inode; | |
852 | int error; | |
853 | ||
854 | error = inode_change_ok(inode, iattr); | |
855 | if (error) | |
856 | return error; | |
857 | ||
858 | error = inode_setattr(inode, iattr); | |
859 | return error; | |
860 | } | |
e6af00f1 | 861 | |
d9c740d2 BH |
862 | static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF( |
863 | EXOFS_APAGE_FS_DATA, | |
864 | EXOFS_ATTR_INODE_FILE_LAYOUT, | |
865 | 0); | |
866 | static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF( | |
867 | EXOFS_APAGE_FS_DATA, | |
868 | EXOFS_ATTR_INODE_DIR_LAYOUT, | |
869 | 0); | |
870 | ||
e6af00f1 BH |
871 | /* |
872 | * Read an inode from the OSD, and return it as is. We also return the size | |
06886a5a | 873 | * attribute in the 'obj_size' argument. |
e6af00f1 BH |
874 | */ |
875 | static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi, | |
06886a5a | 876 | struct exofs_fcb *inode, uint64_t *obj_size) |
e6af00f1 BH |
877 | { |
878 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
d9c740d2 BH |
879 | struct osd_attr attrs[] = { |
880 | [0] = g_attr_inode_data, | |
881 | [1] = g_attr_inode_file_layout, | |
882 | [2] = g_attr_inode_dir_layout, | |
883 | [3] = g_attr_logical_length, | |
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); |
e6af00f1 BH |
906 | if (ret) |
907 | goto out; | |
908 | ||
06886a5a | 909 | ret = extract_attr_from_ios(ios, &attrs[0]); |
e6af00f1 | 910 | if (ret) { |
06886a5a | 911 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); |
e6af00f1 BH |
912 | goto out; |
913 | } | |
06886a5a BH |
914 | WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE); |
915 | memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE); | |
e6af00f1 | 916 | |
06886a5a | 917 | ret = extract_attr_from_ios(ios, &attrs[1]); |
d9c740d2 BH |
918 | if (ret) { |
919 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); | |
920 | goto out; | |
921 | } | |
922 | if (attrs[1].len) { | |
923 | layout = attrs[1].val_ptr; | |
924 | if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) { | |
925 | EXOFS_ERR("%s: unsupported files layout %d\n", | |
926 | __func__, layout->gen_func); | |
927 | ret = -ENOTSUPP; | |
928 | goto out; | |
929 | } | |
930 | } | |
931 | ||
932 | ret = extract_attr_from_ios(ios, &attrs[2]); | |
933 | if (ret) { | |
934 | EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__); | |
935 | goto out; | |
936 | } | |
937 | if (attrs[2].len) { | |
938 | layout = attrs[2].val_ptr; | |
939 | if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) { | |
940 | EXOFS_ERR("%s: unsupported meta-data layout %d\n", | |
941 | __func__, layout->gen_func); | |
942 | ret = -ENOTSUPP; | |
943 | goto out; | |
944 | } | |
945 | } | |
946 | ||
947 | *obj_size = ~0; | |
948 | ret = extract_attr_from_ios(ios, &attrs[3]); | |
e6af00f1 | 949 | if (ret) { |
06886a5a BH |
950 | EXOFS_ERR("%s: extract_attr of logical_length failed\n", |
951 | __func__); | |
e6af00f1 BH |
952 | goto out; |
953 | } | |
d9c740d2 | 954 | *obj_size = get_unaligned_be64(attrs[3].val_ptr); |
e6af00f1 BH |
955 | |
956 | out: | |
06886a5a | 957 | exofs_put_io_state(ios); |
e6af00f1 BH |
958 | return ret; |
959 | } | |
960 | ||
9cfdc7aa BH |
961 | static void __oi_init(struct exofs_i_info *oi) |
962 | { | |
963 | init_waitqueue_head(&oi->i_wq); | |
964 | oi->i_flags = 0; | |
965 | } | |
e6af00f1 BH |
966 | /* |
967 | * Fill in an inode read from the OSD and set it up for use | |
968 | */ | |
969 | struct inode *exofs_iget(struct super_block *sb, unsigned long ino) | |
970 | { | |
971 | struct exofs_i_info *oi; | |
972 | struct exofs_fcb fcb; | |
973 | struct inode *inode; | |
06886a5a | 974 | uint64_t obj_size; |
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 */ | |
06886a5a | 986 | ret = exofs_get_inode(sb, oi, &fcb, &obj_size); |
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 | ||
06886a5a | 1007 | if ((inode->i_size != obj_size) && |
e6af00f1 | 1008 | (!exofs_inode_is_fast_symlink(inode))) { |
58311c43 | 1009 | EXOFS_ERR("WARNING: Size of inode=%llu != object=%llu\n", |
06886a5a BH |
1010 | inode->i_size, _LLU(obj_size)); |
1011 | /* FIXME: call exofs_inode_recovery() */ | |
e6af00f1 | 1012 | } |
e6af00f1 BH |
1013 | |
1014 | oi->i_dir_start_lookup = 0; | |
1015 | ||
1016 | if ((inode->i_nlink == 0) && (inode->i_mode == 0)) { | |
1017 | ret = -ESTALE; | |
1018 | goto bad_inode; | |
1019 | } | |
1020 | ||
1021 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
1022 | if (fcb.i_data[0]) | |
1023 | inode->i_rdev = | |
1024 | old_decode_dev(le32_to_cpu(fcb.i_data[0])); | |
1025 | else | |
1026 | inode->i_rdev = | |
1027 | new_decode_dev(le32_to_cpu(fcb.i_data[1])); | |
1028 | } else { | |
1029 | memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data)); | |
1030 | } | |
1031 | ||
1032 | if (S_ISREG(inode->i_mode)) { | |
1033 | inode->i_op = &exofs_file_inode_operations; | |
1034 | inode->i_fop = &exofs_file_operations; | |
1035 | inode->i_mapping->a_ops = &exofs_aops; | |
1036 | } else if (S_ISDIR(inode->i_mode)) { | |
1037 | inode->i_op = &exofs_dir_inode_operations; | |
1038 | inode->i_fop = &exofs_dir_operations; | |
1039 | inode->i_mapping->a_ops = &exofs_aops; | |
1040 | } else if (S_ISLNK(inode->i_mode)) { | |
1041 | if (exofs_inode_is_fast_symlink(inode)) | |
1042 | inode->i_op = &exofs_fast_symlink_inode_operations; | |
1043 | else { | |
1044 | inode->i_op = &exofs_symlink_inode_operations; | |
1045 | inode->i_mapping->a_ops = &exofs_aops; | |
1046 | } | |
1047 | } else { | |
1048 | inode->i_op = &exofs_special_inode_operations; | |
1049 | if (fcb.i_data[0]) | |
1050 | init_special_inode(inode, inode->i_mode, | |
1051 | old_decode_dev(le32_to_cpu(fcb.i_data[0]))); | |
1052 | else | |
1053 | init_special_inode(inode, inode->i_mode, | |
1054 | new_decode_dev(le32_to_cpu(fcb.i_data[1]))); | |
1055 | } | |
1056 | ||
1057 | unlock_new_inode(inode); | |
1058 | return inode; | |
1059 | ||
1060 | bad_inode: | |
1061 | iget_failed(inode); | |
1062 | return ERR_PTR(ret); | |
1063 | } | |
1064 | ||
1065 | int __exofs_wait_obj_created(struct exofs_i_info *oi) | |
1066 | { | |
1067 | if (!obj_created(oi)) { | |
1068 | BUG_ON(!obj_2bcreated(oi)); | |
1069 | wait_event(oi->i_wq, obj_created(oi)); | |
1070 | } | |
1071 | return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0; | |
1072 | } | |
1073 | /* | |
1074 | * Callback function from exofs_new_inode(). The important thing is that we | |
1075 | * set the obj_created flag so that other methods know that the object exists on | |
1076 | * the OSD. | |
1077 | */ | |
06886a5a | 1078 | static void create_done(struct exofs_io_state *ios, void *p) |
e6af00f1 BH |
1079 | { |
1080 | struct inode *inode = p; | |
1081 | struct exofs_i_info *oi = exofs_i(inode); | |
1082 | struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; | |
1083 | int ret; | |
1084 | ||
06886a5a BH |
1085 | ret = exofs_check_io(ios, NULL); |
1086 | exofs_put_io_state(ios); | |
1087 | ||
e6af00f1 BH |
1088 | atomic_dec(&sbi->s_curr_pending); |
1089 | ||
1090 | if (unlikely(ret)) { | |
1091 | EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx", | |
45d3abcb | 1092 | _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid)); |
06886a5a BH |
1093 | /*TODO: When FS is corrupted creation can fail, object already |
1094 | * exist. Get rid of this asynchronous creation, if exist | |
1095 | * increment the obj counter and try the next object. Until we | |
1096 | * succeed. All these dangling objects will be made into lost | |
1097 | * files by chkfs.exofs | |
1098 | */ | |
1099 | } | |
1100 | ||
1101 | set_obj_created(oi); | |
e6af00f1 BH |
1102 | |
1103 | atomic_dec(&inode->i_count); | |
1104 | wake_up(&oi->i_wq); | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Set up a new inode and create an object for it on the OSD | |
1109 | */ | |
1110 | struct inode *exofs_new_inode(struct inode *dir, int mode) | |
1111 | { | |
1112 | struct super_block *sb; | |
1113 | struct inode *inode; | |
1114 | struct exofs_i_info *oi; | |
1115 | struct exofs_sb_info *sbi; | |
06886a5a | 1116 | struct exofs_io_state *ios; |
e6af00f1 BH |
1117 | int ret; |
1118 | ||
1119 | sb = dir->i_sb; | |
1120 | inode = new_inode(sb); | |
1121 | if (!inode) | |
1122 | return ERR_PTR(-ENOMEM); | |
1123 | ||
1124 | oi = exofs_i(inode); | |
9cfdc7aa | 1125 | __oi_init(oi); |
e6af00f1 | 1126 | |
e6af00f1 BH |
1127 | set_obj_2bcreated(oi); |
1128 | ||
1129 | sbi = sb->s_fs_info; | |
1130 | ||
1131 | sb->s_dirt = 1; | |
1132 | inode->i_uid = current->cred->fsuid; | |
1133 | if (dir->i_mode & S_ISGID) { | |
1134 | inode->i_gid = dir->i_gid; | |
1135 | if (S_ISDIR(mode)) | |
1136 | mode |= S_ISGID; | |
1137 | } else { | |
1138 | inode->i_gid = current->cred->fsgid; | |
1139 | } | |
1140 | inode->i_mode = mode; | |
1141 | ||
1142 | inode->i_ino = sbi->s_nextid++; | |
1143 | inode->i_blkbits = EXOFS_BLKSHIFT; | |
1144 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
1145 | oi->i_commit_size = inode->i_size = 0; | |
1146 | spin_lock(&sbi->s_next_gen_lock); | |
1147 | inode->i_generation = sbi->s_next_generation++; | |
1148 | spin_unlock(&sbi->s_next_gen_lock); | |
1149 | insert_inode_hash(inode); | |
1150 | ||
1151 | mark_inode_dirty(inode); | |
1152 | ||
45d3abcb | 1153 | ret = exofs_get_io_state(&sbi->layout, &ios); |
06886a5a BH |
1154 | if (unlikely(ret)) { |
1155 | EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n"); | |
1156 | return ERR_PTR(ret); | |
e6af00f1 BH |
1157 | } |
1158 | ||
06886a5a BH |
1159 | ios->obj.id = exofs_oi_objno(oi); |
1160 | exofs_make_credential(oi->i_cred, &ios->obj); | |
e6af00f1 BH |
1161 | |
1162 | /* increment the refcount so that the inode will still be around when we | |
1163 | * reach the callback | |
1164 | */ | |
1165 | atomic_inc(&inode->i_count); | |
1166 | ||
06886a5a BH |
1167 | ios->done = create_done; |
1168 | ios->private = inode; | |
1169 | ios->cred = oi->i_cred; | |
1170 | ret = exofs_sbi_create(ios); | |
e6af00f1 BH |
1171 | if (ret) { |
1172 | atomic_dec(&inode->i_count); | |
06886a5a BH |
1173 | exofs_put_io_state(ios); |
1174 | return ERR_PTR(ret); | |
e6af00f1 BH |
1175 | } |
1176 | atomic_inc(&sbi->s_curr_pending); | |
1177 | ||
1178 | return inode; | |
1179 | } | |
ba9e5e98 BH |
1180 | |
1181 | /* | |
1182 | * struct to pass two arguments to update_inode's callback | |
1183 | */ | |
1184 | struct updatei_args { | |
1185 | struct exofs_sb_info *sbi; | |
1186 | struct exofs_fcb fcb; | |
1187 | }; | |
1188 | ||
1189 | /* | |
1190 | * Callback function from exofs_update_inode(). | |
1191 | */ | |
06886a5a | 1192 | static void updatei_done(struct exofs_io_state *ios, void *p) |
ba9e5e98 BH |
1193 | { |
1194 | struct updatei_args *args = p; | |
1195 | ||
06886a5a | 1196 | exofs_put_io_state(ios); |
ba9e5e98 BH |
1197 | |
1198 | atomic_dec(&args->sbi->s_curr_pending); | |
1199 | ||
1200 | kfree(args); | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * Write the inode to the OSD. Just fill up the struct, and set the attribute | |
1205 | * synchronously or asynchronously depending on the do_sync flag. | |
1206 | */ | |
1207 | static int exofs_update_inode(struct inode *inode, int do_sync) | |
1208 | { | |
1209 | struct exofs_i_info *oi = exofs_i(inode); | |
1210 | struct super_block *sb = inode->i_sb; | |
1211 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
06886a5a | 1212 | struct exofs_io_state *ios; |
ba9e5e98 BH |
1213 | struct osd_attr attr; |
1214 | struct exofs_fcb *fcb; | |
1215 | struct updatei_args *args; | |
1216 | int ret; | |
1217 | ||
1218 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
34ce4e7c BH |
1219 | if (!args) { |
1220 | EXOFS_DBGMSG("Faild kzalloc of args\n"); | |
ba9e5e98 | 1221 | return -ENOMEM; |
34ce4e7c | 1222 | } |
ba9e5e98 BH |
1223 | |
1224 | fcb = &args->fcb; | |
1225 | ||
1226 | fcb->i_mode = cpu_to_le16(inode->i_mode); | |
1227 | fcb->i_uid = cpu_to_le32(inode->i_uid); | |
1228 | fcb->i_gid = cpu_to_le32(inode->i_gid); | |
1229 | fcb->i_links_count = cpu_to_le16(inode->i_nlink); | |
1230 | fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); | |
1231 | fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec); | |
1232 | fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); | |
1233 | oi->i_commit_size = i_size_read(inode); | |
1234 | fcb->i_size = cpu_to_le64(oi->i_commit_size); | |
1235 | fcb->i_generation = cpu_to_le32(inode->i_generation); | |
1236 | ||
1237 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
1238 | if (old_valid_dev(inode->i_rdev)) { | |
1239 | fcb->i_data[0] = | |
1240 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
1241 | fcb->i_data[1] = 0; | |
1242 | } else { | |
1243 | fcb->i_data[0] = 0; | |
1244 | fcb->i_data[1] = | |
1245 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
1246 | fcb->i_data[2] = 0; | |
1247 | } | |
1248 | } else | |
1249 | memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data)); | |
1250 | ||
45d3abcb | 1251 | ret = exofs_get_io_state(&sbi->layout, &ios); |
06886a5a BH |
1252 | if (unlikely(ret)) { |
1253 | EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__); | |
ba9e5e98 BH |
1254 | goto free_args; |
1255 | } | |
1256 | ||
ba9e5e98 BH |
1257 | attr = g_attr_inode_data; |
1258 | attr.val_ptr = fcb; | |
06886a5a BH |
1259 | ios->out_attr_len = 1; |
1260 | ios->out_attr = &attr; | |
ba9e5e98 BH |
1261 | |
1262 | if (!obj_created(oi)) { | |
1263 | EXOFS_DBGMSG("!obj_created\n"); | |
1264 | BUG_ON(!obj_2bcreated(oi)); | |
1265 | wait_event(oi->i_wq, obj_created(oi)); | |
1266 | EXOFS_DBGMSG("wait_event done\n"); | |
1267 | } | |
1268 | ||
06886a5a | 1269 | if (!do_sync) { |
ba9e5e98 | 1270 | args->sbi = sbi; |
06886a5a BH |
1271 | ios->done = updatei_done; |
1272 | ios->private = args; | |
1273 | } | |
ba9e5e98 | 1274 | |
06886a5a BH |
1275 | ret = exofs_oi_write(oi, ios); |
1276 | if (!do_sync && !ret) { | |
ba9e5e98 BH |
1277 | atomic_inc(&sbi->s_curr_pending); |
1278 | goto out; /* deallocation in updatei_done */ | |
1279 | } | |
1280 | ||
06886a5a | 1281 | exofs_put_io_state(ios); |
ba9e5e98 BH |
1282 | free_args: |
1283 | kfree(args); | |
1284 | out: | |
34ce4e7c BH |
1285 | EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n", |
1286 | inode->i_ino, do_sync, ret); | |
ba9e5e98 BH |
1287 | return ret; |
1288 | } | |
1289 | ||
1290 | int exofs_write_inode(struct inode *inode, int wait) | |
1291 | { | |
1292 | return exofs_update_inode(inode, wait); | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * Callback function from exofs_delete_inode() - don't have much cleaning up to | |
1297 | * do. | |
1298 | */ | |
06886a5a | 1299 | static void delete_done(struct exofs_io_state *ios, void *p) |
ba9e5e98 | 1300 | { |
06886a5a BH |
1301 | struct exofs_sb_info *sbi = p; |
1302 | ||
1303 | exofs_put_io_state(ios); | |
1304 | ||
ba9e5e98 BH |
1305 | atomic_dec(&sbi->s_curr_pending); |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * Called when the refcount of an inode reaches zero. We remove the object | |
1310 | * from the OSD here. We make sure the object was created before we try and | |
1311 | * delete it. | |
1312 | */ | |
1313 | void exofs_delete_inode(struct inode *inode) | |
1314 | { | |
1315 | struct exofs_i_info *oi = exofs_i(inode); | |
1316 | struct super_block *sb = inode->i_sb; | |
1317 | struct exofs_sb_info *sbi = sb->s_fs_info; | |
06886a5a | 1318 | struct exofs_io_state *ios; |
ba9e5e98 BH |
1319 | int ret; |
1320 | ||
1321 | truncate_inode_pages(&inode->i_data, 0); | |
1322 | ||
1323 | if (is_bad_inode(inode)) | |
1324 | goto no_delete; | |
1325 | ||
1326 | mark_inode_dirty(inode); | |
1327 | exofs_update_inode(inode, inode_needs_sync(inode)); | |
1328 | ||
1329 | inode->i_size = 0; | |
1330 | if (inode->i_blocks) | |
1331 | exofs_truncate(inode); | |
1332 | ||
1333 | clear_inode(inode); | |
1334 | ||
45d3abcb | 1335 | ret = exofs_get_io_state(&sbi->layout, &ios); |
06886a5a BH |
1336 | if (unlikely(ret)) { |
1337 | EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__); | |
ba9e5e98 BH |
1338 | return; |
1339 | } | |
1340 | ||
ba9e5e98 BH |
1341 | /* if we are deleting an obj that hasn't been created yet, wait */ |
1342 | if (!obj_created(oi)) { | |
1343 | BUG_ON(!obj_2bcreated(oi)); | |
1344 | wait_event(oi->i_wq, obj_created(oi)); | |
1345 | } | |
1346 | ||
06886a5a BH |
1347 | ios->obj.id = exofs_oi_objno(oi); |
1348 | ios->done = delete_done; | |
1349 | ios->private = sbi; | |
1350 | ios->cred = oi->i_cred; | |
1351 | ret = exofs_sbi_remove(ios); | |
ba9e5e98 | 1352 | if (ret) { |
06886a5a BH |
1353 | EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__); |
1354 | exofs_put_io_state(ios); | |
ba9e5e98 BH |
1355 | return; |
1356 | } | |
1357 | atomic_inc(&sbi->s_curr_pending); | |
1358 | ||
1359 | return; | |
1360 | ||
1361 | no_delete: | |
1362 | clear_inode(inode); | |
1363 | } |