blob: 4bb6ef822e46d8daba3b312afbb4258aab7ea1ee [file] [log] [blame]
Boaz Harroshe8062712008-10-27 18:37:02 +02001/*
2 * Copyright (C) 2005, 2006
Boaz Harrosh27d2e142009-06-14 17:23:09 +03003 * Avishay Traeger (avishay@gmail.com)
Boaz Harroshe8062712008-10-27 18:37:02 +02004 * 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
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090034#include <linux/slab.h>
Boaz Harroshe8062712008-10-27 18:37:02 +020035#include <linux/writeback.h>
36#include <linux/buffer_head.h>
Boaz Harroshbeaec072008-10-27 19:31:34 +020037#include <scsi/scsi_device.h>
Boaz Harroshe8062712008-10-27 18:37:02 +020038
39#include "exofs.h"
40
Boaz Harroshfe33cc12009-11-01 18:28:14 +020041#define EXOFS_DBGMSG2(M...) do {} while (0)
42
Boaz Harrosh06886a52009-11-08 14:54:08 +020043enum { BIO_MAX_PAGES_KMALLOC =
44 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
Boaz Harrosh86093aa2010-01-28 18:24:06 +020045 MAX_PAGES_KMALLOC =
46 PAGE_SIZE / sizeof(struct page *),
Boaz Harrosh06886a52009-11-08 14:54:08 +020047};
48
Boaz Harroshbeaec072008-10-27 19:31:34 +020049struct page_collect {
50 struct exofs_sb_info *sbi;
Boaz Harroshbeaec072008-10-27 19:31:34 +020051 struct inode *inode;
52 unsigned expected_pages;
Boaz Harrosh06886a52009-11-08 14:54:08 +020053 struct exofs_io_state *ios;
Boaz Harroshbeaec072008-10-27 19:31:34 +020054
Boaz Harrosh86093aa2010-01-28 18:24:06 +020055 struct page **pages;
56 unsigned alloc_pages;
Boaz Harroshbeaec072008-10-27 19:31:34 +020057 unsigned nr_pages;
58 unsigned long length;
59 loff_t pg_first; /* keep 64bit also in 32-arches */
60};
61
62static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
Boaz Harrosh06886a52009-11-08 14:54:08 +020063 struct inode *inode)
Boaz Harroshbeaec072008-10-27 19:31:34 +020064{
65 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
Boaz Harroshbeaec072008-10-27 19:31:34 +020066
67 pcol->sbi = sbi;
Boaz Harroshbeaec072008-10-27 19:31:34 +020068 pcol->inode = inode;
69 pcol->expected_pages = expected_pages;
70
Boaz Harrosh06886a52009-11-08 14:54:08 +020071 pcol->ios = NULL;
Boaz Harrosh86093aa2010-01-28 18:24:06 +020072 pcol->pages = NULL;
73 pcol->alloc_pages = 0;
Boaz Harroshbeaec072008-10-27 19:31:34 +020074 pcol->nr_pages = 0;
75 pcol->length = 0;
76 pcol->pg_first = -1;
Boaz Harroshbeaec072008-10-27 19:31:34 +020077}
78
79static void _pcol_reset(struct page_collect *pcol)
80{
81 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
82
Boaz Harrosh86093aa2010-01-28 18:24:06 +020083 pcol->pages = NULL;
84 pcol->alloc_pages = 0;
Boaz Harroshbeaec072008-10-27 19:31:34 +020085 pcol->nr_pages = 0;
86 pcol->length = 0;
87 pcol->pg_first = -1;
Boaz Harrosh06886a52009-11-08 14:54:08 +020088 pcol->ios = NULL;
Boaz Harroshbeaec072008-10-27 19:31:34 +020089
90 /* this is probably the end of the loop but in writes
91 * it might not end here. don't be left with nothing
92 */
93 if (!pcol->expected_pages)
Boaz Harrosh86093aa2010-01-28 18:24:06 +020094 pcol->expected_pages = MAX_PAGES_KMALLOC;
Boaz Harroshbeaec072008-10-27 19:31:34 +020095}
96
97static int pcol_try_alloc(struct page_collect *pcol)
98{
Boaz Harrosh86093aa2010-01-28 18:24:06 +020099 unsigned pages = min_t(unsigned, pcol->expected_pages,
100 MAX_PAGES_KMALLOC);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200101
102 if (!pcol->ios) { /* First time allocate io_state */
Boaz Harrosh45d3abc2010-01-28 11:46:16 +0200103 int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200104
105 if (ret)
106 return ret;
107 }
Boaz Harroshbeaec072008-10-27 19:31:34 +0200108
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200109 /* TODO: easily support bio chaining */
110 pages = min_t(unsigned, pages,
111 pcol->sbi->layout.group_width * BIO_MAX_PAGES_KMALLOC);
112
Boaz Harroshbeaec072008-10-27 19:31:34 +0200113 for (; pages; pages >>= 1) {
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200114 pcol->pages = kmalloc(pages * sizeof(struct page *),
115 GFP_KERNEL);
116 if (likely(pcol->pages)) {
117 pcol->alloc_pages = pages;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200118 return 0;
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200119 }
Boaz Harroshbeaec072008-10-27 19:31:34 +0200120 }
121
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200122 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
Boaz Harroshbeaec072008-10-27 19:31:34 +0200123 pcol->expected_pages);
124 return -ENOMEM;
125}
126
127static void pcol_free(struct page_collect *pcol)
128{
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200129 kfree(pcol->pages);
130 pcol->pages = NULL;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200131
132 if (pcol->ios) {
133 exofs_put_io_state(pcol->ios);
134 pcol->ios = NULL;
135 }
Boaz Harroshbeaec072008-10-27 19:31:34 +0200136}
137
138static int pcol_add_page(struct page_collect *pcol, struct page *page,
139 unsigned len)
140{
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200141 if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
Boaz Harroshbeaec072008-10-27 19:31:34 +0200142 return -ENOMEM;
143
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200144 pcol->pages[pcol->nr_pages++] = page;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200145 pcol->length += len;
146 return 0;
147}
148
149static int update_read_page(struct page *page, int ret)
150{
151 if (ret == 0) {
152 /* Everything is OK */
153 SetPageUptodate(page);
154 if (PageError(page))
155 ClearPageError(page);
156 } else if (ret == -EFAULT) {
157 /* In this case we were trying to read something that wasn't on
158 * disk yet - return a page full of zeroes. This should be OK,
159 * because the object should be empty (if there was a write
160 * before this read, the read would be waiting with the page
161 * locked */
162 clear_highpage(page);
163
164 SetPageUptodate(page);
165 if (PageError(page))
166 ClearPageError(page);
167 ret = 0; /* recovered error */
168 EXOFS_DBGMSG("recovered read error\n");
169 } else /* Error */
170 SetPageError(page);
171
172 return ret;
173}
174
175static void update_write_page(struct page *page, int ret)
176{
177 if (ret) {
178 mapping_set_error(page->mapping, ret);
179 SetPageError(page);
180 }
181 end_page_writeback(page);
182}
183
184/* Called at the end of reads, to optionally unlock pages and update their
185 * status.
186 */
Boaz Harrosh06886a52009-11-08 14:54:08 +0200187static int __readpages_done(struct page_collect *pcol, bool do_unlock)
Boaz Harroshbeaec072008-10-27 19:31:34 +0200188{
Boaz Harroshbeaec072008-10-27 19:31:34 +0200189 int i;
190 u64 resid;
191 u64 good_bytes;
192 u64 length = 0;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200193 int ret = exofs_check_io(pcol->ios, &resid);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200194
195 if (likely(!ret))
196 good_bytes = pcol->length;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200197 else
198 good_bytes = pcol->length - resid;
199
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200200 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
Boaz Harroshbeaec072008-10-27 19:31:34 +0200201 " length=0x%lx nr_pages=%u\n",
202 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
203 pcol->nr_pages);
204
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200205 for (i = 0; i < pcol->nr_pages; i++) {
206 struct page *page = pcol->pages[i];
Boaz Harroshbeaec072008-10-27 19:31:34 +0200207 struct inode *inode = page->mapping->host;
208 int page_stat;
209
210 if (inode != pcol->inode)
211 continue; /* osd might add more pages at end */
212
213 if (likely(length < good_bytes))
214 page_stat = 0;
215 else
216 page_stat = ret;
217
Boaz Harroshfe33cc12009-11-01 18:28:14 +0200218 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
Boaz Harroshbeaec072008-10-27 19:31:34 +0200219 inode->i_ino, page->index,
220 page_stat ? "bad_bytes" : "good_bytes");
221
222 ret = update_read_page(page, page_stat);
223 if (do_unlock)
224 unlock_page(page);
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200225 length += PAGE_SIZE;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200226 }
227
228 pcol_free(pcol);
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200229 EXOFS_DBGMSG2("readpages_done END\n");
Boaz Harroshbeaec072008-10-27 19:31:34 +0200230 return ret;
231}
232
233/* callback of async reads */
Boaz Harrosh06886a52009-11-08 14:54:08 +0200234static void readpages_done(struct exofs_io_state *ios, void *p)
Boaz Harroshbeaec072008-10-27 19:31:34 +0200235{
236 struct page_collect *pcol = p;
237
Boaz Harrosh06886a52009-11-08 14:54:08 +0200238 __readpages_done(pcol, true);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200239 atomic_dec(&pcol->sbi->s_curr_pending);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200240 kfree(pcol);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200241}
242
243static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
244{
Boaz Harroshbeaec072008-10-27 19:31:34 +0200245 int i;
246
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200247 for (i = 0; i < pcol->nr_pages; i++) {
248 struct page *page = pcol->pages[i];
Boaz Harroshbeaec072008-10-27 19:31:34 +0200249
250 if (rw == READ)
251 update_read_page(page, ret);
252 else
253 update_write_page(page, ret);
254
255 unlock_page(page);
256 }
Boaz Harroshbeaec072008-10-27 19:31:34 +0200257}
258
259static int read_exec(struct page_collect *pcol, bool is_sync)
260{
261 struct exofs_i_info *oi = exofs_i(pcol->inode);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200262 struct exofs_io_state *ios = pcol->ios;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200263 struct page_collect *pcol_copy = NULL;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200264 int ret;
265
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200266 if (!pcol->pages)
Boaz Harroshbeaec072008-10-27 19:31:34 +0200267 return 0;
268
269 /* see comment in _readpage() about sync reads */
270 WARN_ON(is_sync && (pcol->nr_pages != 1));
271
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200272 ios->pages = pcol->pages;
273 ios->nr_pages = pcol->nr_pages;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200274 ios->length = pcol->length;
275 ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200276
277 if (is_sync) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200278 exofs_oi_read(oi, pcol->ios);
279 return __readpages_done(pcol, false);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200280 }
281
282 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
283 if (!pcol_copy) {
284 ret = -ENOMEM;
285 goto err;
286 }
287
288 *pcol_copy = *pcol;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200289 ios->done = readpages_done;
290 ios->private = pcol_copy;
291 ret = exofs_oi_read(oi, ios);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200292 if (unlikely(ret))
293 goto err;
294
295 atomic_inc(&pcol->sbi->s_curr_pending);
296
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200297 EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
Boaz Harrosh06886a52009-11-08 14:54:08 +0200298 ios->obj.id, _LLU(ios->offset), pcol->length);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200299
300 /* pages ownership was passed to pcol_copy */
301 _pcol_reset(pcol);
302 return 0;
303
304err:
305 if (!is_sync)
306 _unlock_pcol_pages(pcol, ret, READ);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200307
308 pcol_free(pcol);
Boaz Harroshb76a3f92009-06-08 19:28:41 +0300309
Boaz Harroshbeaec072008-10-27 19:31:34 +0200310 kfree(pcol_copy);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200311 return ret;
312}
313
314/* readpage_strip is called either directly from readpage() or by the VFS from
315 * within read_cache_pages(), to add one more page to be read. It will try to
316 * collect as many contiguous pages as posible. If a discontinuity is
317 * encountered, or it runs out of resources, it will submit the previous segment
318 * and will start a new collection. Eventually caller must submit the last
319 * segment if present.
320 */
321static int readpage_strip(void *data, struct page *page)
322{
323 struct page_collect *pcol = data;
324 struct inode *inode = pcol->inode;
325 struct exofs_i_info *oi = exofs_i(inode);
326 loff_t i_size = i_size_read(inode);
327 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
328 size_t len;
329 int ret;
330
331 /* FIXME: Just for debugging, will be removed */
332 if (PageUptodate(page))
333 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
334 page->index);
335
336 if (page->index < end_index)
337 len = PAGE_CACHE_SIZE;
338 else if (page->index == end_index)
339 len = i_size & ~PAGE_CACHE_MASK;
340 else
341 len = 0;
342
343 if (!len || !obj_created(oi)) {
344 /* this will be out of bounds, or doesn't exist yet.
345 * Current page is cleared and the request is split
346 */
347 clear_highpage(page);
348
349 SetPageUptodate(page);
350 if (PageError(page))
351 ClearPageError(page);
352
353 unlock_page(page);
354 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
355 " splitting\n", inode->i_ino, page->index);
356
357 return read_exec(pcol, false);
358 }
359
360try_again:
361
362 if (unlikely(pcol->pg_first == -1)) {
363 pcol->pg_first = page->index;
364 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
365 page->index)) {
366 /* Discontinuity detected, split the request */
367 ret = read_exec(pcol, false);
368 if (unlikely(ret))
369 goto fail;
370 goto try_again;
371 }
372
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200373 if (!pcol->pages) {
Boaz Harroshbeaec072008-10-27 19:31:34 +0200374 ret = pcol_try_alloc(pcol);
375 if (unlikely(ret))
376 goto fail;
377 }
378
379 if (len != PAGE_CACHE_SIZE)
380 zero_user(page, len, PAGE_CACHE_SIZE - len);
381
Boaz Harroshfe33cc12009-11-01 18:28:14 +0200382 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
Boaz Harroshbeaec072008-10-27 19:31:34 +0200383 inode->i_ino, page->index, len);
384
385 ret = pcol_add_page(pcol, page, len);
386 if (ret) {
Boaz Harroshfe33cc12009-11-01 18:28:14 +0200387 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
Boaz Harroshbeaec072008-10-27 19:31:34 +0200388 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
389 page, len, pcol->nr_pages, pcol->length);
390
391 /* split the request, and start again with current page */
392 ret = read_exec(pcol, false);
393 if (unlikely(ret))
394 goto fail;
395
396 goto try_again;
397 }
398
399 return 0;
400
401fail:
402 /* SetPageError(page); ??? */
403 unlock_page(page);
404 return ret;
405}
406
407static int exofs_readpages(struct file *file, struct address_space *mapping,
408 struct list_head *pages, unsigned nr_pages)
409{
410 struct page_collect pcol;
411 int ret;
412
413 _pcol_init(&pcol, nr_pages, mapping->host);
414
415 ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
416 if (ret) {
417 EXOFS_ERR("read_cache_pages => %d\n", ret);
418 return ret;
419 }
420
421 return read_exec(&pcol, false);
422}
423
424static int _readpage(struct page *page, bool is_sync)
425{
426 struct page_collect pcol;
427 int ret;
428
429 _pcol_init(&pcol, 1, page->mapping->host);
430
Boaz Harrosh06886a52009-11-08 14:54:08 +0200431 /* readpage_strip might call read_exec(,is_sync==false) at several
432 * places but not if we have a single page.
Boaz Harroshbeaec072008-10-27 19:31:34 +0200433 */
434 ret = readpage_strip(&pcol, page);
435 if (ret) {
436 EXOFS_ERR("_readpage => %d\n", ret);
437 return ret;
438 }
439
440 return read_exec(&pcol, is_sync);
441}
442
443/*
444 * We don't need the file
445 */
446static int exofs_readpage(struct file *file, struct page *page)
447{
448 return _readpage(page, false);
449}
450
Boaz Harrosh06886a52009-11-08 14:54:08 +0200451/* Callback for osd_write. All writes are asynchronous */
452static void writepages_done(struct exofs_io_state *ios, void *p)
Boaz Harroshbeaec072008-10-27 19:31:34 +0200453{
454 struct page_collect *pcol = p;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200455 int i;
456 u64 resid;
457 u64 good_bytes;
458 u64 length = 0;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200459 int ret = exofs_check_io(ios, &resid);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200460
Boaz Harroshbeaec072008-10-27 19:31:34 +0200461 atomic_dec(&pcol->sbi->s_curr_pending);
462
463 if (likely(!ret))
464 good_bytes = pcol->length;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200465 else
466 good_bytes = pcol->length - resid;
467
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200468 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
Boaz Harroshbeaec072008-10-27 19:31:34 +0200469 " length=0x%lx nr_pages=%u\n",
470 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
471 pcol->nr_pages);
472
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200473 for (i = 0; i < pcol->nr_pages; i++) {
474 struct page *page = pcol->pages[i];
Boaz Harroshbeaec072008-10-27 19:31:34 +0200475 struct inode *inode = page->mapping->host;
476 int page_stat;
477
478 if (inode != pcol->inode)
479 continue; /* osd might add more pages to a bio */
480
481 if (likely(length < good_bytes))
482 page_stat = 0;
483 else
484 page_stat = ret;
485
486 update_write_page(page, page_stat);
487 unlock_page(page);
Boaz Harroshfe33cc12009-11-01 18:28:14 +0200488 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
Boaz Harroshbeaec072008-10-27 19:31:34 +0200489 inode->i_ino, page->index, page_stat);
490
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200491 length += PAGE_SIZE;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200492 }
493
494 pcol_free(pcol);
495 kfree(pcol);
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200496 EXOFS_DBGMSG2("writepages_done END\n");
Boaz Harroshbeaec072008-10-27 19:31:34 +0200497}
498
499static int write_exec(struct page_collect *pcol)
500{
501 struct exofs_i_info *oi = exofs_i(pcol->inode);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200502 struct exofs_io_state *ios = pcol->ios;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200503 struct page_collect *pcol_copy = NULL;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200504 int ret;
505
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200506 if (!pcol->pages)
Boaz Harroshbeaec072008-10-27 19:31:34 +0200507 return 0;
508
Boaz Harroshbeaec072008-10-27 19:31:34 +0200509 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
510 if (!pcol_copy) {
511 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n");
512 ret = -ENOMEM;
513 goto err;
514 }
515
516 *pcol_copy = *pcol;
517
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200518 ios->pages = pcol_copy->pages;
519 ios->nr_pages = pcol_copy->nr_pages;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200520 ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT;
521 ios->length = pcol_copy->length;
522 ios->done = writepages_done;
523 ios->private = pcol_copy;
524
525 ret = exofs_oi_write(oi, ios);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200526 if (unlikely(ret)) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200527 EXOFS_ERR("write_exec: exofs_oi_write() Faild\n");
Boaz Harroshbeaec072008-10-27 19:31:34 +0200528 goto err;
529 }
530
531 atomic_inc(&pcol->sbi->s_curr_pending);
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200532 EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
Boaz Harrosh06886a52009-11-08 14:54:08 +0200533 pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset),
Boaz Harroshbeaec072008-10-27 19:31:34 +0200534 pcol->length);
535 /* pages ownership was passed to pcol_copy */
536 _pcol_reset(pcol);
537 return 0;
538
539err:
540 _unlock_pcol_pages(pcol, ret, WRITE);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200541 pcol_free(pcol);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200542 kfree(pcol_copy);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200543
Boaz Harroshbeaec072008-10-27 19:31:34 +0200544 return ret;
545}
546
547/* writepage_strip is called either directly from writepage() or by the VFS from
548 * within write_cache_pages(), to add one more page to be written to storage.
549 * It will try to collect as many contiguous pages as possible. If a
550 * discontinuity is encountered or it runs out of resources it will submit the
551 * previous segment and will start a new collection.
552 * Eventually caller must submit the last segment if present.
553 */
554static int writepage_strip(struct page *page,
555 struct writeback_control *wbc_unused, void *data)
556{
557 struct page_collect *pcol = data;
558 struct inode *inode = pcol->inode;
559 struct exofs_i_info *oi = exofs_i(inode);
560 loff_t i_size = i_size_read(inode);
561 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
562 size_t len;
563 int ret;
564
565 BUG_ON(!PageLocked(page));
566
567 ret = wait_obj_created(oi);
568 if (unlikely(ret))
569 goto fail;
570
571 if (page->index < end_index)
572 /* in this case, the page is within the limits of the file */
573 len = PAGE_CACHE_SIZE;
574 else {
575 len = i_size & ~PAGE_CACHE_MASK;
576
577 if (page->index > end_index || !len) {
578 /* in this case, the page is outside the limits
579 * (truncate in progress)
580 */
581 ret = write_exec(pcol);
582 if (unlikely(ret))
583 goto fail;
584 if (PageError(page))
585 ClearPageError(page);
586 unlock_page(page);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200587 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
588 "outside the limits\n",
589 inode->i_ino, page->index);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200590 return 0;
591 }
592 }
593
594try_again:
595
596 if (unlikely(pcol->pg_first == -1)) {
597 pcol->pg_first = page->index;
598 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
599 page->index)) {
600 /* Discontinuity detected, split the request */
601 ret = write_exec(pcol);
602 if (unlikely(ret))
603 goto fail;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200604
605 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
606 inode->i_ino, page->index);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200607 goto try_again;
608 }
609
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200610 if (!pcol->pages) {
Boaz Harroshbeaec072008-10-27 19:31:34 +0200611 ret = pcol_try_alloc(pcol);
612 if (unlikely(ret))
613 goto fail;
614 }
615
Boaz Harroshfe33cc12009-11-01 18:28:14 +0200616 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
Boaz Harroshbeaec072008-10-27 19:31:34 +0200617 inode->i_ino, page->index, len);
618
619 ret = pcol_add_page(pcol, page, len);
620 if (unlikely(ret)) {
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200621 EXOFS_DBGMSG2("Failed pcol_add_page "
Boaz Harroshbeaec072008-10-27 19:31:34 +0200622 "nr_pages=%u total_length=0x%lx\n",
623 pcol->nr_pages, pcol->length);
624
625 /* split the request, next loop will start again */
626 ret = write_exec(pcol);
627 if (unlikely(ret)) {
628 EXOFS_DBGMSG("write_exec faild => %d", ret);
629 goto fail;
630 }
631
632 goto try_again;
633 }
634
635 BUG_ON(PageWriteback(page));
636 set_page_writeback(page);
637
638 return 0;
639
640fail:
Boaz Harrosh06886a52009-11-08 14:54:08 +0200641 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
642 inode->i_ino, page->index, ret);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200643 set_bit(AS_EIO, &page->mapping->flags);
644 unlock_page(page);
645 return ret;
646}
647
648static int exofs_writepages(struct address_space *mapping,
649 struct writeback_control *wbc)
650{
651 struct page_collect pcol;
652 long start, end, expected_pages;
653 int ret;
654
655 start = wbc->range_start >> PAGE_CACHE_SHIFT;
656 end = (wbc->range_end == LLONG_MAX) ?
657 start + mapping->nrpages :
658 wbc->range_end >> PAGE_CACHE_SHIFT;
659
660 if (start || end)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200661 expected_pages = end - start + 1;
Boaz Harroshbeaec072008-10-27 19:31:34 +0200662 else
663 expected_pages = mapping->nrpages;
664
Boaz Harrosh06886a52009-11-08 14:54:08 +0200665 if (expected_pages < 32L)
666 expected_pages = 32L;
667
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200668 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
Boaz Harrosh06886a52009-11-08 14:54:08 +0200669 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
Boaz Harroshbeaec072008-10-27 19:31:34 +0200670 mapping->host->i_ino, wbc->range_start, wbc->range_end,
Boaz Harrosh06886a52009-11-08 14:54:08 +0200671 mapping->nrpages, start, end, expected_pages);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200672
673 _pcol_init(&pcol, expected_pages, mapping->host);
674
675 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
676 if (ret) {
677 EXOFS_ERR("write_cache_pages => %d\n", ret);
678 return ret;
679 }
680
681 return write_exec(&pcol);
682}
683
684static int exofs_writepage(struct page *page, struct writeback_control *wbc)
685{
686 struct page_collect pcol;
687 int ret;
688
689 _pcol_init(&pcol, 1, page->mapping->host);
690
691 ret = writepage_strip(page, NULL, &pcol);
692 if (ret) {
693 EXOFS_ERR("exofs_writepage => %d\n", ret);
694 return ret;
695 }
696
697 return write_exec(&pcol);
698}
699
700int exofs_write_begin(struct file *file, struct address_space *mapping,
701 loff_t pos, unsigned len, unsigned flags,
702 struct page **pagep, void **fsdata)
703{
704 int ret = 0;
705 struct page *page;
706
707 page = *pagep;
708 if (page == NULL) {
709 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
710 fsdata);
711 if (ret) {
712 EXOFS_DBGMSG("simple_write_begin faild\n");
713 return ret;
714 }
715
716 page = *pagep;
717 }
718
719 /* read modify write */
720 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
721 ret = _readpage(page, true);
722 if (ret) {
723 /*SetPageError was done by _readpage. Is it ok?*/
724 unlock_page(page);
725 EXOFS_DBGMSG("__readpage_filler faild\n");
726 }
727 }
728
729 return ret;
730}
731
732static int exofs_write_begin_export(struct file *file,
733 struct address_space *mapping,
734 loff_t pos, unsigned len, unsigned flags,
735 struct page **pagep, void **fsdata)
736{
737 *pagep = NULL;
738
739 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
740 fsdata);
741}
742
Boaz Harroshefd124b2009-12-27 17:01:42 +0200743static int exofs_write_end(struct file *file, struct address_space *mapping,
744 loff_t pos, unsigned len, unsigned copied,
745 struct page *page, void *fsdata)
746{
747 struct inode *inode = mapping->host;
748 /* According to comment in simple_write_end i_mutex is held */
749 loff_t i_size = inode->i_size;
750 int ret;
751
752 ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
753 if (i_size != inode->i_size)
754 mark_inode_dirty(inode);
755 return ret;
756}
757
Boaz Harrosh200b0702010-03-22 11:23:40 +0200758static int exofs_releasepage(struct page *page, gfp_t gfp)
759{
760 EXOFS_DBGMSG("page 0x%lx\n", page->index);
761 WARN_ON(1);
762 return try_to_free_buffers(page);
763}
764
765static void exofs_invalidatepage(struct page *page, unsigned long offset)
766{
767 EXOFS_DBGMSG("page_has_buffers=>%d\n", page_has_buffers(page));
768 WARN_ON(1);
769
770 block_invalidatepage(page, offset);
771}
772
Boaz Harroshbeaec072008-10-27 19:31:34 +0200773const struct address_space_operations exofs_aops = {
774 .readpage = exofs_readpage,
775 .readpages = exofs_readpages,
776 .writepage = exofs_writepage,
777 .writepages = exofs_writepages,
778 .write_begin = exofs_write_begin_export,
Boaz Harroshefd124b2009-12-27 17:01:42 +0200779 .write_end = exofs_write_end,
Boaz Harrosh200b0702010-03-22 11:23:40 +0200780 .releasepage = exofs_releasepage,
781 .set_page_dirty = __set_page_dirty_nobuffers,
782 .invalidatepage = exofs_invalidatepage,
783
784 /* Not implemented Yet */
785 .bmap = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
786 .direct_IO = NULL, /* TODO: Should be trivial to do */
787
788 /* With these NULL has special meaning or default is not exported */
789 .sync_page = NULL,
790 .get_xip_mem = NULL,
791 .migratepage = NULL,
792 .launder_page = NULL,
793 .is_partially_uptodate = NULL,
794 .error_remove_page = NULL,
Boaz Harroshbeaec072008-10-27 19:31:34 +0200795};
796
Boaz Harroshe8062712008-10-27 18:37:02 +0200797/******************************************************************************
798 * INODE OPERATIONS
799 *****************************************************************************/
800
801/*
802 * Test whether an inode is a fast symlink.
803 */
804static inline int exofs_inode_is_fast_symlink(struct inode *inode)
805{
806 struct exofs_i_info *oi = exofs_i(inode);
807
808 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
809}
810
811/*
812 * get_block_t - Fill in a buffer_head
813 * An OSD takes care of block allocation so we just fake an allocation by
814 * putting in the inode's sector_t in the buffer_head.
815 * TODO: What about the case of create==0 and @iblock does not exist in the
816 * object?
817 */
818static int exofs_get_block(struct inode *inode, sector_t iblock,
819 struct buffer_head *bh_result, int create)
820{
821 map_bh(bh_result, inode->i_sb, iblock);
822 return 0;
823}
824
825const struct osd_attr g_attr_logical_length = ATTR_DEF(
826 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
827
Boaz Harrosh06886a52009-11-08 14:54:08 +0200828static int _do_truncate(struct inode *inode)
829{
830 struct exofs_i_info *oi = exofs_i(inode);
831 loff_t isize = i_size_read(inode);
832 int ret;
833
834 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
835
836 nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
837
838 ret = exofs_oi_truncate(oi, (u64)isize);
839 EXOFS_DBGMSG("(0x%lx) size=0x%llx\n", inode->i_ino, isize);
840 return ret;
841}
842
Boaz Harroshe8062712008-10-27 18:37:02 +0200843/*
844 * Truncate a file to the specified size - all we have to do is set the size
845 * attribute. We make sure the object exists first.
846 */
847void exofs_truncate(struct inode *inode)
848{
Boaz Harroshe8062712008-10-27 18:37:02 +0200849 struct exofs_i_info *oi = exofs_i(inode);
Boaz Harroshe8062712008-10-27 18:37:02 +0200850 int ret;
851
852 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
853 || S_ISLNK(inode->i_mode)))
854 return;
855 if (exofs_inode_is_fast_symlink(inode))
856 return;
857 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
858 return;
Boaz Harroshe8062712008-10-27 18:37:02 +0200859
860 /* if we are about to truncate an object, and it hasn't been
861 * created yet, wait
862 */
863 if (unlikely(wait_obj_created(oi)))
864 goto fail;
865
Boaz Harrosh06886a52009-11-08 14:54:08 +0200866 ret = _do_truncate(inode);
Boaz Harroshe8062712008-10-27 18:37:02 +0200867 if (ret)
868 goto fail;
869
870out:
871 mark_inode_dirty(inode);
872 return;
873fail:
874 make_bad_inode(inode);
875 goto out;
876}
877
878/*
879 * Set inode attributes - just call generic functions.
880 */
881int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
882{
883 struct inode *inode = dentry->d_inode;
884 int error;
885
886 error = inode_change_ok(inode, iattr);
887 if (error)
888 return error;
889
890 error = inode_setattr(inode, iattr);
891 return error;
892}
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200893
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200894static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
895 EXOFS_APAGE_FS_DATA,
896 EXOFS_ATTR_INODE_FILE_LAYOUT,
897 0);
898static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
899 EXOFS_APAGE_FS_DATA,
900 EXOFS_ATTR_INODE_DIR_LAYOUT,
901 0);
902
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200903/*
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200904 * Read the Linux inode info from the OSD, and return it as is. In exofs the
905 * inode info is in an application specific page/attribute of the osd-object.
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200906 */
907static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200908 struct exofs_fcb *inode)
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200909{
910 struct exofs_sb_info *sbi = sb->s_fs_info;
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200911 struct osd_attr attrs[] = {
912 [0] = g_attr_inode_data,
913 [1] = g_attr_inode_file_layout,
914 [2] = g_attr_inode_dir_layout,
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200915 };
Boaz Harrosh06886a52009-11-08 14:54:08 +0200916 struct exofs_io_state *ios;
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200917 struct exofs_on_disk_inode_layout *layout;
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200918 int ret;
919
Boaz Harrosh45d3abc2010-01-28 11:46:16 +0200920 ret = exofs_get_io_state(&sbi->layout, &ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200921 if (unlikely(ret)) {
922 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
923 return ret;
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200924 }
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200925
Boaz Harrosh06886a52009-11-08 14:54:08 +0200926 ios->obj.id = exofs_oi_objno(oi);
927 exofs_make_credential(oi->i_cred, &ios->obj);
928 ios->cred = oi->i_cred;
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200929
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200930 attrs[1].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
931 attrs[2].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
932
Boaz Harrosh06886a52009-11-08 14:54:08 +0200933 ios->in_attr = attrs;
934 ios->in_attr_len = ARRAY_SIZE(attrs);
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200935
Boaz Harrosh06886a52009-11-08 14:54:08 +0200936 ret = exofs_sbi_read(ios);
Boaz Harrosh96391e22010-02-09 11:43:21 +0200937 if (unlikely(ret)) {
938 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
939 _LLU(ios->obj.id), ret);
940 memset(inode, 0, sizeof(*inode));
941 inode->i_mode = 0040000 | (0777 & ~022);
942 /* If object is lost on target we might as well enable it's
943 * delete.
944 */
945 if ((ret == -ENOENT) || (ret == -EINVAL))
946 ret = 0;
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200947 goto out;
Boaz Harrosh96391e22010-02-09 11:43:21 +0200948 }
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200949
Boaz Harrosh06886a52009-11-08 14:54:08 +0200950 ret = extract_attr_from_ios(ios, &attrs[0]);
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200951 if (ret) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200952 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200953 goto out;
954 }
Boaz Harrosh06886a52009-11-08 14:54:08 +0200955 WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
956 memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200957
Boaz Harrosh06886a52009-11-08 14:54:08 +0200958 ret = extract_attr_from_ios(ios, &attrs[1]);
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200959 if (ret) {
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200960 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
961 goto out;
962 }
963 if (attrs[1].len) {
964 layout = attrs[1].val_ptr;
965 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
966 EXOFS_ERR("%s: unsupported files layout %d\n",
967 __func__, layout->gen_func);
968 ret = -ENOTSUPP;
969 goto out;
970 }
971 }
972
973 ret = extract_attr_from_ios(ios, &attrs[2]);
974 if (ret) {
975 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
976 goto out;
977 }
978 if (attrs[2].len) {
979 layout = attrs[2].val_ptr;
980 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
981 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
982 __func__, layout->gen_func);
983 ret = -ENOTSUPP;
984 goto out;
985 }
986 }
987
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200988out:
Boaz Harrosh06886a52009-11-08 14:54:08 +0200989 exofs_put_io_state(ios);
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200990 return ret;
991}
992
Boaz Harrosh9cfdc7a2009-08-04 20:40:29 +0300993static void __oi_init(struct exofs_i_info *oi)
994{
995 init_waitqueue_head(&oi->i_wq);
996 oi->i_flags = 0;
997}
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200998/*
999 * Fill in an inode read from the OSD and set it up for use
1000 */
1001struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
1002{
1003 struct exofs_i_info *oi;
1004 struct exofs_fcb fcb;
1005 struct inode *inode;
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001006 int ret;
1007
1008 inode = iget_locked(sb, ino);
1009 if (!inode)
1010 return ERR_PTR(-ENOMEM);
1011 if (!(inode->i_state & I_NEW))
1012 return inode;
1013 oi = exofs_i(inode);
Boaz Harrosh9cfdc7a2009-08-04 20:40:29 +03001014 __oi_init(oi);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001015
1016 /* read the inode from the osd */
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001017 ret = exofs_get_inode(sb, oi, &fcb);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001018 if (ret)
1019 goto bad_inode;
1020
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001021 set_obj_created(oi);
1022
1023 /* copy stuff from on-disk struct to in-memory struct */
1024 inode->i_mode = le16_to_cpu(fcb.i_mode);
1025 inode->i_uid = le32_to_cpu(fcb.i_uid);
1026 inode->i_gid = le32_to_cpu(fcb.i_gid);
1027 inode->i_nlink = le16_to_cpu(fcb.i_links_count);
1028 inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1029 inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1030 inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1031 inode->i_ctime.tv_nsec =
1032 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1033 oi->i_commit_size = le64_to_cpu(fcb.i_size);
1034 i_size_write(inode, oi->i_commit_size);
1035 inode->i_blkbits = EXOFS_BLKSHIFT;
1036 inode->i_generation = le32_to_cpu(fcb.i_generation);
1037
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001038 oi->i_dir_start_lookup = 0;
1039
1040 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1041 ret = -ESTALE;
1042 goto bad_inode;
1043 }
1044
1045 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1046 if (fcb.i_data[0])
1047 inode->i_rdev =
1048 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1049 else
1050 inode->i_rdev =
1051 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1052 } else {
1053 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1054 }
1055
1056 if (S_ISREG(inode->i_mode)) {
1057 inode->i_op = &exofs_file_inode_operations;
1058 inode->i_fop = &exofs_file_operations;
1059 inode->i_mapping->a_ops = &exofs_aops;
1060 } else if (S_ISDIR(inode->i_mode)) {
1061 inode->i_op = &exofs_dir_inode_operations;
1062 inode->i_fop = &exofs_dir_operations;
1063 inode->i_mapping->a_ops = &exofs_aops;
1064 } else if (S_ISLNK(inode->i_mode)) {
1065 if (exofs_inode_is_fast_symlink(inode))
1066 inode->i_op = &exofs_fast_symlink_inode_operations;
1067 else {
1068 inode->i_op = &exofs_symlink_inode_operations;
1069 inode->i_mapping->a_ops = &exofs_aops;
1070 }
1071 } else {
1072 inode->i_op = &exofs_special_inode_operations;
1073 if (fcb.i_data[0])
1074 init_special_inode(inode, inode->i_mode,
1075 old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1076 else
1077 init_special_inode(inode, inode->i_mode,
1078 new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1079 }
1080
1081 unlock_new_inode(inode);
1082 return inode;
1083
1084bad_inode:
1085 iget_failed(inode);
1086 return ERR_PTR(ret);
1087}
1088
1089int __exofs_wait_obj_created(struct exofs_i_info *oi)
1090{
1091 if (!obj_created(oi)) {
1092 BUG_ON(!obj_2bcreated(oi));
1093 wait_event(oi->i_wq, obj_created(oi));
1094 }
1095 return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1096}
1097/*
1098 * Callback function from exofs_new_inode(). The important thing is that we
1099 * set the obj_created flag so that other methods know that the object exists on
1100 * the OSD.
1101 */
Boaz Harrosh06886a52009-11-08 14:54:08 +02001102static void create_done(struct exofs_io_state *ios, void *p)
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001103{
1104 struct inode *inode = p;
1105 struct exofs_i_info *oi = exofs_i(inode);
1106 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1107 int ret;
1108
Boaz Harrosh06886a52009-11-08 14:54:08 +02001109 ret = exofs_check_io(ios, NULL);
1110 exofs_put_io_state(ios);
1111
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001112 atomic_dec(&sbi->s_curr_pending);
1113
1114 if (unlikely(ret)) {
1115 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
Boaz Harrosh45d3abc2010-01-28 11:46:16 +02001116 _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid));
Boaz Harrosh06886a52009-11-08 14:54:08 +02001117 /*TODO: When FS is corrupted creation can fail, object already
1118 * exist. Get rid of this asynchronous creation, if exist
1119 * increment the obj counter and try the next object. Until we
1120 * succeed. All these dangling objects will be made into lost
1121 * files by chkfs.exofs
1122 */
1123 }
1124
1125 set_obj_created(oi);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001126
1127 atomic_dec(&inode->i_count);
1128 wake_up(&oi->i_wq);
1129}
1130
1131/*
1132 * Set up a new inode and create an object for it on the OSD
1133 */
1134struct inode *exofs_new_inode(struct inode *dir, int mode)
1135{
1136 struct super_block *sb;
1137 struct inode *inode;
1138 struct exofs_i_info *oi;
1139 struct exofs_sb_info *sbi;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001140 struct exofs_io_state *ios;
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001141 int ret;
1142
1143 sb = dir->i_sb;
1144 inode = new_inode(sb);
1145 if (!inode)
1146 return ERR_PTR(-ENOMEM);
1147
1148 oi = exofs_i(inode);
Boaz Harrosh9cfdc7a2009-08-04 20:40:29 +03001149 __oi_init(oi);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001150
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001151 set_obj_2bcreated(oi);
1152
1153 sbi = sb->s_fs_info;
1154
1155 sb->s_dirt = 1;
Dmitry Monakhove00117f2010-03-04 17:31:48 +03001156 inode_init_owner(inode, dir, mode);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001157 inode->i_ino = sbi->s_nextid++;
1158 inode->i_blkbits = EXOFS_BLKSHIFT;
1159 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1160 oi->i_commit_size = inode->i_size = 0;
1161 spin_lock(&sbi->s_next_gen_lock);
1162 inode->i_generation = sbi->s_next_generation++;
1163 spin_unlock(&sbi->s_next_gen_lock);
1164 insert_inode_hash(inode);
1165
1166 mark_inode_dirty(inode);
1167
Boaz Harrosh45d3abc2010-01-28 11:46:16 +02001168 ret = exofs_get_io_state(&sbi->layout, &ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +02001169 if (unlikely(ret)) {
1170 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1171 return ERR_PTR(ret);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001172 }
1173
Boaz Harrosh06886a52009-11-08 14:54:08 +02001174 ios->obj.id = exofs_oi_objno(oi);
1175 exofs_make_credential(oi->i_cred, &ios->obj);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001176
1177 /* increment the refcount so that the inode will still be around when we
1178 * reach the callback
1179 */
1180 atomic_inc(&inode->i_count);
1181
Boaz Harrosh06886a52009-11-08 14:54:08 +02001182 ios->done = create_done;
1183 ios->private = inode;
1184 ios->cred = oi->i_cred;
1185 ret = exofs_sbi_create(ios);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001186 if (ret) {
1187 atomic_dec(&inode->i_count);
Boaz Harrosh06886a52009-11-08 14:54:08 +02001188 exofs_put_io_state(ios);
1189 return ERR_PTR(ret);
Boaz Harroshe6af00f2008-10-28 15:38:12 +02001190 }
1191 atomic_inc(&sbi->s_curr_pending);
1192
1193 return inode;
1194}
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001195
1196/*
1197 * struct to pass two arguments to update_inode's callback
1198 */
1199struct updatei_args {
1200 struct exofs_sb_info *sbi;
1201 struct exofs_fcb fcb;
1202};
1203
1204/*
1205 * Callback function from exofs_update_inode().
1206 */
Boaz Harrosh06886a52009-11-08 14:54:08 +02001207static void updatei_done(struct exofs_io_state *ios, void *p)
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001208{
1209 struct updatei_args *args = p;
1210
Boaz Harrosh06886a52009-11-08 14:54:08 +02001211 exofs_put_io_state(ios);
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001212
1213 atomic_dec(&args->sbi->s_curr_pending);
1214
1215 kfree(args);
1216}
1217
1218/*
1219 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1220 * synchronously or asynchronously depending on the do_sync flag.
1221 */
1222static int exofs_update_inode(struct inode *inode, int do_sync)
1223{
1224 struct exofs_i_info *oi = exofs_i(inode);
1225 struct super_block *sb = inode->i_sb;
1226 struct exofs_sb_info *sbi = sb->s_fs_info;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001227 struct exofs_io_state *ios;
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001228 struct osd_attr attr;
1229 struct exofs_fcb *fcb;
1230 struct updatei_args *args;
1231 int ret;
1232
1233 args = kzalloc(sizeof(*args), GFP_KERNEL);
Boaz Harrosh34ce4e72009-12-15 19:34:17 +02001234 if (!args) {
1235 EXOFS_DBGMSG("Faild kzalloc of args\n");
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001236 return -ENOMEM;
Boaz Harrosh34ce4e72009-12-15 19:34:17 +02001237 }
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001238
1239 fcb = &args->fcb;
1240
1241 fcb->i_mode = cpu_to_le16(inode->i_mode);
1242 fcb->i_uid = cpu_to_le32(inode->i_uid);
1243 fcb->i_gid = cpu_to_le32(inode->i_gid);
1244 fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1245 fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1246 fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1247 fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1248 oi->i_commit_size = i_size_read(inode);
1249 fcb->i_size = cpu_to_le64(oi->i_commit_size);
1250 fcb->i_generation = cpu_to_le32(inode->i_generation);
1251
1252 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1253 if (old_valid_dev(inode->i_rdev)) {
1254 fcb->i_data[0] =
1255 cpu_to_le32(old_encode_dev(inode->i_rdev));
1256 fcb->i_data[1] = 0;
1257 } else {
1258 fcb->i_data[0] = 0;
1259 fcb->i_data[1] =
1260 cpu_to_le32(new_encode_dev(inode->i_rdev));
1261 fcb->i_data[2] = 0;
1262 }
1263 } else
1264 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1265
Boaz Harrosh45d3abc2010-01-28 11:46:16 +02001266 ret = exofs_get_io_state(&sbi->layout, &ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +02001267 if (unlikely(ret)) {
1268 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001269 goto free_args;
1270 }
1271
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001272 attr = g_attr_inode_data;
1273 attr.val_ptr = fcb;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001274 ios->out_attr_len = 1;
1275 ios->out_attr = &attr;
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001276
1277 if (!obj_created(oi)) {
1278 EXOFS_DBGMSG("!obj_created\n");
1279 BUG_ON(!obj_2bcreated(oi));
1280 wait_event(oi->i_wq, obj_created(oi));
1281 EXOFS_DBGMSG("wait_event done\n");
1282 }
1283
Boaz Harrosh06886a52009-11-08 14:54:08 +02001284 if (!do_sync) {
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001285 args->sbi = sbi;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001286 ios->done = updatei_done;
1287 ios->private = args;
1288 }
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001289
Boaz Harrosh06886a52009-11-08 14:54:08 +02001290 ret = exofs_oi_write(oi, ios);
1291 if (!do_sync && !ret) {
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001292 atomic_inc(&sbi->s_curr_pending);
1293 goto out; /* deallocation in updatei_done */
1294 }
1295
Boaz Harrosh06886a52009-11-08 14:54:08 +02001296 exofs_put_io_state(ios);
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001297free_args:
1298 kfree(args);
1299out:
Boaz Harrosh34ce4e72009-12-15 19:34:17 +02001300 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1301 inode->i_ino, do_sync, ret);
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001302 return ret;
1303}
1304
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001305int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001306{
Christoph Hellwiga9185b42010-03-05 09:21:37 +01001307 return exofs_update_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001308}
1309
1310/*
1311 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1312 * do.
1313 */
Boaz Harrosh06886a52009-11-08 14:54:08 +02001314static void delete_done(struct exofs_io_state *ios, void *p)
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001315{
Boaz Harrosh06886a52009-11-08 14:54:08 +02001316 struct exofs_sb_info *sbi = p;
1317
1318 exofs_put_io_state(ios);
1319
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001320 atomic_dec(&sbi->s_curr_pending);
1321}
1322
1323/*
1324 * Called when the refcount of an inode reaches zero. We remove the object
1325 * from the OSD here. We make sure the object was created before we try and
1326 * delete it.
1327 */
1328void exofs_delete_inode(struct inode *inode)
1329{
1330 struct exofs_i_info *oi = exofs_i(inode);
1331 struct super_block *sb = inode->i_sb;
1332 struct exofs_sb_info *sbi = sb->s_fs_info;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001333 struct exofs_io_state *ios;
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001334 int ret;
1335
1336 truncate_inode_pages(&inode->i_data, 0);
1337
1338 if (is_bad_inode(inode))
1339 goto no_delete;
1340
1341 mark_inode_dirty(inode);
1342 exofs_update_inode(inode, inode_needs_sync(inode));
1343
1344 inode->i_size = 0;
1345 if (inode->i_blocks)
1346 exofs_truncate(inode);
1347
1348 clear_inode(inode);
1349
Boaz Harrosh45d3abc2010-01-28 11:46:16 +02001350 ret = exofs_get_io_state(&sbi->layout, &ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +02001351 if (unlikely(ret)) {
1352 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__);
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001353 return;
1354 }
1355
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001356 /* if we are deleting an obj that hasn't been created yet, wait */
1357 if (!obj_created(oi)) {
1358 BUG_ON(!obj_2bcreated(oi));
1359 wait_event(oi->i_wq, obj_created(oi));
1360 }
1361
Boaz Harrosh06886a52009-11-08 14:54:08 +02001362 ios->obj.id = exofs_oi_objno(oi);
1363 ios->done = delete_done;
1364 ios->private = sbi;
1365 ios->cred = oi->i_cred;
1366 ret = exofs_sbi_remove(ios);
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001367 if (ret) {
Boaz Harrosh06886a52009-11-08 14:54:08 +02001368 EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__);
1369 exofs_put_io_state(ios);
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001370 return;
1371 }
1372 atomic_inc(&sbi->s_curr_pending);
1373
1374 return;
1375
1376no_delete:
1377 clear_inode(inode);
1378}