Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
5db11c21 MM |
2 | /* |
3 | * (C) 2001 Clemson University and The University of Chicago | |
afd9fb2a | 4 | * Copyright 2018 Omnibond Systems, L.L.C. |
5db11c21 MM |
5 | * |
6 | * See COPYING in top-level directory. | |
7 | */ | |
8 | ||
9 | /* | |
10 | * Linux VFS inode operations. | |
11 | */ | |
12 | ||
ccdf7741 | 13 | #include <linux/blkdev.h> |
1f26b062 | 14 | #include <linux/fileattr.h> |
5db11c21 | 15 | #include "protocol.h" |
575e9461 MM |
16 | #include "orangefs-kernel.h" |
17 | #include "orangefs-bufmap.h" | |
5db11c21 | 18 | |
52e2d0a3 MB |
19 | static int orangefs_writepage_locked(struct page *page, |
20 | struct writeback_control *wbc) | |
85ac799c MB |
21 | { |
22 | struct inode *inode = page->mapping->host; | |
52e2d0a3 | 23 | struct orangefs_write_range *wr = NULL; |
85ac799c MB |
24 | struct iov_iter iter; |
25 | struct bio_vec bv; | |
26 | size_t len, wlen; | |
27 | ssize_t ret; | |
28 | loff_t off; | |
29 | ||
30 | set_page_writeback(page); | |
31 | ||
85ac799c | 32 | len = i_size_read(inode); |
52e2d0a3 MB |
33 | if (PagePrivate(page)) { |
34 | wr = (struct orangefs_write_range *)page_private(page); | |
8f04e1be | 35 | WARN_ON(wr->pos >= len); |
52e2d0a3 MB |
36 | off = wr->pos; |
37 | if (off + wr->len > len) | |
38 | wlen = len - off; | |
39 | else | |
40 | wlen = wr->len; | |
41 | } else { | |
42 | WARN_ON(1); | |
43 | off = page_offset(page); | |
44 | if (off + PAGE_SIZE > len) | |
45 | wlen = len - off; | |
46 | else | |
47 | wlen = PAGE_SIZE; | |
85ac799c | 48 | } |
2a40be81 | 49 | /* Should've been handled in orangefs_invalidate_folio. */ |
52e2d0a3 | 50 | WARN_ON(off == len || off + wlen > len); |
85ac799c | 51 | |
52e2d0a3 | 52 | WARN_ON(wlen == 0); |
8ead80b2 | 53 | bvec_set_page(&bv, page, wlen, off % PAGE_SIZE); |
de4eda9d | 54 | iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, wlen); |
85ac799c MB |
55 | |
56 | ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen, | |
f9bbb682 | 57 | len, wr, NULL, NULL); |
85ac799c MB |
58 | if (ret < 0) { |
59 | SetPageError(page); | |
60 | mapping_set_error(page->mapping, ret); | |
61 | } else { | |
62 | ret = 0; | |
63 | } | |
4c42be38 | 64 | kfree(detach_page_private(page)); |
52e2d0a3 MB |
65 | return ret; |
66 | } | |
67 | ||
68 | static int orangefs_writepage(struct page *page, struct writeback_control *wbc) | |
69 | { | |
70 | int ret; | |
71 | ret = orangefs_writepage_locked(page, wbc); | |
85ac799c MB |
72 | unlock_page(page); |
73 | end_page_writeback(page); | |
74 | return ret; | |
75 | } | |
76 | ||
8f04e1be MB |
77 | struct orangefs_writepages { |
78 | loff_t off; | |
79 | size_t len; | |
80 | kuid_t uid; | |
81 | kgid_t gid; | |
82 | int maxpages; | |
83 | int npages; | |
84 | struct page **pages; | |
85 | struct bio_vec *bv; | |
86 | }; | |
87 | ||
88 | static int orangefs_writepages_work(struct orangefs_writepages *ow, | |
89 | struct writeback_control *wbc) | |
90 | { | |
91 | struct inode *inode = ow->pages[0]->mapping->host; | |
92 | struct orangefs_write_range *wrp, wr; | |
93 | struct iov_iter iter; | |
94 | ssize_t ret; | |
95 | size_t len; | |
96 | loff_t off; | |
97 | int i; | |
98 | ||
99 | len = i_size_read(inode); | |
100 | ||
101 | for (i = 0; i < ow->npages; i++) { | |
102 | set_page_writeback(ow->pages[i]); | |
8ead80b2 CH |
103 | bvec_set_page(&ow->bv[i], ow->pages[i], |
104 | min(page_offset(ow->pages[i]) + PAGE_SIZE, | |
105 | ow->off + ow->len) - | |
106 | max(ow->off, page_offset(ow->pages[i])), | |
107 | i == 0 ? ow->off - page_offset(ow->pages[i]) : 0); | |
8f04e1be | 108 | } |
de4eda9d | 109 | iov_iter_bvec(&iter, ITER_SOURCE, ow->bv, ow->npages, ow->len); |
8f04e1be MB |
110 | |
111 | WARN_ON(ow->off >= len); | |
112 | if (ow->off + ow->len > len) | |
113 | ow->len = len - ow->off; | |
114 | ||
115 | off = ow->off; | |
116 | wr.uid = ow->uid; | |
117 | wr.gid = ow->gid; | |
118 | ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len, | |
f9bbb682 | 119 | 0, &wr, NULL, NULL); |
8f04e1be MB |
120 | if (ret < 0) { |
121 | for (i = 0; i < ow->npages; i++) { | |
122 | SetPageError(ow->pages[i]); | |
123 | mapping_set_error(ow->pages[i]->mapping, ret); | |
124 | if (PagePrivate(ow->pages[i])) { | |
125 | wrp = (struct orangefs_write_range *) | |
126 | page_private(ow->pages[i]); | |
127 | ClearPagePrivate(ow->pages[i]); | |
128 | put_page(ow->pages[i]); | |
129 | kfree(wrp); | |
130 | } | |
131 | end_page_writeback(ow->pages[i]); | |
132 | unlock_page(ow->pages[i]); | |
133 | } | |
134 | } else { | |
135 | ret = 0; | |
136 | for (i = 0; i < ow->npages; i++) { | |
137 | if (PagePrivate(ow->pages[i])) { | |
138 | wrp = (struct orangefs_write_range *) | |
139 | page_private(ow->pages[i]); | |
140 | ClearPagePrivate(ow->pages[i]); | |
141 | put_page(ow->pages[i]); | |
142 | kfree(wrp); | |
143 | } | |
144 | end_page_writeback(ow->pages[i]); | |
145 | unlock_page(ow->pages[i]); | |
146 | } | |
147 | } | |
148 | return ret; | |
149 | } | |
150 | ||
d585bdbe MWO |
151 | static int orangefs_writepages_callback(struct folio *folio, |
152 | struct writeback_control *wbc, void *data) | |
8f04e1be MB |
153 | { |
154 | struct orangefs_writepages *ow = data; | |
d585bdbe | 155 | struct orangefs_write_range *wr = folio->private; |
8f04e1be MB |
156 | int ret; |
157 | ||
d585bdbe MWO |
158 | if (!wr) { |
159 | folio_unlock(folio); | |
8f04e1be MB |
160 | /* It's not private so there's nothing to write, right? */ |
161 | printk("writepages_callback not private!\n"); | |
162 | BUG(); | |
163 | return 0; | |
164 | } | |
8f04e1be MB |
165 | |
166 | ret = -1; | |
167 | if (ow->npages == 0) { | |
168 | ow->off = wr->pos; | |
169 | ow->len = wr->len; | |
170 | ow->uid = wr->uid; | |
171 | ow->gid = wr->gid; | |
d585bdbe | 172 | ow->pages[ow->npages++] = &folio->page; |
8f04e1be MB |
173 | ret = 0; |
174 | goto done; | |
175 | } | |
176 | if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) { | |
177 | orangefs_writepages_work(ow, wbc); | |
178 | ow->npages = 0; | |
179 | ret = -1; | |
180 | goto done; | |
181 | } | |
182 | if (ow->off + ow->len == wr->pos) { | |
183 | ow->len += wr->len; | |
d585bdbe | 184 | ow->pages[ow->npages++] = &folio->page; |
8f04e1be MB |
185 | ret = 0; |
186 | goto done; | |
187 | } | |
188 | done: | |
189 | if (ret == -1) { | |
190 | if (ow->npages) { | |
191 | orangefs_writepages_work(ow, wbc); | |
192 | ow->npages = 0; | |
193 | } | |
d585bdbe MWO |
194 | ret = orangefs_writepage_locked(&folio->page, wbc); |
195 | mapping_set_error(folio->mapping, ret); | |
196 | folio_unlock(folio); | |
197 | folio_end_writeback(folio); | |
8f04e1be MB |
198 | } else { |
199 | if (ow->npages == ow->maxpages) { | |
200 | orangefs_writepages_work(ow, wbc); | |
201 | ow->npages = 0; | |
202 | } | |
203 | } | |
204 | return ret; | |
205 | } | |
206 | ||
207 | static int orangefs_writepages(struct address_space *mapping, | |
208 | struct writeback_control *wbc) | |
209 | { | |
210 | struct orangefs_writepages *ow; | |
211 | struct blk_plug plug; | |
212 | int ret; | |
213 | ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL); | |
214 | if (!ow) | |
215 | return -ENOMEM; | |
216 | ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE; | |
217 | ow->pages = kcalloc(ow->maxpages, sizeof(struct page *), GFP_KERNEL); | |
218 | if (!ow->pages) { | |
219 | kfree(ow); | |
220 | return -ENOMEM; | |
221 | } | |
222 | ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL); | |
223 | if (!ow->bv) { | |
224 | kfree(ow->pages); | |
225 | kfree(ow); | |
226 | return -ENOMEM; | |
227 | } | |
228 | blk_start_plug(&plug); | |
229 | ret = write_cache_pages(mapping, wbc, orangefs_writepages_callback, ow); | |
230 | if (ow->npages) | |
231 | ret = orangefs_writepages_work(ow, wbc); | |
232 | blk_finish_plug(&plug); | |
233 | kfree(ow->pages); | |
234 | kfree(ow->bv); | |
235 | kfree(ow); | |
236 | return ret; | |
237 | } | |
238 | ||
eabf038f | 239 | static int orangefs_launder_folio(struct folio *); |
dd59a647 | 240 | |
0c4b7cad MM |
241 | static void orangefs_readahead(struct readahead_control *rac) |
242 | { | |
243 | loff_t offset; | |
244 | struct iov_iter iter; | |
24523e45 | 245 | struct inode *inode = rac->mapping->host; |
0c4b7cad | 246 | struct xarray *i_pages; |
cd01049d | 247 | struct folio *folio; |
0c4b7cad MM |
248 | loff_t new_start = readahead_pos(rac); |
249 | int ret; | |
250 | size_t new_len = 0; | |
251 | ||
252 | loff_t bytes_remaining = inode->i_size - readahead_pos(rac); | |
253 | loff_t pages_remaining = bytes_remaining / PAGE_SIZE; | |
254 | ||
255 | if (pages_remaining >= 1024) | |
256 | new_len = 4194304; | |
257 | else if (pages_remaining > readahead_count(rac)) | |
258 | new_len = bytes_remaining; | |
259 | ||
260 | if (new_len) | |
261 | readahead_expand(rac, new_start, new_len); | |
262 | ||
263 | offset = readahead_pos(rac); | |
24523e45 | 264 | i_pages = &rac->mapping->i_pages; |
0c4b7cad | 265 | |
de4eda9d | 266 | iov_iter_xarray(&iter, ITER_DEST, i_pages, offset, readahead_length(rac)); |
0c4b7cad MM |
267 | |
268 | /* read in the pages. */ | |
269 | if ((ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, | |
270 | &offset, &iter, readahead_length(rac), | |
24523e45 | 271 | inode->i_size, NULL, NULL, rac->file)) < 0) |
0c4b7cad MM |
272 | gossip_debug(GOSSIP_FILE_DEBUG, |
273 | "%s: wait_for_direct_io failed. \n", __func__); | |
274 | else | |
275 | ret = 0; | |
276 | ||
277 | /* clean up. */ | |
cd01049d PR |
278 | while ((folio = readahead_folio(rac))) { |
279 | if (!ret) | |
280 | folio_mark_uptodate(folio); | |
281 | folio_unlock(folio); | |
0c4b7cad MM |
282 | } |
283 | } | |
284 | ||
1a641788 | 285 | static int orangefs_read_folio(struct file *file, struct folio *folio) |
5db11c21 | 286 | { |
1a641788 | 287 | struct inode *inode = folio->mapping->host; |
c453dcfc MB |
288 | struct iov_iter iter; |
289 | struct bio_vec bv; | |
290 | ssize_t ret; | |
1a641788 | 291 | loff_t off; /* offset of this folio in the file */ |
dd59a647 | 292 | |
eabf038f MWO |
293 | if (folio_test_dirty(folio)) |
294 | orangefs_launder_folio(folio); | |
c453dcfc | 295 | |
1a641788 | 296 | off = folio_pos(folio); |
8ead80b2 | 297 | bvec_set_folio(&bv, folio, folio_size(folio), 0); |
de4eda9d | 298 | iov_iter_bvec(&iter, ITER_DEST, &bv, 1, folio_size(folio)); |
c453dcfc MB |
299 | |
300 | ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter, | |
1a641788 | 301 | folio_size(folio), inode->i_size, NULL, NULL, file); |
9329883a | 302 | /* this will only zero remaining unread portions of the folio data */ |
c453dcfc | 303 | iov_iter_zero(~0U, &iter); |
5db11c21 | 304 | /* takes care of potential aliasing */ |
1a641788 | 305 | flush_dcache_folio(folio); |
c453dcfc | 306 | if (ret < 0) { |
1a641788 | 307 | folio_set_error(folio); |
5db11c21 | 308 | } else { |
1a641788 | 309 | folio_mark_uptodate(folio); |
5db11c21 MM |
310 | ret = 0; |
311 | } | |
1a641788 MWO |
312 | /* unlock the folio after the ->read_folio() routine completes */ |
313 | folio_unlock(folio); | |
0c4b7cad | 314 | return ret; |
5db11c21 MM |
315 | } |
316 | ||
52e2d0a3 | 317 | static int orangefs_write_begin(struct file *file, |
9d6b0cd7 MWO |
318 | struct address_space *mapping, loff_t pos, unsigned len, |
319 | struct page **pagep, void **fsdata) | |
52e2d0a3 MB |
320 | { |
321 | struct orangefs_write_range *wr; | |
eabf038f | 322 | struct folio *folio; |
52e2d0a3 MB |
323 | struct page *page; |
324 | pgoff_t index; | |
325 | int ret; | |
326 | ||
327 | index = pos >> PAGE_SHIFT; | |
328 | ||
b7446e7c | 329 | page = grab_cache_page_write_begin(mapping, index); |
52e2d0a3 MB |
330 | if (!page) |
331 | return -ENOMEM; | |
332 | ||
333 | *pagep = page; | |
eabf038f | 334 | folio = page_folio(page); |
52e2d0a3 | 335 | |
eabf038f | 336 | if (folio_test_dirty(folio) && !folio_test_private(folio)) { |
52e2d0a3 MB |
337 | /* |
338 | * Should be impossible. If it happens, launder the page | |
339 | * since we don't know what's dirty. This will WARN in | |
340 | * orangefs_writepage_locked. | |
341 | */ | |
eabf038f | 342 | ret = orangefs_launder_folio(folio); |
52e2d0a3 MB |
343 | if (ret) |
344 | return ret; | |
345 | } | |
eabf038f | 346 | if (folio_test_private(folio)) { |
52e2d0a3 | 347 | struct orangefs_write_range *wr; |
eabf038f | 348 | wr = folio_get_private(folio); |
52e2d0a3 MB |
349 | if (wr->pos + wr->len == pos && |
350 | uid_eq(wr->uid, current_fsuid()) && | |
351 | gid_eq(wr->gid, current_fsgid())) { | |
352 | wr->len += len; | |
353 | goto okay; | |
354 | } else { | |
eabf038f | 355 | ret = orangefs_launder_folio(folio); |
52e2d0a3 MB |
356 | if (ret) |
357 | return ret; | |
358 | } | |
52e2d0a3 MB |
359 | } |
360 | ||
361 | wr = kmalloc(sizeof *wr, GFP_KERNEL); | |
362 | if (!wr) | |
363 | return -ENOMEM; | |
364 | ||
365 | wr->pos = pos; | |
366 | wr->len = len; | |
367 | wr->uid = current_fsuid(); | |
368 | wr->gid = current_fsgid(); | |
eabf038f | 369 | folio_attach_private(folio, wr); |
52e2d0a3 | 370 | okay: |
52e2d0a3 MB |
371 | return 0; |
372 | } | |
373 | ||
85ac799c MB |
374 | static int orangefs_write_end(struct file *file, struct address_space *mapping, |
375 | loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata) | |
376 | { | |
8f04e1be MB |
377 | struct inode *inode = page->mapping->host; |
378 | loff_t last_pos = pos + copied; | |
379 | ||
380 | /* | |
381 | * No need to use i_size_read() here, the i_size | |
382 | * cannot change under us because we hold the i_mutex. | |
383 | */ | |
384 | if (last_pos > inode->i_size) | |
385 | i_size_write(inode, last_pos); | |
386 | ||
387 | /* zero the stale part of the page if we did a short copy */ | |
388 | if (!PageUptodate(page)) { | |
389 | unsigned from = pos & (PAGE_SIZE - 1); | |
390 | if (copied < len) { | |
391 | zero_user(page, from + copied, len - copied); | |
392 | } | |
393 | /* Set fully written pages uptodate. */ | |
394 | if (pos == page_offset(page) && | |
395 | (len == PAGE_SIZE || pos + len == inode->i_size)) { | |
396 | zero_user_segment(page, from + copied, PAGE_SIZE); | |
397 | SetPageUptodate(page); | |
398 | } | |
399 | } | |
400 | ||
401 | set_page_dirty(page); | |
402 | unlock_page(page); | |
403 | put_page(page); | |
404 | ||
85ac799c | 405 | mark_inode_dirty_sync(file_inode(file)); |
8f04e1be | 406 | return copied; |
85ac799c MB |
407 | } |
408 | ||
2a40be81 MWO |
409 | static void orangefs_invalidate_folio(struct folio *folio, |
410 | size_t offset, size_t length) | |
5db11c21 | 411 | { |
2a40be81 | 412 | struct orangefs_write_range *wr = folio_get_private(folio); |
52e2d0a3 MB |
413 | |
414 | if (offset == 0 && length == PAGE_SIZE) { | |
2a40be81 | 415 | kfree(folio_detach_private(folio)); |
8f04e1be | 416 | return; |
52e2d0a3 | 417 | /* write range entirely within invalidate range (or equal) */ |
2a40be81 MWO |
418 | } else if (folio_pos(folio) + offset <= wr->pos && |
419 | wr->pos + wr->len <= folio_pos(folio) + offset + length) { | |
420 | kfree(folio_detach_private(folio)); | |
52e2d0a3 | 421 | /* XXX is this right? only caller in fs */ |
2a40be81 | 422 | folio_cancel_dirty(folio); |
8f04e1be | 423 | return; |
52e2d0a3 | 424 | /* invalidate range chops off end of write range */ |
2a40be81 MWO |
425 | } else if (wr->pos < folio_pos(folio) + offset && |
426 | wr->pos + wr->len <= folio_pos(folio) + offset + length && | |
427 | folio_pos(folio) + offset < wr->pos + wr->len) { | |
52e2d0a3 | 428 | size_t x; |
2a40be81 | 429 | x = wr->pos + wr->len - (folio_pos(folio) + offset); |
52e2d0a3 MB |
430 | WARN_ON(x > wr->len); |
431 | wr->len -= x; | |
432 | wr->uid = current_fsuid(); | |
433 | wr->gid = current_fsgid(); | |
434 | /* invalidate range chops off beginning of write range */ | |
2a40be81 MWO |
435 | } else if (folio_pos(folio) + offset <= wr->pos && |
436 | folio_pos(folio) + offset + length < wr->pos + wr->len && | |
437 | wr->pos < folio_pos(folio) + offset + length) { | |
52e2d0a3 | 438 | size_t x; |
2a40be81 | 439 | x = folio_pos(folio) + offset + length - wr->pos; |
52e2d0a3 MB |
440 | WARN_ON(x > wr->len); |
441 | wr->pos += x; | |
442 | wr->len -= x; | |
443 | wr->uid = current_fsuid(); | |
444 | wr->gid = current_fsgid(); | |
445 | /* invalidate range entirely within write range (punch hole) */ | |
2a40be81 MWO |
446 | } else if (wr->pos < folio_pos(folio) + offset && |
447 | folio_pos(folio) + offset + length < wr->pos + wr->len) { | |
52e2d0a3 MB |
448 | /* XXX what do we do here... should not WARN_ON */ |
449 | WARN_ON(1); | |
450 | /* punch hole */ | |
451 | /* | |
452 | * should we just ignore this and write it out anyway? | |
453 | * it hardly makes sense | |
454 | */ | |
8f04e1be | 455 | return; |
52e2d0a3 MB |
456 | /* non-overlapping ranges */ |
457 | } else { | |
458 | /* WARN if they do overlap */ | |
2a40be81 MWO |
459 | if (!((folio_pos(folio) + offset + length <= wr->pos) ^ |
460 | (wr->pos + wr->len <= folio_pos(folio) + offset))) { | |
52e2d0a3 | 461 | WARN_ON(1); |
2a40be81 MWO |
462 | printk("invalidate range offset %llu length %zu\n", |
463 | folio_pos(folio) + offset, length); | |
52e2d0a3 MB |
464 | printk("write range offset %llu length %zu\n", |
465 | wr->pos, wr->len); | |
466 | } | |
8f04e1be | 467 | return; |
52e2d0a3 | 468 | } |
8f04e1be MB |
469 | |
470 | /* | |
471 | * Above there are returns where wr is freed or where we WARN. | |
472 | * Thus the following runs if wr was modified above. | |
473 | */ | |
474 | ||
eabf038f | 475 | orangefs_launder_folio(folio); |
52e2d0a3 | 476 | } |
5db11c21 | 477 | |
4993474a | 478 | static bool orangefs_release_folio(struct folio *folio, gfp_t foo) |
52e2d0a3 | 479 | { |
4993474a | 480 | return !folio_test_private(folio); |
52e2d0a3 | 481 | } |
5db11c21 | 482 | |
c78ac80e | 483 | static void orangefs_free_folio(struct folio *folio) |
52e2d0a3 | 484 | { |
c78ac80e | 485 | kfree(folio_detach_private(folio)); |
5db11c21 MM |
486 | } |
487 | ||
eabf038f | 488 | static int orangefs_launder_folio(struct folio *folio) |
5db11c21 | 489 | { |
52e2d0a3 MB |
490 | int r = 0; |
491 | struct writeback_control wbc = { | |
492 | .sync_mode = WB_SYNC_ALL, | |
493 | .nr_to_write = 0, | |
494 | }; | |
eabf038f MWO |
495 | folio_wait_writeback(folio); |
496 | if (folio_clear_dirty_for_io(folio)) { | |
497 | r = orangefs_writepage_locked(&folio->page, &wbc); | |
498 | folio_end_writeback(folio); | |
52e2d0a3 MB |
499 | } |
500 | return r; | |
5db11c21 MM |
501 | } |
502 | ||
3903f150 MM |
503 | static ssize_t orangefs_direct_IO(struct kiocb *iocb, |
504 | struct iov_iter *iter) | |
505 | { | |
3e9dfc6e MB |
506 | /* |
507 | * Comment from original do_readv_writev: | |
508 | * Common entry point for read/write/readv/writev | |
509 | * This function will dispatch it to either the direct I/O | |
510 | * or buffered I/O path depending on the mount options and/or | |
511 | * augmented/extended metadata attached to the file. | |
512 | * Note: File extended attributes override any mount options. | |
513 | */ | |
c453dcfc | 514 | struct file *file = iocb->ki_filp; |
3e9dfc6e MB |
515 | loff_t pos = iocb->ki_pos; |
516 | enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ? | |
517 | ORANGEFS_IO_WRITE : ORANGEFS_IO_READ; | |
518 | loff_t *offset = &pos; | |
519 | struct inode *inode = file->f_mapping->host; | |
520 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); | |
521 | struct orangefs_khandle *handle = &orangefs_inode->refn.khandle; | |
522 | size_t count = iov_iter_count(iter); | |
3e9dfc6e MB |
523 | ssize_t total_count = 0; |
524 | ssize_t ret = -EINVAL; | |
3e9dfc6e MB |
525 | |
526 | gossip_debug(GOSSIP_FILE_DEBUG, | |
527 | "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n", | |
528 | __func__, | |
529 | handle, | |
530 | (int)count); | |
531 | ||
532 | if (type == ORANGEFS_IO_WRITE) { | |
533 | gossip_debug(GOSSIP_FILE_DEBUG, | |
534 | "%s(%pU): proceeding with offset : %llu, " | |
535 | "size %d\n", | |
536 | __func__, | |
537 | handle, | |
538 | llu(*offset), | |
539 | (int)count); | |
540 | } | |
541 | ||
542 | if (count == 0) { | |
543 | ret = 0; | |
544 | goto out; | |
545 | } | |
546 | ||
547 | while (iov_iter_count(iter)) { | |
548 | size_t each_count = iov_iter_count(iter); | |
549 | size_t amt_complete; | |
3e9dfc6e MB |
550 | |
551 | /* how much to transfer in this loop iteration */ | |
552 | if (each_count > orangefs_bufmap_size_query()) | |
553 | each_count = orangefs_bufmap_size_query(); | |
554 | ||
555 | gossip_debug(GOSSIP_FILE_DEBUG, | |
556 | "%s(%pU): size of each_count(%d)\n", | |
557 | __func__, | |
558 | handle, | |
559 | (int)each_count); | |
560 | gossip_debug(GOSSIP_FILE_DEBUG, | |
561 | "%s(%pU): BEFORE wait_for_io: offset is %d\n", | |
562 | __func__, | |
563 | handle, | |
564 | (int)*offset); | |
565 | ||
566 | ret = wait_for_direct_io(type, inode, offset, iter, | |
f9bbb682 | 567 | each_count, 0, NULL, NULL, file); |
3e9dfc6e MB |
568 | gossip_debug(GOSSIP_FILE_DEBUG, |
569 | "%s(%pU): return from wait_for_io:%d\n", | |
570 | __func__, | |
571 | handle, | |
572 | (int)ret); | |
573 | ||
574 | if (ret < 0) | |
575 | goto out; | |
576 | ||
577 | *offset += ret; | |
578 | total_count += ret; | |
579 | amt_complete = ret; | |
580 | ||
581 | gossip_debug(GOSSIP_FILE_DEBUG, | |
582 | "%s(%pU): AFTER wait_for_io: offset is %d\n", | |
583 | __func__, | |
584 | handle, | |
585 | (int)*offset); | |
586 | ||
587 | /* | |
588 | * if we got a short I/O operations, | |
589 | * fall out and return what we got so far | |
590 | */ | |
591 | if (amt_complete < each_count) | |
592 | break; | |
593 | } /*end while */ | |
594 | ||
595 | out: | |
596 | if (total_count > 0) | |
597 | ret = total_count; | |
598 | if (ret > 0) { | |
599 | if (type == ORANGEFS_IO_READ) { | |
600 | file_accessed(file); | |
601 | } else { | |
602 | file_update_time(file); | |
603 | if (*offset > i_size_read(inode)) | |
604 | i_size_write(inode, *offset); | |
605 | } | |
606 | } | |
607 | ||
608 | gossip_debug(GOSSIP_FILE_DEBUG, | |
609 | "%s(%pU): Value(%d) returned.\n", | |
610 | __func__, | |
611 | handle, | |
612 | (int)ret); | |
613 | ||
614 | return ret; | |
3903f150 | 615 | } |
5db11c21 | 616 | |
8bb8aefd | 617 | /** ORANGEFS2 implementation of address space operations */ |
bdd6f083 | 618 | static const struct address_space_operations orangefs_address_operations = { |
85ac799c | 619 | .writepage = orangefs_writepage, |
0c4b7cad | 620 | .readahead = orangefs_readahead, |
1a641788 | 621 | .read_folio = orangefs_read_folio, |
8f04e1be | 622 | .writepages = orangefs_writepages, |
187c82cb | 623 | .dirty_folio = filemap_dirty_folio, |
52e2d0a3 | 624 | .write_begin = orangefs_write_begin, |
85ac799c | 625 | .write_end = orangefs_write_end, |
2a40be81 | 626 | .invalidate_folio = orangefs_invalidate_folio, |
4993474a | 627 | .release_folio = orangefs_release_folio, |
c78ac80e | 628 | .free_folio = orangefs_free_folio, |
eabf038f | 629 | .launder_folio = orangefs_launder_folio, |
3903f150 | 630 | .direct_IO = orangefs_direct_IO, |
5db11c21 MM |
631 | }; |
632 | ||
52e2d0a3 MB |
633 | vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf) |
634 | { | |
eabf038f | 635 | struct folio *folio = page_folio(vmf->page); |
52e2d0a3 | 636 | struct inode *inode = file_inode(vmf->vma->vm_file); |
8f04e1be MB |
637 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
638 | unsigned long *bitlock = &orangefs_inode->bitlock; | |
639 | vm_fault_t ret; | |
52e2d0a3 MB |
640 | struct orangefs_write_range *wr; |
641 | ||
8f04e1be MB |
642 | sb_start_pagefault(inode->i_sb); |
643 | ||
644 | if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) { | |
645 | ret = VM_FAULT_RETRY; | |
646 | goto out; | |
647 | } | |
648 | ||
eabf038f MWO |
649 | folio_lock(folio); |
650 | if (folio_test_dirty(folio) && !folio_test_private(folio)) { | |
52e2d0a3 | 651 | /* |
eabf038f | 652 | * Should be impossible. If it happens, launder the folio |
52e2d0a3 MB |
653 | * since we don't know what's dirty. This will WARN in |
654 | * orangefs_writepage_locked. | |
655 | */ | |
eabf038f | 656 | if (orangefs_launder_folio(folio)) { |
8f04e1be | 657 | ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; |
52e2d0a3 MB |
658 | goto out; |
659 | } | |
660 | } | |
eabf038f MWO |
661 | if (folio_test_private(folio)) { |
662 | wr = folio_get_private(folio); | |
52e2d0a3 MB |
663 | if (uid_eq(wr->uid, current_fsuid()) && |
664 | gid_eq(wr->gid, current_fsgid())) { | |
eabf038f | 665 | wr->pos = page_offset(vmf->page); |
52e2d0a3 MB |
666 | wr->len = PAGE_SIZE; |
667 | goto okay; | |
668 | } else { | |
eabf038f | 669 | if (orangefs_launder_folio(folio)) { |
8f04e1be | 670 | ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; |
52e2d0a3 MB |
671 | goto out; |
672 | } | |
673 | } | |
674 | } | |
675 | wr = kmalloc(sizeof *wr, GFP_KERNEL); | |
676 | if (!wr) { | |
8f04e1be | 677 | ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; |
52e2d0a3 MB |
678 | goto out; |
679 | } | |
eabf038f | 680 | wr->pos = page_offset(vmf->page); |
52e2d0a3 MB |
681 | wr->len = PAGE_SIZE; |
682 | wr->uid = current_fsuid(); | |
683 | wr->gid = current_fsgid(); | |
eabf038f | 684 | folio_attach_private(folio, wr); |
52e2d0a3 MB |
685 | okay: |
686 | ||
52e2d0a3 | 687 | file_update_time(vmf->vma->vm_file); |
eabf038f MWO |
688 | if (folio->mapping != inode->i_mapping) { |
689 | folio_unlock(folio); | |
8f04e1be | 690 | ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE; |
52e2d0a3 MB |
691 | goto out; |
692 | } | |
693 | ||
694 | /* | |
eabf038f | 695 | * We mark the folio dirty already here so that when freeze is in |
52e2d0a3 | 696 | * progress, we are guaranteed that writeback during freezing will |
eabf038f | 697 | * see the dirty folio and writeprotect it again. |
52e2d0a3 | 698 | */ |
eabf038f MWO |
699 | folio_mark_dirty(folio); |
700 | folio_wait_stable(folio); | |
8f04e1be | 701 | ret = VM_FAULT_LOCKED; |
52e2d0a3 MB |
702 | out: |
703 | sb_end_pagefault(inode->i_sb); | |
704 | return ret; | |
705 | } | |
706 | ||
8bb8aefd | 707 | static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr) |
5db11c21 | 708 | { |
8bb8aefd YL |
709 | struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); |
710 | struct orangefs_kernel_op_s *new_op; | |
fecd86aa | 711 | loff_t orig_size; |
5db11c21 MM |
712 | int ret = -EINVAL; |
713 | ||
714 | gossip_debug(GOSSIP_INODE_DEBUG, | |
715 | "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n", | |
716 | __func__, | |
717 | get_khandle_from_ino(inode), | |
8bb8aefd YL |
718 | &orangefs_inode->refn.khandle, |
719 | orangefs_inode->refn.fs_id, | |
5db11c21 MM |
720 | iattr->ia_size); |
721 | ||
fecd86aa | 722 | /* Ensure that we have a up to date size, so we know if it changed. */ |
8b60785c | 723 | ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE); |
fecd86aa MB |
724 | if (ret == -ESTALE) |
725 | ret = -EIO; | |
726 | if (ret) { | |
727 | gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n", | |
728 | __func__, ret); | |
729 | return ret; | |
730 | } | |
731 | orig_size = i_size_read(inode); | |
732 | ||
33713cd0 MB |
733 | /* This is truncate_setsize in a different order. */ |
734 | truncate_pagecache(inode, iattr->ia_size); | |
735 | i_size_write(inode, iattr->ia_size); | |
736 | if (iattr->ia_size > orig_size) | |
737 | pagecache_isize_extended(inode, orig_size, iattr->ia_size); | |
5db11c21 | 738 | |
8bb8aefd | 739 | new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE); |
5db11c21 MM |
740 | if (!new_op) |
741 | return -ENOMEM; | |
742 | ||
8bb8aefd | 743 | new_op->upcall.req.truncate.refn = orangefs_inode->refn; |
5db11c21 MM |
744 | new_op->upcall.req.truncate.size = (__s64) iattr->ia_size; |
745 | ||
95f5f88f MM |
746 | ret = service_operation(new_op, |
747 | __func__, | |
748 | get_interruptible_flag(inode)); | |
5db11c21 MM |
749 | |
750 | /* | |
751 | * the truncate has no downcall members to retrieve, but | |
752 | * the status value tells us if it went through ok or not | |
753 | */ | |
95f5f88f | 754 | gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret); |
5db11c21 MM |
755 | |
756 | op_release(new_op); | |
757 | ||
758 | if (ret != 0) | |
759 | return ret; | |
760 | ||
f83140c1 | 761 | if (orig_size != i_size_read(inode)) |
5db11c21 | 762 | iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; |
5db11c21 MM |
763 | |
764 | return ret; | |
765 | } | |
766 | ||
afd9fb2a | 767 | int __orangefs_setattr(struct inode *inode, struct iattr *iattr) |
5db11c21 | 768 | { |
df2d7337 | 769 | int ret; |
5db11c21 | 770 | |
afd9fb2a MB |
771 | if (iattr->ia_valid & ATTR_MODE) { |
772 | if (iattr->ia_mode & (S_ISVTX)) { | |
773 | if (is_root_handle(inode)) { | |
774 | /* | |
775 | * allow sticky bit to be set on root (since | |
776 | * it shows up that way by default anyhow), | |
777 | * but don't show it to the server | |
778 | */ | |
779 | iattr->ia_mode -= S_ISVTX; | |
780 | } else { | |
781 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
782 | "User attempted to set sticky bit on non-root directory; returning EINVAL.\n"); | |
8f04e1be MB |
783 | ret = -EINVAL; |
784 | goto out; | |
afd9fb2a MB |
785 | } |
786 | } | |
787 | if (iattr->ia_mode & (S_ISUID)) { | |
788 | gossip_debug(GOSSIP_UTILS_DEBUG, | |
789 | "Attempting to set setuid bit (not supported); returning EINVAL.\n"); | |
8f04e1be MB |
790 | ret = -EINVAL; |
791 | goto out; | |
afd9fb2a MB |
792 | } |
793 | } | |
5db11c21 | 794 | |
53950ef5 | 795 | if (iattr->ia_valid & ATTR_SIZE) { |
8bb8aefd | 796 | ret = orangefs_setattr_size(inode, iattr); |
5db11c21 MM |
797 | if (ret) |
798 | goto out; | |
799 | } | |
800 | ||
afd9fb2a MB |
801 | again: |
802 | spin_lock(&inode->i_lock); | |
803 | if (ORANGEFS_I(inode)->attr_valid) { | |
804 | if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) && | |
805 | gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) { | |
806 | ORANGEFS_I(inode)->attr_valid = iattr->ia_valid; | |
807 | } else { | |
808 | spin_unlock(&inode->i_lock); | |
809 | write_inode_now(inode, 1); | |
810 | goto again; | |
811 | } | |
812 | } else { | |
813 | ORANGEFS_I(inode)->attr_valid = iattr->ia_valid; | |
814 | ORANGEFS_I(inode)->attr_uid = current_fsuid(); | |
815 | ORANGEFS_I(inode)->attr_gid = current_fsgid(); | |
816 | } | |
c1632a0f | 817 | setattr_copy(&nop_mnt_idmap, inode, iattr); |
afd9fb2a | 818 | spin_unlock(&inode->i_lock); |
5db11c21 MM |
819 | mark_inode_dirty(inode); |
820 | ||
df2d7337 | 821 | ret = 0; |
5db11c21 | 822 | out: |
afd9fb2a MB |
823 | return ret; |
824 | } | |
825 | ||
138060ba | 826 | int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr) |
4053d250 CB |
827 | { |
828 | int ret; | |
138060ba | 829 | struct inode *inode = d_inode(dentry); |
4053d250 CB |
830 | |
831 | ret = __orangefs_setattr(inode, iattr); | |
832 | /* change mode on a file that has ACLs */ | |
833 | if (!ret && (iattr->ia_valid & ATTR_MODE)) | |
13e83a49 | 834 | ret = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode); |
4053d250 CB |
835 | return ret; |
836 | } | |
837 | ||
afd9fb2a MB |
838 | /* |
839 | * Change attributes of an object referenced by dentry. | |
840 | */ | |
c1632a0f | 841 | int orangefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, |
549c7297 | 842 | struct iattr *iattr) |
afd9fb2a MB |
843 | { |
844 | int ret; | |
845 | gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n", | |
846 | dentry); | |
c1632a0f | 847 | ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr); |
afd9fb2a MB |
848 | if (ret) |
849 | goto out; | |
138060ba | 850 | ret = __orangefs_setattr_mode(dentry, iattr); |
afd9fb2a MB |
851 | sync_inode_metadata(d_inode(dentry), 1); |
852 | out: | |
853 | gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", | |
854 | ret); | |
5db11c21 MM |
855 | return ret; |
856 | } | |
857 | ||
858 | /* | |
859 | * Obtain attributes of an object given a dentry | |
860 | */ | |
b74d24f7 | 861 | int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path, |
549c7297 | 862 | struct kstat *stat, u32 request_mask, unsigned int flags) |
5db11c21 | 863 | { |
e6b998ab | 864 | int ret; |
a528d35e | 865 | struct inode *inode = path->dentry->d_inode; |
5db11c21 MM |
866 | |
867 | gossip_debug(GOSSIP_INODE_DEBUG, | |
5e4f606e MB |
868 | "orangefs_getattr: called on %pd mask %u\n", |
869 | path->dentry, request_mask); | |
5db11c21 | 870 | |
8b60785c MB |
871 | ret = orangefs_inode_getattr(inode, |
872 | request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0); | |
5db11c21 | 873 | if (ret == 0) { |
0d72b928 | 874 | generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); |
a7d3e78a | 875 | |
5db11c21 | 876 | /* override block size reported to stat */ |
5678b5d6 CH |
877 | if (!(request_mask & STATX_SIZE)) |
878 | stat->result_mask &= ~STATX_SIZE; | |
7f54910f | 879 | |
4f911138 | 880 | generic_fill_statx_attr(inode, stat); |
5db11c21 MM |
881 | } |
882 | return ret; | |
883 | } | |
884 | ||
4609e1f1 | 885 | int orangefs_permission(struct mnt_idmap *idmap, |
549c7297 | 886 | struct inode *inode, int mask) |
933287da MB |
887 | { |
888 | int ret; | |
889 | ||
890 | if (mask & MAY_NOT_BLOCK) | |
891 | return -ECHILD; | |
892 | ||
893 | gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__); | |
894 | ||
895 | /* Make sure the permission (and other common attrs) are up to date. */ | |
8b60785c | 896 | ret = orangefs_inode_getattr(inode, 0); |
933287da MB |
897 | if (ret < 0) |
898 | return ret; | |
899 | ||
4609e1f1 | 900 | return generic_permission(&nop_mnt_idmap, inode, mask); |
933287da MB |
901 | } |
902 | ||
913e9928 | 903 | int orangefs_update_time(struct inode *inode, int flags) |
a55f2d86 MB |
904 | { |
905 | struct iattr iattr; | |
541d4c79 | 906 | |
a55f2d86 MB |
907 | gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n", |
908 | get_khandle_from_ino(inode)); | |
541d4c79 | 909 | flags = generic_update_time(inode, flags); |
a55f2d86 MB |
910 | memset(&iattr, 0, sizeof iattr); |
911 | if (flags & S_ATIME) | |
912 | iattr.ia_valid |= ATTR_ATIME; | |
913 | if (flags & S_CTIME) | |
914 | iattr.ia_valid |= ATTR_CTIME; | |
915 | if (flags & S_MTIME) | |
916 | iattr.ia_valid |= ATTR_MTIME; | |
afd9fb2a | 917 | return __orangefs_setattr(inode, &iattr); |
a55f2d86 MB |
918 | } |
919 | ||
1f26b062 MS |
920 | static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa) |
921 | { | |
922 | u64 val = 0; | |
923 | int ret; | |
924 | ||
925 | gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__, | |
926 | dentry); | |
927 | ||
928 | ret = orangefs_inode_getxattr(d_inode(dentry), | |
929 | "user.pvfs2.meta_hint", | |
930 | &val, sizeof(val)); | |
931 | if (ret < 0 && ret != -ENODATA) | |
932 | return ret; | |
933 | ||
934 | gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val); | |
935 | ||
936 | fileattr_fill_flags(fa, val); | |
937 | return 0; | |
938 | } | |
939 | ||
8782a9ae | 940 | static int orangefs_fileattr_set(struct mnt_idmap *idmap, |
1f26b062 MS |
941 | struct dentry *dentry, struct fileattr *fa) |
942 | { | |
943 | u64 val = 0; | |
944 | ||
945 | gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__, | |
946 | dentry); | |
947 | /* | |
948 | * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is | |
949 | * turned on for a file. The user is not allowed to turn on this bit, | |
950 | * but the bit is present if the user first gets the flags and then | |
951 | * updates the flags with some new settings. So, we ignore it in the | |
952 | * following edit. bligon. | |
953 | */ | |
954 | if (fileattr_has_fsx(fa) || | |
955 | (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) { | |
956 | gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n", | |
957 | __func__); | |
958 | return -EOPNOTSUPP; | |
959 | } | |
960 | val = fa->flags; | |
961 | gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val); | |
962 | return orangefs_inode_setxattr(d_inode(dentry), | |
963 | "user.pvfs2.meta_hint", | |
964 | &val, sizeof(val), 0); | |
965 | } | |
966 | ||
95f5f88f | 967 | /* ORANGEFS2 implementation of VFS inode operations for files */ |
bdd6f083 | 968 | static const struct inode_operations orangefs_file_inode_operations = { |
cac2f8b8 | 969 | .get_inode_acl = orangefs_get_acl, |
8bb8aefd YL |
970 | .set_acl = orangefs_set_acl, |
971 | .setattr = orangefs_setattr, | |
972 | .getattr = orangefs_getattr, | |
8bb8aefd | 973 | .listxattr = orangefs_listxattr, |
933287da | 974 | .permission = orangefs_permission, |
a55f2d86 | 975 | .update_time = orangefs_update_time, |
1f26b062 MS |
976 | .fileattr_get = orangefs_fileattr_get, |
977 | .fileattr_set = orangefs_fileattr_set, | |
5db11c21 MM |
978 | }; |
979 | ||
8bb8aefd | 980 | static int orangefs_init_iops(struct inode *inode) |
5db11c21 | 981 | { |
8bb8aefd | 982 | inode->i_mapping->a_ops = &orangefs_address_operations; |
5db11c21 MM |
983 | |
984 | switch (inode->i_mode & S_IFMT) { | |
985 | case S_IFREG: | |
8bb8aefd YL |
986 | inode->i_op = &orangefs_file_inode_operations; |
987 | inode->i_fop = &orangefs_file_operations; | |
5db11c21 MM |
988 | break; |
989 | case S_IFLNK: | |
8bb8aefd | 990 | inode->i_op = &orangefs_symlink_inode_operations; |
5db11c21 MM |
991 | break; |
992 | case S_IFDIR: | |
8bb8aefd YL |
993 | inode->i_op = &orangefs_dir_inode_operations; |
994 | inode->i_fop = &orangefs_dir_operations; | |
5db11c21 MM |
995 | break; |
996 | default: | |
997 | gossip_debug(GOSSIP_INODE_DEBUG, | |
998 | "%s: unsupported mode\n", | |
999 | __func__); | |
1000 | return -EINVAL; | |
1001 | } | |
1002 | ||
1003 | return 0; | |
1004 | } | |
1005 | ||
1006 | /* | |
95f5f88f MM |
1007 | * Given an ORANGEFS object identifier (fsid, handle), convert it into |
1008 | * a ino_t type that will be used as a hash-index from where the handle will | |
5db11c21 MM |
1009 | * be searched for in the VFS hash table of inodes. |
1010 | */ | |
8bb8aefd | 1011 | static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref) |
5db11c21 MM |
1012 | { |
1013 | if (!ref) | |
1014 | return 0; | |
8bb8aefd | 1015 | return orangefs_khandle_to_ino(&(ref->khandle)); |
5db11c21 MM |
1016 | } |
1017 | ||
1018 | /* | |
1019 | * Called to set up an inode from iget5_locked. | |
1020 | */ | |
8bb8aefd | 1021 | static int orangefs_set_inode(struct inode *inode, void *data) |
5db11c21 | 1022 | { |
8bb8aefd | 1023 | struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data; |
a4c680a0 MB |
1024 | ORANGEFS_I(inode)->refn.fs_id = ref->fs_id; |
1025 | ORANGEFS_I(inode)->refn.khandle = ref->khandle; | |
afd9fb2a | 1026 | ORANGEFS_I(inode)->attr_valid = 0; |
fc2e2e9c | 1027 | hash_init(ORANGEFS_I(inode)->xattr_cache); |
8f04e1be MB |
1028 | ORANGEFS_I(inode)->mapping_time = jiffies - 1; |
1029 | ORANGEFS_I(inode)->bitlock = 0; | |
5db11c21 MM |
1030 | return 0; |
1031 | } | |
1032 | ||
1033 | /* | |
1034 | * Called to determine if handles match. | |
1035 | */ | |
8bb8aefd | 1036 | static int orangefs_test_inode(struct inode *inode, void *data) |
5db11c21 | 1037 | { |
8bb8aefd YL |
1038 | struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data; |
1039 | struct orangefs_inode_s *orangefs_inode = NULL; | |
5db11c21 | 1040 | |
8bb8aefd | 1041 | orangefs_inode = ORANGEFS_I(inode); |
95f5f88f MM |
1042 | /* test handles and fs_ids... */ |
1043 | return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), | |
1044 | &(ref->khandle)) && | |
1045 | orangefs_inode->refn.fs_id == ref->fs_id); | |
5db11c21 MM |
1046 | } |
1047 | ||
1048 | /* | |
8bb8aefd | 1049 | * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS |
5db11c21 MM |
1050 | * file handle. |
1051 | * | |
1052 | * @sb: the file system super block instance. | |
95f5f88f | 1053 | * @ref: The ORANGEFS object for which we are trying to locate an inode. |
5db11c21 | 1054 | */ |
95f5f88f MM |
1055 | struct inode *orangefs_iget(struct super_block *sb, |
1056 | struct orangefs_object_kref *ref) | |
5db11c21 MM |
1057 | { |
1058 | struct inode *inode = NULL; | |
1059 | unsigned long hash; | |
1060 | int error; | |
1061 | ||
8bb8aefd | 1062 | hash = orangefs_handle_hash(ref); |
95f5f88f MM |
1063 | inode = iget5_locked(sb, |
1064 | hash, | |
1065 | orangefs_test_inode, | |
1066 | orangefs_set_inode, | |
1067 | ref); | |
b5d72cdc MM |
1068 | |
1069 | if (!inode) | |
1070 | return ERR_PTR(-ENOMEM); | |
1071 | ||
1072 | if (!(inode->i_state & I_NEW)) | |
5db11c21 MM |
1073 | return inode; |
1074 | ||
8b60785c | 1075 | error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW); |
5db11c21 MM |
1076 | if (error) { |
1077 | iget_failed(inode); | |
1078 | return ERR_PTR(error); | |
1079 | } | |
1080 | ||
1081 | inode->i_ino = hash; /* needed for stat etc */ | |
8bb8aefd | 1082 | orangefs_init_iops(inode); |
5db11c21 MM |
1083 | unlock_new_inode(inode); |
1084 | ||
1085 | gossip_debug(GOSSIP_INODE_DEBUG, | |
1086 | "iget handle %pU, fsid %d hash %ld i_ino %lu\n", | |
1087 | &ref->khandle, | |
1088 | ref->fs_id, | |
1089 | hash, | |
1090 | inode->i_ino); | |
1091 | ||
1092 | return inode; | |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * Allocate an inode for a newly created file and insert it into the inode hash. | |
1097 | */ | |
8bb8aefd | 1098 | struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir, |
4053d250 | 1099 | umode_t mode, dev_t dev, struct orangefs_object_kref *ref) |
5db11c21 | 1100 | { |
4053d250 | 1101 | struct posix_acl *acl = NULL, *default_acl = NULL; |
8bb8aefd | 1102 | unsigned long hash = orangefs_handle_hash(ref); |
5db11c21 MM |
1103 | struct inode *inode; |
1104 | int error; | |
1105 | ||
1106 | gossip_debug(GOSSIP_INODE_DEBUG, | |
5253487e MM |
1107 | "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n", |
1108 | __func__, | |
5db11c21 MM |
1109 | sb, |
1110 | MAJOR(dev), | |
1111 | MINOR(dev), | |
1112 | mode); | |
1113 | ||
1114 | inode = new_inode(sb); | |
1115 | if (!inode) | |
56249998 | 1116 | return ERR_PTR(-ENOMEM); |
5db11c21 | 1117 | |
4053d250 CB |
1118 | error = posix_acl_create(dir, &mode, &default_acl, &acl); |
1119 | if (error) | |
1120 | goto out_iput; | |
1121 | ||
8bb8aefd | 1122 | orangefs_set_inode(inode, ref); |
5db11c21 MM |
1123 | inode->i_ino = hash; /* needed for stat etc */ |
1124 | ||
8b60785c | 1125 | error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW); |
5db11c21 MM |
1126 | if (error) |
1127 | goto out_iput; | |
1128 | ||
8bb8aefd | 1129 | orangefs_init_iops(inode); |
5db11c21 MM |
1130 | inode->i_rdev = dev; |
1131 | ||
4053d250 CB |
1132 | if (default_acl) { |
1133 | error = __orangefs_set_acl(inode, default_acl, | |
1134 | ACL_TYPE_DEFAULT); | |
1135 | if (error) | |
1136 | goto out_iput; | |
1137 | } | |
1138 | ||
1139 | if (acl) { | |
1140 | error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS); | |
1141 | if (error) | |
1142 | goto out_iput; | |
1143 | } | |
1144 | ||
8bb8aefd | 1145 | error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref); |
5db11c21 MM |
1146 | if (error < 0) |
1147 | goto out_iput; | |
1148 | ||
1149 | gossip_debug(GOSSIP_INODE_DEBUG, | |
1150 | "Initializing ACL's for inode %pU\n", | |
1151 | get_khandle_from_ino(inode)); | |
e40df428 CB |
1152 | if (mode != inode->i_mode) { |
1153 | struct iattr iattr = { | |
1154 | .ia_mode = mode, | |
1155 | .ia_valid = ATTR_MODE, | |
1156 | }; | |
1157 | inode->i_mode = mode; | |
1158 | __orangefs_setattr(inode, &iattr); | |
1159 | __posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode); | |
1160 | } | |
4053d250 CB |
1161 | posix_acl_release(acl); |
1162 | posix_acl_release(default_acl); | |
5db11c21 MM |
1163 | return inode; |
1164 | ||
1165 | out_iput: | |
1166 | iput(inode); | |
4053d250 CB |
1167 | posix_acl_release(acl); |
1168 | posix_acl_release(default_acl); | |
5db11c21 MM |
1169 | return ERR_PTR(error); |
1170 | } |