Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
1d3576fd SW |
2 | |
3 | #include <linux/backing-dev.h> | |
4 | #include <linux/fs.h> | |
5 | #include <linux/mm.h> | |
6 | #include <linux/pagemap.h> | |
7 | #include <linux/writeback.h> /* generic_writepages */ | |
5a0e3ad6 | 8 | #include <linux/slab.h> |
1d3576fd SW |
9 | #include <linux/pagevec.h> |
10 | #include <linux/task_io_accounting_ops.h> | |
11 | ||
12 | #include "super.h" | |
3d14c5d2 YS |
13 | #include "mds_client.h" |
14 | #include <linux/ceph/osd_client.h> | |
1d3576fd SW |
15 | |
16 | /* | |
17 | * Ceph address space ops. | |
18 | * | |
19 | * There are a few funny things going on here. | |
20 | * | |
21 | * The page->private field is used to reference a struct | |
22 | * ceph_snap_context for _every_ dirty page. This indicates which | |
23 | * snapshot the page was logically dirtied in, and thus which snap | |
24 | * context needs to be associated with the osd write during writeback. | |
25 | * | |
26 | * Similarly, struct ceph_inode_info maintains a set of counters to | |
25985edc | 27 | * count dirty pages on the inode. In the absence of snapshots, |
1d3576fd SW |
28 | * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count. |
29 | * | |
30 | * When a snapshot is taken (that is, when the client receives | |
31 | * notification that a snapshot was taken), each inode with caps and | |
32 | * with dirty pages (dirty pages implies there is a cap) gets a new | |
33 | * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending | |
34 | * order, new snaps go to the tail). The i_wrbuffer_ref_head count is | |
35 | * moved to capsnap->dirty. (Unless a sync write is currently in | |
36 | * progress. In that case, the capsnap is said to be "pending", new | |
37 | * writes cannot start, and the capsnap isn't "finalized" until the | |
38 | * write completes (or fails) and a final size/mtime for the inode for | |
39 | * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0. | |
40 | * | |
41 | * On writeback, we must submit writes to the osd IN SNAP ORDER. So, | |
42 | * we look for the first capsnap in i_cap_snaps and write out pages in | |
43 | * that snap context _only_. Then we move on to the next capsnap, | |
44 | * eventually reaching the "live" or "head" context (i.e., pages that | |
45 | * are not yet snapped) and are writing the most recently dirtied | |
46 | * pages. | |
47 | * | |
48 | * Invalidate and so forth must take care to ensure the dirty page | |
49 | * accounting is preserved. | |
50 | */ | |
51 | ||
2baba250 YS |
52 | #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10)) |
53 | #define CONGESTION_OFF_THRESH(congestion_kb) \ | |
54 | (CONGESTION_ON_THRESH(congestion_kb) - \ | |
55 | (CONGESTION_ON_THRESH(congestion_kb) >> 2)) | |
56 | ||
61600ef8 YZ |
57 | static inline struct ceph_snap_context *page_snap_context(struct page *page) |
58 | { | |
59 | if (PagePrivate(page)) | |
60 | return (void *)page->private; | |
61 | return NULL; | |
62 | } | |
1d3576fd SW |
63 | |
64 | /* | |
65 | * Dirty a page. Optimistically adjust accounting, on the assumption | |
66 | * that we won't race with invalidate. If we do, readjust. | |
67 | */ | |
68 | static int ceph_set_page_dirty(struct page *page) | |
69 | { | |
70 | struct address_space *mapping = page->mapping; | |
71 | struct inode *inode; | |
72 | struct ceph_inode_info *ci; | |
73 | int undo = 0; | |
74 | struct ceph_snap_context *snapc; | |
75 | ||
76 | if (unlikely(!mapping)) | |
77 | return !TestSetPageDirty(page); | |
78 | ||
79 | if (TestSetPageDirty(page)) { | |
80 | dout("%p set_page_dirty %p idx %lu -- already dirty\n", | |
81 | mapping->host, page, page->index); | |
82 | return 0; | |
83 | } | |
84 | ||
85 | inode = mapping->host; | |
86 | ci = ceph_inode(inode); | |
87 | ||
88 | /* | |
89 | * Note that we're grabbing a snapc ref here without holding | |
90 | * any locks! | |
91 | */ | |
92 | snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context); | |
93 | ||
94 | /* dirty the head */ | |
be655596 | 95 | spin_lock(&ci->i_ceph_lock); |
7d8cb26d | 96 | if (ci->i_head_snapc == NULL) |
1d3576fd SW |
97 | ci->i_head_snapc = ceph_get_snap_context(snapc); |
98 | ++ci->i_wrbuffer_ref_head; | |
99 | if (ci->i_wrbuffer_ref == 0) | |
0444d76a | 100 | ihold(inode); |
1d3576fd SW |
101 | ++ci->i_wrbuffer_ref; |
102 | dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d " | |
103 | "snapc %p seq %lld (%d snaps)\n", | |
104 | mapping->host, page, page->index, | |
105 | ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1, | |
106 | ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head, | |
107 | snapc, snapc->seq, snapc->num_snaps); | |
be655596 | 108 | spin_unlock(&ci->i_ceph_lock); |
1d3576fd SW |
109 | |
110 | /* now adjust page */ | |
111 | spin_lock_irq(&mapping->tree_lock); | |
112 | if (page->mapping) { /* Race with truncate? */ | |
113 | WARN_ON_ONCE(!PageUptodate(page)); | |
679ceace | 114 | account_page_dirtied(page, page->mapping); |
1d3576fd SW |
115 | radix_tree_tag_set(&mapping->page_tree, |
116 | page_index(page), PAGECACHE_TAG_DIRTY); | |
117 | ||
118 | /* | |
119 | * Reference snap context in page->private. Also set | |
120 | * PagePrivate so that we get invalidatepage callback. | |
121 | */ | |
122 | page->private = (unsigned long)snapc; | |
123 | SetPagePrivate(page); | |
124 | } else { | |
125 | dout("ANON set_page_dirty %p (raced truncate?)\n", page); | |
126 | undo = 1; | |
127 | } | |
128 | ||
129 | spin_unlock_irq(&mapping->tree_lock); | |
130 | ||
131 | if (undo) | |
132 | /* whoops, we failed to dirty the page */ | |
133 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); | |
134 | ||
135 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
136 | ||
137 | BUG_ON(!PageDirty(page)); | |
138 | return 1; | |
139 | } | |
140 | ||
141 | /* | |
142 | * If we are truncating the full page (i.e. offset == 0), adjust the | |
143 | * dirty page counters appropriately. Only called if there is private | |
144 | * data on the page. | |
145 | */ | |
146 | static void ceph_invalidatepage(struct page *page, unsigned long offset) | |
147 | { | |
4ce1e9ad | 148 | struct inode *inode; |
1d3576fd | 149 | struct ceph_inode_info *ci; |
61600ef8 | 150 | struct ceph_snap_context *snapc = page_snap_context(page); |
1d3576fd SW |
151 | |
152 | BUG_ON(!PageLocked(page)); | |
1d3576fd SW |
153 | BUG_ON(!PagePrivate(page)); |
154 | BUG_ON(!page->mapping); | |
155 | ||
4ce1e9ad AB |
156 | inode = page->mapping->host; |
157 | ||
1d3576fd SW |
158 | /* |
159 | * We can get non-dirty pages here due to races between | |
160 | * set_page_dirty and truncate_complete_page; just spit out a | |
161 | * warning, in case we end up with accounting problems later. | |
162 | */ | |
163 | if (!PageDirty(page)) | |
164 | pr_err("%p invalidatepage %p page not dirty\n", inode, page); | |
165 | ||
166 | if (offset == 0) | |
167 | ClearPageChecked(page); | |
168 | ||
169 | ci = ceph_inode(inode); | |
170 | if (offset == 0) { | |
171 | dout("%p invalidatepage %p idx %lu full dirty page %lu\n", | |
172 | inode, page, page->index, offset); | |
173 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); | |
174 | ceph_put_snap_context(snapc); | |
175 | page->private = 0; | |
176 | ClearPagePrivate(page); | |
177 | } else { | |
178 | dout("%p invalidatepage %p idx %lu partial dirty page\n", | |
179 | inode, page, page->index); | |
180 | } | |
181 | } | |
182 | ||
183 | /* just a sanity check */ | |
184 | static int ceph_releasepage(struct page *page, gfp_t g) | |
185 | { | |
186 | struct inode *inode = page->mapping ? page->mapping->host : NULL; | |
187 | dout("%p releasepage %p idx %lu\n", inode, page, page->index); | |
188 | WARN_ON(PageDirty(page)); | |
1d3576fd SW |
189 | WARN_ON(PagePrivate(page)); |
190 | return 0; | |
191 | } | |
192 | ||
193 | /* | |
194 | * read a single page, without unlocking it. | |
195 | */ | |
196 | static int readpage_nounlock(struct file *filp, struct page *page) | |
197 | { | |
496ad9aa | 198 | struct inode *inode = file_inode(filp); |
1d3576fd | 199 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
200 | struct ceph_osd_client *osdc = |
201 | &ceph_inode_to_client(inode)->client->osdc; | |
1d3576fd SW |
202 | int err = 0; |
203 | u64 len = PAGE_CACHE_SIZE; | |
204 | ||
205 | dout("readpage inode %p file %p page %p index %lu\n", | |
206 | inode, filp, page, page->index); | |
207 | err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout, | |
6285bc23 | 208 | (u64) page_offset(page), &len, |
1d3576fd | 209 | ci->i_truncate_seq, ci->i_truncate_size, |
b7495fc2 | 210 | &page, 1, 0); |
1d3576fd SW |
211 | if (err == -ENOENT) |
212 | err = 0; | |
213 | if (err < 0) { | |
214 | SetPageError(page); | |
215 | goto out; | |
216 | } else if (err < PAGE_CACHE_SIZE) { | |
217 | /* zero fill remainder of page */ | |
218 | zero_user_segment(page, err, PAGE_CACHE_SIZE); | |
219 | } | |
220 | SetPageUptodate(page); | |
221 | ||
222 | out: | |
223 | return err < 0 ? err : 0; | |
224 | } | |
225 | ||
226 | static int ceph_readpage(struct file *filp, struct page *page) | |
227 | { | |
228 | int r = readpage_nounlock(filp, page); | |
229 | unlock_page(page); | |
230 | return r; | |
231 | } | |
232 | ||
233 | /* | |
7c272194 | 234 | * Finish an async read(ahead) op. |
1d3576fd | 235 | */ |
7c272194 | 236 | static void finish_read(struct ceph_osd_request *req, struct ceph_msg *msg) |
1d3576fd | 237 | { |
7c272194 | 238 | struct inode *inode = req->r_inode; |
1b83bef2 SW |
239 | int rc = req->r_result; |
240 | int bytes = le32_to_cpu(msg->hdr.data_len); | |
7c272194 | 241 | int i; |
1d3576fd | 242 | |
7c272194 SW |
243 | dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes); |
244 | ||
245 | /* unlock all pages, zeroing any data we didn't read */ | |
246 | for (i = 0; i < req->r_num_pages; i++, bytes -= PAGE_CACHE_SIZE) { | |
247 | struct page *page = req->r_pages[i]; | |
248 | ||
249 | if (bytes < (int)PAGE_CACHE_SIZE) { | |
250 | /* zero (remainder of) page */ | |
251 | int s = bytes < 0 ? 0 : bytes; | |
252 | zero_user_segment(page, s, PAGE_CACHE_SIZE); | |
1d3576fd | 253 | } |
7c272194 SW |
254 | dout("finish_read %p uptodate %p idx %lu\n", inode, page, |
255 | page->index); | |
256 | flush_dcache_page(page); | |
257 | SetPageUptodate(page); | |
258 | unlock_page(page); | |
259 | page_cache_release(page); | |
1d3576fd | 260 | } |
7c272194 | 261 | kfree(req->r_pages); |
1d3576fd SW |
262 | } |
263 | ||
8884d53d DZ |
264 | static void ceph_unlock_page_vector(struct page **pages, int num_pages) |
265 | { | |
266 | int i; | |
267 | ||
268 | for (i = 0; i < num_pages; i++) | |
269 | unlock_page(pages[i]); | |
270 | } | |
271 | ||
1d3576fd | 272 | /* |
7c272194 SW |
273 | * start an async read(ahead) operation. return nr_pages we submitted |
274 | * a read for on success, or negative error code. | |
1d3576fd | 275 | */ |
0d66a487 | 276 | static int start_read(struct inode *inode, struct list_head *page_list, int max) |
1d3576fd | 277 | { |
3d14c5d2 YS |
278 | struct ceph_osd_client *osdc = |
279 | &ceph_inode_to_client(inode)->client->osdc; | |
7c272194 SW |
280 | struct ceph_inode_info *ci = ceph_inode(inode); |
281 | struct page *page = list_entry(page_list->prev, struct page, lru); | |
282 | struct ceph_osd_request *req; | |
283 | u64 off; | |
1d3576fd | 284 | u64 len; |
7c272194 SW |
285 | int i; |
286 | struct page **pages; | |
287 | pgoff_t next_index; | |
288 | int nr_pages = 0; | |
289 | int ret; | |
1d3576fd | 290 | |
6285bc23 | 291 | off = (u64) page_offset(page); |
1d3576fd | 292 | |
7c272194 SW |
293 | /* count pages */ |
294 | next_index = page->index; | |
295 | list_for_each_entry_reverse(page, page_list, lru) { | |
296 | if (page->index != next_index) | |
297 | break; | |
298 | nr_pages++; | |
299 | next_index++; | |
0d66a487 SW |
300 | if (max && nr_pages == max) |
301 | break; | |
7c272194 | 302 | } |
1d3576fd | 303 | len = nr_pages << PAGE_CACHE_SHIFT; |
7c272194 SW |
304 | dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages, |
305 | off, len); | |
306 | ||
307 | req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), | |
308 | off, &len, | |
309 | CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, | |
310 | NULL, 0, | |
311 | ci->i_truncate_seq, ci->i_truncate_size, | |
153e5167 | 312 | NULL, false); |
6816282d SW |
313 | if (IS_ERR(req)) |
314 | return PTR_ERR(req); | |
1d3576fd | 315 | |
7c272194 | 316 | /* build page vector */ |
cf7b7e14 | 317 | nr_pages = calc_pages_for(0, len); |
7c272194 SW |
318 | pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS); |
319 | ret = -ENOMEM; | |
320 | if (!pages) | |
321 | goto out; | |
322 | for (i = 0; i < nr_pages; ++i) { | |
323 | page = list_entry(page_list->prev, struct page, lru); | |
324 | BUG_ON(PageLocked(page)); | |
1d3576fd | 325 | list_del(&page->lru); |
7c272194 SW |
326 | |
327 | dout("start_read %p adding %p idx %lu\n", inode, page, | |
328 | page->index); | |
329 | if (add_to_page_cache_lru(page, &inode->i_data, page->index, | |
213c99ee | 330 | GFP_NOFS)) { |
1d3576fd | 331 | page_cache_release(page); |
7c272194 | 332 | dout("start_read %p add_to_page_cache failed %p\n", |
1d3576fd | 333 | inode, page); |
7c272194 SW |
334 | nr_pages = i; |
335 | goto out_pages; | |
1d3576fd | 336 | } |
7c272194 | 337 | pages[i] = page; |
1d3576fd | 338 | } |
7c272194 SW |
339 | req->r_pages = pages; |
340 | req->r_num_pages = nr_pages; | |
153e5167 | 341 | req->r_page_alignment = 0; |
7c272194 SW |
342 | req->r_callback = finish_read; |
343 | req->r_inode = inode; | |
344 | ||
345 | dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len); | |
346 | ret = ceph_osdc_start_request(osdc, req, false); | |
347 | if (ret < 0) | |
348 | goto out_pages; | |
349 | ceph_osdc_put_request(req); | |
350 | return nr_pages; | |
351 | ||
352 | out_pages: | |
8884d53d | 353 | ceph_unlock_page_vector(pages, nr_pages); |
7c272194 | 354 | ceph_release_page_vector(pages, nr_pages); |
7c272194 SW |
355 | out: |
356 | ceph_osdc_put_request(req); | |
357 | return ret; | |
358 | } | |
1d3576fd | 359 | |
7c272194 SW |
360 | |
361 | /* | |
362 | * Read multiple pages. Leave pages we don't read + unlock in page_list; | |
363 | * the caller (VM) cleans them up. | |
364 | */ | |
365 | static int ceph_readpages(struct file *file, struct address_space *mapping, | |
366 | struct list_head *page_list, unsigned nr_pages) | |
367 | { | |
496ad9aa | 368 | struct inode *inode = file_inode(file); |
0d66a487 | 369 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
7c272194 | 370 | int rc = 0; |
0d66a487 SW |
371 | int max = 0; |
372 | ||
373 | if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE) | |
374 | max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1) | |
375 | >> PAGE_SHIFT; | |
7c272194 | 376 | |
0d66a487 SW |
377 | dout("readpages %p file %p nr_pages %d max %d\n", inode, file, nr_pages, |
378 | max); | |
7c272194 | 379 | while (!list_empty(page_list)) { |
0d66a487 | 380 | rc = start_read(inode, page_list, max); |
7c272194 SW |
381 | if (rc < 0) |
382 | goto out; | |
383 | BUG_ON(rc == 0); | |
384 | } | |
1d3576fd | 385 | out: |
7c272194 | 386 | dout("readpages %p file %p ret %d\n", inode, file, rc); |
1d3576fd SW |
387 | return rc; |
388 | } | |
389 | ||
390 | /* | |
391 | * Get ref for the oldest snapc for an inode with dirty data... that is, the | |
392 | * only snap context we are allowed to write back. | |
1d3576fd | 393 | */ |
6298a337 SW |
394 | static struct ceph_snap_context *get_oldest_context(struct inode *inode, |
395 | u64 *snap_size) | |
1d3576fd SW |
396 | { |
397 | struct ceph_inode_info *ci = ceph_inode(inode); | |
398 | struct ceph_snap_context *snapc = NULL; | |
399 | struct ceph_cap_snap *capsnap = NULL; | |
400 | ||
be655596 | 401 | spin_lock(&ci->i_ceph_lock); |
1d3576fd SW |
402 | list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) { |
403 | dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap, | |
404 | capsnap->context, capsnap->dirty_pages); | |
405 | if (capsnap->dirty_pages) { | |
406 | snapc = ceph_get_snap_context(capsnap->context); | |
407 | if (snap_size) | |
408 | *snap_size = capsnap->size; | |
409 | break; | |
410 | } | |
411 | } | |
7d8cb26d | 412 | if (!snapc && ci->i_wrbuffer_ref_head) { |
80e755fe | 413 | snapc = ceph_get_snap_context(ci->i_head_snapc); |
1d3576fd SW |
414 | dout(" head snapc %p has %d dirty pages\n", |
415 | snapc, ci->i_wrbuffer_ref_head); | |
416 | } | |
be655596 | 417 | spin_unlock(&ci->i_ceph_lock); |
1d3576fd SW |
418 | return snapc; |
419 | } | |
420 | ||
421 | /* | |
422 | * Write a single page, but leave the page locked. | |
423 | * | |
424 | * If we get a write error, set the page error bit, but still adjust the | |
425 | * dirty page accounting (i.e., page is no longer dirty). | |
426 | */ | |
427 | static int writepage_nounlock(struct page *page, struct writeback_control *wbc) | |
428 | { | |
429 | struct inode *inode; | |
430 | struct ceph_inode_info *ci; | |
3d14c5d2 | 431 | struct ceph_fs_client *fsc; |
1d3576fd | 432 | struct ceph_osd_client *osdc; |
6285bc23 | 433 | loff_t page_off = page_offset(page); |
1d3576fd SW |
434 | int len = PAGE_CACHE_SIZE; |
435 | loff_t i_size; | |
436 | int err = 0; | |
6298a337 | 437 | struct ceph_snap_context *snapc, *oldest; |
1d3576fd | 438 | u64 snap_size = 0; |
2baba250 | 439 | long writeback_stat; |
1d3576fd SW |
440 | |
441 | dout("writepage %p idx %lu\n", page, page->index); | |
442 | ||
443 | if (!page->mapping || !page->mapping->host) { | |
444 | dout("writepage %p - no mapping\n", page); | |
445 | return -EFAULT; | |
446 | } | |
447 | inode = page->mapping->host; | |
448 | ci = ceph_inode(inode); | |
3d14c5d2 YS |
449 | fsc = ceph_inode_to_client(inode); |
450 | osdc = &fsc->client->osdc; | |
1d3576fd SW |
451 | |
452 | /* verify this is a writeable snap context */ | |
61600ef8 | 453 | snapc = page_snap_context(page); |
1d3576fd SW |
454 | if (snapc == NULL) { |
455 | dout("writepage %p page %p not dirty?\n", inode, page); | |
456 | goto out; | |
457 | } | |
6298a337 SW |
458 | oldest = get_oldest_context(inode, &snap_size); |
459 | if (snapc->seq > oldest->seq) { | |
1d3576fd | 460 | dout("writepage %p page %p snapc %p not writeable - noop\n", |
61600ef8 | 461 | inode, page, snapc); |
1d3576fd SW |
462 | /* we should only noop if called by kswapd */ |
463 | WARN_ON((current->flags & PF_MEMALLOC) == 0); | |
6298a337 | 464 | ceph_put_snap_context(oldest); |
1d3576fd SW |
465 | goto out; |
466 | } | |
6298a337 | 467 | ceph_put_snap_context(oldest); |
1d3576fd SW |
468 | |
469 | /* is this a partial page at end of file? */ | |
470 | if (snap_size) | |
471 | i_size = snap_size; | |
472 | else | |
473 | i_size = i_size_read(inode); | |
474 | if (i_size < page_off + len) | |
475 | len = i_size - page_off; | |
476 | ||
ae00d4f3 SW |
477 | dout("writepage %p page %p index %lu on %llu~%u snapc %p\n", |
478 | inode, page, page->index, page_off, len, snapc); | |
1d3576fd | 479 | |
3d14c5d2 | 480 | writeback_stat = atomic_long_inc_return(&fsc->writeback_count); |
2baba250 | 481 | if (writeback_stat > |
3d14c5d2 YS |
482 | CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb)) |
483 | set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC); | |
2baba250 | 484 | |
1d3576fd SW |
485 | set_page_writeback(page); |
486 | err = ceph_osdc_writepages(osdc, ceph_vino(inode), | |
487 | &ci->i_layout, snapc, | |
488 | page_off, len, | |
489 | ci->i_truncate_seq, ci->i_truncate_size, | |
24808826 | 490 | &inode->i_mtime, &page, 1); |
1d3576fd SW |
491 | if (err < 0) { |
492 | dout("writepage setting page/mapping error %d %p\n", err, page); | |
493 | SetPageError(page); | |
494 | mapping_set_error(&inode->i_data, err); | |
495 | if (wbc) | |
496 | wbc->pages_skipped++; | |
497 | } else { | |
498 | dout("writepage cleaned page %p\n", page); | |
499 | err = 0; /* vfs expects us to return 0 */ | |
500 | } | |
501 | page->private = 0; | |
502 | ClearPagePrivate(page); | |
503 | end_page_writeback(page); | |
504 | ceph_put_wrbuffer_cap_refs(ci, 1, snapc); | |
6298a337 | 505 | ceph_put_snap_context(snapc); /* page's reference */ |
1d3576fd SW |
506 | out: |
507 | return err; | |
508 | } | |
509 | ||
510 | static int ceph_writepage(struct page *page, struct writeback_control *wbc) | |
511 | { | |
dbd646a8 YS |
512 | int err; |
513 | struct inode *inode = page->mapping->host; | |
514 | BUG_ON(!inode); | |
70b666c3 | 515 | ihold(inode); |
dbd646a8 | 516 | err = writepage_nounlock(page, wbc); |
1d3576fd | 517 | unlock_page(page); |
dbd646a8 | 518 | iput(inode); |
1d3576fd SW |
519 | return err; |
520 | } | |
521 | ||
522 | ||
523 | /* | |
524 | * lame release_pages helper. release_pages() isn't exported to | |
525 | * modules. | |
526 | */ | |
527 | static void ceph_release_pages(struct page **pages, int num) | |
528 | { | |
529 | struct pagevec pvec; | |
530 | int i; | |
531 | ||
532 | pagevec_init(&pvec, 0); | |
533 | for (i = 0; i < num; i++) { | |
534 | if (pagevec_add(&pvec, pages[i]) == 0) | |
535 | pagevec_release(&pvec); | |
536 | } | |
537 | pagevec_release(&pvec); | |
538 | } | |
539 | ||
540 | ||
541 | /* | |
542 | * async writeback completion handler. | |
543 | * | |
544 | * If we get an error, set the mapping error bit, but not the individual | |
545 | * page error bits. | |
546 | */ | |
547 | static void writepages_finish(struct ceph_osd_request *req, | |
548 | struct ceph_msg *msg) | |
549 | { | |
550 | struct inode *inode = req->r_inode; | |
1d3576fd SW |
551 | struct ceph_inode_info *ci = ceph_inode(inode); |
552 | unsigned wrote; | |
1d3576fd SW |
553 | struct page *page; |
554 | int i; | |
555 | struct ceph_snap_context *snapc = req->r_snapc; | |
556 | struct address_space *mapping = inode->i_mapping; | |
1b83bef2 SW |
557 | int rc = req->r_result; |
558 | u64 bytes = le64_to_cpu(req->r_request_ops[0].extent.length); | |
3d14c5d2 | 559 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
2baba250 | 560 | long writeback_stat; |
7ff899da | 561 | unsigned issued = ceph_caps_issued(ci); |
1d3576fd | 562 | |
1d3576fd | 563 | if (rc >= 0) { |
79788c69 SW |
564 | /* |
565 | * Assume we wrote the pages we originally sent. The | |
566 | * osd might reply with fewer pages if our writeback | |
567 | * raced with a truncation and was adjusted at the osd, | |
568 | * so don't believe the reply. | |
569 | */ | |
570 | wrote = req->r_num_pages; | |
1d3576fd SW |
571 | } else { |
572 | wrote = 0; | |
573 | mapping_set_error(mapping, rc); | |
574 | } | |
575 | dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n", | |
576 | inode, rc, bytes, wrote); | |
577 | ||
578 | /* clean all pages */ | |
579 | for (i = 0; i < req->r_num_pages; i++) { | |
580 | page = req->r_pages[i]; | |
581 | BUG_ON(!page); | |
582 | WARN_ON(!PageUptodate(page)); | |
583 | ||
2baba250 | 584 | writeback_stat = |
3d14c5d2 | 585 | atomic_long_dec_return(&fsc->writeback_count); |
2baba250 | 586 | if (writeback_stat < |
3d14c5d2 YS |
587 | CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb)) |
588 | clear_bdi_congested(&fsc->backing_dev_info, | |
2baba250 YS |
589 | BLK_RW_ASYNC); |
590 | ||
61600ef8 | 591 | ceph_put_snap_context(page_snap_context(page)); |
1d3576fd SW |
592 | page->private = 0; |
593 | ClearPagePrivate(page); | |
1d3576fd SW |
594 | dout("unlocking %d %p\n", i, page); |
595 | end_page_writeback(page); | |
e63dc5c7 YS |
596 | |
597 | /* | |
598 | * We lost the cache cap, need to truncate the page before | |
599 | * it is unlocked, otherwise we'd truncate it later in the | |
600 | * page truncation thread, possibly losing some data that | |
601 | * raced its way in | |
602 | */ | |
2962507c | 603 | if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) |
e63dc5c7 YS |
604 | generic_error_remove_page(inode->i_mapping, page); |
605 | ||
1d3576fd SW |
606 | unlock_page(page); |
607 | } | |
608 | dout("%p wrote+cleaned %d pages\n", inode, wrote); | |
609 | ceph_put_wrbuffer_cap_refs(ci, req->r_num_pages, snapc); | |
610 | ||
611 | ceph_release_pages(req->r_pages, req->r_num_pages); | |
612 | if (req->r_pages_from_pool) | |
613 | mempool_free(req->r_pages, | |
640ef79d | 614 | ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool); |
1d3576fd SW |
615 | else |
616 | kfree(req->r_pages); | |
617 | ceph_osdc_put_request(req); | |
618 | } | |
619 | ||
620 | /* | |
621 | * allocate a page vec, either directly, or if necessary, via a the | |
622 | * mempool. we avoid the mempool if we can because req->r_num_pages | |
623 | * may be less than the maximum write size. | |
624 | */ | |
3d14c5d2 | 625 | static void alloc_page_vec(struct ceph_fs_client *fsc, |
1d3576fd SW |
626 | struct ceph_osd_request *req) |
627 | { | |
628 | req->r_pages = kmalloc(sizeof(struct page *) * req->r_num_pages, | |
629 | GFP_NOFS); | |
630 | if (!req->r_pages) { | |
3d14c5d2 | 631 | req->r_pages = mempool_alloc(fsc->wb_pagevec_pool, GFP_NOFS); |
1d3576fd SW |
632 | req->r_pages_from_pool = 1; |
633 | WARN_ON(!req->r_pages); | |
634 | } | |
635 | } | |
636 | ||
637 | /* | |
638 | * initiate async writeback | |
639 | */ | |
640 | static int ceph_writepages_start(struct address_space *mapping, | |
641 | struct writeback_control *wbc) | |
642 | { | |
643 | struct inode *inode = mapping->host; | |
1d3576fd | 644 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 645 | struct ceph_fs_client *fsc; |
1d3576fd SW |
646 | pgoff_t index, start, end; |
647 | int range_whole = 0; | |
648 | int should_loop = 1; | |
649 | pgoff_t max_pages = 0, max_pages_ever = 0; | |
80e755fe | 650 | struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc; |
1d3576fd SW |
651 | struct pagevec pvec; |
652 | int done = 0; | |
653 | int rc = 0; | |
654 | unsigned wsize = 1 << inode->i_blkbits; | |
655 | struct ceph_osd_request *req = NULL; | |
656 | int do_sync; | |
657 | u64 snap_size = 0; | |
658 | ||
659 | /* | |
660 | * Include a 'sync' in the OSD request if this is a data | |
661 | * integrity write (e.g., O_SYNC write or fsync()), or if our | |
662 | * cap is being revoked. | |
663 | */ | |
664 | do_sync = wbc->sync_mode == WB_SYNC_ALL; | |
665 | if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER)) | |
666 | do_sync = 1; | |
667 | dout("writepages_start %p dosync=%d (mode=%s)\n", | |
668 | inode, do_sync, | |
669 | wbc->sync_mode == WB_SYNC_NONE ? "NONE" : | |
670 | (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD")); | |
671 | ||
3d14c5d2 YS |
672 | fsc = ceph_inode_to_client(inode); |
673 | if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) { | |
1d3576fd SW |
674 | pr_warning("writepage_start %p on forced umount\n", inode); |
675 | return -EIO; /* we're in a forced umount, don't write! */ | |
676 | } | |
3d14c5d2 YS |
677 | if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize) |
678 | wsize = fsc->mount_options->wsize; | |
1d3576fd SW |
679 | if (wsize < PAGE_CACHE_SIZE) |
680 | wsize = PAGE_CACHE_SIZE; | |
681 | max_pages_ever = wsize >> PAGE_CACHE_SHIFT; | |
682 | ||
683 | pagevec_init(&pvec, 0); | |
684 | ||
1d3576fd SW |
685 | /* where to start/end? */ |
686 | if (wbc->range_cyclic) { | |
687 | start = mapping->writeback_index; /* Start from prev offset */ | |
688 | end = -1; | |
689 | dout(" cyclic, start at %lu\n", start); | |
690 | } else { | |
691 | start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
692 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
693 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
694 | range_whole = 1; | |
695 | should_loop = 0; | |
696 | dout(" not cyclic, %lu to %lu\n", start, end); | |
697 | } | |
698 | index = start; | |
699 | ||
700 | retry: | |
701 | /* find oldest snap context with dirty data */ | |
702 | ceph_put_snap_context(snapc); | |
703 | snapc = get_oldest_context(inode, &snap_size); | |
704 | if (!snapc) { | |
705 | /* hmm, why does writepages get called when there | |
706 | is no dirty data? */ | |
707 | dout(" no snap context with dirty data?\n"); | |
708 | goto out; | |
709 | } | |
710 | dout(" oldest snapc is %p seq %lld (%d snaps)\n", | |
711 | snapc, snapc->seq, snapc->num_snaps); | |
712 | if (last_snapc && snapc != last_snapc) { | |
713 | /* if we switched to a newer snapc, restart our scan at the | |
714 | * start of the original file range. */ | |
715 | dout(" snapc differs from last pass, restarting at %lu\n", | |
716 | index); | |
717 | index = start; | |
718 | } | |
719 | last_snapc = snapc; | |
720 | ||
721 | while (!done && index <= end) { | |
722 | unsigned i; | |
723 | int first; | |
724 | pgoff_t next; | |
725 | int pvec_pages, locked_pages; | |
726 | struct page *page; | |
727 | int want; | |
728 | u64 offset, len; | |
2baba250 | 729 | long writeback_stat; |
1d3576fd SW |
730 | |
731 | next = 0; | |
732 | locked_pages = 0; | |
733 | max_pages = max_pages_ever; | |
734 | ||
735 | get_more_pages: | |
736 | first = -1; | |
737 | want = min(end - index, | |
738 | min((pgoff_t)PAGEVEC_SIZE, | |
739 | max_pages - (pgoff_t)locked_pages) - 1) | |
740 | + 1; | |
741 | pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
742 | PAGECACHE_TAG_DIRTY, | |
743 | want); | |
744 | dout("pagevec_lookup_tag got %d\n", pvec_pages); | |
745 | if (!pvec_pages && !locked_pages) | |
746 | break; | |
747 | for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) { | |
748 | page = pvec.pages[i]; | |
749 | dout("? %p idx %lu\n", page, page->index); | |
750 | if (locked_pages == 0) | |
751 | lock_page(page); /* first page */ | |
752 | else if (!trylock_page(page)) | |
753 | break; | |
754 | ||
755 | /* only dirty pages, or our accounting breaks */ | |
756 | if (unlikely(!PageDirty(page)) || | |
757 | unlikely(page->mapping != mapping)) { | |
758 | dout("!dirty or !mapping %p\n", page); | |
759 | unlock_page(page); | |
760 | break; | |
761 | } | |
762 | if (!wbc->range_cyclic && page->index > end) { | |
763 | dout("end of range %p\n", page); | |
764 | done = 1; | |
765 | unlock_page(page); | |
766 | break; | |
767 | } | |
768 | if (next && (page->index != next)) { | |
769 | dout("not consecutive %p\n", page); | |
770 | unlock_page(page); | |
771 | break; | |
772 | } | |
773 | if (wbc->sync_mode != WB_SYNC_NONE) { | |
774 | dout("waiting on writeback %p\n", page); | |
775 | wait_on_page_writeback(page); | |
776 | } | |
777 | if ((snap_size && page_offset(page) > snap_size) || | |
778 | (!snap_size && | |
779 | page_offset(page) > i_size_read(inode))) { | |
780 | dout("%p page eof %llu\n", page, snap_size ? | |
781 | snap_size : i_size_read(inode)); | |
782 | done = 1; | |
783 | unlock_page(page); | |
784 | break; | |
785 | } | |
786 | if (PageWriteback(page)) { | |
787 | dout("%p under writeback\n", page); | |
788 | unlock_page(page); | |
789 | break; | |
790 | } | |
791 | ||
792 | /* only if matching snap context */ | |
61600ef8 | 793 | pgsnapc = page_snap_context(page); |
80e755fe SW |
794 | if (pgsnapc->seq > snapc->seq) { |
795 | dout("page snapc %p %lld > oldest %p %lld\n", | |
796 | pgsnapc, pgsnapc->seq, snapc, snapc->seq); | |
1d3576fd SW |
797 | unlock_page(page); |
798 | if (!locked_pages) | |
799 | continue; /* keep looking for snap */ | |
800 | break; | |
801 | } | |
802 | ||
803 | if (!clear_page_dirty_for_io(page)) { | |
804 | dout("%p !clear_page_dirty_for_io\n", page); | |
805 | unlock_page(page); | |
806 | break; | |
807 | } | |
808 | ||
809 | /* ok */ | |
810 | if (locked_pages == 0) { | |
811 | /* prepare async write request */ | |
6285bc23 | 812 | offset = (u64) page_offset(page); |
1d3576fd | 813 | len = wsize; |
3d14c5d2 | 814 | req = ceph_osdc_new_request(&fsc->client->osdc, |
1d3576fd SW |
815 | &ci->i_layout, |
816 | ceph_vino(inode), | |
817 | offset, &len, | |
818 | CEPH_OSD_OP_WRITE, | |
819 | CEPH_OSD_FLAG_WRITE | | |
820 | CEPH_OSD_FLAG_ONDISK, | |
821 | snapc, do_sync, | |
822 | ci->i_truncate_seq, | |
823 | ci->i_truncate_size, | |
153e5167 | 824 | &inode->i_mtime, true); |
8c71897b | 825 | |
6816282d SW |
826 | if (IS_ERR(req)) { |
827 | rc = PTR_ERR(req); | |
8c71897b HC |
828 | unlock_page(page); |
829 | break; | |
830 | } | |
831 | ||
153e5167 AE |
832 | req->r_num_pages = calc_pages_for(0, len); |
833 | req->r_page_alignment = 0; | |
1d3576fd SW |
834 | max_pages = req->r_num_pages; |
835 | ||
3d14c5d2 | 836 | alloc_page_vec(fsc, req); |
1d3576fd SW |
837 | req->r_callback = writepages_finish; |
838 | req->r_inode = inode; | |
1d3576fd SW |
839 | } |
840 | ||
841 | /* note position of first page in pvec */ | |
842 | if (first < 0) | |
843 | first = i; | |
844 | dout("%p will write page %p idx %lu\n", | |
845 | inode, page, page->index); | |
2baba250 | 846 | |
213c99ee | 847 | writeback_stat = |
3d14c5d2 | 848 | atomic_long_inc_return(&fsc->writeback_count); |
213c99ee | 849 | if (writeback_stat > CONGESTION_ON_THRESH( |
3d14c5d2 YS |
850 | fsc->mount_options->congestion_kb)) { |
851 | set_bdi_congested(&fsc->backing_dev_info, | |
213c99ee | 852 | BLK_RW_ASYNC); |
2baba250 YS |
853 | } |
854 | ||
1d3576fd SW |
855 | set_page_writeback(page); |
856 | req->r_pages[locked_pages] = page; | |
857 | locked_pages++; | |
858 | next = page->index + 1; | |
859 | } | |
860 | ||
861 | /* did we get anything? */ | |
862 | if (!locked_pages) | |
863 | goto release_pvec_pages; | |
864 | if (i) { | |
865 | int j; | |
866 | BUG_ON(!locked_pages || first < 0); | |
867 | ||
868 | if (pvec_pages && i == pvec_pages && | |
869 | locked_pages < max_pages) { | |
870 | dout("reached end pvec, trying for more\n"); | |
871 | pagevec_reinit(&pvec); | |
872 | goto get_more_pages; | |
873 | } | |
874 | ||
875 | /* shift unused pages over in the pvec... we | |
876 | * will need to release them below. */ | |
877 | for (j = i; j < pvec_pages; j++) { | |
878 | dout(" pvec leftover page %p\n", | |
879 | pvec.pages[j]); | |
880 | pvec.pages[j-i+first] = pvec.pages[j]; | |
881 | } | |
882 | pvec.nr -= i-first; | |
883 | } | |
884 | ||
885 | /* submit the write */ | |
886 | offset = req->r_pages[0]->index << PAGE_CACHE_SHIFT; | |
887 | len = min((snap_size ? snap_size : i_size_read(inode)) - offset, | |
888 | (u64)locked_pages << PAGE_CACHE_SHIFT); | |
889 | dout("writepages got %d pages at %llu~%llu\n", | |
890 | locked_pages, offset, len); | |
891 | ||
892 | /* revise final length, page count */ | |
893 | req->r_num_pages = locked_pages; | |
1b83bef2 SW |
894 | req->r_request_ops[0].extent.length = cpu_to_le64(len); |
895 | req->r_request_ops[0].payload_len = cpu_to_le32(len); | |
1d3576fd SW |
896 | req->r_request->hdr.data_len = cpu_to_le32(len); |
897 | ||
9d6fcb08 SW |
898 | rc = ceph_osdc_start_request(&fsc->client->osdc, req, true); |
899 | BUG_ON(rc); | |
1d3576fd SW |
900 | req = NULL; |
901 | ||
902 | /* continue? */ | |
903 | index = next; | |
904 | wbc->nr_to_write -= locked_pages; | |
905 | if (wbc->nr_to_write <= 0) | |
906 | done = 1; | |
907 | ||
908 | release_pvec_pages: | |
909 | dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr, | |
910 | pvec.nr ? pvec.pages[0] : NULL); | |
911 | pagevec_release(&pvec); | |
912 | ||
913 | if (locked_pages && !done) | |
914 | goto retry; | |
915 | } | |
916 | ||
917 | if (should_loop && !done) { | |
918 | /* more to do; loop back to beginning of file */ | |
919 | dout("writepages looping back to beginning of file\n"); | |
920 | should_loop = 0; | |
921 | index = 0; | |
922 | goto retry; | |
923 | } | |
924 | ||
925 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
926 | mapping->writeback_index = index; | |
927 | ||
928 | out: | |
929 | if (req) | |
930 | ceph_osdc_put_request(req); | |
1d3576fd SW |
931 | ceph_put_snap_context(snapc); |
932 | dout("writepages done, rc = %d\n", rc); | |
1d3576fd SW |
933 | return rc; |
934 | } | |
935 | ||
936 | ||
937 | ||
938 | /* | |
939 | * See if a given @snapc is either writeable, or already written. | |
940 | */ | |
941 | static int context_is_writeable_or_written(struct inode *inode, | |
942 | struct ceph_snap_context *snapc) | |
943 | { | |
944 | struct ceph_snap_context *oldest = get_oldest_context(inode, NULL); | |
6298a337 SW |
945 | int ret = !oldest || snapc->seq <= oldest->seq; |
946 | ||
947 | ceph_put_snap_context(oldest); | |
948 | return ret; | |
1d3576fd SW |
949 | } |
950 | ||
951 | /* | |
952 | * We are only allowed to write into/dirty the page if the page is | |
953 | * clean, or already dirty within the same snap context. | |
8f883c24 SW |
954 | * |
955 | * called with page locked. | |
956 | * return success with page locked, | |
957 | * or any failure (incl -EAGAIN) with page unlocked. | |
1d3576fd | 958 | */ |
4af6b225 YS |
959 | static int ceph_update_writeable_page(struct file *file, |
960 | loff_t pos, unsigned len, | |
961 | struct page *page) | |
1d3576fd | 962 | { |
496ad9aa | 963 | struct inode *inode = file_inode(file); |
1d3576fd | 964 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 | 965 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
1d3576fd SW |
966 | loff_t page_off = pos & PAGE_CACHE_MASK; |
967 | int pos_in_page = pos & ~PAGE_CACHE_MASK; | |
968 | int end_in_page = pos_in_page + len; | |
969 | loff_t i_size; | |
1d3576fd | 970 | int r; |
80e755fe | 971 | struct ceph_snap_context *snapc, *oldest; |
1d3576fd | 972 | |
1d3576fd SW |
973 | retry_locked: |
974 | /* writepages currently holds page lock, but if we change that later, */ | |
975 | wait_on_page_writeback(page); | |
976 | ||
977 | /* check snap context */ | |
978 | BUG_ON(!ci->i_snap_realm); | |
979 | down_read(&mdsc->snap_rwsem); | |
980 | BUG_ON(!ci->i_snap_realm->cached_context); | |
61600ef8 | 981 | snapc = page_snap_context(page); |
80e755fe | 982 | if (snapc && snapc != ci->i_head_snapc) { |
1d3576fd SW |
983 | /* |
984 | * this page is already dirty in another (older) snap | |
985 | * context! is it writeable now? | |
986 | */ | |
80e755fe | 987 | oldest = get_oldest_context(inode, NULL); |
1d3576fd SW |
988 | up_read(&mdsc->snap_rwsem); |
989 | ||
80e755fe | 990 | if (snapc->seq > oldest->seq) { |
6298a337 | 991 | ceph_put_snap_context(oldest); |
1d3576fd | 992 | dout(" page %p snapc %p not current or oldest\n", |
6298a337 | 993 | page, snapc); |
1d3576fd SW |
994 | /* |
995 | * queue for writeback, and wait for snapc to | |
996 | * be writeable or written | |
997 | */ | |
6298a337 | 998 | snapc = ceph_get_snap_context(snapc); |
1d3576fd | 999 | unlock_page(page); |
3c6f6b79 | 1000 | ceph_queue_writeback(inode); |
8f883c24 | 1001 | r = wait_event_interruptible(ci->i_cap_wq, |
1d3576fd SW |
1002 | context_is_writeable_or_written(inode, snapc)); |
1003 | ceph_put_snap_context(snapc); | |
8f883c24 SW |
1004 | if (r == -ERESTARTSYS) |
1005 | return r; | |
4af6b225 | 1006 | return -EAGAIN; |
1d3576fd | 1007 | } |
6298a337 | 1008 | ceph_put_snap_context(oldest); |
1d3576fd SW |
1009 | |
1010 | /* yay, writeable, do it now (without dropping page lock) */ | |
1011 | dout(" page %p snapc %p not current, but oldest\n", | |
1012 | page, snapc); | |
1013 | if (!clear_page_dirty_for_io(page)) | |
1014 | goto retry_locked; | |
1015 | r = writepage_nounlock(page, NULL); | |
1016 | if (r < 0) | |
1017 | goto fail_nosnap; | |
1018 | goto retry_locked; | |
1019 | } | |
1020 | ||
1021 | if (PageUptodate(page)) { | |
1022 | dout(" page %p already uptodate\n", page); | |
1023 | return 0; | |
1024 | } | |
1025 | ||
1026 | /* full page? */ | |
1027 | if (pos_in_page == 0 && len == PAGE_CACHE_SIZE) | |
1028 | return 0; | |
1029 | ||
1030 | /* past end of file? */ | |
1031 | i_size = inode->i_size; /* caller holds i_mutex */ | |
1032 | ||
1033 | if (i_size + len > inode->i_sb->s_maxbytes) { | |
1034 | /* file is too big */ | |
1035 | r = -EINVAL; | |
1036 | goto fail; | |
1037 | } | |
1038 | ||
1039 | if (page_off >= i_size || | |
1040 | (pos_in_page == 0 && (pos+len) >= i_size && | |
1041 | end_in_page - pos_in_page != PAGE_CACHE_SIZE)) { | |
1042 | dout(" zeroing %p 0 - %d and %d - %d\n", | |
1043 | page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE); | |
1044 | zero_user_segments(page, | |
1045 | 0, pos_in_page, | |
1046 | end_in_page, PAGE_CACHE_SIZE); | |
1047 | return 0; | |
1048 | } | |
1049 | ||
1050 | /* we need to read it. */ | |
1051 | up_read(&mdsc->snap_rwsem); | |
1052 | r = readpage_nounlock(file, page); | |
1053 | if (r < 0) | |
1054 | goto fail_nosnap; | |
1055 | goto retry_locked; | |
1056 | ||
1057 | fail: | |
1058 | up_read(&mdsc->snap_rwsem); | |
1059 | fail_nosnap: | |
1060 | unlock_page(page); | |
1061 | return r; | |
1062 | } | |
1063 | ||
4af6b225 YS |
1064 | /* |
1065 | * We are only allowed to write into/dirty the page if the page is | |
1066 | * clean, or already dirty within the same snap context. | |
1067 | */ | |
1068 | static int ceph_write_begin(struct file *file, struct address_space *mapping, | |
1069 | loff_t pos, unsigned len, unsigned flags, | |
1070 | struct page **pagep, void **fsdata) | |
1071 | { | |
496ad9aa | 1072 | struct inode *inode = file_inode(file); |
4af6b225 YS |
1073 | struct page *page; |
1074 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
7971bd92 | 1075 | int r; |
4af6b225 YS |
1076 | |
1077 | do { | |
8f883c24 | 1078 | /* get a page */ |
4af6b225 | 1079 | page = grab_cache_page_write_begin(mapping, index, 0); |
7971bd92 SW |
1080 | if (!page) |
1081 | return -ENOMEM; | |
1082 | *pagep = page; | |
4af6b225 YS |
1083 | |
1084 | dout("write_begin file %p inode %p page %p %d~%d\n", file, | |
213c99ee | 1085 | inode, page, (int)pos, (int)len); |
4af6b225 YS |
1086 | |
1087 | r = ceph_update_writeable_page(file, pos, len, page); | |
1088 | } while (r == -EAGAIN); | |
1089 | ||
1090 | return r; | |
1091 | } | |
1092 | ||
1d3576fd SW |
1093 | /* |
1094 | * we don't do anything in here that simple_write_end doesn't do | |
1095 | * except adjust dirty page accounting and drop read lock on | |
1096 | * mdsc->snap_rwsem. | |
1097 | */ | |
1098 | static int ceph_write_end(struct file *file, struct address_space *mapping, | |
1099 | loff_t pos, unsigned len, unsigned copied, | |
1100 | struct page *page, void *fsdata) | |
1101 | { | |
496ad9aa | 1102 | struct inode *inode = file_inode(file); |
3d14c5d2 YS |
1103 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
1104 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
1d3576fd SW |
1105 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); |
1106 | int check_cap = 0; | |
1107 | ||
1108 | dout("write_end file %p inode %p page %p %d~%d (%d)\n", file, | |
1109 | inode, page, (int)pos, (int)copied, (int)len); | |
1110 | ||
1111 | /* zero the stale part of the page if we did a short copy */ | |
1112 | if (copied < len) | |
1113 | zero_user_segment(page, from+copied, len); | |
1114 | ||
1115 | /* did file size increase? */ | |
1116 | /* (no need for i_size_read(); we caller holds i_mutex */ | |
1117 | if (pos+copied > inode->i_size) | |
1118 | check_cap = ceph_inode_set_size(inode, pos+copied); | |
1119 | ||
1120 | if (!PageUptodate(page)) | |
1121 | SetPageUptodate(page); | |
1122 | ||
1123 | set_page_dirty(page); | |
1124 | ||
1125 | unlock_page(page); | |
1126 | up_read(&mdsc->snap_rwsem); | |
1127 | page_cache_release(page); | |
1128 | ||
1129 | if (check_cap) | |
1130 | ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL); | |
1131 | ||
1132 | return copied; | |
1133 | } | |
1134 | ||
1135 | /* | |
1136 | * we set .direct_IO to indicate direct io is supported, but since we | |
1137 | * intercept O_DIRECT reads and writes early, this function should | |
1138 | * never get called. | |
1139 | */ | |
1140 | static ssize_t ceph_direct_io(int rw, struct kiocb *iocb, | |
1141 | const struct iovec *iov, | |
1142 | loff_t pos, unsigned long nr_segs) | |
1143 | { | |
1144 | WARN_ON(1); | |
1145 | return -EINVAL; | |
1146 | } | |
1147 | ||
1148 | const struct address_space_operations ceph_aops = { | |
1149 | .readpage = ceph_readpage, | |
1150 | .readpages = ceph_readpages, | |
1151 | .writepage = ceph_writepage, | |
1152 | .writepages = ceph_writepages_start, | |
1153 | .write_begin = ceph_write_begin, | |
1154 | .write_end = ceph_write_end, | |
1155 | .set_page_dirty = ceph_set_page_dirty, | |
1156 | .invalidatepage = ceph_invalidatepage, | |
1157 | .releasepage = ceph_releasepage, | |
1158 | .direct_IO = ceph_direct_io, | |
1159 | }; | |
1160 | ||
1161 | ||
1162 | /* | |
1163 | * vm ops | |
1164 | */ | |
1165 | ||
1166 | /* | |
1167 | * Reuse write_begin here for simplicity. | |
1168 | */ | |
1169 | static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) | |
1170 | { | |
496ad9aa | 1171 | struct inode *inode = file_inode(vma->vm_file); |
1d3576fd | 1172 | struct page *page = vmf->page; |
3d14c5d2 | 1173 | struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; |
6285bc23 | 1174 | loff_t off = page_offset(page); |
1d3576fd | 1175 | loff_t size, len; |
1d3576fd SW |
1176 | int ret; |
1177 | ||
3ca9c3bd JK |
1178 | /* Update time before taking page lock */ |
1179 | file_update_time(vma->vm_file); | |
1180 | ||
1d3576fd SW |
1181 | size = i_size_read(inode); |
1182 | if (off + PAGE_CACHE_SIZE <= size) | |
1183 | len = PAGE_CACHE_SIZE; | |
1184 | else | |
1185 | len = size & ~PAGE_CACHE_MASK; | |
1186 | ||
1187 | dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode, | |
1188 | off, len, page, page->index); | |
4af6b225 YS |
1189 | |
1190 | lock_page(page); | |
1191 | ||
1192 | ret = VM_FAULT_NOPAGE; | |
1193 | if ((off > size) || | |
1194 | (page->mapping != inode->i_mapping)) | |
1195 | goto out; | |
1196 | ||
1197 | ret = ceph_update_writeable_page(vma->vm_file, off, len, page); | |
1198 | if (ret == 0) { | |
1199 | /* success. we'll keep the page locked. */ | |
1d3576fd SW |
1200 | set_page_dirty(page); |
1201 | up_read(&mdsc->snap_rwsem); | |
1d3576fd SW |
1202 | ret = VM_FAULT_LOCKED; |
1203 | } else { | |
4af6b225 YS |
1204 | if (ret == -ENOMEM) |
1205 | ret = VM_FAULT_OOM; | |
1206 | else | |
1207 | ret = VM_FAULT_SIGBUS; | |
1d3576fd | 1208 | } |
4af6b225 | 1209 | out: |
1d3576fd | 1210 | dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret); |
4af6b225 YS |
1211 | if (ret != VM_FAULT_LOCKED) |
1212 | unlock_page(page); | |
1d3576fd SW |
1213 | return ret; |
1214 | } | |
1215 | ||
1216 | static struct vm_operations_struct ceph_vmops = { | |
1217 | .fault = filemap_fault, | |
1218 | .page_mkwrite = ceph_page_mkwrite, | |
0b173bc4 | 1219 | .remap_pages = generic_file_remap_pages, |
1d3576fd SW |
1220 | }; |
1221 | ||
1222 | int ceph_mmap(struct file *file, struct vm_area_struct *vma) | |
1223 | { | |
1224 | struct address_space *mapping = file->f_mapping; | |
1225 | ||
1226 | if (!mapping->a_ops->readpage) | |
1227 | return -ENOEXEC; | |
1228 | file_accessed(file); | |
1229 | vma->vm_ops = &ceph_vmops; | |
1d3576fd SW |
1230 | return 0; |
1231 | } |